You are on page 1of 370




Consulting Editor


University of Wisconsin - Madison




Massachusetts Institute of Technology

Boston University


Massachusetts Institute of Technology


Harvard Business School



Distributors for North America:

Kluwer Academic Publishers
101 Philip Drive
Assinippi Park
Norwell, Massachusetts 02061, USA
Distributors for the UK and Ireland:
Kluwer Academic Publishers
Falcon House, Queen Square
Distributors for all other eountries:
Kluwer Academic Publishers
Distribution Centre
Post Office Box 322
3300 AH Dordrecht, THE NETHERLANDS

Library of Congress Cataloging-in-Publication Data

Spot pricing of electricity/by Fred C. Schweppe ... let al.].
p. cm. - (The Kluwer international series in engineering and computer science. Power electronics
and power systems)
Includes index
ISBN-13: 978-1-4612-8950-0

e-ISBN-13: 978-1-4613-1683-1

DOl: 10.1007/978-1-4613-1683-1
I. Electric utilities-Rates. 2. Commodity exchanges.
I. Schweppe, Fred c., 1933II. Series.
HG6047.E43S66 1987
338.4'336362 - dcl9


Copyright 1988 by Kluwer Academic Publishers. Fourth Printing 2000.

Softcover reprint of the hardcover 1st edition 2000
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or
transmitted in any form or by any means, mechanical, photocopying, recording, or otherwise,
without the prior written permission of the publisher, Kluwer Academic Publishers, 101 Philip
Drive, Assinippi Park, Norwell, Massachusetts 02061

To Mary Alice Sanderson, long live the sponge;
and to our children, Carl, Christina, Constantine, Daniel, David, Edmund, Fritz (Charles),
Kristen, and Malaika.
Shortly before completion of this book Fred C.
Schweppe, our friend, colleague and senior
author died suddenly. Fred created spot pricing
and proved, again, that "The forecast is always


when using the PDF file,

add 17 to the number here




Goal of Book
Three Steps to an Energy Marketplace
How Will Customers Respond?
Encl"li,+, Marketplace Operation: A Developed Country
Energy Marketplace Operation: A Developing Country
Discussion of Chapter I
Supplemcnt to Chapter I: Summary of Issues
Historical Notes and References: Chapter I





Preface to Part I: The Energy Marketplace


Behavior of Hourly Spot Prices



Definition of Hourly Spot Price

Components of Hourly Spot Prices
Operating Cost Components
Quality of Supply Components
Aggregated Network
Revenue Reconciliation Components
Buy-Back Rates
Expected Price Trajectories
Price Duration Curves





viii Contents

2.10 Customer Response

2. II Discusion of Chapter 2
Historical Notes and References: Chapter 2



Energy Marketplace Transactions

3.1 Criteria for Choice of Transactions

3.2 Customer Classes
3.3 Price-Only Transactions
3.4 Price-Quantity Transactions
3.5 Long-Term Contracts
3.6 Optional and Custom-Tailored Transactions
3.7 Why No Demand Charge?
3.8 Relationship to Present-Day Transactions
3.9 Customer-Owned Generation: Avoided Costs
3.10 Special Customer Treatment
3.11 Wheeling Rates
3.12 Discussion of Chapter 3
Historical Notes and References: Chapter 3





Energy Marketplace: Operation

Energy Marketplace: Planning
Customer: Operation
Customer: Planning
Calculation of Hourly Spot Prices
Utility: Operation
Utility: Planning
Regulatory Commission: Operation and Planning
Discussion of Chapter 4
Historical Notes and References: Chapter 4


A Possible Future: Deregulation



A Deregulated Energy Marketplace

Short-Term Operation and Control
Long-Term Operation and Planning
A Scenario
Discussion of Chapter 5
Historical Notes and References: Chapter 5




Generation Only
Generation Fuel and Variable Maintenace: ).(t)
Generation Quality of Supply, YQs(t): Cost Function Approach
Generation Quality of Supply, YQs(t): Market Clearing Approach
Generation Self-Dispatch
Multiple Time Periods
Discussion of Chapter 6
Historical Notes and References: Chapter 6








Generation and Network


7.1 Problem Formulation: Real Power Only

7.2 General Result
7.3 Network Loss: '1u(t)
7.4 Network Maintenance: '1M.k(t)
7.5 Network Quality of Supply: '1QS.k(t)
7.6 Two-Bus Example
7.7 Price Difference Across a Line
7.8 Customer-Owned Generation: Self-Dispatch
7.9 Aggregated Networks
7.10 Reactive Energy and Voltage Magnitudes
7.11 Discussion of Chapter 7
Historical Notes and References: Chapter 7


Revenue Reconciliation



Modify Spot Prices: Aggregate Reconciliation

Buy-Back Rate
Revolving Fund
Modify Spot Prices: Decomposed Reconciliation
Fixed Charges
Nonlinear Pricing
Revenue Neutrality
Discussion of Chapter 8
Historical Notes',and References: Chapter 8


$pot Price Basei Rates



Predetermined Price-Only Transactions

Price-Quantity Transactions
Lorig-Term Contracts
Special i::ustomer Contracts
Wheeling Rates
Discussion of Chapter 9
Historical Notes and References: Chapter 9



Optimal Investment Conditions



Overall Problem Formulation

Generator Investment Conditions
Customer Investment Conditions
Transmission Investment Conditions
Revenue Reconciliation for Optimum Systems
Long-Run Versus Short-Run Marginal Cost Pricing
Discussion of Chapter 10
Historical Notes and References: Chapter 10







x Contents





Power System Analysis and Control



Network Flows
Local Controllers
Mathematical Models for System Dynamics
Power System Dynamics
Further Reading


Power System Operation

Short-Term Load Forecasting
System Economics
System Security
BA Automatic Generation Control (AGC)
B.5 Interconnected Systems
Further Reading


Power System Planning



Multiple Attribute Decision Making Under Uncertainty

Long Range Load Forecasting Models
Production Cost Models
Financial Models
Generation Expansion Programs
Network Expansion Programs
Feedback Couplings
Further Reading


DC Load "low



D.I General Relationships

D.2 A Potpourri of Results
D.3 Three-Bus Example



Customer Response Model Structures

E.I Single Period

E.2 Multiple Period





Basic Concepts
FAPER Designs
Variations on the Basic Concept
Analysis of Multiple FAPER Response
Incentives to Install FAPERS


Expected Behavior of Spot Prices




24-Hour Trajectories
Price Duration Curves
Impact of Customer Response on Variable Energy Costs


Interchange of Derivative and Expectation Operators





Xhe .title page of this book lists four names, but as the bibliography shows,
spOt pricing has had many more authors. The folIowing is an attempt to
acl<nowledge partially the contributions of others, in roughly chronological
order. N~tu\l"alIy, none of those mentioned here is responsible for what we have
done with their ideas.
The impodance of changing the coupling that exists between a utility and its
industrial customers emerged during a research project (1977-78) done for New
England Electric System in close cooperation with some of its industrial
customers. Ed Gulachenski of New England Electric was the prime mover
behind the research effort.
Many of the initial formulations and papers associated with spot pricing
arose during a series of informal "Homeostatic Control" workshops conducted
at MIT during 1977-78. James Kirtley of MIT's Electric Power Systems Engineering Laboratory, EPSEL (now part of the Laboratory Electromagnetic and
Electronic Systems, LEES), was one of the key early contributors with his, Tom
Sterling's and Ron William's local communication network methodology and
microshedding concepts (which are called price-quantity transactions in this
book). Hugh Outhred of the University of New South Wales was another key
initial idea generator, and he has continued to evolve the ideas in a faraway land
south of the equator. Fred Pickel and Alan Cox helped co-author the initial
Homeostatic Control paper. Other members of the original Homeostatic
Control working group included Lenny Gould and Steve Umans.

xiv Acknowledgments

The overall effort was given a key boost when Henry Jacoby and Loren Cox,
Directors of MIT's Center for Energy Policy Research, organized the Boxborough Conference on Homeostatic Control in 1979. If we had listened more
carefully to Loren's advice over the years, spot pricing might have progressed
much faster than it has. David White, Director of MIT's Energy Laboratory,
was an early and continuously sustaining supporter of the ideas.
Early research efforts were supported under the sponsorship of Charlie Smith
and Ray Dunlop at DOE, which provided a quick, if short, start. The PSC of
Wisconsin offered the next installment. Rod Stevenson at the University of
Wisconsin helped formulate and complete this early study on the impact of spot
pricing for industrial applications.
In California, the moral support of Rusty Schweickart, then Chairman of the
California Energy Commission, and John Bryson, then Chairman of the California Department of Public Utilities, led to two key projects, one funded jointly
by Pacific Gas and Electric and Southern California Edison with AI Garcia and
Jack Runnels as two extremely helpful project monitors. A foll\ow-up, was
funded by the California Energy Commission guided and assisted by John
The California studies started a continuing and extremely productive collaboration with John Flory of Utility Customer Interfaces. The world needs more
individuals of his caliber.
At one point in the spot pricing evolution (as developed for regulated utilities), we were pulled into the great debate on deregulation. This led to interesting
and stimulating discussions with a variety of people such as Leonard Hyman of
Merrill Lynch Pierce Fenner and Smith, Joe Pace ofNERA, and William Berry,
president of Virginia Electric. Interactions with Richard Schmalensee and Paul
Joskow of MIT helped formulate our approach to deregulation.
When we first met Robert Peddie, then Chairman of the Southeastern Distribution Board in England, he was already well along with his exciting, visionary
approach to residential customer-utility interactions, called the CALMU
system. Subsequent discussions and interactions with him and his CALMU
system had a significant impact on our own efforts.
Bernie Hastings of Detroit Edison did not sit down to write theoretical spot
pricing papers or to do simulation studies. Instead he simply went ahead and
saw to the actual implementation of spot prices for some of Detroit Edison's
industrial customers. This was the first implementation of the basic ideas of spot
John Phillips, Chairman of the California Energy Coalition, is a very special
individual whose support, ideas, and enthusiasm, combined with his strong
advocacy of utility-customer partnering, aided spot pricing in ways that can
never be quantified.
The Integrated Communications Systems Inc. (ICS) demonstration of their
Transtext system for Georgia Power residential customers provided us with an
opportunity to evaluate the variable energy costs (i.e. spot prices) for many of


the participating utilities. These numerical studies of diverse utilities have

greatly increased our own understanding of the real issues and concerns. We
thank Doug Bulleit and Paul Spaduzzi of ICS for the opportunity to work with
them and with their participating utilities.
Meta Systems of Cambridge, Massachusetts provided us with the opportunity to do spot pricing studies (such as for ICS) in a consulting, rather than
academic, mode of operation, with all the subsequent advantages to the
sponsors of faster response and the ability to keep data confidential. Meta
Systems funded the final stages of word processing for this book and should play
a key role in future spot price applications.
A DOE-funded project on wheeling helped solidify and expand some of the
basis spot pricing concepts. The project monitor, Jeff Skeer, provided critical
comments on various drafts of the wheeling report that have had an impact on
the book itself. An EPRI funded project, monitored by Larry Carmichael, has
been very valuable in the development of alogrithms for response by residential
customers to energy marketplace spot prices. The Division of Research of the
Harvard Business School supported the fourth author for work on spatial
pricing and customer response.
One very important group of individuals left out of the above chronological
ordering are the faculty and staff of MIT's LEES/EPSEL who were impacted
by spot pricing, even though they were not directly involved in it. Spot pricing
has;11ever been a favorite funding area for research agencies. As a result there
havibeen some very lean years. Those of us involved in the effort usually did
nOl~brjng our fair sha~ of financial support into the laboratory. The faculty and
staff;of the laboratories (and in particular the various Directors: Gerry Wilson,
Tom Lee and Jim Mekher) have to be thanked for putting up with being leaned
on for many _years - as does Barbara Smith, who somehow kept us all rolling.
Mary Alice Slmderson managed to make sense out of the unintelligible, many
The last (but definitely not least) group left out of the chronological order is
the students who have been involved in the effort. Students make or break a
research university like MIT.
The most important Ph.D. thesis relative to spot pricing was written by the
fourth author of this book while he was at MIT [1982]. This thesis contains.
many of the basic ideas found in this book as well as discussions of customer
response. Richard Schmalensee of the Sloan School of Management was the
thesis supervisor and Paul Joskow of the Department of Economics was a thesis
reader who insisted that high standards be maintained. We thank both these
men for the important role they played.
The original industrial load modeling research that motivated the spot pricing
concepts was done by the Yon gut Manichaikul for his Ph.D. thesis [1978], which
combined mathematical analysis with the results of his talking to a lot of
industrial plant operators and the rubbing grease off of motor name plates.
Mike Ruane's Ph.D. thesis [1980] provided a general stochastic framework for



written to serve as a single, integrated sourcebook on the theory and implementation of a spot price based energy marketplace.
The book is written for electric power engineers interested in operation,
planning, and load management; for economists interested in electric power
regulation and pricing; and for utility regulators and overall policy makers. It
is a "how to" book, written so it can be used by those who are mainly interested
in the application of the concepts and techniques. Detailed mathematical derivations are also provided for those who are interested.
A spot price based energy marketplace has many benefits for both the electric
utility and its customers. These benefits include improvements in operating
efficiency, reductions in needed capital investments, and customer options on
the type (reliability) of electricity to be bought. A spot priced based energy
marketplace is a win-win situation for both the regulated utility and its
customers. The customer's lifestyles improve because the customers are receiving more service from the use of electric energy per dollar spent. The utility has
a more controllable, less uncertain world in which to operate.
A spot price based energy marketplace can be implemented using today's
proven technologies. However, its existence stimulates the development of new
microelectronic technologies and hence enables further exploitation of the
microelectronic revolution in communication and computation.
The spot price based energy marketplace concepts were originally developed
to meet the present and future needs of the complex, interconnected, sophisticated power systems of developed countries. However, the basic ideas are also
applicable to the smaller, rapidly growing, less sophisticated power systems
often found in other parts of the world.
The spot price based energy marketplace was developed to be applicable to
present-day structures wherein a privately owned utility is regulated by some
government agency, or the utility is government owned and operated. However,
the energy marketplace introduces the possibility of various degrees of deregulation wherein some generation is provided by privately owned, less regulated
Spot pricing is the natural evolution of existing techniques for power system
operation, planning, load management and the economic theory of marginal
cost pricing. This book relates spot pricing to existing rate structures, direct load
control techniques, interruptible contracts, interutility sales, and power system
Organization of Book

Chapter I provides an overview of spot pricing and the issues to be covered. The
rest of the main text is subdivided into two parts.
Part I (Chapters 2 through 5) consists of three main chapters that show how
hourly spot prices behave, discuss various types of energy marketplace transactions, and address implementation of a spot price based energy marketplace
from both the utility and customer points of view. The final chapter of Part I


briefly explores the possible evolution of the energy marketplace (as developed
for a regulated utility) into a system of deregulated generation.
Part II is a sequence of chapters (6 through 10) that develop the theory of spot
prices starting with simple cases and progressing toward more complex and
realistic situations.
Appendices provide background material and supplemental detail.
The book is written so that it can be read in different sequences. All readers
should begin with Chapter I. The text is organized to discuss behavior and
implementation (Part I) before going through the technical theory (Part II). This
is the appropriate sequence for readers interested primarily in broad applicability and implementation issues, since they do not have to wade through a lot of
equations. However, some readers may find it preferable to read Part n before
Part I to get the detailed theory first.
Readers with a limited background in power systems operation and planning
should begin by glancing at Appendices A, B, and C. These appendices provide
a brief overview of the main power system concepts affecting an energy marketplace.



This chapter provides an overview of the book and the spot price based energy
Section 1.1 discuspes the goals of the book. The three steps to an energy
marketplace are introduced in Section 1.2. (These three steps are expanded in
Chapters 2, 3 and 4 respectively.) Section 1.3 addresses customer response to an
energy marijetplace. Sections 1.4 and 1.5 discuss energy marketplace operation
for developed and developing countries. A supplement to Chapter 1 provides a
summary of issues relevant to a spot price based energy marketplace.

The electric utility industry today is undergoing rapid and irreversible changes.
Volatile fuel costs, less predictable load growth, a more complex regulatory
environment and a deceleration in conventional technical progress are important examples of these changes. Yet the need for growth in productivity and
efficiency, and for increased flexibility to handle future uncertainties, is stronger
and more challenging than ever. The utility industry, which has matured into a
toO-billion-dollar industry in the United States and constitutes a substantial
economic sector in all industrialized countries, must evolve to meet this challenge.
New directions for the utility industry are being sought by many interested
parties in the government, the private sector, and the universities. One such
direction has been widespread interest in utility-customer cooperation through

J. Spot pricing of electricity

innovative rates characterized by broader options and better use of information

on utility costs and customer needs.
The goal of this book is to provide a theoretically sound, yet practical
foundation for the implementation of utility-customer transactions based on
today's needs. Our goal is to meet four criteria:

Freedom of Choice: Provide customers with options on the cost and reliability of supply and how they choose to use electric energy.
Economic Efficiency: Motivate customers to adjust their own electric energy
usage patterns to match utility marginal costs.
Equity:l Reduce customer cross-subsidies-i.e. a customer's charges are based
on the utility's costs to serve that customer.
Utility Control, Operation and Planning: Consider the engineering requirements for controlling, operating and planning an electric power system.

Present-Day Transactions

Most transactions between utilities and their customers today fall far short of
meeting these four criteria. Flat and time-of-use rates are related to actual
marginal costs in only a gross, average sense. The demand charge is an anachronism with at most a tenuous relationship to marginal costs; it usually encourages counterproductive customer behavior. Load management approaches
often involve direct utility control (i.e., the big brother approach) and
sometimes encourage wasteful customer behavior. Cross-subsidization between
customer classes and within a given class is rampant. Utility-customer transactions are typically specified without considering the engineering problems of
controlling, operating, and planning an electri\ power system.
The failure of most present-day utility-customer transactions to meet today's
needs can be traced to their historical foundations. They were established by
individuals unconcerned with power system control and operation, during times
when communication and computation were very expensive, when there was less
incentive to use electricity in an efficient fashion, and when cross-subsidies were
of limited concern to society.
The four basic criteria cannot be achieved by putting "band-aids" on the
present-day approaches. They cannot be achieved by adding modern microelectronics whose functions are based on present-day approaches.
The four basic criteria can be achieved only by returning to the first principles
of economics and engineering and by viewing the utility and its customers as a
single integrated system. The result of this integration is the spot price based
energy marketplace, which is the subject of this book.
Energy Marketplace Transactions

Students taking an introductory course in economics are taught that the best
way to achieve

I. Overview


Economic efficiency
Customer freedom of choice

is the use of marketplace where market clearing prices are determined by supply
and demand. For example, when a lot of fresh produce (e.g., lettuce) is available,
the prices go down and consumption goes up. During other times of the year,
or after a hard freeze, the supply goes down and prices go up to reduce
Five ingredients for a successful marketplace are

A supply side with varying supply costs that increase with demand
A demand side with varying demands which can adapt to price changes
A market mechanism for buying and selling
No monopsonistic behavior on the demand side
No monopolistic behavior on the supply side

Electric energy is an ideal textbook-type commodity relative to the first four

ingredients, i.e.:
I: Figures 1.1.1 and 1.1.2 illustrate the effect of changing supply-demand
..conditions on supply costs. Figure 1.1.1 shows measured marginal fuel costs
, while Figure 1.1.2 shows total marginal costs (for a different utility) which
:also includes nonfuel effects such as quality of supply/reliability premiums
and recovery of embedded capital costs (revenue reconciliation). These
marginal.~cost variations are highly correlated with variations in demand
2. Industrial processes and space conditioning are only two examples from a
host of demands that are price responsive even to current rate structures.
3. The existing market mechanism (billing for electric power, etc.) can be
adapted. Industrial customers who already have recording demand meters
needed no new hardware to implement a marketplace with prices varying
each hour (assuming they also have a telephone). The details of creating a
market mechanism are the subject of this book.
4. Monopsonistic behavior is difficult on the demand side because the number
of customers ranges from thousands to millions.
Relative to the fifth ingredient, electric energy is not an ideal textbook example
for a marketplace because the supply side is a monopoly that is either government regulated or government owned. However rate-of-return regulation is
used today to limit the ability and incentives to behave monopolistically. This
would continue with spot pricing.
Throughout almost all of this book, we define

6 1. Spot pricing of electricity

Week 1












Week 2







Figure 1.1.1. Measured marginal fuel costs.

Energy Marketplace The buying and selling of electric energy between independent
customers and a regulated or government owned utility.

The one exception is Chapter 5, where we discuss the possibilities of deregulation of generation.
The energy marketplace is designed explicitly to include engineering issues
associated with power system control and operation. Hence all four of the
original criteria (efficiency, freedom of choice, equity, and control/operation)
are met.
In the energy marketplace, all utility-customer transactions are based in a
self-consistent fashion on a single quantity, the hourly spot price. The hourly
spot price is determined by the demand at that hour and the hourly varying costs
and capabilities of the generation, transmission and distribution systems. The

1. Overview 7

MUs/kWh .-----y-EA-R-'-PE-'K-O-'y-S-S-UM-M-AR-Y----.













LY2; \1\:\






Mils/kWh .-----y-E'-R-2P-=-:AK-=D-:Ay....,S-:UM-M-:A=-Ry,-----,

MUs/kWh .-----=yE-:AR:-2-T-=YP-::'CA-L-::"DA-:y-,S-UM-:M-:AR:-Y----,









Figure 1.1.2. Total marginal costs.

hourly spot price is defined In terms of marginal costs subject to revenue

reconciliation. 2
In the energy marketplace, there is closed-loop feedback between the utility
a'lld its customers. Jhe whole electric power system (generation, transmission,
distribution, and customers) is controlled and operated in an integrated fashion,
without removing tne customers' freedom of choice. This is made possible by the
d1versity in customers' characteristics, desires and needs .
. The . benefits of well-designed, real-time, utility-customer feedback are
clear, or will be after reading this book. However, so are the metering and
communic~tions costs associated with conveying the necessary information.
Therefore, the energy marketplace transactions are designed to match benefits
to transactions costs. Some customers (e.g., large industrial) might see prices
updated each hour, while other customers (e.g., residential) might normally see
prices updated each billing period. However, a residential customer who wants
to exploit hourly price variation has the option of seeing more frequent price
updates, provided the customer pays the additional costs.
What is the Difference?

The initial reaction of many people to spot pricing is that the major difference
between spot prices and present-day transactions is that spot prices have
complex time variations. Actually, present-day prices also exhibit complex time
variations, so the major difference is in the nature of the price variations, not
their presence. As one example, consider Figure 1.1.3, which plots the historical
variation of prices (jkWh) for one utility. The hourly cost (spot price) variations of Figures 1.1.1 and 1.1.2 simply exhibit finer time detail. As a second
example, consider an industrial customer with a 5/kWh energy charge and a


I. Spot pricing of electricity




















Figure 1.1.3. Example of monthly variations in residential prices, under conventional rates.

5 $/kW demand charge based on the energy used during the customer's peak
hour during the month. The corresponding energy rate paid by the industrial
customer is plotted in Figure 1.1.4. It displays a dramatic time variation which
bears little resemblance to either Figure 1.1.1 or 1.1.2.
A second difference lies in the nature of the, relationship between the utility
and its customers. In a spot price based energy marketplace, the utility and its
customers are partners working together to achieve the maximum benefit from
electric energy usage at minimum cost. The amount of such partnering found in
present-day utility-customer relationships is small at best.
Present-day flat and time-of-use rates are specified by formulas based on
expected costs, rates of return on equity, etc. An hourly spot price is based on
the same principles, but the formula is solved every hour by computers instead
of once a year or once per month (for fuel adjustment clauses). Present-day
direct load control and present-day interruptable contracts are special cases of
energy marketplace transactions. Present-day power system operation often
involves a real-time spot market for purchases and sales between utilities. The
spot price based energy marketplace simply extends this spot market to include
the customers. Thus it can be concluded that
The spot price based energy marketplace is the logical evolution of present-day rates and
load management techniques, married with present-day practices of power system operation, the concept of utility--customer partnering, and the availability of inexpensive
communications and computation equipment.

I. Overview

Rate Paid
by Customer

Energy Charge: 5 /kWh

Demand Charge: 5 $/kW



fipire 1.1.4. Effective

Hour of Peak Specified by Customer




variations resulting from demand charge, under conventional rates.


A spot pric~ based energy marketplace that meets the four criteria of Section 1.1
can be achieved in three steps:
Step I: Define hourly spot prices and evaluate their behavior.
Step II: Specify an appropriate set of utility-<:ustomer transactions based on
the hourly spot price and associated transactions costs.
Step III: Implement the energy marketplace considering the needs and capabilities of both the utility and the customers.
Step I: Define Hourly Spot Prices

The fundamental quantity underlying the energy marketplace is the hourly spot
price. 3
The hourly spot price is defined in terms of marginal costs subject to revenue
reconciliation (i.e. recovery of operating and embedded capital costs).
The value of the spot price at any hour depends on the hourly variations of

Generation fuel costs and capacities

10 1. Spot pricing of electricity


Transmission distribution network losses and capacities

Aggregated customer demand patterns

The hourly spot price is a random process (e.g. see Figures 1.1.1 and 1.1.2). Its
future value cannot be predicted perfectly because of random equipment
outages and demand variations.
An hourly spot price can be determined for a utility which is buying energy
from, as well as seJling energy to its cutomers. The buy-back hourly spot price
can be either greater or less than the selling hourly spot price.
Step II: Specify Utility-Customer Transactions

All utility-<:ustomer transactions are based on the hourly spot price defined in
Step I. Three general types of transactions are

Long-Term Contracts

Examples of price-only transactions are


One-Hour Update: Customers see prices ($/kWh) varying each hour, predicted and communicated one hour in advance.
24-Hour Update: Customers see prices ($jkWh) varying every hour, predicted
and communicated 24 hours in advance.
Time-of-Use (TOU) and Flat Rates: 4 Rates ($jkWh) are calculated using
predictions of hourly spot price behavior one billing period in advance.

The choice of what type of price-only transactions to offer is based on a tradeoff

between the transactions costs (metering, communication, computation, etc.)
and the benefits obtained from the transactions. For example, large industrial
customers might see one-hour update prices while small residential customers
see flat rates.
Price-quantity transactions involve a short-term utility-<:ustomer contract.
For example, a customer may choose to receive a specified quantity of energy
at a lower price with a contract to drop all or a part of such usage on a signal
from the utility. Price-quantity transactions include as special cases present-day
interruptible contracts and direct load control. However, the energy marketplace transactions are more robust and can better match the customer's needs
and capabilities. Use of price--quantity transactions instead of price-only transaction can be motivated by

A need for fast acting, accurate load control (seconds to minutes rather than
hours) to maintain power system security.
A desire to reduce transactions costs below those of rapidly varying price-only

I. Overview

transactions. Some price-quantity transactions are cheaper to implement

than some price-only transactions.
Long-term contracts are fixed price and fixed quantity. Thus they are like
present-day commodity contracts, and like present-day long-term contracts
between utilities they can extend hours, day, months and years into the future.
As an example, suppose that on January I an industrial customer contracts for
1000 kWh of energy to be delivered between 10 and 11 AM on July I, for
lOjkWh. If, when July I finally comes, the price between 10 and II AM is
actually 9 /kWh, the actual cash flow is as follows:
If Customer Uses
2000 kWh

The Customer Pays

$100 + $90 = $190
$100 - $90 = $10

Thus the incremental cost of the customer's usage that hour is 9/kWh. Such
transactions enable a customer to buy an insurance policy by locking in
1000kWh at 1O/kWh, even though the customer still sees the spot price for
actual usage.
A variation on this long-term contract is the option wherein the customer buys
the.'fight to buy up to a fixed amount of energy at a fixed price. The customer
ex~rcises the option ,if the hourly spot price turns out to be greater than the
o(9tiqn price.
Step III: fmplement Energy Marketplace

Implementation of the energy marketplace involves the utility and its customers
operatmg as' partners.
Utility implementation concerns include real-time calculation/prediction of
hourly spot prices, metering-{;ommunications-billing, and system control
center operation using the new control signal called price. The impact of the
energy marketplace on utility long-term investment decisions is also of concern.
Customers who choose to exploit the energy marketplace potentials must
implement the appropriate response systems, which could range from simple
manual response to sophisticated digital control. The utility can provide the
control mechanism as a service to customers.

A key question for a spot price based energy marketplace is: How will customers
respond (change their usage patterns) to price changes? Two ways to try to
answer this question are

Use direct observations of actual responses

Use models of response


I. Spot pricing of electricity

Unfortunately, not enough direct observations of the right type are available
today. There are a lot of data (and studies) on customer response to time-of-use
rates, which are a special case of a spot price based transaction. However, there
is a large difference between a time-of-use rate's behavior and that of, for
example, Figures 1.1.1 and 1.1.2. This difference combined with the nonlinear
nature of customer response makes an extrapolation of response from time-ofuse results a vague guideline at best. Interruptible rates and certain types of
direct load control are also special cases of spot price based transactions, but
their method of implementation makes it very difficult to draw conclusions for
an energy marketplace type operation. There are various implementations of
one- and 24-hour-type spot prices in operation, but the data coming from them
are still too limited to provide a basis for any general conclusion.
When enough direct observations are not yet available, it is necessary to
resort to the use of models. Models can range in sophistication from statistically
based computer simulations or optimizations to simple mental pictures obtained
by having an industrial plant manager spend a few hours showing one around
the plant and explaining the processes and operating constraints.
We will now provide some discussion on what we know of customer response
in the industrial and residential sectors (discussions on response mechanisms are
given in Section 4.3).
Industrial Response

We have modeled (at various levels of detail) over 20 industrial customers in

different parts of the U.S.A. This experience showed that five issues to be
considered when talking to an industrial customer are



negative initial reaction syndrome

importance of nonphysical and noneco~omic factors
difference between a demonstration and the real world
importance of advance knowledge
nonlinear character of the response

The initial reaction of an industrial plant manager to the idea of facing a

one-hour or 24-hour update spot price can be expected to be negative. However,
in most cases this negative reaction can be turned around. For example, a
negative reaction to the possibility of very high energy rates at certain times can
be countered by the existence of very low energy rates at other times and the
death of the demand charge. As a second example, an initial negative reaction
based on the concept of moving entire work shifts around can often be countered by discussing large electricity-using devices whose operation can be rescheduled without affecting more than a few workers. As a third example, an
initial negative reaction about uncertainty associated with rapidly varying rates
(hours to days) can be replaced by an appreciation that a fluctuating market is
a place where money can be made. One customer who actually was on a

I. Overview


one-half-hour update of spot prices complained that during some seasons of the
year the price didn't vary enough to make any money.
When we started looking at industrial response, we naively assumed that
response depended only on the physical character of the plant and the
economics of its operation. However, in the real world other factors also play
a key role. These factors included the nature of union contracts, the management of the company (locally owned or part of a big corporation), and the
character of the work force (social background), the location of the plant (was
it safe to work there at night). We studied two plants with identical SIC codes
located within twenty miles of each other. However, they had completely
different response capabilities.
It became very clear during our interactions that industrial response to a
one-year test or demonstration program would be much less than for a realworld spot price based energy marketplace that won't go away in 12 months.
Many customers could greatly improve their response capabilities by relatively
minor additions of new control, process, etc., equipment. However, such modifications could not be cost justified for a mere demonstration or test.
Industrial response can be greatly increased if the customers know a few
hours to a day in advance when exceptionally high or low prices will or are
expected to occur. Such lead or warning time allows rescheduling of highenergy-consuming processes with much less cost. The use of a 24-hour update
pr:~vides a "hard price" a day in advance. For a one-hour spot price, it is
i~()rtant for the utility to provide a forecast of future prices (the same forecasts
which formed the blise for updating the 24-hour spot price). The importance of
w?!r~ing.or lead another reason why it is difficult to extrapolate from time
of use or interruptible response to energy marketplace response.
Industrial response is usually a very nonlinear function of price. Doubling the
price from $ to 10/kWh may create little response, while doubling it from 10
to 20 /kWh could cause sizable reaction because of the threshold effects. It is
also clear from our studies that price differentials during a day can be equally
if not more important than the absolute value of the price.
Residential Response

Residential response is entirely different from industrial response. Our studies

and ponderings on residential response to a spot price based energy marketplace
lead to the conclusion that the key issue is the character of the electronic support
system that is provided. One cannot expect the typical residential customer to
interact effectively with a spot price based energy marketplace without electronic support. If a residential customer has available a well-designed microprocessor-based information and control system with user-friendly interface, extensive
response can be expected. Residential customers without such support available
should, in general, see only very simple spot price type rates.
The cost of microelectronics is continuing to fall at a rapid rate. Therefore a
k~y issue in residential response is how long it will take before the costs of such


I. Spot pricing of electricity

electronics become less than the benefits (to the customer and/or utility) of the
resulting response. Some might argue that such a time has already arrived.
The importance of warning time and the nonlinearity of response apply to
both residential and industrial customers.
Example of Customer Benefits

Why is it that putting customers onto a spot price makes them better off and
simultaneously makes the utility (or the utility's other customers) better off? Of
course one of the precepts of free market economies is that prices based on a
marketplace send better signals than do fixed price. But that is a theoretical
argument, and while it can be made rigorous (as we do in Chapter 6), it is not
very intuitive initially.
Therefore let's look at a simple example: an industrial customer who has to
pump a lot of water (or other fluids) into storage tanks. For the sake of
familiarity, consider a municipal water department which has to fill its water
tanks daily. Suppose the customer is initially on a two level time-of-day rate with
prices which vary over time but in a predetermined and rigid way. (If the
customer is initially on a flat rate or a rate with a demand charge, then there are
even more benefits from spot pricing in the following example.) Then it will
pump water every night and every weekend, at the same hours every week. If
it cannot fill the tanks completely during the low price period (which in some
time-of-day rates is necessarily less than 12 hours long), it will wait until the
tanks get to the low limit, then resume pumping. This will often mean that it
resumes pumping in the afternoon or early evening, just at the tail end of the
high price period. The customer will keep the same timing of usage every week
that a particular rate is in effect, and every day of the week, without adjusting
for generator outages, changes in electricity generating costs, etc.
Now switch this customer to a 24 hour upaate spot price. At worst, the
customer can continue its old pumping schedule as if it were still on predetermined time-of-day prices. Assuming that the old time-of-day rate and the spot
price based rate were calculated based on comparable formulas, its bill for
electricity would then stay the same and the cost for the utility to serve it would
stay the same. (If the old rate had a hidden cross subsidy, in either direction,
then the bill could change by the amount of that cross-subsidy.)
However, our spot pricing customer will start adjusting its pumping
schedule each day, based on the hourly prices for the following day. It will plan
to pump in all the lowest cost hours, which may not necessarily be consecutive.
Because each day's schedule will be different, it wilI start heavy pumping at a
different time each day. On days when the overall spot price turns out to be low
(often but not always these will be weekends) it may pump almost all day, to fill
up longer term storage areas, and give more flexibility on high priced days. (This
requires some kind of forecasting, but no matter how unsophisticated the
forecasting method, over the course of a year the customer is never worse off
than just ignoring this opportunity to store for more than a day.) Different

I. Overview


seasons of the year and different generator outage situations will automatically
lead to appropriate behavior by the customer, without elaborate formal mechanisms. (e.g. look at Figure 1.1.1 and 1.1.2.)
On some days the customer may have unusually high demand for pumping
services (e.g. due to a fire). On these days it will know that it can get energy when
it needs it, albeit perhaps at a high price. To the extent it has some discretion
(e.g. the fire has been put out, but the tanks are still below a safe level) it can
choose to reduce demand during the peak peak priced hours. All such decisions
are entirely up to the customer, not the utility.
Obviously the customer's benefits from spot pricing depend on how much
operating flexibility it has (or can get, by making selective investments). They
also depend on the hour to hour and day to day price variation - the higher the
variation, the more the customer stands to save by shifting its usage.
How about the utility's costs? If the customer does not respond to spot prices,
the utility sees the same demand pattern as before and is no worse off. However
if the customer makes any response at all, the utility is guaranteed to be made
better off because all responses will shift energy use from times of high spot
prices to times of low spot prices.
We could construct similar examples for other situations, but the basic point
is clear. If the customer does not change its behavior, then neither it nor the
utility are worse off. (Except perhaps for shifts in cross subsidies. In principle
. thl:lSpot prices could be adjusted to preserve cross subsidies, and we will mention
thltt at an appropria,te point.) The more the customer changes behavior, the
better off both parties are. And highly variable spot prices make the benefits for
bo~hparties bigger.
Why No Numbers?

We have devbted a lot of time and effort to trying to understand how industrial
and residential customers will respond. We firmly believe large response will
occur. (Otherwise we wouldn't have written this book.) However, readers who
go beyond Chapter I will find that we have chosen not to present explicit
numbers or even case studies on customer response. One reason for this decision
was the desire to keep the size of the book under contro\. However, the main
reason is that our understandings have not reached the point where we can
provide numerics that can be extrapolated to cover a utility service territory. We
are like the classical blind men feeling different parts of an elephant. We have
a lot of information but still are not certain what the overall elephant looks like.
However, we do know that it is big.

A simplified example of a sophisticated energy marketplace in operation in a

developed country is used to illustrate its basic functions.
Figure 1.4.1 summarizes the main energy marketplace transactions and information flows. For simplicity, only one customer is shown. We will discuss
each box in Figure 1.4.1 separately.

Hi I. Spot pricing of electricity

Generation transmission distribution system. The energy marketplace has a

long-range impact on generation and network capacity requirements. In many
situations, the result is a reduction in installed capacity (per unit of demand) and
less reliance on generation peaking units.
Power system operation. All of the basic present-day power system operating
functions continue in an energy marketplace, but some of the functions are
modified. Short-term-demand forecasts now include the effects of price. Unit
commitment logics incorporate the effect of price feedback. Operating reserve
requirements are carried by the load or generation, whichever is least expensive.
All of these changes reduce total system costs.
Meter. A recording meter measures and stores hourly energy usage for each
customer. It is read once a month and used to compute the monthly bill.
Price-only transactions. At ten minutes before the hour, the one-hour update
spot price is computed based on a 70-minute forecast of system operation
conditions (fuel costs, losses, generation reserve margin, and network capabilities) and demand (taking into account price effects). The one-hour update spot
price is automatically communicated in digital form over telephone lines to the
computers of those customers who are on the one-hour update. It holds for one
At 3 PM of each afternoon, the hourly spot prices for the 24-hour period
starting at 3 AM of the next morning are computed using forecasts of operating








Price Only

Price Quantity


Short Term

Long Term Price Forecast


Long Term Contracts

Figure 1.4.1. Functions and information flow in an energy marketplace.

I. Overview


conditions. Customers on a 24-hour update price can receive the 24 numbers by

telephoning the utility any time after 4 PM. Prices are available in verbal or
digital format. Next-day prices are also communicated using newspapers and
Price-quantity transactions. The price-quantity transactions are available
only to those customers on a one-hour update spot price. At ten minutes before
each hour, an interruptible energy rate for the subsequent hour is computed,
based on system operating conditions and forecasts of how much interruptible
energy will be purchased by customers. It is communicated in digital form over
telephone to those customers who request it. Customers who choose to buy
interruptible energy do so by communicating the secure energy level they want,
so that all usage above that level is at the interruptible price. If the utility
experiences short-term emergency operating conditions (such as a major generation forced outage), the reduction required from each customer is computed and
communicated via telephone to the customer's computer.
Short-term price forecast. Forecasts of the hourly spot price for each hour of
the next week are made available if requested by the customer via telephone, in
either verbal or digital format. Naturally these are only forecasts and can be
Billing. Customer bills are computed each month by summing the metered
and recorded hourly energy use and multiplying by the spot price that existed
ai" the corresponding hour.
"::Long-term price hrecasts. Monthly and yearly price forecasts are made by the
tltiliJY itself or by in'dependent information consultants who are not part of the
e~ctric. utili ty.
Long-term contracts. Long-term fixed-price-fixed-quantity contracts are
written through energy brokers who contract directly with the customers. These
transactions have no effect on power system operation.
Now consider the customer boxes of Figure 104.1. Three different customers
will be discussed:


A large industrial customer who sees one-hour update spot prices, purchases
interruptible energy, and has a long-term contract.
A sophisticated residential customer seeing 24-hour update spot prices.
A simple residential customer seeing 24-hour update spot prices. 6

Large industrial customer (foundry). This industrial customer is a foundry with

many metalworking machines (1 kW to 100 kW), electrical metal melting
(10 MW to melt, 1 MW to maintain molten), lighting, and space conditioning in
the office. This customer has a sophisticated computer system which is used for
both production scheduling (hourly for the next day and week) and direct
control of scheduled processes. There are two individuals responsible for this
customer's interaction with the energy marketplace: a production scheduler and
a production manager. The production manager is concerned with long-term


I. Spot pricing of electricity

(monthly to yearly) issues and makes longterm contracts by combining long

term price forecasts with forecasts of this customer's future electricity needs and
cash flows. The production scheduler is concerned with hourly and daily opera
tion and provides the inputs into this customer's control and scheduling
The scheduling of plant operation is done by combining the shortterm price
forecasts with the specified product mix that the plant is to produce. Metal
melting is always scheduled to occur at the time of lowest forecast prices. When
there is sufficient variation in the spot price, some of the largest metalworking
machines (that require only a few operators) are scheduled to times of low
prices. The office space conditioner is scheduled to make use of preheating and
precooling the building (predictions of future outside temperature is an input to
the optimization logic). When prices get extremely low at night, some of the
metal curing processes that are normally done using natural gas are switched to
electric energy. The price-quantity transactions are adjusted so that metal
melting is always done on an interruptible price-quantity basis. During times
when there is a large difference between the secure and interruptible prices, the
secure level of energy purchase is reduced so that some of tht> other end uses are
also on the interruptible rate (unless the plant has a very tight production
schedule to meet). All of those scheduling operations are done automatically by
the computer using input instructions and data provided by the production
The direct control function of the computer makes minor adjustments on the
actual hourby-hour operations depending on how the hourly spot prices
actually behave (i.e., corrects for differences between actual spot prices and the
short-term forecasts). When the utility calls for an interruption, the computer
drops the overall energy usage to the specified secure level by turning off the
processes which had already been classified as being interruptible.
The plant management is happiest when the spot prices have large time
variations, as it is these variations that offer an opportunity to save money.
A sophisticated residential customer. This sophisticated residential customer
under the 24-hour update spot price has a small, special-purpose computer
which automatically dials the utility once each day to get the 24 prices for the
next day. The computer then controls the space conditioning and water heating
to meet this customer's desires (as told to or learned by the computer) with
minimum cost. If the prices rise above a critical level specified by this customer,
the computer warns the customer so manual control of usage can be undertaken
if desired. The computer acts as an expert system to help the customer diagnose
his/her own needs/desires and make rational decisions during normal and very
high price variation days.
A simple residential customer. This simple residential customer also sees a
24-hour spot price but has no computer. Most of the time, this simple customer
ignores the price variations and operates as usual. However, during the few
critical times of the year when the price gets very high, the customer exercises

I. Overview


manual control. This customer learns of such times by reading the newspaper
and/or listening to announcements made by the utility over the radio and TV.

The example implementation of Section 1.4 emphasizes utility and customer

computers talking to each other. This is reasonable for high-technology parts of
the world where both the electric power system and some of the customers are
very sophisticated. Our second example considers a developing part of the world
where the electric power system is being expanded as fast as possible to meet and
fuel the needs of a rapidly changing, less sophisticated society.
Instead of repeating the box-by-box discussion of Figure 1.4.1, we discuss the
main differences between this case and the sophisticated implementation of
Section 1.4.
In a developing-country implementation, only large industrial and the largest
commercial customers see 24-hour and one-hour update spot prices. The 24hour update is used for most cases. Residential customers (who have meters) see
flat rates which mayor may not be updated each billing period. 7
For the developing country, all long-term contracts and price forecasts are
provided by the utility itself.
The biggest difference is that for the developing country, new industrial and
commercial facilities are built knowing, from the start, that they will be seeing
energy marketplace spot prices. Thus their basic design incorporates control
swilehes, storage capabilities, fuel switching capabilities, etc., that enable fast
anG~large response to~changing prices. With a large labor force, fancy digital
hardwarejs not needed for control. This yields a much more price-responsive
industrial load than found in the developed country, where industries were built
assumingth<:<y would always be furnished highly reliable power at low fixed
rates. As a re'Sult the energy marketplace has a much faster and larger impact
on the choice of generation mix and the amount of generation and transmission
to be built.
To illustrate the impact on expansion planning, assume a policy decision is
made to channel limited financial resources toward renewable generation and
remote hydro sites with relatively weak transmission links to major load centers.
By present-day standards used in developed countries, the resulting system is
unreliable in the sense that it cannot always meet the demand. However, having
an industrial--{;ommericalload that is very price-responsive completely changes
the meaning of reliability. Times of limited capacity are no longer handled by
rotating blackouts. Instead industrial and commercial customers see high prices
and reduce electrical usage in ways that do not have major financial impacts on
their own operations. Industries that must run at a specific time can have all the
energy they need, albeit at the high price. s

This chapter has provided an overview of a spot pricing system. The authors of

20 1. Spot pricing of electricity

the book strongly believe that a spot price based energy marketplace is the
logical and inevitable successor of the present-day system. The key question is
not if it will happen, but when, how and to what degree.
We would like to make a general comment on the appearance of some of our
results. Certain of our formulas lead to price behavior which is quite different
than conventional utility rates, and sometimes is even counter-intuitive. For
example, line overload conditions on a single transmission line can affect spot
prices throughout a network, raising some prices and lowering others. Another
example is that spot prices for customers and generators are entirely symmetric.
Results such as this reflect the physical and economic reality of electric power
systems. In fact in many cases power system practice has been similar to spot
prices for inter-utility sales, but customers have not been allowed to participate
on the same basis. For example, inter-utility sales of economy energy have long
been based on hourly prices, and have treated buying and selling symmetrically.
In other cases, the anomalies are due to regulatory issues such as revenue
reconciliation. They occur in any pricing system, although spot pricing
sometimes makes them more visible by removing hidden cross-subsidies.
One side effect of the way spot prices reflect physical and economic reality is
that spot pricing can be used to evaluate non-spot rates and transactions. For
example, if spot prices paid to a small power producer would over time give it
higher payments than the present-day rate, then at present the small power
producer is cross-subsidizing the utility's other customers.

An energy marketplace is based on the same fundamental principles that

underlie the present-day utility system. However, its operation and effect are
radically different. As a result, many issues arise. These issues are summarized
here. They will be discussed in detail in subsequent chapters.
This material is included as a supplement to the main text because not all
readers will want to wade through it at this time. However, many readers will
eventually find it useful to have a concise summary, and we decided to include
it at the beginning rather than the end of the book.
Table I.S.I lists the issues to be discussed.
In the following discussion on the issues of Table I.S.I, the energy marketplace is compared with present-day transactions such as
Flat Rates: $jkWh fixed for all hours. Specified roughly one year in advance.
Time-Of-Use (TOU) Energy Rates: Rates ($jkWh) vary with hour of day, day
of week, season of year. Specified one year in advance.
Direct Load Control (DLC): Utility has ability to directly turn off or limit kW
usage of individual customer devices or groups of devices. Incentives specified
one year in advance.
Demand Charge: $jkW at time of customer peak demand during month
(possibly with yearly rachet). Specified one year in advance.

1. Overview


Table I.S.1. Issues Relevant to Energy Marketplace Implementation

Changing the Status Quo

Transactions Costs
Utility Operating Costs
Utility System Operation
Utility Capital Costs to Meet Demand
Utility Sustem Planning
Utility Demand Forecasting
Utility Revenue Stability
Customer Options
Customer Education
Customer Understanding
Customer Response
Customer Cross-Subsidies
Role of Regulatory Commission
Impact on Microelectronic Industry

Changing the Status Quo

Introduction of the energy marketplace constitutes a major change in the status

quo. Individuals as we\1 as institutions usua\1y resist change. Utility rate makers
and regulatory personnel are comfortable with present-day procedures. Utility
operators and planners are used to making decisions without detailed consideration of rates and customer reactions. Customers may not be happy with the bills
thty presently pay, but often suspect that change is just another way for the
uttfity to extract more money. Customers who benefit from cross-subsidization
undr present-day rates will be extremely unhappy with an energy marketplace
wb'ereiRthey have to pay a fairer share of the costs. Industrial customers may
h~l.Ve built their plants in ways that require modifications to be able to respond
effectively to price changes.
The costS: and problems associated with changing the present system constitute the single largest obstacle to the implementation of an energy marketplace.
Fortunately, an energy marketplace can be implemented by a gradual transition from the present system. For example, implementation can start with
large industrial customers and be extended to sma\1er customers on a voluntary
basis. The benefits of the marketplace can thus become familiar to customers
and the utility before its more comprehensive adoption.
Developing countries do not have as severe a status quo problem as that faced
by developed countries.

Transactions Costs


The costs of changing the status quo will be large, but transient in nature. The
transactions costs of the additional metering, communication and computation
needed for an energy marketplace .will continue through time. Thus one basic
principle underlying the specification of energy marketplace transactions is that
the associated costs never exceed the additional resulting benefits, i.e., transac-


I. Spot pricing of electricity

tional costs determine how customers are matched to types of transactions. The
microelectronic revolution is driving transactional costs down at a rapid rate.
Utility Operating Costs

These costs include fuel, losses and maintenance associated with generation,
transmission and distribution. Ideally, customer-utility transactions result in
customers shifting some usage to times when the operating costs are low. Flat
rates provide no such motivation. TOU rates can result in some desirable
demand shifting, but because of the unpredictability of the actual times of utility
low and high costs a year in advance (see Figures 1.1.1 and 1.1.2), TOU rates
can also cause undesirable load shifts. OLe could have a sizeable impact on
utility operating costs if it can be exercised as a part of the normal system
operation (i.e. if there is no limit on its usage). However, OLe often involves
contracts with customers which limit the number of utility control actions, so
the reduction in annual operating costs is greatly reduced. Demand charges
motivate customers to reduce their own monthly peak in a manner which
usually has little or no effect on the utility's operating costs.
A spot price based energy marketplace can yield major reductions in utility
operating costs because customers see prices directly related to the actual
operating costs occurring at that time.
Utility System Operation

The control and operation of an electric utility system is done by a highly

sophisticated combination of computer hardware and software, communications, and human judgment. Both economics and system security are of concern.
Flat and TOU rates usually have little positive impact on control and operation,
but TOU rates can complicate system operations by causing sharp jumps in
system demand at times of TOU changes. Some types of OLe impose heavy
demands on system operators. Good models which predict OLe impacts on
demand can be difficult to develop. If there are limits on the number of possible
OLe actions, the system operators become gamblers who have to guess when
to use a limited resource in an uncertain world. This also holds true for many
present-day interruptible contracts.
The energy marketplace's ability to vary prices in response to actual conditions provides system operators with a powerful new control signal. Good
models for predicting short-term demand response are obtainable. Spot prices
are control signals which are available all the time and which can get as large
as needed, as often as needed, while still reducing overall customer bills.
To maintain system security, power systems always try to operate with
sufficient operating reserve. The energy marketplace provides a mechanism for
moving some or all of these operating reserves from the generators to the
Utility Capital Costs to Meet Demand

Ideally, customer-utility transactions reduce the required investments for new

I. Overview 23












generation, transmission and distribution by providing very strong incentives

for customers to reduce usage during those few critical hours of a year when the
utility system is approaching its capacity.9 Customers on TOU rates usually
have little incentive to reduce demand at such critical times because the swing
in TOU rates is too small, and the high TOU rates are in effect for many hours
a week. Similarly, demand charges provide an incentive for customers to behave
in a way which has a beneficial impact on utility capital requirements only in
those rare instances when the customer's monthly peak hour happens to correspond to a critical time. DLC can provide the utility with short-term load relief
only during those times when the particular appliances under control are being
used by the customers (e.g., residential air conditioning cycling may not be very
useful during the winter). DLC often has very limited value in long-term
emergency situations (e.g., oil embargoes, coal strikes, and nuclear moratoriums), because of the limitations imposed by customer contracts on the
maximum frequency of its use.
An energy marketplace enables major utility capital investment savings for
both the short-term and long-term emergency conditions. The size of the pricing
signals during the few critical hours can be very large compared to TOU
variations because TOU rates correspond to average conditions. Demand relief
can be obtained during Fall and Spring, if it is needed because of unexpected
equipment outages combined with maintenance scheduling. In major long-term
emtrgencies, spot pricing can be viewed as a method of rationing by price. This
iscwore desirable forLboth the utility and its customers than the use of rotating
blackouts, brownouts, or any other arbitrary rationing scheme.

Utility System Planning

Uncertainty in long-term load growth, fuel costs and availability, capital costs,
environment'al standards, etc. cause longfalls or shortfalls in utility capital
investments. Spot pricing provides feedback that acts as a buffer against uncertainty. Spot pricing is effective during times of capacity excess as well as shortage
(short- or long-term) because low prices can stimulate use of electric energy.


Utility Demand Prediction

An electric utility needs to predict future load behavior on time scales ranging
from hours to years.
Predictions of short-term (hours) response to DLC requires a sophisticated
model for each individual applicance type under control (e.g., an air conditionmodel depends on time of day, temperature, and season of year). Developof DLC response models is difficult because DLe tends to operate only
and hence only a limited data base is available. Prediction of short-term
response and of short-term spot pricing response is very similar. Spot price
does require a somewhat more sophisticated model than TOU
but the marketplace transactions provide extensive data on both
:usltonler desires and usage patterns which enables the modeling to be done.

24 1. Spot pricing of electricity

Even more important is the fact that spot prices provide closed loop feedback
control which reduces the impact of uncertainty.1O
The major determinants of long-term (many years) load behavior are exogenous quantities such as the general state of the economy, demographic trends,
and the price of electricity relative to gas and oil. Long-term predictions of OLC
impacts present special problems because many of the OLC techniques are
vulnerable to customer countermeasures, which customers learn over time as the
OLC is exercised more and more. The detailed information on customer usage
patterns and priorities provided by marketplace transactions is a major aid to
the long-term modeling problem.
Utility Revenue Stability

In an ideal world, a regulated utility receives revenues which cover its costs plus
a reasonable rate of return on investment. In the real world, life is not so simple.
TOU rates have caused revenue stability problems because utilities have found
it difficult to specify TOU rates that yield the specified revenue targets. An
energy marketplace introduces similar problems. However, the ability to continuously adjust prices to market conditions makes it possible to greatly reduce
revenue stability problems associated with modeling customer response. These
problems will decline as data becomes available to develop the necessary
Customer Options

Customers have a growing desire to make their own decisions on the type of
electric service they purchase. OLC and interruptible contracts are steps in this
direction. However, an energy marketplace provides customers with a richer
spectrum of options and allows more freedo,m of choice.
Customer Education

An energy marketplace results in increased customer awareness of the value of

electric energy. When customers have to decide how to respond, they automatically begin to consider the value of the services provided by electric energy. This
educational process greatly improves the relationship between the customers,
the utility and the regulatory commission.
Customer Bills

Customers almost always say they want to reduce their monthly electricity bills.
Actually they are willing to pay a higher bill if it is associated with tangible
benefits they understand, such as increased comfort or production. Spot pricing
gives customers maximum control over their bills.
The time average of TOU rates is higher than the time average of spot prices.
Thus customers with flat usage see a reduction of their bills in the energy

I. Overview 25

Customers (like utilities) do not want to face uncertainty. Customers would

like to be able to know what their bills are going to be in the future. In an energy
marketplace, the customer's ability to control can be used to reduce bill uncertainty if desired. The energy marketplace also provides a mechanism (via
long-term fixed-price-fixed-quantity contracts) for a customer to buy a type of
insurance policy if uncertainty is particularly important. The uncertainty in
hourly spot prices does not translate directly into an inability to predict future
monthly bills with reasonable accuracy, because the summing of hourly costs
into monthly bills introduces a lot of averaging. Present-day fuel adjustment
clauses can introduce more uncertainty in monthly bills than spot pricing.
Customer Understanding



Customers need to understand what is going on. Customers can readily understand TOU rates for energy, but are often confused by demand charges and
DLC contracts which include limitations and conditions on utility control.
Spot pricing is a concept customers can understand because they are used to
seeing prices which fluctuate in response to supply and demand conditions (e.g.,
the price offresh produce and airline tickets). An energy marketplace can result
in a smaller and more easily understood set of transactions than that presently
used by many utilities. Today, many utility rate books are thick documents that
take training (and a lot of patience) to understand.
<;ustomer Response

An i~sue closely related to customer understanding is the need for customers to





re~ond,., DLC has the advantage that the customer does not have to make or
implement'real-time decisions. However, DLC can have a major negative
impact on ,c:ustomer lifestyle unless override capabilities are provided (e.g.,
residential atr conditioner cycling during a hot day when the customer is having
an important party at home). Modern electronics can make it easy for a
customer to respond to spot price variations. Since the customer is in final
control, major impacts on lifestyle such as might occur due to the loss of air
conditioning at particularly inappropriate times do not happen under spot
pricing. The utility can provide control services for residential and small commercial customers where the customer chooses the control strategy to be
.T/"\II/"\\l"Pfl while the utility provides the hardware and software for its imple-

Industrial and large commercial customers with energy management systems

modify them to respond to spot prices.
;lIlstOirner Cross-Subsidies

equity issue of energy subsidies between customer classes and between

within a given class is becoming ever more important. Present-day

26 I. Spot pricing of electricity

transactions provide a limited vehicle for reducing such cross-subsidies. An

energy marketplace cannot completely eliminate cross-subsidies, but it can
greatly reduce them. If cross-subsidies such as life-line rates are decreed to be
socially desirable, they can be incorporated into an energy marketplace.
Role of Regulatory Commission

Investor-owned utilities operate under regulatory commissions. An energy marketplace changes the role of the regulatory commission.
In an energy marketplace, the regulatory commission no longer determines
the actual values of the rates. Instead the commission determines the formulas
which are used to compute the spot prices. This is a logical evolution from
present practice, such as fuel adjustment clauses. It has the advantage that the
commission is oper~ting at a higher level of decision making which is more
appropriate to its basic function.
An energy marketplace gives the commission more options on how to provide
the utility with incentives for efficient operation.
Impact on Microelectronics Industry

The microelectronic revolution has given birth to many new, dynamic companies which are trying to introduce new communication, computation and
control systems into the electricity market. However, many vendors have
become extremely frustrated with the present-day electric utility situation
because it is not clear what sort of product is marketable. The establishment of
an energy marketplace will specify the ground rules for such vendors and
provide them the opportunity to exploit their capabilities and offer innovative
products to the customers. This will result in reduced costs for transactions and
customer response.
Discussion of Supplement

All of the above utility, customer, regulatory and vendor issues have to be
considered when deciding whether to follow the energy marketplace path and,
if so, what directions to take. We believe that careful weighing of the costs and
benefits will show an energy marketplace to be a clear winner in virtually all

As discussed in Section I. I, a spot price based energy marketplace can be viewed

as the logical evolution of present-day practice in any of the three fields: rate
making, load management, and power system operation. As a result, the background literature base underlying spot pricing is huge. At the end of each
chapter we will use these "Historical Notes and References" to help the reader
sort through the literature. However, we readily admit to failing to reference
all the related work. In some areas we simply chose not to take the space to
provide a comprehensive survey. Furthermore, we try hard but we are only
human; i.e. we don't know everything.

I. Overview











We do provide at the end of the book an Annotated Bibliography of our own

reports and papers. Thus a small part of the overall subject is reasonably well
To our knowledge, Vickrey [1971] was the first to explicitly propose a spot
price based type of s~stem similar to Section 1.1. He called it responsive pricing.
We do not use his terminology because, unfortunately we were well into our
"spot pricing" research before we learned of his work.
Another term coming into usage in the literature is "real time" rates (or
prices). The hourly spot price based rates discussed in this book are real-time
rates. However, a real-time rate is the same as a spot price based rate only if it
is based in a self-consistent, logical fashion on hourly (or some similar shorttime span) spot prices, i.e., marginal cost with revenue reconciliation. Hence,
real-time rates that are not the same as spot price based rates can and do exist.
However, in many cases, one cannot tell the difference without exploring the
basis under which the real-time rate is established.
At the present time, the electric utility industry in the U.S. and Europe seems
to be moving (continuously, albeit slowly) towards implementation of an energy
marketplace of the type discussed in Section 1.2. Tabors, Schweppe, and Caramanis [1988) summarizes existing rates and load management schemes that are
either directly spot price based or closely related. The report will undoubtedly
be o,ut of date by the time this book is published. However, it is a place to start
f<ft:lhose interested ill implementation experiences. Garcia and Runnels [1984]
previde a utility per~pective on spot pricing.
'We make no attempt in this book to survey the literature on existing response
m6deling as per Section 1.3. However, the Annotated Bibliography does cover
our own work in the area.
The deve1<i'ping country discussion of Section 1.5 is taken from Schweppe and
Tabors [1984].




1. There are, of course, other definitions for "equity."

2. We believe that Vickrey (1971) was the first to propose a pricing scheme that was close or
equivalent to our spot pricing concept for electric energy. He called it "responsive pricing."
Chapter 2 lists other developers of versions of responsive pricing.
3. The use of an hour as the fundamental time unit is convenient, but not essential. Definitions
could range from minutes to several hours.
4. These energy marketplace TOU and flat rates differ from conventional ones because they vary
each billing period and a demand charge is not used.
5. Of course, more sophisticated and expensive communication systems can also be used.
6. Many residential customers will see spot price based flat or time-of-use prices which are updated
each billing period, but 24-hour update residential customers make a more interesting example.
7. The choice of what types of residential metering systems to install is made keeping in mind the
fact that, eventually, developing countries become developed and later, the energy marketplace
will be extended to the residential sectors. Thus an electronic based system might be employed
to avoid the "trap" developed countries now face of having a major investment sunk into
already supplanted technology, in this case simple meters.


1. Spot pricing of electricity

8. A variation on this approach is to replace the use of prices to allocate a scarce resource by
allocating each industry energy credits, where a kWh used during critical times requires the use
of more credits. This leads to the possibility of bartering among customers (see Williams, 1984).
9. Such times often do not correspond to times of peak system demand because of maintenance
scheduling and random equipment outages.
10. In theory, feedback can sometimes aggravate the effect of uncertainty and even lead to an
unstable system. However, spot prices are computed in a fashion which provides desirable





I of this book is designed to provide the reader with an understanding

. spot price based energy marketplace without having to go through the
UO;;Lall<;U equations which form its basis. The three main chapters are
t::lla[)lter 2: Ifehavior of Hourly Spot Prices
,ctlap,ter 3: Energy Marketplace Transactions
4: Implementation
organization foJlows the three steps to an energy marketplace discussed in
I. FinalJy,
5: A Possible Future: Deregulation
the possible evolution of the energy marketplace (as developed for
utilities) into a partiaJly deregulated world.



The hourly spot price for electric energy is the foundation of our approach
an energy marketplace. This chapter provides a review of hourly spot price
ci)ncepts. Mathematical details can be found in Part II. The Annotated Bibliogiaphy'contains an extensive list of references and discusses their contents.
The hourly spot price is determined by the supply/demand conditions that
exist at thai hour. In particular it depends on that hour's:



Demand (in total and by location)

Generation availability and costs (including purchases from other utilities)
Transmission/distribution network availability and losses

This chapter provides the reader with an understanding of the composition and
interpretation of hourly spot prices.
Sections 2.1 through 2.4 define the hourly spot price and its components.
Section 2.5 discusses the special case where an aggregated network model is used
instead of explicitly considering individual network lines. Section 2.6 discusses
revenue reconciliation. Section 2.7 discusses buy-back rates. Sections 2.8 and 2.9
discuss two measures of hourly spot price behavior: expected price trajectories
and the price duration curve. The chapter concludes with Section 2.10, which
considers customer response.


32 I. The energy marketplace


Pk(t): Hourly spot price for kth customer during hour t ($/kwh)
dk(t): Demand of kth customer during hour t (kwh)
d(t): Total demand of all customers during hour t (kwh)
del) = Lk dk (t)

An hourly spot price can be quantified in various ways. The basic approach
used in this book is
Pk (I): Marginal (or incremental') cost of providing electric energy to cllstomer k during
hour ( taking into consideration both operating and capital costs ($/kwh).

Definition of Marginal Cost

The hourly spot price (without revenue reconciliation) is given by the marginal
cost, i.e.
(Total cost of providing electric energy to all customers
now and through the future]

(2.1. I)

The derivative of (2.1.1) is evaluated subject to constraints such as

Energy Balance: Total generation equals total demand plus losses.
Generation Limits: Total demand during hour t cannot exceed the capacity of
all the power plants available at hour t.
Kirchoff's Laws: Energy Rows and losses on a -network are specified by physical
Line Flow Limits: Energy Rows over a particular line cannot exceed specified
limits without causing system operating problems.
Revenue Reconciliation

The basic definition of (2.1.1) involves only marginal costs without consideration of revenue reconciliation - i.e. without consideration of embedded capital
costs and rate of return on investment.
A key property of marginal cost based spot prices (equation 2.1.1) is
They tend to recover both operating and capital costs.
Since generation is assumed to be dispatched optimally, marginal costs exceed
average variable operating costs. Thus, charging customers at marginal costs
yields revenues that exceed total variable operating costs; and this difference can
be applied towards the capital costs. An optimum power system's marginal costs
yields revenues which exactly match operating and capital costs. Unfortunately,

2. Behavior of hourly spot prices 33

in the real world, this difference will usually either over- or underrecover capital
costS.2 Mechanisms for revenue reconciliation are discussed in Section 2.6.
NUMERICAL EXAMPLE. Consider a utility with two generators where

Generator I
Generator 2

Operating Costs

Assume the generators are optimally dispatched (i.e. Generator I is used until
demand exceeds 1000 MW). Then, ignoring losses and capital costs, it follows

If Demand
If Demand

1000MW, then:
1100 MW, then:

Operating Cost

Average Op.

Spot Price




Maximum demand = 1100 MW

Benefit customers receive from using electric energy is 5 (kWh

short-term social welfare as

S1;ivr't- T~rm Social We:lfare =

Customer Benefit - Utility Operating Costs

Assume customers behave in their own best interest, i.e. are not willing to pay
1O(kWh fCir a benefit worth only 5 (kWh. Then it follows that




If customers pay spot

prices, the demand will be
cut-off an 000 MW and:
If customers pay average
operating costs, then:




Social Welfare









Hence short-term social welfare is higher if customers see spot prices instead of
average operating costs. Now change the conditions and assume the customer
benefit is 15 (kWh instead of 5 jkWh. Then for both pricing schemes
Demand = 1100 MW
Customer Benefit = $165,000

34 I. The energy marketplace

Utility Cost = $30,000

Social Welfare = $135,000
The utility's revenues (price times demand) are

If customers pay spot prices, then:

If customers pay average operating
cost, then:

Utility Revenue

Utility Revenue
Minus Costs ($)




The $80,000 profit made by the utility under spot pricing can be used to pay the
capital costs of the two generators; however, this may either over- or underrecover the actual capital costs of the plants. We will return to this issue in
Section 2.6.

The hourly spot price associated with the kth customer during hour t is viewed
as the sum of individual components defined byJ,4
[Generation Marginal Fuel]

Pk(t) = Yr(t)



[Generation Marginal Maintenance]


[Generation Quality of Supply]

YR (t)

[Generation Revenue Reconciliation]

~L.k (t)

[Network Marginal Losses]


[Network Quality of Supply]


[Network Revenue Reconciliation]


Quality of supply components arise when generation or network capacity limits

are being approached. Thus they serve as curtailment premiums or reliability
surcharges. The components of (2.2.1) are often combined into groups such as





'lk (I)



[System Lambda]s


[Marginal Value of Generation]

[Marginal Value of Network Operation]



The hourly spot price is decomposed into components as in (2.2.1) because

each component of (2.2.1) has a physical/economic interpretation. However, it
is important to note that (as will be shown shortly)
The various components of (2.2.1) are not necessarily independent of each other.

2. Behavior of hourly spot prices 35

For example, the loss component 'fJL.k(t) depends on system lambda, A(t).
The network components of the hourly spot price depend on the customer
index k because different customers are located at different parts of the network.
This spatial pricing results from the differences in line losses and the fact that
individual lines can become overloaded in one part of the network while over
the remaining lines flows are sustainable. In practice, spot prices will be the same
for most customers in a specific class, e.g., service at 13 kV in the Sanderson
Township. Section 2.5 introduces aggregated network models which eliminate
some of the complications associated with spatial pricing.

The operating cost components of the hourly spot prices are usually the largest.
They are:
Generation Fuel and Operations: A(t)
Network Losses: 1)L,k (I)

Generation Marginal Fuel and Maintenance: let)

The system lambda, A(t), component of the hourly spot price is the derivative
of generation total fuel and maintenance costs with respect to demand this hour.
Ideally it is obtainecLas the output of the unit commitment generation dispatch
logics used in many modern electric power system control centers. System
lam~da (as it is defirted in this book) includes the effect of purchases and sales
th' utili!y may make with external, interconnected utilities.
In general, A(t) tends to increase with increasing total demand d(t). The A vs.
demand curve varies over time since it depends on forced (and scheduled) power
plant outag1s, water availability, purchase-sale opportunities, load-following
costs as affected by ramp and must-run constraints, etc. When these intertempora! constraints are important, A(t) may be heavily influenced by planned
future events. For example, in a system with a lot of hydro storage, the incremental generator is sometimes a hydro unit. The system lambda at those times
is calculated based on the anticipated running cost of a fossil unit which will
need to generate more when the reservoirs are empty.
NUMERICAL EXAMPLE. Consider a 14-generator system (no transmission losses)
with maximum demand of 2000 MW where
1 Nuclear
I Coal
12 Gas Turbines

100 each

Operating Cost

Figure 2.3. J shows how the availability of the nuclear and coal plants affects the

A vs. demand curve.


I. The energy marketplace

Case 1: 80th Nuclear and Coal Available


Gas Turbine

10 f-







Case 2: Coal Not Available

Gas Turbine






Case 3: Nuclear Not Available

Gas Turbine




800 1000

1800 2000

Figure 2.3.1. Effect of plant availability on A vs. demand curve.

Network Losses: '1L.k(t)

This component arises from the energy losses resulting from transmission and
distribution. It is shown in Chapter 7 that, assuming a quadratic dependence of
losses on line flows, 1]L,k(t) is given by

}'Qs(t)] fJdk(t)


2. Behavior of hourly spot prices 37


I'Qs(t)] ~ 2R;z;(t) adk(t)

dk(t): Demand of kth customer during hour t (kWh)

z;(t): Energy flowing over line i during hour t (kWh)
L(t) : Total losses during hour t (kWh)

L, L,[z,(t)]

losses in line



R;: Constant depending on resistance of line i

The effect of dk(t) on total losses and/or individual line losses depends on the
physical location of the kth customer in the network. This dependence (as
determined by Kirchoff's laws) is discussed in detail in Chapter 7 and in
Appendices A and D.
The marginal network loss component can be quite important at times of high
demand even if annual percentage losses are relatively small.
NUMERICAL EXAMPLE. Consider a two-bus, one-line system with all generation
on ~ne bus and demand d(t) on the other. There is only one flow z(t), so

aztO '=





and (2.3.1) yields

ryl(t) =



Assume annual line losses over a year are 5% of annual flows. Thus
Annual Losses = 5% Annual Flows







L z(t)


Assume the demand variation over a year is such that

500 MWh for 5000 hours

1000 MWh for 3000 hours

38 I. The energy marketplace

1500MWh for 760 hours

Then the value of R can be back-calculated as:

R =

(0.05)[(500)(5000) + (1000)(3000) + (1500)(760)]

(50W(5000) + (1000)2(3000) + (l50W(760)
(0.05)(66.4)(10 5 )
59.6(108 )

5.5(10- 5 )

Therefore when the line is heavily loaded, i.e. z




1500 MWh,

+ 1'Qs(I)]2(5.5)(l0-5)(1500)
+ 1'Qs(I)][O.165]

When the line is lightly loaded, i.e. z

I1L (t) =

500 MWh,

1'QS (1))[0.055]

Thus for annual losses of 5%, the marginal network loss component ranges
from 5.5% to 16.5% of A(t) + YQs(t). Note that if line load were constant over
the years, marginal network losses would be twice the average loss, or 10%.

The generation and network quality of supply components, YQs(!) and IJQS.k(t),
can be quantified in different ways. All approaches yield behaviors characterized
by very small or zero levels most of the time with a large, rapid increase when
the generation or network capacity is being ,approached. During such critical
times, these quality of supply components dominate the hourly spot price.
Generation Quality of Supply: l'Qs(t)

get): Total generation during hour t (kWh)
get) = Total Demand + Total Losses
= d(t) + L(t)
gcrit.y(t): A critical generation level based on available generation capacity and
operating reserve requirements

The utility tries to operate to maintain the constraint

Three methods described in Chapter 6 for quantifying YQs(t) are

2. Behavior of hourly spot prices 39

Market Clearing Price: Set I'Qs(t) to be the value that causes customers to
reduce usage until g(t) = gcrit.y(t). This value depends on the amount of load
reduction required.
Value of Unserved Energy: Set I'Qs(t) such that the resulting spot prices equal
the cost to the customer of unserved energy, averaged over customers end
Allocation of'Peaking Plant Capital: Base YQs(t) on the annualized capital
cost of a new peaking plant.

For the peaking plant cost allocation method, one approach discussed in
Chapter 6 yields
(2.4.1 )


L (1;(1)



Aos,;': Annualized capital cost of peaking plant ($/kW)




LOLP;.(t): Loss of load probability due to generation at hour

ii;,= LOLH;.: Annual loss of load hours

Note that multiplication of av(t) by any constant does not change YQs(t)
'.Jhe loss of load probability LOLPy(t) in (2.4.1) is evaluated for hour t at the
be-ginning of hour t. ~f it is assumed that all events can be predicted perfectly one
h(),ur in advance, then


LOLP .. (I) =



g(t) > gc,iq(t)


However, a more reasonable model has LOLPy(t) varying smoothly between 0

and I as g(t) approaches gcrit,y(t)
The market clearing price approach is recommended for an ideal world.
However, the other two approaches can be easier to implement in the real world.
The value of unserved energy is related to but not necessarily equal to the
market clearing price. 7
NUMERICAL EXAMPLE. Consider a utility whose marginal fuel and maintenance
costs (i.e. A) reach 10 jkWh when demand is large. Ignore all network costs.

900 MWh


1000 MWh

if p(t) = ).(t) = 10/kWh

To illustrate the market clearing price approach, assume all customers are seeing
spot prices and that a 10% drop in load requires a 20jkWh increase in price
(i.e., an elasticity of - 0.05). Then YQs (t) = 20 jkWh. To illustrate the value of


I. The energy marketplace

the unserved energy approach, assume the cost to the customer of unserved
energy is 100 /kWh. Then when generation starts to exceed its critical level,
')IQs (t) goes up to 90 /kWh so thaL1.(t) + ')IQs (I) = 100 /kWh. Finally, consider
an illustration of the peaking plant capital approach. If Annualized Plant
Cost = 100$/kW
LOLH y = 200 hours,
')IQs(t) = [50/
kWh] x LOLPy(t).
Network Quality of Supply:


By analogy with 1Qs(t), I]QS.k(t) becomes large in magnitude when the capacity
of the network to transport energy is being approached. Assume one particular
line, say line i, with flow z;(t) is overloading. One structural form for 'lQS,k(t)
derived in Chapter 7 is
110s,k ()
t =


odk (l)


()os."Al) has units of marginal cost, $jkWh

where two ways to model the marginal cost 8Qs .ry,,(t) are

Market clearing: 8Qs .,I ,; is adjusted until customers and generators respond to
change the usage and generation patterns so that the line overload goes away.
The derivative of a cost function with respect to z;(1) where the cost function
is zero until z;(t) approaches the line overload level.

Spot prices are affected at locations throughout the network even if only one line
(line i) is being overloaded,
NUMERICAL EXAMPLE. Consider two buses with one lossless line connecting
them. Assume Bus 1 has a base-load generator with marginal fuel costs of
5/kWh while Bus 2 has a peaking plant with marginal fuel costs of 1O/kWh.,
Assume all load is at Bus 2 and the load is less than the base load unit's capacity.
Thus if the line can carryall of the demand, the peaking plant generation is zero

A. = 5/kWh, I]QS.k = 0
At both busses p = 5/kWh

If the line capacity is less than the demand and the peaking plant has to be run
to meet the demand, the resulting prices are

At base load bus

At peaking plant bus

which is achieved by using

p = 5/kWh


= 5 /kWh, while 'lQS,

= O.

2, Behavior of hourly spot prices 41



An Upper Bound on the Hourly Spot Price

The generation and network quality of supply components can yield very high
spot prices under some conditions. If the energy marketplace contains a mixture
of customers seeing hourly spot prices and customers seeing long-term,
predetermined prices, then the outage costs of the predetermined price
customers set an upper bound on the hourly spot price. Thus
When the hourly spot price reaches their outage costs, the predetermined price customers
are "dropped" rather than raising the hourly spot price any further.


This upper bounding is discussed more in Section 9.1.

;,k (t)



j to


The network loss '1L,k (t) of (2.3.1) and network quality of supply '1QS,k (t) of
(2.4,2) are based on an explicit network model which considers individual lines
(and transformers, etc,). In practice it is often desirable to use a more aggregated
network model for the transmission system. Even if all individual transmission
lines are modeled, some aggregation of lower voltage distribution lines will
almost always be desired.
To illustrate the approach, assume all transmission and distribution lines are
t(),.be aggregated together. This can be viewed as a hypothetical two-bus system
'~here all generation is on one bus and all demand d(t) is on the other. Therefore
"ill customers see ~the same spot price; i.e., Pk(t) = pet), '1L.k (I) = '1L (t),


;; Thenetwork loss component of (2.3.1) becomes (see Chapter 7)




where the losses L(t) are the function of total demand d(t), total generation get),
and possibly other quantities such as purchases and sales from other utilities.
The network quality of supply term of (2.4.2) is now based on the use of an
aggregate cost function. Since individual line flows are not modeled, market
clearing or a cost function based on individual lines cannot be used. One
approach discussed in Chapter 7 is closely related in philosophy to the "annualized cost of a peaking plant" method of computing the generation quality of
supply YQs(t). It is

11Qs(1) =

AQs ." a"y)



AQs ,),: Annualized capital cost of network expansion made to maintain system reliab
ility ($jkW)

LOLP~(t): Loss of load probability due to effect of demand on the network's

ability to meet demand


I. The energy marketplace


L a~(t)



A reasonable model for



k[d(t) - dcd',~]

d(t) > dcrit.~



dcric ,,: Parameter chosen by engineering judgment (say 80 to 95% of dmax )

Parameter whose value does not effect ~Qs(t)


If the demand d(t) is viewed as a random variable with a flat probability density
between d crit." and dma., then it is shown in Section 7.9 that (2.5.2) (2.5.3) become
for del) ~ dcrit.q



Prob[d(t) > detll .,,]

Then for d(t)



~ dcrit.~,

(2.5.4) yields

2( 100)[d(t) - 0.9 dmax ]

8765(0.1 )dmax (0.05)

so for del)
~Qs (t)


[::~ -

0.9] $/kWh


45 /k Wh


Electric utilities are usually run by a government agency or by private industry

that is regulated by a government agency. Hence there is usually a need for
revenue reconciliation to insure that they do not make or lose (too much)
money. Approaches to revenue reconciliation discussed in Chapter 8 include

Use of surcharge or refund

2. Behavior of hourly spot prices 43



Use of revolving funds

Modifying the spot price

For an ideal world, the use of revolving funds or certain types of surcharge/
refund is recommended, as they do not change the hourly spot price and hence
attain the maximum possible social benefits. However, such approaches present
many practical implementation problems. Therefore we will discuss here the
approach of modifying the spot price through the use of revenue reconciliation

Pk(t): Hourly spot price without revenue reconciliation component




[Generation Components]

[Network Components)

II! (I)


(2.6.1 )


The basic idea is to modify the prices paid by the customers so that the
utility's revenue over some time interval (say one year) covers its operating and
eniJ';,edded capital costs plus a reasonable rate of return on investment. This
gives rise to the gen~ration and network revenue reconciliation components,
y~~O'and IJR.k(t) respectively.
0ne sttucture that~is a special case of the Ramsey pricing or weighted least
squares theory of Chapter 8 is a multiplicative modification, i.e .

YR (I)

m'lA (I)


which yields an hourly spot price with revenue reconciliation given by


The reconciliation multiplier m in (2.6.2) is a constant which is adjusted until

the expected annual revenues equal the annual target revenue. Thus the value
of m is specified by the condition





I =:: I


(Annual Target Revenue)



where the left-hand side should actually be written in terms of the expected value
of lJdt)dk(t). If demand response to price is considered, dk(t) is a function of Pk(t)
which is a function of m, so an iterative solution of (2.6.3) for m is required.


I. The energy marketplace

In Chapter 8 other multiplicative forms are discussed. In one form revenue

reconciliation is obtained separately for generation and network capital costs so

where m and m~ can have different signs. Another decomposes revenue reconciliation "even further down to individual lines or voltage classes of lines. Still
another approach yields revenue reconciliation multipliers that vary with k, e.g.
depends on the kth customer's demand elasticity. Some forms discussed in
Chapter 8 have m's which depend on dk(t), i.e., represent nonlinear pricing
modifications. We emphasize use of constant multipliers in this book because
they simplify the discussions. In practice, a more sophisticated approach can be
chosen for actual implementation.

Ifa utility buys back electric energy from its customers, an hourly buy-back spot
price is needed. An important point developed in Part n is
The operating and quality of supply hourly spot price components are independent of
whether the customer is buying from or selling to the utility.

However, revenue reconciliation destroys this symmetry. Revenue reconciliation increases the hourly buy-back spot price when the utility is overrecovering
revenue and vice versa.
In Section 2.6, one hourly spot price with revenue reconciliation is given by
Pk(t) =





This same basic formula applies to both buying and selling, except the value of
m changes. Define

Value when utility is selling to customer

Value when utility is buying from customer

For the special case leading to (2.6.2), it is shown in Chapter 8 that after a few

so that if
Spot price for selling energy to customer
Pbuy.k(t): Spot price for buying energy from customers

Psell.k (t):

2. Behavior of hourly spot prices 45

P"II.,(t) =



(1 - m)[y(l)

Pbuy.k(l) =


P,dl = ~ =



If m





which is a 50% difference.

An Unfortunate Complication
The use of (2.7.2) can lead to unpleasant complications in some situations. As
an example, consider a customer with demand d(l) and own generation get)
where d(t) > g(t). If the customer has a single meter which measures a net
purchase of d(t) - get) from the utility, the customer's bill is

Bill = (\



tik(t)][d(t) - g(t)]

If the customer has two meters, one for demand d(t) and one for generation get),
customer's net bill is

Net Bill

= (I


- (I - m)(y(t)



(Generator Buyback]



Unfortunately (2.7.3) and (2.7.4) are not equal unless m = 0, i.e., unless revenue
reconciliatiOin components are not needed.
This problem is not basic to spot pricing; it is simply a consequence of using
a simple approach to revenue reconciliation. Other approaches to revenue
reconciliation discussed in Chapter 8 such as the use of revolving funds solve this
problem but introduce other complications. The world would be nicer if revenue
reconciliation could be ignored.

The preceding sections discussed the individual components of the hourly spot
price. It is now time to put them together and look at the behavior of the spot
price itself.
The hourly spot price is a random process because demand and outages are
random. Figure 2.8.1 presents expected spot price trajectories for 24 hours
assuming no spatial (k) dependence, and that demand is independent of price.
These trajectories are the conditional expectation of the spot price given a
specific demand level; i.e. the effects of generation outages have been averaged


I. The energy marketplace

Mils/kWh .-----y-EA-A-'-=-=PECCAK"'"'O-AY-,:S-:S....,UM-:M-:AR::c
y- - - - - ,






























Figure 2.8.1. Hourly spot price.

The expected 24-hour price trajectories as in Figure 2.8.1 are computed by a

two-step process:
, First, a curve of E{p(t)/d(t)} versus d(t) is computed where E{p(t)/d(t)} is the
expected value of pet) given d(t).
o Then the curve is "multiplied" by 24-hour trajectories of demand d(t) versus
The computation of E{p(t)/d(t)} requires the 'evaluation of E{A.(t)/d(t)}. This can
be done using the results of production cost computer program (see Appendix
C). The actual procedure used depends on the nature of the program used.
Appendix G presents the details of how the trajectories of Figure 2.8.1 were
actually computed.
The 24-hour spot price trajectories of Figure 2.8.1 correspond to the forecasted typical and peak days of the same utilitl in two consecutive years. These
years were selected because during year I the utility was expected to have a
capacity shortage, while during year 2 installation of significant new baseloaded capacity was expected to change the shortage to an excess.
The following conclusions can be drawn from Figures 2.8.1:

The expected hourly spot prices exhibit very wide variations with time.
The behavior of spot prices may change significantly with changes in installed
It is tempting to look at Figure 2.8.1 and also conclude that hourly spot prices

2. Behavior of hourly spot prices 47



Price is Greater than 1O~/kWh for 2000 Hours




Figure 2.9.1. Alfannual price duration curve.



te,nQ to be smallest -in the Spring and Fall. However, such a conclusion is not
n~essarily true for1wo reasons:
Maintenance Scheduling Was Ignored: Most maintenance is done in Spring
and Fa\l~Therefore prices during Spring and Fall will usually be higher than
shown here.
" Averages Over Generation Outages Were Used: The actual p(t) can get quite
large in Spring or Fall if major forced outages occur.

Because of generation outages, the highest spot prices will usually not occur
at the hour of the highest demand. Furthermore, actual spot price trajectories
will not always be as smooth as those of Figure 2.8.1, since a generator outage
during hour 1 can cause the spot price to "jump" the next hour.

The 24-hour price trajectories of Section 2.8 give one picture of hourly spot price
behavior. A different picture (taken from a different angle) is given by looking
at a price duration curve which defines the probability distribution of the hourly
An annual price duration curve can be thought of as summarizing the
information included in 365, 24-hour expected price trajectories; see Figure

48 I. The energy marketplace

2.9.1. It is analogous to the load duration curve. Price duration curves can be
obtained for a year or a subyearly period.
A price duration curve can be calculated using three different methodologies:

Aggregation of expected spot price trajectories

Use of Monte Carlo simulation
Use of probability convolutions

They yield similar but not identical results.

Aggregation of Hourly Spot Price Trajectories

The conceptually simplest way to compute an annual price duration curve starts
by computing expected hourly price trajectories for 365 days and then summing
the number of hours in the year that the expected value exceeds some specified
Monte Carlo Simulation

The hourly trajectories of Section 2.8 are computed using the expected value of
the marginal costs where the expectation is over generation outages. Thus a
price duration curve computed directly from such hourly trajectories (as above)
does not display all of the actual characteristics of the hourly spot price variations.
A more accurate procedure is to compute hourly spot price trajectories by
Monte Carlo simulation, wherein the explicit values for the generation outages
for a given day are first determined using random numbers and then the 24-hour
price trajectory for that day is computed. Plotting the resulting 24-hour trajectories would not be very meaningful, but by doing enough random sampling, the
random hourly price trajectories can be averaged into the actual price duration
curve. The Monte Carlo price curve will have a higher peak than the one which
just summarizes 365 expected 24-hour trajectories.
Monte Carlo is the most versatile and powerful approach because detailed
time dynamics (such as customer response to spot prices using thermal storage)
can be incorporated. Unfortunately, the Monte Carlo approach can require a
large amount of computer time.
Use of Probability Convolution

Probability convolution is less demanding than Monte Carlo in terms of both

computation and input data requirements. Its major disadvantage is that it is
not able to model detailed time dynamics.
The probability convolution approach employs the technique of Equivalent
Load Duration Curves to provide estimates of the proportion oftime (duration)
that a specific generating unit is on the margin, i.e. the proportion of time that
system lambda equals that unit's variable energy costs. Since system lambda is
a major component of the spot price, this is an important step toward obtaining

2. Behavior of hourly spot prices 49

n be

the price duration curve. The remaining components are obtained by considering their conditional expectation given the level of system lambda (since their
correlation with system lambda is generally high, the error introduced is usually
In the 24-hour price trajectories of Section 2.8, spot prices were based on the
expected value of system lambda conditional upon the load level. Here spot
prices are based on the expected value of load conditional upon system lambda


In practice, customers respond to spot prices by changing their demand.

Customer response modeling is a complex subject and detailed discussions are
not given in this book. However, a brief discussion is provided here on two
issues. Please remember that in the near term, the major impact of spot pricing
will be on industrial and commercial customers.
Nature of Individual Customer Responses

e of
IS a

Customers have no desire for electric energy per se (although people can get a
charge out of it). Customers do desire the services provided by electric energy.
Customers respond to different prices in two basic ways:




t is

t is

o+;Modify Usage: ltprice in a given hour is very high, they may reduce usage
.:at that hour because the value of some of the services is less than the price.
. Alternatively, if price is very low, they may increase usage to receive services
~::l:h~y !,ormally wouldn't buy.
Reschedule Usage: If the price is high during some hours of the day and low
during other hours, customers may reschedule usage so their desires are met
but at dift!erent times. Such rescheduling usually imposes some customer costs
(or reduction in benefits received from the service) so the amount of rescheduling depends on the price difference.
Some examples of customer response are
o Space Conditioning: Reset thermostat. Pre-heat or cool at times of low price
to make use of a building's thermal storage capability. Use thermal bricks or
ice to store thermal energy purchased at times of low prices.
o Water Heating: Heat at times of low price.
o Water Pumping: Fill t.anks at times of low price.
o Reschedule Production: Reschedule hours of industrial production for
processes with high energy usage and low labor cost to times of low price
(assuming sufficient product storage is available).
o Fuel Switch: Use electric energy when its price is low and switch to oil or
natural gas when the price of electric energy is high.
o Cooking: Change the menu if the price is high or defer the meal until the price
drops (not necessarily a "cost-effective" response).


l. The energy marketplace


Resulting Demand

Demand (kWh)

Supply Curve: Marginal Cost (Price) Increases When Demand Increases

Demand Curve: Demand Decreases When Price Increases

Figure 2.10.1. Intersection of hourly supply and demand curve.

Further discussions related to customer response are given in Chapter 4.

Aggregate Customer Modeling and Impact

At the individual customer level, the response to price is a very nonlinear,

complex phenomenon which depends on the individual customer's needs and
capabilities. Fortunately, utilities do not need to be able to model (predict) the
response of individual customers to price. -The utility is only concerned with
aggregate customer response. Because of the diversity between customers, relatively simple response models can be used. Various structural forms for aggregate customer responses are given in Appendix E.
The classical way to solve for the impact of customer response (if cross-elasticities through time are ignored) is to find the intersection of the supply and
demand curves, where

Supply Curve: Plot of p vs. d showing how supply costs and hence prices
increase with demand.
Demand Curve: Plot of d vs. p showing how demand decreases as price

If both curves are plotted on the same axis, their intersection yields the resulting
values of both price p and demand d. This is illustrated in Figure 2.10.1. Both
curves change each hour in response to weather, time of day, outages, purchases
from other utilities, etc.

2. Behavior of hourly spot prices 51


Chapter 2 provided an understanding of the various components of the hourly

spot price and how they and the spot price itself behave. This lays the foundation for the next chapter which discusses the various types of transactions that
can be based on the hourly spot price.
The hourly spot price has many components. System lambda A(t) is the only
one many people think of when marginal costs are discussed. However, the
other components are also very important. The nonlinear structure of losses
amplifies their impact on marginal costs during times of high demand. The
quality of supply terms (generation and network) can completely dominate all
other components during times when the power system's capacity is being
approached. Revenue reconciliation is crucial if the energy marketplace is to
operate for regulated or government-owned utilities which do not want to show
excess profit or loss.





This chapter only presents overall ideas. Details and the underlying theory are
given in Chapters 6, 7, and 8 of Part II. For further reading on our approach,
see the notes and references at the end of these chapters and also the Annotated
Our basic approach defines an hourly spot price which depends on equipment
ot1~ges and load levels at that hour, as well as long-run revenue reconciliation.
T~'se hourly spot pt;ices, which are set after events are revealed, provide the
ba~is.for the actual energy marketplace transactions.
~any.authors have rejected the idea that prices can be set after events are
revealed. Fbr example, Crew and Kleindorfer [1980, p. 55] write:
Where the fin~ sets the price, it may do so either exallte or expost ... For the case of the
regulator setting the price expost, he or she would either have to allow a market-clearing
price or have some deliberate arrangement for setting the price above or below the price.
Were the regulator [to allow) the market clearing price, he would, in effect, be giving up
his right to regulate price.
Turvey and Anderson [1977, p. 298] are even more adamant in their rejection
of spot pricing:
'" for a wide class of random disturbances (but not for all), it is not possible to respond
to the resultant random excess or shortage of capacity by adjusting prices ... Failure of
a generating plant on Thursday cannot be followed by a higher price on Friday, and the
price in January cannot be raised when it becomes apparent that January is colder than
usual. Even though telecontrol makes the necessary metering technically possible, it
would be expensive, and ... there would be difficulties in informing consumers of the new
price. It would also be scarcely possible to estimate its market clearing level. Sudden and
random price fluctuations would in any case impose considerable costs and irritations on
consumers. Hence responsive pricing that always restrains demand to capacity is not
practicable, and some interruptions [are thus desirable].

52 I. The energy marketplace

Their rejection thus appears to be based on the belief that the transactions costs
of spot pricing would outweigh any possible benefits. At least for large
customers, we obviously do not agree. Of course, the technology for decentralized communication and control has improved drastically since these words
were written.
Other authors have accepted to some extent the concept of setting prices after
events have occurred. As discussed at the end of Chapter I, spot pricing of
public utility services was apparently first proposed by Vickrey [1971], under the
name "responsive pricing." His original article presented general discussions
using as examples mainly long-distance telephones and airlines. The,emphasis
was on curtailment premiums, rather than on marginal production cost changes
over time. Later manuscripts on electricity developed the ideas in more detail,
including some discussion of optimal investment criteria [Vickrey, 1979, p. 12],
metering requirements and designs, pricing of reactive energy, and short-run
marginal operating costs (system lambda). He proposed that utilities be free to
set prices however they want over time, subject only to limits on total profits.
In retrospect, many of his ideas were quite far sighted.
Vickrey's essential insight was that prices can be set after some variables are
observed, and optimal prices should reflect this. Since his original article,
different versions of this basic idea have been developed independently and
under different names, with varying levels of rigor. The other approaches


"State preference" approach to pricing electricity [Littlechild, 1972], a formal

stochastic model of both pricing and investment under static conditions. Both
operating costs and capacity constraints are modeled, but with homogeneous
fixed coefficient technology, i.e., only one kind of capital.
"Time varying congestion tolls" for a highway or communications network
[Agnew, 1973; 1977]. A formal deterministic optimal control model incorporating only capacity constraints and delays. No discussion of investment.
"Real time pricing" of electricity [Rand, 1979]. Informal; no specific
"Load adaptive pricing" of electricity [Luh et aI., 1982]. A game theoretic
model; nonlinear prices aJlowed. Quadratic production costs assumed, with
no capacity constraints and no investment. Their formulation allows for
games between one utility and one consumer which is not a pure price taker.
"Flexible pricing" of electricity [Kepner and Reinbergs, 1980]. Informal.
"Dynamic Pricing" [Peddie, 1986]

A series of articles on spot pricing and Interactive Load Control appeared in

Electric Review in 1981 and 1982 [Berrie, 1981/82].
The Credit and Load Management Systems (CALMS) is an important system
and hardware development in England [Peddie, 1982a, 1982b, 1983] that has
major implications for spot pricing. The key component, the Credit and Load
Management Unit (CALMU) is a microprocessor-based metering control and

2. Behavior of hourly spot prices



display system designed for residential use. A new version can accept a spot
price data stream.
Blackmon [1985] evaluates feasibility and potential benefits of establishing a
futures market for electricity related to the ideas of Section 3.5.



g of
. the

e to



\. At this level of the discussion, marginal means the same thing as incremental. In actual imple
mentation there is a difference between the two terms (see Chapter 4, Section 4,5),
2, Considering the massive uncertainty utility planners have to face, it is almost an accident if the
generation mix, etc, happen to be "optimum" at any given time,
3. A notational convention should be clear; y's and IJ'S are used for generation and networkrelated
quantities respectively,
4, The theory of Part II includes another component related to network maintenance, but it is
ignored here because it is difficult to model and usually not very important (see Section 7.4),
5, }.(I) as defined in this book includes the effect of purchases and sales with other utilities, Hence
it is not always equivalent to the "lambda" used in economic dispatch logics (see Appendix B),
6. R, has units or (MWh) - I and not the resistance units of ohms because of the use of "per unit
voltages," See Appendix D.
7. In general, the market clearing price is less than the cost of unserved energy because market
clearing implies that the customers select the least valuable load first. Unserved energy costs
assume nonselective blackout costs. which include very valuable loads,
8. The curves of Figure 2,8, I are the result of actual studies of a particular utility, Other utilities we
have studied have the same amount of variation but of course with different characteristics,


hourly spot price is the basis of the energy marketplace because it

provides the foundation for all transactions. Chapter I introduced three diffe~nt types of transactions. This chapter discusses these transactions in more
detail and relates them to present-day transactions rates.
The number of different types of transactions that can occur between a utility
and its custdmers is limited primarily by one's imagination. There is no unique
set of transactions that is best for all situations. The three basic types of
transactiolls to be discussed are
Price-Only: Customer can use all the electrical energy desired at a quoted energy
price ($jkWh).
Price-Quantity: Customer agrees to adapt usage to meet the utility's needs
under prespecified conditions, in return for financial reward.
Long-Term Contracts: Customers engage in long-term, fixed price, fixed
quantity contracts with the utility, other customers, or brokers.
Section 3.1 defines the criteria to be used in choosing transactions while
Section 3.2 discusses the various types of customer classes. Sections 3.3, 3.4 and
3.5 discuss the three basic types of transactions. Section 3.6 considers the
possibility of giving the customers a choice of transactions and/or having
transactions that are custom tailored to the customer's needs. Section 3.7
discusses the reason why demand charges are not part of the spot price based

56 . I:Tlleenergimarketplace

energy marketplace. Then Section 3.8 relates spot price based transactions to:
present-day transactions. The chapter concludes with Sections 3.8,3.9 and 3.10,
which discuss customer-owned generation, special rates for special customers,
and wheeling rates.

The ideal criterion to be considered when choosing a set of transactions is

Choose those types of transactions that yield the best possibJe cost-benefit tradeoffs for
the particular utility and its customers.

Of course such a criterion cannot be used until the costs of transactions and their
benefits are understood and estimated.
The costs of electricity sales fall into two categories: physical costs, and
transactions costs. Physical costs were dealt with in Chapter 2, and are for
energy itself (per kWh). Transactions costs are the costs of computing and
communicating the information and money that go with the transactions.
Transactions Costs

Electricity sales in an energy marketplace result in transactions costs for both

the utility and its customers. Examples of utility costs include

Costs of Computation: Prices; bills

Costs of Communication: Forecasts of future prices to customers; actual
prices to customers; actual prices to meters; meter readings to billing
computer; bills to customer
Costs of Metering. To put these costs in perspective, a recording meter
capable of handling hourly data costs on tne order of several hundred dollars.
Note that the meter does not have to receive prices each day, because the bill
needs to be computed only at the end of the billing cycle, and at the utility's
office. The cost of communicating an hourly spot price to a customer's
computer is less than $1,000, even if a new computer is needed. By comparison, a customer with an average usage of I MW on a system with an
average price of 5/kWh, will pay about $400,000 per year for energy. Thus
an hourly spot price has small transactions costs relative to usage by a large

Customers can face transactions costs associated with installing computation

and internal communication hardware to exploit changing prices. The transition
from the present system to an energy marketplace will also cause costs associated with training utility personnel and educating customers to deal with new
types of transactions.

3. Energy marketplace transactions



: for



costs can be quantified reasonably well by engineering analysis.

however, are much more complex. Marketplace transactions should
satisfy criteria such as
" Economic Efficiency: Customers should receive signals which motivate them
to behave in a socially desirable fashion; e.g. ideally their demand usage
pattern should be as if they were seeing the hourly spot price.
o Equity: Cross-subsidies between customers should be as small as possible
(unless something like a lifeline rate is deemed to be socially desirable).
o Freedom of Choice: Customers should have a high degree of freedom to
choose their own behavior patterns.
o Customer Acceptance and Understanding: Customers should be able to
understand the nature of the transactions and be convinced that they are fair.
o Utility Control, Operation, and Planning: The utility's job of running the
power system should be made easier.
o Customer Control, Operation, and Planning: The customers' reaction to
transactions should not have to be unwieldy or unnecessarily complex.
The benefits of different types of transactions are measured in terms of how well
th~~~atisfy the abov~ criteria. Unfortunately the criteria are often conflicting.
FQct~xample, a flat rate (no variation in time) satisfies the last criterion but not
th~&criterion of econ~mic efficiency, nor of utility operation.

Role of Forecasts

The last two criteria deal with issues concerned with the operation and control
of the e1ectriQ power system and the customer's own usage devices. An ability
to forecast the future with some degree of accuracy is critical for both functions.
In particular,




The electric utility must be able to forecast approximately how the aggregate
customer demand will respond to price.
The customer must have available a forecast of how prices will behave in the

The use of forecasts available does not imply that perfect foresight is required,
i.e. that the forecasts have to be error free. It does imply that information on
expected future behavior is essential for making control, operating and planning
decisions. This is also true today.

In Chapter 2, a spatially varying (due to the network effects) hourly spot price
was discussed. Thus, in theory, each customer could have a different hourly spot

58 I. The energy marketplace

price. In practice, some large industrial customers may have their own prices,
but most customers will just be divided into classes. Subsequent discussions
often refer to the "price seen by the kth customer." This price will be the same
for all customers in their class.
Three customer classes used in present-day rates are residential, commercial,
and industrial. In an energy marketplace, one logical basis for defining customer
classes is the voltage level at which the customer receives service. Customers who
receive service at 13 kV impose fewer capital costs and cause fewer network
losses than customers fed at a line voltage level such as 440 V. Thus the higher
voltage customers should see lower prices (all other things being equal).
A related basis for specifying customer classes could be geographical location,
if the network capital costs and losses vary widely throughout the utility's
service territory.
As will be seen, a customer's demand characteristics can also form the basis
for the definition of customer classes.

For a price-only sale of electric energy, the utility quotes, in advance, a fixed
price for energy ($/kWh) where the quote is valid for some specified period of
time. The customer can buy any amount of electric energy at the quoted price.
Most present-day transactions are of this type, and this will continue in virtually
any implementation of spot pricing.
Characteristics of Price-Only

There are many possible price-only transactions types. The four discussed most
often in this book are

One-Hour Update: An energy price valid Jor the next hour is quoted at the
beginning of the hour.
24-Hour Update: On the afternoon of the present day (say at 4 PM), the 24
prices are quoted that will hold each hour for a period starting early the next
morning and lasting until the same time the subsequent morning (say from
3AM to 3AM).
Billing Period Update Flat: A flat energy rate (i.e. constant in time) valid for
the subsequent billing period is quoted at the beginning of the billing period,
e.g. each month.
Billing Period Update TOU: A time-of-use (TOU) type energy rate (i.e. varies
at prespecified times of day and days of week) is quoted at the beginning of
the billing period.

Yearly update flat and/or TOU rates could be defined analogously to the above.
The preceding four types illustrate two key characteristics of price-only

Update Cycle Length: The length of time a quoted price or set of prices is
valid. The interval between new price announcements.

3. Energy marketplace transactions 59

Period Definition: The number of separate prices that are quoted within the
update cycle.
example, a 24-hour update has an update cycle length of 24 hours and
definition of24. A billing period TOU update has a period definition that
on the definition of peak, off-peak and shoulder times, I and an update
of one billing period.
Two other general characteristics of price-only transactions are



Advance Warning: Length of time before the start of an update cycle that the
prices are quoted.
Number of Price Levels: Prices may be constrained so that they can be set
only at prespecified levels.



A one-hour update might have an advance warning time of 5 or 10 minutes

while a 24-hour update quoted at 4 PM to start at 3 AM the next morning would
have an advance warning time of I hour. Price-only transactions can have any
number of price levels. If a finite number of price levels are not prespecified, the
transaction is said to have a continuous price level.


Spe!:ification of Price-Only: Effect of Update Cycle Length


the effect of the update cycle length, consider a 24-hour update

where prices change each hour even though they are prespecified many hours in
ad~nce."How should such prices be set?

Pk(t!r): Price:, for kth customer at hour t defined at hour r ($/kWh).

For example, a 24-hour update price quoted at 4 PM for the hour between 9 and
10 AM the next day would have r = 4 PM and t = 9 AM.
The mathematical theory developed in Chapter 9 yields




= E{Pk(t)I,) + Covariance Term

(3.3.1 )

The first term of (3.3.1) is

E{Pk(t)I,}: Conditional expectation of the hourly spot price Pk(t) given all available

information at hour


In less mathematical terminology, E{Pk(t)lr} is the best guess of the value of the
hourly spot price at t that can be made at the earlier time r. This is not
unexpected. Of course with a one hour update, t = rand Pk(t/r) = Pk(t).
The second term of (3.3.1) is called the covariance term. It is much less
intuitive, very messy in appearance, and in general less important. In fact, it can
often be dropped. However, it is part of the general theory so it will be discussed.

60 1. The energy marketplace

The explicit equation is

Covanance Term =

Cov (1)

Cov (t): Conditional covariance between Pk (t) and d; (t)

Cov (t)

E{[Pk(t) -

d: (t)

E{Pk(t)lr}][d;(t) -



(closely related to demand elasticity.)


dk (t): Demand of customer k during hour t.

E{d[(t)lr): Conditional expectation of diet) given all available information at hour 1".

In less mathematical terminology, the covariance term, Cov (t), is a measure of

the relationship between variations in PkCt) and the derivative d;Ct); it is positive
if a PkCt) that is larger than its expected value E{PkCt)lr} OCcurs when the
derivative dIU) is greater than its expected value E{d';Ctlr}.
The covariance term could become important when a long update cycle is
used. Given sufficient information on customer demand response, the covariance term provides another basis for defining customer classes which see
different prices. For example, customers with air conditioning who are on a
billing period update might be considered a separate class paying higher rates
if their covariance terms were determined to be positive.
NUMERICAL EXAMPLE. Assume that the only uncertainty existing at hour r is
the possibility of a large coal-fired power plant being forced out by the time hour
t arrives. Assume that if this occurs, a much more expensive gas turbine has to
be used. Assume Pk (t) = p(t) where


If coal plant is available during hour t

If coal plant is not available during hour [
and gas turbine has to be used

q: Probability coal plant is available during hour t given all information available at hour



(I -

q) 10

10 - 5q


It is reasonable to assume that the forced outage of a generation plant is

independent of any random behavior of the demand. Hence
Cov(t) =

3. Energy marketplace transactions 61

so (3.3.1) yields


= 5.25 jkWh .
Assume the only uncertainty at time r is the outside
temperature that will exist during hour t. Assume that if the temperature is high,
the overall demand will be high and the gas turbine will have to be used. If the
temperature is low, only the coal plant will be needed. Thus

If q


10 - 5q

p(tlr) =

0.95, then pet Ir)


5 /kWh if temperature is low

pet) =

10/kWh if temperature is high

e of
e is

q: Probability temperature is low during hour t given all weather information available
at hour r.

Then, as in the forced outage example,


10 - 5q

r IS

s to

H&i>wever, now assU}J1e the kth customer has an air conditioner which is run
w.!i'cn the temperature is high, so during hour t

). . II



kWh if air conditioning is off

IS on
2. kWh'f'
I air cond


2(1 - q) =

2 - q

In order to illustrate (3.3.2), assume that


Then after some algebra, (3.3.2) yields

Cov (t) =

5q(l - q)

So (3.3.1) becomes

For q = 0.5, p(tlr)

= 7.5 +


= 8.33/kWh,

so the covariance

J. The energy marketplace


term is not negligible. However, if the derivative dk(t) had been assumed to be
a constant, independent of dk(t), then from (3.3.2), Cov (t) = 0 and
PkCt/r) = 7.5/kWh. 2
Specification of Price-Only: Effect of Period Definition

Equation (3.3. J) is fine as long as the price changes each hour, i.e. has a very
detailed period definition. However, billing period flat or TOU prices do not
vary each hour. Thus (3.3.1) has to be modified for many types of period
The principles are easiest seen for a flat billing period update. This is not a
recommended type of transaction, since it loses most of the value of time varying
prices. However it is very common today, hence is an example of how to use spot
prices to calculate present day rates. Define

Constant price specified at beginning of month to hold for all hours of the month.

It is shown in Chapter 9 that under some very reasonable assumptions on the

derivative d'(t)

= 720


L: p(tlr) + Covariance Term



Number of hours of month =

r: Time



is computed

p(tlr): As given by (3.3.1)

The covariance term of (3.3.3) is similar to the covariance term of (3.3.1). It

could conceivably be important for long update cycles such as the one-month
billing period being considered here. We leave presentation of the detailed
equations to Section 9.1 of Chapter 9.
Relationship to Criteria of Section 3.1

In general, both transactions costs and the benefits associated with meeting the
other criteria tend to increase as transactions become closer to continuous
hourly spot prices, i.e.:

Update cycle length is reduced

Period definition is increased
Number of price levels is increased

Since it is desired to maximize benefits minus transactions costs, the choice of

which type of price-only transaction to offer depends on a cost-benefit trade off,
i.e. on the characteristics of the customers and the utility.
Some general patterns emerge. For example, the benefits of transactions go
up with the size of the customer and with the customer's ability to respond,

3. Energy marketplace transactions 63

o be

while transactions costs are almost independent of customer size. Therefore

large electricity users should have transaction types closer to hourly spot prices
than do residential customers.

ot a


). It


A power system can theoretically be controlled through the use of prices or the
use of quantity control. Theoretically, the desired customer behavior could be
obtained by the use of quantity control wherein the utility directly controls all
electric energy usage based on the hourly spot price and customer-provided
information on how they value the services provided by electric energy.
However, under the conditions of the energy marketplace where there are a huge
number of customers with large diversity in usage patterns and needs, the use
of prices is far superior to a quantity control with respect to both reducing transactions costs and increasing benefits. There is, however, a potential role for
certain price-quantity transactions in the energy marketplace.
Price-quantity transactions can sometimes be used to reduce transactions
costs. For example, a 24-hour update price-only transaction has lower transactions costs than a one-hour update but is vulnerable to plant outages or errors
made in forecasting the weather. (If the customers respond to the wrong price
signal if a plant outage occurs or a major weather forecasting mistake is made
at the time the 24-hour update prices were computed and quoted.) The use of
3:i~~-hour update spot price combined with an interruptable contract (a type of
flfice-quantity transaction) enables corrections to be made for especially bad
weaJher forecasts oj. plant outages, with transactions costs that might be lower
t~n those of a one>-hour update price. Whether the transactions costs would
really be much lower is an empirical question. We are skeptical.
A one-hQur update price can be combined with an interruptible contract to
provide the~utility with a way of modifying demand on time intervals of less than
one hour.' This enables the customers to carry the operating reserve necessary
for power system security control, i.e., to respond satisfactorily to sudden loss
of a major generation plant or tie line support. If the customers are to carry the
operating reserve, the utility has to have fast and predictable customer response.
Characteristics of Price-Quantity

The basic idea underlying many price-quantity transactions is for customers to

pledge (or contract) an amount of demand which the utility can control under
certain circumstances. In return, the customer receives a monetary incentive.
Some of the many possibilities for such transactions are as follows:


, go

Type of Pledge:
Fixed amount of energy reduction over some time interval
- Fixed amount of power level reduction
- All energy above a prespecified level
- Reduction of power to a prespecified level, or

64 I. The energy marketplace

- Control of a particular appliance or process

Override Capability: The customer has the right to buy-back the pledge, i.e.
to override any utility control signals, with a subsequent penalty of some sort
- A priori Payment (i.e. payment for the pledge)
- A posteri Payment (i.e. payment only when control is exercised)
- Combined a priori and a posteri Payments
Utility Control Logic: The utility may have information on the relative values
of the energy for different customers which is used to determine control

In addition to the above, characteristics which are analogous to price-only

transactions such as update cycle length, number of levels, advance warning
time, etc., apply as well.
A simple formula, (3.3.1) defined the basic principles for specifying price-only
transactions. Formulas also exist for the specification of price-quantity transactions. However, because of the many extra degrees of freedom present, the
formulas are not as simple. Therefore, we only illustrate the specification of
price-quantity transactions by the use of examples and leave the general
equations to Section 9.2 of Chapter 9.
NUMERICAL EXAMPLE. Consider the example used for price-only where
5/kWh if coal plant is available during hour [

. not aval.) a bl e an d a gas tur b'me h as to be use d

IO/kWh I'f coal plant IS

where q is the probability the coal plant will be available as estimated using all
the information available at time t. If q = 0.95 then as earlier (assuming the
covariance term is zero), (3.3.1) yields p(tlt) = 5.25/kWh. However, if the
possible event that the coal plant is forced out is ignored, p(tlt) = 5jkWh.
Hence, if a customer agrees a priori to reduce demand if the coal plant forced
outage event occurs, that customer can be offered the lower rate of 5 jkWh. If
the coal plant turns out to be available at time t, the customer has saved
0.25 /k Wh for all of the energy that was purchased.
NUMERICAL EXAMPLE. Consider a situation where it costs the utility 0.5 jkWh
to maintain the necessary operating reserves. In this case, the utility could offer
a O.5jkWh discount on its regular energy rate to customers who are willing to
help carry the operating reserve by rapidly reducing load when a sudden,
unexpected loss of a major generator occurs.
In the above two examples, the price-quantity transaction's value was specified by the utility's costs. An alternate procedure is for the utility to auction off
price-quantity contracts so that customers themselves determine the prices. For
the operating reserve example, the utility might start out by offering a 0.1 jkWh
discount; then if not enough customers sign up to meet the utility's needs, the
utility could raise the discount to 0.2 jkWh. This process would continue until

3. Energy marketplace transactions 65

, i.e.

either the customers have pledged enough load to meet the utility'S needs or
until the offered discount has reached O.5/kWh, in which case the utility carries
the operating reserve on its own generators.
A Possible Trap


1 of

The reader should be aware that it is possible to design a price-<luantity

transaction that "looks reasonable" but which does not have the desired effect.
One example will be given to illustrate the types of traps that await the implementor of price-<luantity transactions.
Consider a price-<luantity transaction that places a maximum kW limit on an
individual residence's usage during critical times. Assume, to illustrate the
concept, that the customer's normal peak demand is IOkW with an average
energy usage of 4 kWh. If the maximum kW constraint is 5 kW, a customer with
a sophisticated microprocessor-based controller might be able to meet the 5 kW
constraint by merely rescheduling appliance usage times so that all customer
needs are still met; i.e., the total energy usage over an hour remains at 4 kWh.
Because of the effect of customer diversity, the implementation of a 5 kW peak
constraint on many such customers could have little or none of the desired
effects of reducing the total demand peak that the utility generation has to meet.

Miny of today's utility-customer transactions appear to be price-quantity

~i!e in actuality tI1ey are price-only. Examples of such masquerades are now

gl!en .....

1. If
~ to


INTERRUPTIBLE WITH PENALTY. Consider an interruptible contract where the

utility reduces its regular flat rates while the customer agrees to drop load if
signaled by~the utility. Failure to drop load when signaled results in a financial
penalty to be paid by the customer. This is really a two-level price-only transaction, where the penalty corresponds to a high spot price at the time of
DIRECT CONTROL OF STORAGE. Consider a particular customer storage appliance (such as an oversized hot water heater) which the utility has bought (by
a lower rate) the right to control. The utility controls the appliance to be on
when the utility's costs are low and off when the costs are high (or capacity is
limited). This is similar to the customer controlling the appliance to be on when
the spot price is low and off when the spot price is high (see discussion in Section
4.3 on utility provided tactical control services).

Relationships to Criteria of Section 3.1

In some respects, the choice of what types of price-quantity transactions to offer

is analogous to the choice of price-only transactions. As the degree of sophistication increases, transactions costs and benefits both increase. Therefore the
choice involves a cost-benefit tradeoff. Three special factors must be considered


I. The energy marketplace

Table 3.5.1. Example of Cash Flow From a Long-Term Contract

If Customers Uses

If Q(t) = 9/kWh,
the customer

If Q(t) = II /kWh,
the customer


Pays $100
Pays $100
Pays $10

Pays $100
Pays $100 + 110
Receives $\0


Customer has a Contract for I MWh at IO/MWh.

when making such tradeoff studies. All three argue for the use of prices rather
than price quantity transactions.
First, although one of the motivations for the use of price-quantity was to
reduce transactions costs relative to price-only, it is possible to get so carried
away in designing fancy price-quantity transactions that the resulting transactions costs exceed those of price-only with a fast update cycle length.
Second, another motivation for the use of price-quantity was a desire to be
able to predict customer response more accurately. However, with certain types
of price-quantity transactions, such predictions become rather difficult. For
example, if the price-quantity transaction gives the utility the right to turn off
a customer's air conditioning system, it is necessary for the utility to develop a
model which gives the probability the air conditioning system is on as a function
of time of day, day of week, season of year, and of course, outside temperature.
In general, as the nature of the price-quantity transaction becomes more
sophisticated, it becomes more difficult to predict customer response.
Third, important criteria involve issues of customer understanding and acceptance. A price-only transaction is inherently easier for customers to understand
than most, if not all, price-quantity transactions. Also, a customer might
initially accept the idea of lower prices under price-quantity but then become
very disenchanted after the utility has exercised its control options a few times.
Price-quantity transactions can have a unique role in security control when
some or all of the operating reserve is carried by the customers.

One of the criteria listed in Section 3.1 for evaluating the properties of energy
marketplace transactions is the extent to which they facilitate customer operating and planning decisions. It might be very important for certain customers to
know what their future energy costs are going to be hours, days, months or years
in advance. This desire could be met by offering transactions with long update
cycle times, but this would result in an efficiency reduction. Fortunately there
is a type of long-term contract which enables customers to buy rights to future
energy at a fixed price (e.g. to buy an insurance policy) while still maintaining
the efficiency of short update cycle lengths ranging from hours to days. It is no
coincidence that the following discussion of long term contracts sounds like the
way agricultural forward contracts work. The underlying issues are very similar.

3. Energy marketplace transactions 67



s to

lns, be


Whether the risk hedging provided by electrical contracts is worth their transactions costs remains to be seen.
The basic approach is to offer fixed-price, fixed-quantity long-term contracts
for specific future time intervals. Consider a customer who has bought such a
contract. When the future time finally arrives, the customer can definitely have
the agreed amount of energy at the specified price, independent of the actual
spot price at that time. However, if the hourly spot price turns out to be above
the value in the customer's contract, the customer might choose to use less
energy than in the contract and effectively sell back his/her rights to the utility
at a profit. If the hourly spot price turns out to be much lower than that specified
in the contract, the customer has paid a penalty for having replaced uncertainty
with certainty.
NUMERICAL EXAMPLE. Assume that on January I, an industrial customer
purchases 1 mWh of energy to be delivered between 10 and II AM on July I for
10/kWh. Assume that when July 1st finally comes the price p(t) between 10 and
II AM is either 9 /kWh II /kWh. The actual cash flow depends on the
customer's actual usage on July 1st and is given in Table 3.5.1.
Futures Market

The use of fixed-price, fixed-quantity long-term contracts could evolve into a

futures market, with continual trading of future rights. In the preceding
ex4l/ffiple, the industl;ial customer who on January 1st bought I MWh to be
d~lwered on July 1st at 10/kWh might decide on April 1st that his or her needs
fer electric energy in July have changed, and therefore may try to sell the futures
rig)qts b~ck to the utility or to some other customer.
The utility does not have to be involved in the futures market. All of the
information, and transactions can be conducted through third-party energy
brokers. Thi!s would simplify the utility's problems and reduce the fear that a
utility heavily involved in the futures market would exploit its control over its
own generators and transmission network in order to increase its profits.

'at, to

A variation on the long-term fixed-price--quantity contract is the option. Here

the customer purchases the right to buy, at some specified future time, up to a
fixed amount of energy at a fixed price. When that time finally arrives, the
customer exercises the option if the actual price at the time is greater than the
price specified in the option. Otherwise, the customer purchases energy at the
going, lower price.
The option approach does not have the nice property of the basic fixed-pricefixed-quantity contract of maintaining economic efficiency, i.e., the customer
always making usage decisions based only on the existing hourly spot prices.
However, it still might be a usefull element of an energy marketplace.
NUMERICAL EXAMPLE. Assume that on January 1, the customer pays 1 /kWh
= lO$/MWh for the right to purchase one MWh of energy between 10 and


I. The energy marketplace

II AM on July I at a rate of 1O/kWh. If on July I, p(t) = 8/kWh, the

customer does not exercise the option. But if on July I p(t) = 12 /kWh, the
customer exercises the option and pays 10 /kWh, for all energy used up to one
MWh. The amount paid on January I, i.e., $10, is the same in all cases.

Many types of transactions are possible in a spot price based energy marketplace. A key question is "Which transaction will a particular customer see?"
Three approaches are

Mandatory: The utility/regulatory commission fixes the type of transaction

seen by all customers in a given class.
Optional: Customers have the right to choose from a menu of transactions
specified by the utility/commission.
Custom Tailored: The transactions seen by a particular customer are tailored
to the needs of the customer.

The optional and custom tailored approaches will be discussed.


If a customer has the option to choose among types of transactions, we recommend the following approach.
Specify a particular default price-only transaction for a particular class of
customers on the basis of trading off the transactions costs versus some measure
of the benefits evaluated for the class as a whole. Place all customers in this class
under the default transaction unless they choose to deal under a more sophisticated rate (such as one with a shorter update cycle time). More sophisticated
rates have higher transactions costs and customers who exercise this option have
to pay the additional costs over and above those of the default rate.
As an example of the above, consider a utility whose cost-benefit analysis
indicates industrial and commercial customers should see one- and 24-hour
update prices respectively while residential customers should, as a class, see
billing period update prices. Furthermore, assume that all residential customers
with air conditioners are put into a special class which pays higher monthly rates
because the covariance term of (3.3.1) is positive. Some of these air conditioner
customers might choose to see 24-hour update prices so they could reschedule
their air conditioning away from times of high prices and hence save money on
their electric bills (while also making things better for the utility). Such
customers would have to pay an additional charge to cover the extra communication and metering costs.
In a similar fashion, price-quantity and futures market transactions could be
made available as options for any customer willing to cover the additional
transactions cost.
Naturally the real world is complex enough that other, less simple, approach-

3. Energy marketplace transactions 69

, the
, the
) one



om,s of

omd be


es may be desirable. As an example, in areas with a weak distribution system (i.e.

one subject to contingencies which often prevent meeting the whole demand),
it may be desirable to mandate price-quantity (interruptible) transactions for
certain classes of customers. However, in general the basic approach of allowing
considered. 4 It does lead to some issues of self selection by customers, affecting
revenue reconciliation.
Custom Tailored

In some parts of the world, large industrial customer rates are individually
negotiated between the customer and the utility. Thus each customer sees rates
that are custom tailored to their needs, subject of course to utility cost constraints. Such custom tailoring is not part of most U.s.A. practices, but the
theory of spot pricing makes it a viable alternative with some desirable features.
Under the custom tailoring approach, a large industrial or commercial
customer specifies the type of transaction that best meets its needs. For example,
a customer might choose a 24-hour update with only three pricing periods
corresponding to the shift change times of the customer. As another example,
a customer might choose a price-quantity (interruptable) contract with warning
times, quantities, etc. that match the customer's own mode of operation. One
of the beauties of the overall spot pricing theory is that starting with basic hourly
spC,H prices, the utility can compute the value of a host of different types of
{f\nsactions in a consistent fashion. Thus even if each large industrial-comi:ftercial customer cijooses a different type of transaction, all of the transactions
v.onld be consistent and thus, hopefully, acceptable to the regulatory com~
With custom tailoring, it becomes more practical to include the various
'''covarianc; terms" that are part of the basic theory but that are difficult to
compute for a class of customers.
Custom tailoring implies a lot of overhead administrative expenses. Thus it
is difficult to envision its application to residential or small industrial-<::ommercia} customers.

Demand charges are presently an important source of revenue for many utilities.
However, demand charges have essentially no role in the energy marketplace. s
An explicit example of a demand charge is used to elaborate its disadvantages.
rc: Flat rate energy charge ($/kWh)
Demand charge based on one-hour peak demand during billing period



dk(t): Energy used by kth customer during hour t (kWh)

Billk : Bill kth customer sees at end of billing period ($)


I. The energy marketplace

If the billing period is one month (Le., 720 hours) then


Billk = re

L dk(t)

+ rAmax dk(t)]


max dk(t): Maximum value of dk(t), t = I, ... 720


The units of rd and dk(t) appear to be inconsistent, but a one-hour interval is

being used so all is well. Two main disadvantages of demand charges are
elaborated below using this example.
Demand Charges Do Not Send Good Price Signals

Pequil(t): Time-varying energy prices which are equivalent to (3.7.1) in the

sense that
Billk = '1:.;:1 PCquil(t)dk(t)
Assume re = 5jkWh and rd = 5$jkW. The resulting Pequil(t) was plotted in
Figure 1.1.4 of Chapter I. It bears very little resemblance to say Figure 1.1.1 or
1.1.2. Hence demand charges yield price signals that do not look like the actual
hourly spot price.
Since the price signal sent by a demand charge bears little relation to hourly
spot prices, the criterion of economic efficiency is violated. Customers are not
motivated to adjust their usage patterns to match the utility's capabilities; i.e.,
A demand charge motivates customers to take actions which reduce their bills while
leading to little or no reduction in the utility's costs.

For example, consider an industrial customer who installs an energy control

system that predicts when the customer is approaching a monthly one-hour
peak demand and then reschedules energy usage around that one hour. Such
actions reduce the utility's costs appreciably only if the utility has a needle peak
(i.e. about one hour long) that happens to occur at the time of the customers
peak.6 If a customer's peak happens to occur early in the month, the customer
will take no action the rest of the month. At least one utility, San Diego Gas &
Electric, has a "coincident demand charge" which looks at customer use during
the hour of the utility's peak. This is much better, and in fact is similar to a two
level spot price. However it still does not reflect the true costs to the utility of
customer usage.
Demand charges often have rachet clauses so that the peak demand during
a one-hour period of one month can fix the demand charge for the whole year.
If the demand charge is 5 $/kW, it could happen that running an extra appliance
using one kWh for one hour can cost 5 x 12 = $60. This can make a very
expensive pot of coffee.

3. Energy marketplace transactions 71

Demand Charges Are Not a Good Way to Recover Capital Costs


:val is
~s are

Demand charges are often used to recover capital costs, while energy charges are
used to recover fuel costs. This approach is not applicable in the energy marketplace, which employs rates based on marginal costs. Rates based on marginal
costs can lead to revenues which overrecover (as well as underrecover) relative
to embedded capital costs, rate of return, etc. Thus association of demand
charges with capital can result in negative demand charges - a concept which is
difficult to accept (at least by the authors of this book).
Attempts to Justify a Demand Charge

n the

~ed in

One could try to relate demand charges to the covariance term of (3.3.1) and
argue that customers with a higher monthly one-hour peak usage also have a
demand behavior that yields a positive covariance term. However, we find little
justification for such an argument.
A demand charge can be justified in an energy marketplace if the size of a
customer's peak usage has an impact on the utility's distribution costs, e.g. if the
utility has to install a special transformer or distribution line. Of course such
distribution costs could also be incorporated into an energy charge via the
revenue reconciliation term, or treated as a fixed charge.

e not
;; i.e.,

Jas &
a two
. very

Section 3.7, we <:Iiscussed the role (or lack thereof) of demand charges. We
discuss relatiQnships between the energy marketplace and other types of
"tle'Sent-day transa~tions.
Table 3.8.1. Characteristics of Generic Present-Day Transactions
Type of Pre~nt-Day Transactions

Assumed Characteristics'

Energy Usage
. Flat
Demand Charge
Block Rate

$/kWh with possible

variations for fuel
adjustment clause
See Section 3.7
$/kWh depends on total
kWh per month
Reduced $/kWh in return
for giving utility rights
to control usage
Reduced $/kWh in return for
accepting a demand
limiting device

lnterruptable Contacts

Demand Limiter
" Fixed Limit
,. Variable Limit
c Interlock
Direct Appliance Control
,; Air Conditioners
;. Water Heating
Life-Line Rates

Reduced $/kWh in return for

giving utility right to
control specific appliances
Reduced $/kWh for certain
customer classes

All values except fuel adjustment charges are assumed to be specified one year in advance.

72 I. The energy marketplace

We consider only the generic types of transactions listed in Table 3.8.1, which
gives their assumed names and characteristics. Table 3.8.1 leaves out many
interesting types of present-day transactions, but the scope is sufficient to
provide the reader with a good feel for the relationships between present-day
transactions and those of the energy marketplace. Note that unless the prices are
calculated using the formulas of this book, i.e. are based on hourly spot prices,
the rates are not consistent with our approach, even if the type of transaction
is similar.
Two particular existing innovative rates not included in Table 3.8.1 are rates
depending on the outside temperature or demand charges imposed at the time
of the utility's peak demand rather than the customer's peak. They can be
viewed as a step from the present-day transactions of Table 3.8.1 toward energy
marketplace transactions.
Energy Usage

If a fuel adjustment clause is not used, present-day TOU and flat rates are
price-only transactions with an update cycle interval of one year. A retrospective
(i.e. based on past fuel costs) fuel adjustment clause that is adjusted each billing
period is not the same as a billing period update cycle interval, because a
price-only transaction is based on predictions of future costs.
When a fuel adjustment clause is being used, we believe it is illogical to specify
one part of the rate a year in advance. The customers receive more accurate price
signals if the overall rate is on a billing period update cycle, and the fuel
adjustment clause is dropped as a separate item.
Average Spot Price Versus Flat Rate

The annual average spot price is given by.


Flat energy rate ($/kWh) that would recover the required revenue.

It follows that if

L p(l)d(t) yields the desired revenue






L d(t)


L p(t)d(t)

3. Energy marketplace transactions 73

1t to
~s are
n be

; are
se a


A little manipulation then yields

As long as p(t) and d(t) are positively correlated (i.e., pet) tends to increase as d(t)
then ra at > Pavc'

customers with flat usage patterns (such as some industrial customers) will
spot prices since their bills will go down relative to the flat rate (as defined
Block rates are nonlinear pricing schemes wherein the energy price per kWh
on the customer's total kWh usage.
The revenue reconciliation discussions of Section 8.7 show how nonlinear
price structures can arise from the theory of spot pricing, but the motivation for
present-day block rates does not appear to be based on the same logic.
Present-day block rates are sometimes justified by arguing that large
lOT/' should pay lower rates because they are fed at high voltage and
not pay the capital costs and losses of the lower voltage distribution
However, the use of customer classes based on voltage level of service
is a more desirable approach.
The covariance term of (3.3. J) can also be used as a way to try to define a
rate. For example, if the class of customers with high monthly energy
also tend to h~ve a negative covariance term, then a declining block rate
Unfortunately, monthly energy usage by itself is not a very good
of customer hourly usage patterns.

.. """1-'''11'';)

Interruptable Contracts

Most presedt-day interruptable contracts are special cases of an energy marketplace price-quantity transaction or are disguised price-only transactions.
However, most present-day interruptable contracts are prespecified a year in
advance, while energy marketplace transactions allow for much more frequent
Thinking in terms of an energy marketplace opens up the range of possible
interruptable contracts. With the energy marketplace framework, it becomes
possible to offer customers a flexible set of transactions. Thus customers can
match their needs and ability to respond. This flexibility can be extremely
important for industrial customers.

Demand limiters can have constant settings (e.g. fuses) or the limits can be
imposed under utility control in real time (e.g. Demand Subscription Service).
They are related in concept to demand charges, and many of the critical
discussions given on demand charges also apply here.
Real-time demand limits based on the utility's needs can be effective.

74 I. The energy marketplace

However, like demand charges, they have the disadvantage that as customers
become more sophisticated, they can install control hardware or modify their
behavior to reduce the plan's overall effectiveness.
It is extremely difficult to see how applicance interlocks lead to an appreciable
reduction in the utility's costs. At best, there is an extremely weak link.
The basic felicity underlying many demand-limiting schemes is the belief that
reduction in an individual's peak demand results in a reduction in the peak
demand seen by the utility. Diversity and the law of large numbers destroy this
belief. A general rule, which is true in almost but not all cases, is

Demand limiters reduce the utility's demand only if they reduce the energy
used by individual customers over a period of several hours.

Direct Appliance Control

A customer selling the utility the right to directly control individual appliances
is usually a special type of energy marketplace price-quantity transaction or a
price-only transaction in disguise. The key question is whether or not it is a
desirable type of transaction. The answer depends on the type of appliance. Air
conditioner cycling and water heater controls are discussed to illustrate the
Under air conditioner cycling, the utility can turn the air conditioner off if
deemed necessary. This has three major shortcomings. First, the utility is put in
the position of appearing to play big brother. The utility has no way of knowing
how important the air conditioning is to anyone customer at anyone time. Thus
control might be exercised at a particularly bad time (such as when the customer
is holding a very important dinner party); the result is a very unhappy customer.
Second, the utility has to build a model for the probability that the air conditioner is on as a function of time of day, week and season and of outside
temperature and humidity in order to predict what type of response to expect.
Third, most present-day air conditioning cycling contracts have a clause which
limits the number of times the utility can exercise control. This places a very
heavy burden on the utility system operator, who has to gamble on whether or
not to exercise control at any given time. Air conditioner cycling does not have
a role in an energy marketplace.
Direct utility control of water heaters is, however, quite different in character.
The large amount of thermal storage in most hot water tanks enables the design
of systems where the utility can exercise direct control (both on and oft), which
helps the utility reduce costs without causing much or even any customer
Life-Line Rates

Life-line (or base-line) rates refer to the use of low, non-cost-based prices for
particular classes of low income or otherwise needy customers. The class is
usually distinguished by a low monthly use of electric energy.

3. Energy marketplace transactions 75



::>r a

is a
ff if
r or

; is

The use of such rates is socially motivated by the belief that today, enough
energy to meet certain basic needs should be made available to everyone,
loe'perloent of ability to pay. This could be done by using general tax reven ues.
rates are an alternate procedure wherein needy customers pay artificilow rates and the other customers make up the difference - i.e., a hidden tax.
Life-line rates can definitely be incorporated into an energy marketplace if so
Note that even customers who are on life-line rates should see a billing
update cycle. Nobody should be denied the right to achieve the benefits
matching their usage patterns to actual costs.

revenue reconciliation is ignored, hourly spot prices for the utility selling to
customer are the same as when the utility is buying electric energy generated
a customer (see Section 2.7). Revenue reconciliation can destroy the
lImmptr\l but does not change the basic price behavior. Thus most of the
discussions of this chapter also apply to the transformation of a basic
hourly spot price into a set of transactions which provide good
benefit tradeoffs relative to the criteria of Section 3.1.
price-only and long-term contract transactions discussions apply almost
for word to the buy-back case. The details of some of the price-quantity
"."'!l\t~""".v .. S may chanJ?;e, but the principles do not. For example, during hours
when a private cogenerator is generating electric energy below
because of a lew internal demand for steam, the cogenerator might sell
}~tility<the extra capacity for use in emergency situations.
fUV-t.J<t ......

Capacity Credits

the United~States, the PURPA legislation stated that buy-back rates should
based on avoided costs without clearly defining what avoided costs are.
spot prices provide such a definition.
In present-day transactions, avoided costs are often subdivided into an
energy cost and a capacity credit which is used to reflect impacts on the
ity's capital costs. In the energy marketplace, capacity credits are not used.
pacity credits can be viewed as being incorporated into either the hourly spot
gen~ration quality of supply component, or the revenue reconciliation
01111pOlnem. Use of capacity credits for customer generation is analogous to the
of demand charges for customer usage, and is not appropriate.
These discussions on buy-back rates have implicitly assumed that each incustomer-owned generator is small enough relative to the utility's total
tion that diversity arguments can be used. Thus they can be treated just
individual customer loads where diversity is critical; i.e., no individual load
customer generator) behavior has a noticable effect on the overall spot price.
individual customer-owned generator is too large, the conclusion no longer
(The definition of "too large" is very situation-specific.)

76 I. The energy marketplace


The impact of revenue reconciliation can lead to situations where it might be

desired to give certain customers "special treatment;" e.g., to charge them rates
that at least appear to be inconsistent with those seen by other customers. This
is different from the custom tailored rate discussed in Section 3.6 because the
rates of Section 3.6 are all based on a hourly spot price in a self-consistent
To illustrate the special treatment case, consider a utility with excess generation capacity, which would under recover its operating and capital costs if a spot
price without revenue reconciliation were used. In such a case it can be argued

A potential new customer who is thinking of entering the service territory, or

an old customer who is on the verge of leaving the service territory (or closing
up), could be charged a lower overall rate to entice them to enter or not leave.

The basis of the argument is that if they enter or do not leave, the capital costs
of the generators are spread over a larger base and hence regular customers also
Similar types of arguments can be applied to the case of a utility with
insufficient generating capacity, albeit the sign of rate differential changes.
Similar types of arguments are also applied to customer-owned generation.
Under relatively reasonable assumptions (as discussed in Section 9.4), spot
pricing theory provides a rational basis for establishing such special rates for
special customers. The theory of Section 9.4 combines revenue reconciliation
and long-term customer demand elasticities to show that such special rates are
self-consistent with the spot price based rates being discussed in this chapter.
The concept of charging special customers special rates is controversial, and
leads to lots of complications in actual implementation.

Wheeling is a term used to describe the situation wherein a given utility's

transmission distribution network is used to transfer electric energy between two
other parties. This is a common practice when the two other parties are electric
utilities. It is also being advocated by some for cases where the two parties are
private users and generators or a private party and a different utility.
Spot pricing provides a viable vehicle for defining wheeling rates (i.e. the
prices the wheeling utility charges for use of its network). For example, if the
wheeling is from one point (bus) to another point (bus), a wheeling rate can be
defined in terms of the difference in the spot prices at the two points. Sections
9.5 and 9.6 contain more discussions on wheeling. See also Section 5.4 on

3. Energy marketplace transactions 77



ry, or
; also

s for
s are



s are
. the
f the
to be
4 on

chapter has outlined the very rich menu of transactions that are possible
an hourly spot price based energy marketplace. Different transactions can
because they all have the same common basis, the hourly spot price. The
of which transactions to offer particular classes of customers is based on
between the cost of the transactions and the benefits of having transthat are close to the hourly spot prices.
One key conclusion is that there is symmetry between customer owned loads
customer owned generation. There is no reason to discriminate against one
the other by charging special rates. Also, there are good reasons to let
see hourly spot prices so that they can adjust their generation
to better help the utility.
An obvious question is, what should be the relative roles of price-only,
and fixed-price-fixed-quantity long-term contract transactions?
presently recommend major reliance on price-only transactions with pricetransactions playing only a support role to deal with particular needs
operating reserves. This recommendation applies today particularly to
and large commercial customers, and in the future to much of the
class as well. One reason for this recommendation is the complexity
by price-quantity transactions into the lives of both the customers
utility operators.
;1'-"'iiYP(1-nr,'l'p-fixed-quantity contracts playa different type of role than priceand price-quanttty, if they are applied over a long term such as months or
:cThey provided customers with risk hedging. Second in our opinion most
uantity transactions should be converted to price-only transactions
give customers the ability to change their mind, if they are willing to pay
penalte'. Final comparisons must await more field experience.
- of energy marketplace transactions with present-day transshows both similarities and fundamental differences. The price-only and
energy marketplace transactions can be viewed as the logical
VolilltiCln of present-day time-of-use (TOU) rates and interruptable contracts.
fixed-price-fixed-quantity long-term contracts/futures market seems to
no present-day counterpart. Present-day demand charges, capacity credits,
many demand limiting schemes have essentially no role in an hourly spot
based energy marketplace.
if only the very largest customers are actually on hourly spot prices, spot
principles can be used to set rates for other customers .

chapter discussed in general terms the mapping of hourly spot prices into
marketplace transactions. Details and theory are presented in Chapter 9
II. Readers who want to pursue our ideas further should start with
9. There appears to be little literature on systematic rules for how often


I. The energy marketplace

to change prices over time, despite the vast literature on how to set the level of
time varying prices. Bohn [1982, Section 2.5] presents a simple model and
reviews the existing economics literature. A key insight is the difference between
prespecified changes (which can change once per period) and new price calculations, which occur once each update cycle. Other key issues are transactions
costs and the impact of self-selection, which are analyzed for TOU rates by
Acton and Mitchell [1980]. Various papers discussed in the Annotated Bibliography develop the framework used here and in Chapter 9.
A time-of-use (TOU) rate is a special type of price only transaction as
discussed in Section 3.3. Hence, this is as good a place as any to review this TOU
literature. TOU rates can have many periods, but are updated only annually or
irregularly. They are an important intermediate rate between flat prices and spot
The desirability of TOU rates has been the topic of major research by the
Electric Power Research Institute [1979]. For a good summary of this effort and
discussions of associated problems, see MaIko and Faruqui [1980] and Faruqui
and Maiko [198Ia]. For a good review of U.S. Department of Energy sponsored
residential TOU experiments, see Faruqui and MaIko [I98Jb].
The idea of time-differentiated prices goes back at least to Boiteux [1949] (see
also Vickrey [1955] and Steiner [1957]). Until Brown and Johnson (1969] the
models were purely static and deterministic. During the 1970s various authors
presented prescriptions for TOU pricing in static models with demand uncertainty.
The "standard" TOU pricing models are surveyed in Gellerson and Grosskopf (1980] and Crew and Kleindorfer (1979]. They include Wenders (1976],
Crew and Kleindorfer [1976, 1979, Ch. 4 and 5], Turvey and Anderson [1977,
Ch. 14], and various predecessors. These models include multiple types of
generators and stochastic demand. Some of the limitations of these models are
as follows:

Generating unit availability is practically ignored or modeled by simply

derating unit sizes at all times. This fails to penalize large units properly, and
it gives wrong estimates of the probability that rationing will be needed. It
also gives no guidance for how to evaluate new technologies such as solar and
cogeneration, whose "availabilities" are correlated with demands.
There is no analysis of how or when prices should be recalculated. These
models rule out frequent recalculations (by spot pricing) by assumption. By
assuming infinitely repetitive demand cycles and stable factor prices, they
show no need for annual or less frequent recalculations.
These models treat all investment as occurring at once. Investment is really a
sequential process. True utilities never have the static optimal capital stock of
these models, because conditions change more rapidly than capital stock
turns over. Therefore pricing equations which assume optimal capital stock,
i.e. assume that short-run and long-run marginal costs are equal, have limited

3. Energy marketplace transactions 79

vel of
I and
es by




I spot

'y the
t and

,] (see
I] the

~s of
s are

. and
:d. It
: and





lily a
ck of

practical value. In fact long-run marginal costs can only be calculated con:ditional on a particular scenario or probability distribution of demand and
factor prices. This problem is addressed by Ellis [1981].
The models assume that demands and generating costs are independent from
one hour to another. This is very convenient, since it allows the use of a single
load duration curve or price duration curve. Nonetheless, the availability of
storage [Nguyen, 1976] startup/shutdown costs, etc., and demand rescheduling can have a major impact on optimal investment policies.
The models ignore transmission, which is equivalent to assuming an infinitely
strong transmission system. These models give no insight into how to price
over space.
The models do not use the device of state contingent prices. Therefore, the
investment conditions derived in the models are hard to interpret, although
they are correct (given the limiting assumptions above). For example, Crew
and Kleindorfer [1979, p.77] interpret their results only for the case of
interchanging units which are adjacent in the loading order. Littlechild [1972]
showed the way out of this problem, but his point was apparently missed by
subsequent authors.
Several authors present deterministic, explicitly dynamic models which can be
as deterministic versions of price-only transactions combined with
decisions. Crew and Kleindorfer [1979, Ch. 7] give a continuous time
control model with one type of capital. They get the result that
level of capacity, price is to be set to maximize instantaneous [short run)
returns subject to the given capacity restriction [po 113]. That is, price should
SRMC. 9f course, at optimum capital stock is adjusted so as to equate SRMC and
C .... In the event of ... a fall in demand, [optimal] price is less than LRMC, then
ty would be allowed to decline until equality between price and LRMC were


are thinking here on a time scale of years, not hours; they reject continuous
of prices to reflect the actual level of demand. Nonetheless, their
can be interpreted in terms of hourly price adjustments.
Turvey and Anderson [1978, Ch.17] have a discrete time dynamic model
leads to discontinuous prices, as capital investment is made in lumps.
....""'pvpr, they reject this approach: "It is apparent that, for one reason or
such fluctuations are unacceptable." They also acknowledge that
Imles1tIDIem decisions must be made before price decisions, and with more
about future demands, but they do not incorporate this into their
Ellis [1981] explicitly models sequential investment and pricing decisions. He
""U\o,IU'.1"" that" ... welfare optimal pricing rules differ according to whether
prices must be set either before or after investment decisions are made" [p.2].


I. The energy marketplace

He uses dynamic programming to look at how the character of optimal sequential investment depends on capital stock irreversibility and the sequential revelation of information about future demands.
Price-quantity transactions as discussed in Section 3.4 can be viewed as
procedures which allows customers to choose their own reliability levels.
Marchand [1974] has a model in which customers select and pay for different
reliability. The utility allocates shortages accordingly, when curtailment is
necessary. His approach differs from (and is, except for transactions costs,
inferior to) spot pricing because customers must contract in advance, and
therefore have no real time control over their level of service. Also, customers
not curtailed by the utility have no incentive to adjust demands.
Marchand's proposal has been greatly extended under the name "priority
service". [Oren et ai, 1986; Chao and Wilson, 1987] Although discussed as an
alternative to spot pricing, priority service in fact addresses a very different set
of issues. Customers select priority levels; higher priority customers pay more,
but are cut off later in the event of supply interruptions. Such a scheme is
obviously not designed to deal with cyclic and hourly changes in the system
lambda, but instead to deal with occasional situations where the generation
quality of supply component becomes positive and there is not enough generation capacity available to meet the demand. The existing literature on priority
service also seems oriented toward residential customers. The implicit but
unquantified assumption is that metering and other transactions costs will be
high relative to energy used. We do not attempt a detailed analysis of priority
service in this book. Since it is a form of price-quantity transactions it has the
virtues and defects of that class.
I. Many present-day rates have the appearance of being updated yearly but are not because of the
presence of fuel adjustment clauses, which changes upredictably every quarter or month, and in
many cases are retroactive.
2. Unfortunately, a good model for the derivative of demand with respect to price is not available.
This problem will raise its head many times throughout this book. Conventional rates avoid this
problem by ignoring the covariance issue, causing cross-subsidies and inefficiencies.
3. The use of onc hour as the basic time unit in this book is not essential. For example, it is possible
to use a five-minute update price if desired. However the update cycle length cannot be reduced
too much as then the power system dynamics start to become important and more advanced
pricing theory is required. Berger [1983] discusses some of the issues which arise when system
dynamics come into play. A special case arises when customers receive incentives to install devices
which temporarily control certain of their loads as a function of system frequency. This concept,
which makes use of a FAPER (Frequency Adaptive Power Energy Rescheduler), is discussed in
Appendix F.
4. An implicit underlying assumption is that all transactions are based on hourly spot prices; i.e. are
consistent. A choice between an hourly spot price and a flat rate computed on a different basis
should not be allowed.
5. Recent historical research by Hausman and Neufeld [1984J points out that a strong advocacy of
marginal cost based time-dependent electricity rates existed in the United States as early as the
1882-1915 period when a wide-ranging debate over electricity rates took place. They suggest that
the final adoption of "demand charges" instead of time-of-use rates can be found in the
"economic conditions faced by electric utilities in their early days and in the nature of rate
regulation" [Hausman and Neufeld, 1984, p. 125J.
6. Demand charges are often computed on a 15- or 30-minute basis. This increases the problem.

quenevelaed as
~nt is
, and
as an
n.t set
ne is
t but
ill be
s the


first two steps leading to an energy marketplace are to evaluate the behaY'iRUJ: of the hourly spot prices (as discussed in Chapter 2) and to determine the
ty~es of,transactions'to be offered (as discussed in Chapter 3). The third step is
the actual implementation, which is the subject of the present chapter.
The discussions are divided into four general areas:

of the

md in
ld this


Energy Marketplace (Communications, Metering-Billing)

Regulatory Commission



The discussion is also divided into operating and planning issues:

sed in

Operation: Decisions on how to operate existing system.

Planning: Decisions on new capital investment, or long-term policies.

e. are
ICY of
IS the

t that

n the

The exact pattern is as follows. Sections 4.1 and 4.2 discuss operation and
planning for the energy marketplace, while Sections 4.3 and 4.4 discuss operation and planning issues relative to the customer. Section 4.5 deviates from the
pattern and discusses some of the practical issues associated with calculating an
hourly spot price. Sections 4.6 and 4.7 then discuss operation and planning for

r rate


J. The energy marketplace

Table 4.1.1. Examples of Price-OnlyjPrice-Quantity with Same Hardware





Price-Quantity: Flat Rate Plus

Flat Rate
Peak-Offpeak Rates

Limit on monthly energy use

Limit on energy usage during
peak hours
Limit on energy usage during
peak hours only on critical days













Pea k-Offpea k Ra tes;

Peak rate charged
only on critical days
Peak-Ofrpeak Rates;
Peak Levels specified
each day
Peak-Offpeak Rates;
Rate times specified
each day
24-Hour Update
Peak-Offpeak Rates;
Peak times respecillcd
each hour
One-Hour Updates

Limit on energy usage during

peak and off-peak hours
varies each day
Limit on energy usage during
peak times specified
each day
Limit on energy usage varies
each hour; respecified each day
Limit on energy usage during peak
hours; peak times respecified
each hour
Limit on energy usage varies
each hour; respecified each hour

the utility. The chapter concludes with section 4.8, which discusses operation
and planning issues faced by a regulatory commission.
Note that this chapter is, deliberately, quite comprehensive. For example, it
talks about some types of transactions which we personally doubt will be used
by many utilities. In contrast, most utilities will start with a simple, price-only
hourly spot price or a 24 hour update price. This will be tried with a few large
customers who volunteer for the rate. The experience of implementing these
transactions will give indications about which way to go with broader application. Although in principle one can argue for starting with a clean slate,
regulatory tradition and the inherent uncertainties of trying something new
suggest that this is unlikely, at least in the U.S.

The operation of an energy marketplace is characterized by the types of transactions offered and by the communication and metering-billing systems used to
implement them.
Types of Transactions

Chapter 3 introduced three basic types of energy marketplace transactions:


Price-Only: An energy rate that is quoted in advance. Customers may use all
that they desire at that rate.
Price-Quantity: Customers agree to adjust their usage in some pattern on a
signal from the utility.

4. Implementation

Long-Term Contracts/Futures Market: A fixed price for a fixed quantity to

be delivered in the future.




Three of the factors that characterize these transactions are

Cycle Length: Length of time between the respecifications of new prices or
Period Definition: Number of individual periods within the cycle length.
o Number of Levels: Number of different price or quantity levels.
Table 4.1.1 summarizes some of the many possible price-only and price-quantransactions, assuming the billing period is one month. The price-only and
g price-quantity entries in the table are duals in the sense that each
be implemented with the same type of hardware for communication,
and billing. I To illustrate this duality, consider the first two rows of
4.1.1. The first row requires a single-register meter, while the second one
be implemented with a two-register meter with clock. It must be emphasized




19 peak



pIe, it
. used
: new

.nsaced to


Ise all

on a

duality in Table 4.1.1 does not imply similar customer response, acceptance,

One aspect of price-quantity transactions which is not found in their pricearts is the issue of what happens if the customer does not fulfill the
quantity change when so requested by the utility. Three possible


Failure to comply causes dollar penalty

Failure to comply causes loss of right to participate in future price-quantity
Failure to comply causes shutoff of all service
dollar penalty actually converts the price-quantity contract into a priceonly transaction. The loss of right to future participation in price-quantity
tracts requires no extra metering or billing hardware. If the consequence of
failure to comply is to shut off all service, extra high power switches are required
the hardware duality no longer exists in a strict sense.
The billing period update, 24-hour update, and one-hour update price-only
discussed in Chapter 3 are the first, sixth and last rows of the
price-only column of Table 4.1.1.
Because of the hardware duality between many price-quantity and price-only
implementation discussions concentrate on price-only transac;tions. However, there are some price-quantity transactions which do not have
price-only counterparts, such as
Direct Load Control: Individual appliances are switched on and off under
utility control.


I. The energy marketplace

Table 4.1.2. Examples of Possible Communication Media

Existing Systems
o Newspaper
o Commercial Radio/TV Programs
o Telephone Calls
- Customer-instigated
- Utility-instigated
- Digital data, synthesized voice, or person to person

Special Systems

Utility Radio Broadcasting

- Modulation of carrier of commercial station
- Own transmitter
- Satellite
Telephone Circuits
- Leased line
- Carrier which does not interfere with normal telephone usage
Power Line
- Ripple control
- Low-frequency carrier (less than 6000 Hz) to go through transformers
- High-frequency carrier
- Other modulation scheme
Cable TV

Demand Subscription Service: A time-varying demand limit is imposed by the


Implementation issues associated with these price-quantity transactions are

discussed separately. Price-quantity contracts also introduce the possibility of
bidding among the customers.
Table 4.1.1 covers only two or an infinite number of price levels. In practice,
use of three to five levels may often be more desirable than two levels but the
ideas extend very naturally.
The subsequent discussions use one hour as the shortest time interval of
concern. 2
Communication Media Available

Energy marketplace operation involves two types of communication:


Communication of Prices to Customers: Both the latest price or quantities

signal, and forecasts of their expected future behavior.
Communication Required for Billing and Metering.

Table 4.1.2 lists some of the different types of communication media that could
be used. The Existing Systems of Table 4.1.2 require no special hardware while
the Special Systems do. Communication can be either one-way or two-way.

4. Implementation


4.1.3. Use of Existing Systems for Price-Only Communication with Customers

Cycle Length of Price Transaction





communication is often not needed for energy marketplace functions

is usually desired from the point of view of reliability.

ommu,nic:ati(m to Customers

price-only transactions and their price-quantity duals, it is desired to

icate both a price level and the times at which it applies.
discussions that follow, the time intervals are assumed to be fixed and the
is variable. An alternate approach is to consider the length of the time
variable for fixed levels.
cycle length is one of the key characteristics which determines the type
communication media to be used. Table 4.1.3 indicates types of Existing
that can be used for communicating both forecasted and actual prices
transa~tions. Any of the Special systems of Table 4.1.2 could also
customer undef:a price~only transaction has the option of ignoring the price
The energy marketplace can be set up so it is each customer's
to learn what the prices are. This is not true in the case of many
ty contracts where it is essential that the utility send a signal to the
custorner al the times when the utility exercises its quantity control options.
the only Existing system that could be used for price-quantity transacis utility-initiated telephone calls.
Direct load control and Demand Subscription Service require Special Systems
reach the switch on the appliance being controlled, or the meter and switch
. total customer service. In short, the more complex the transaction
type, the fewer inexpensive communication options are available.

For price-only transactions, the utility has to monitor the customer's usage and
then compute a monthly bill (assuming the billing period is one month) given

Monthly Bill of kth customer =

Pk(t) dk(t)

k =

1 ... 720

{ = I


Implementation of this metering-billing function involves a tradeoff between

storage and communication. For example, two extreme ways a price-only
one-hour update can be implemented are


J. The energy marketplace

Store the energy usage for each hour of the month (720 storage registers)
which are then retrieved by a meter reader and returned to a central utility
computer to calculate the monthly bill.
a Communicate the hourly price to the meter, which has one storage register
that accumulates the bill (product of price times demand); retrieved by the
meter reader once a month.
There are many possible variations between these extremes. For example,
transactions Types 3, 5 and 7 of Table 4.1.1 can be implemented by a meter
which has two storage registers and simply receives a binary signal by some
special communication system. If a two-way electronic communication system
exists, the need for a meter reader can be eliminated.

Bidding for Price-Quantity Transactions

As discussed in Chapter 3, one possible implementation procedure associated

with price-quantity is for the customers to bid for the amount of interruptible
service they desire and the price they are willing to pay for it. This introduces
an additional communication burden which could be done via telephone
(person to person or computer to computer) or via special two-way communication systems.
A special market mechanism is required for implementation. It can take
various forms.
Integration with Other Functions

Discussions thus far have assumed that the energy marketplace communications
and metering-billing functions are to be implemented as a stand-alone system.
However, in practice they could be integrated with other types of functions.
Many utilities are considering distribution automation and control whereby
electronic communication systems are established on the distribution system for
monitoring flows, controlling switches and setting relays, etc. If such a communication system reaches the customer's meter, it could conceivably also be
used for the energy marketplace transactions.
There is a major revolution brewing in establishing electronic communication
between an individual residence and the outside world. The functions being
considered include


Pay Cable TV: Customers pay for individual programs

Home Banking
Home Security Systems: Monitor the home for fire and burglary and communicate to the appropriate outside agencies
Home Shopping
Computer Networking
Utility Meter Reading (Electric, Gas, Water)

Such applications can have a major effect on how energy marketplace transac-

4. Implementation 87


are implemented in the residential sector. In many cases, the same com'on medium and terminal equipment can be shared.


Term Contracts/Futures Market


{ the

''''\'"'~t''r ...... contracts place no extra burden on the real-time communicationmetering and billing systems (unless, of course, the length of the contract is only
a few hours or days). As with the bidding for price-quantity contracts, there
many possible ways the market mechanisms can be developed. For example,
I~ ~ 'f6~'''' contracts could be offered that


As discussed in Chapter 3, a futures market in long-term contracts can operate

independent of the utility, and there are many advantages to such an arrangePossible arrangements include



om) be

Are based on prediction of future costs

Are sold to the highest bidder

A broker system which enables customers' investors, and other parties to

establish the market value of future quantities of electric energy.
An insurance system wherein regional or national companies offer customers
fixed-quantity insurance policies.
S,ssiblya formal market, in one of the established commodity market places.


planning questions associated with energy marketplace implementation are


Which Tyr1es of Transactions to Offer

How to Implement Them




a set of energy marketplace transactions involves consideration of

both the utility's and the customers' needs.
Figure 4.2.1 contains an oversimplified daily load curve consisting of a peak
.and valley and summarizes the conventional wisdom on the relationships
between the utility's operating problems and the types of peak and valley
changes that would be beneficial to the utility.
The conventional wisdom underlying Figure 4.2.1 actua\1y does not apply in
a strict sense to a spot price based energy marketplace for several reasons. First,
if there is a generation shortage, the goal is not rea\1y to reduce peak demands
rather to keep demand from exceeding available generation; such problems
need not occur at times of peak demands (because of plant outages). Second and
even more basic, the utility wants customers to shift demand from times of high


I. The energy marketplace






If Utility's Situation Is
A: Generation Capacity Shortage
B: Generation Fuel Shortage
C: Fuel Cost Varies Widely with Demand Level
D: Excess Base Load Generation Capacity
E: Excess All Type Generation Capacity
Then Optimum Load Modification is to

Move Peak
Move Valley



A, B, C

C, D, E

Figure 4.2.1. Dependence of load modification goals on utility's situation.

prices to times of low prices only if it is in the customer's interest to do so. If

the services provided by electric energy are so valuable that customers are
willing to pay a very high price, the utility should concentrate on providing the
energy and not worrying about load shifting (an unlikely but theoretically
possible event). However, Figure 4.2. I is useful in illustrating the fact that there
is much more to load management than simply trying to "chop off" the peak
When choosing a set of transactions which accomplish the desired goals, it
must be remembered that a utility's needs change over the years. History has
shown in the United States that years of undercapacity are often followed by
years of overcapacity. Such variations will probably continue in the future. The
price-only transactions and some of the price-quantity transactions can be used
to cover all of the conditions summarized in Figure 4.2.1. However, particular
types of price-quantity contracts such as direct load control of individual
appliances and demand limits are more restricted in the types of conditions
where they are effective.
From the utility's perspective, the choice of types of transactions to be offered
is a tradeoff between implementation costs and the amount of customer
response. This cost-benefit tradeoff is done using standard system planning
The choice of energy marketplace transctions to be offered must also consider
the customer's desires and needs. Conceptually such needs can be incorporated
by modeling the customers' response, but in practice it is better to consider the
customer's point of view in a more explicit fashion. Issues of concern to

4. Implementation 89



Maximizing the perceived benefit received from using electric energy

Maximizing value added for electricity (for a firm)
Minimizing the complexity of the transactions
Minimizing the monthly bill
.Minimizing the uncertainty in bills and availability of energy.
Maintaining maximum control over their own fortunes
Developing a feeling of partnership with their utility (i.e., the utility is not the
of these issues are difficult to include in formal cost-benefit analysis (as
is usually defined) but they cannot be ignored.
of Customer Classes

hourly spot price, Pk(t), is the spot price for customer k during hour t. In
!)rac:uce, the k index usually refers to customer classes and rarely to individual
(except possibly for some very large industrials). For example, a given
cover all residential customers in some geographic region. For commerand industrial customers, the index k may refer to voltage level of service,
again with possible geographic dependence .
. definition of these customer classes influences the degree of sophistication
in calculating the hourly spot price (see Section 4.5).
We Get There From Here?

key issue in choosing energy marketplace transactions is the question of

to evolve from the present-day system into a full energy marketplace. There
four basic* stages leading towards the eventual full energy marketplace



Demonstration and Test: An experimentation and learning phase used to

gather data and experience.
Initial Implementation: Applied only to selected classes of customers.
Commitment to a Spot Price Based Energy Marketplace: This is a major
milestone wherein the utility and regulatory commission announce to the
customers a firm commitment to the establishment of a spot price based
energy marketplace for all customers.
Full Scale Implementation.

a commitment to the energy marketplace has been made, it will be

eSSefl1ltal to embark on a major educational process so that all customers can
the opportunities that will exist within the energy marketplace and how
they can best take advantage of them. It will also be necessary for the electric
utility operators, planners, etc. to learn what type of impacts the energy marketwill have on their operation and planning functions.
During the initial implementation phase, hourly spot price based transactions


J. The energy marketplace

are offered to restricted classes of customers. Revenue reconciliation can then

done only for these classes. However, it will be necessary, as soon as possib
to make all rates consistent by basing them on the hourly spot prices. If this
not done, there can be major discrepancies in equity between the costs borne
those customers under spot pricing and other customers.
As discussed in Chapter 3, the recommended approach is to establish
mandatory minimum sophistication level of price-only transactions for differe
customer classes (e.g. industrial customers see one-hour or 24-hour updat
while residential customers see billing period updates). Customers should ha
the option to choose price-only transactions with faster cycle lengths or mo
detailed levels and to engage in price-quantity transactions. The addition
costs are either assigned to the customers in terms of fixed charges or factor
in the price--<.:ontracts. The mandatory versus optional transaction issue shoul.
also play an important role in the transition period.
The logical place to begin the implementation of sophisticated energy market
place spot price based transactions is with the large industrial or commerci
users, since the classical cost-benefit tradeoffs are maximum for this clas
However, residential customers who have sophisticated electronic communica
tion for other purposes also constitute an important area for initial implementa~
tion of sophisticated spot price transactions. Furthermore, the psychological
impacts on the residential customers of having more control of their own liv
may justify their seeing sophisticated spot prices even if a strict dollar-base
cost-benefit tradeoff does not. The importance of this last point must b
emphasized. Narrow engineering--economic cost-benefit analysis based strictly
on dollars can be very misleading relative to what is socially desirable in the Ion
As discussed in Section 4.1, there are many types of communication medi
which can be used to implement energy-marketplace transactions, and no 0
approach is superior for all situations. However, it should be noted that the
telephone call based systems have major advantages during the transition
period, since the customers can choose sophisticated transactions without
having to wait for the utility to install special communication systems in their
geographic area. 4

Control and decision making devices-logics that customers can implement to'
exploit the potential of the energy marketplace are now discussed.
Approaches to Control and Decision Making

Customer response to energy marketplace signals can be divided into two levels:

Strategic Decision Making: Determines the overall customer response policy.

For example, for space conditioning control, a strategic decision is storage.
Tactical Control: Combines real-time energy marketplace signals with other

4. Implementation 91

such as temperatures, production schedules, etc. Decides the

real-time customer response.
main tools used to perform these decision and control functions are
beings and digital computers. Strategic decision making is a human
, which can be aided by computer simulation and analyses.
control can be done by human beings, by digital computers, or by
combinations thereof. A digital computer's role in tactical control can
forms such as
binary decision logic such as a switch which turns off a particular
whenever the spot price exceeds a certain level.
digital control by micro or mini computers such as are used for various
processes, commercial office buildings etc.
" .. rt.~"~t .. m type computer-aided human decision making, such as for an
plant where the computer provides bookkeeping facilities and
aids to help the plant manager schedule production based on both
needs and the cost of electricity.
of Customer Response

examples of customer response discussed in Chapters I, 2, and 3 can

into some general categories. When making such a categorization, it
to dra\V a careful distinction between the service provided by
energy and the actual usage of electric energy. For end uses such as
the service provided by electric energy, i.e. light, is provided at exactly
time that electric energy is consumed. However, for other types of end
as~space conditioning, the electrical energy may be converted into
energy (heat or cold) at one hour which is then used to provide the
to the customer (Le., desired temperature) at some other hour. Keeping
distinction in mind, some general categories are

of usage always involves some type of storage. Examples are

Storage, Inherent: Preheating and cooling of the building shell or
Storage: Batteries.
Pressure and Gravity: Water pumping, air compressing, gas storage.
Product Storage: Rescheduling of industrial production by storing semi-

92 I. The energy marketplace

finished or finished p r o d u c t s . ) ;
Working Hours: Shifting work hours or assigning workers to other tasks for'!
a few hours
Product storage is a very important type of energy storage which is sometimes]!
overlooked. Shift rescheduling is energy storge that is like product storage, butJ
it is usually much less c o s t - e f f e c t i v e . , ;

Customer Response to Price-only Transactions

When discussing customer response, it is helpful to aggregate hourly spot price~;

components into

Normal Operating Components: Hourly spot price components associated

with normal operation when the system's capacity is not being approached,
i.e. A.(t), IJdt), YR (t), IJR (t).
Quality of Supply Components: The YQs(t) and IJQs(t) components which are
very small or zero most of the time, but dominate the normal components
during times when the system's generation or network capacity is being

Customers can respond to normal price variations by


Ignoring the Variations: Customer finds that the normal variations are insufficient to justify any response.
Treating Them as Prespecified Time-of-Use: Customer responds to normal
variations only in an expected value sense, where expectation is over months
and makes no attempt to respond to the day-to-day variations.
Real-Time Response: Customer responds in real time to both changes in the
forecasts of future spot prices and the}r actual variations.

The customers who respond in real time to normal spot price variations often
view their tactical control as being divided into two parts.

Long-term (say a day to weeks) control decisions based on forecasted spot

price behavior and forecasts of other variables such as weather and production needs.
Short-term (say hourly) deviation from long-term schedule, to respond to
unexpected changes in hourly spot price, weather, etc.

For many small industrial/commercial and almost all residential customers,

real-time response to normal spot price variations requires some sort of digital
computer aid for tactical control, since the amount of energy usage and costs
involved usually do not justify the use of human decision making in a repetitive
daily pattern. However, for large industrial/commercial customers, humanbased tactical control can be done each day. (Computers can be used to aid such
human decision making).
Now consider customer response when quality of supply components cause

4. Implementation 93



high prices. s If services provided by electricity are extremely valuable to a

at a particular time, that customer may choose not to respond, but this
1i,1"ll>rl>nt than ignoring the spot price. Customers usually cannot respond in
.pre:sp(~cilied manner to the quality of supply components because, for most
tU<lI1I\,,,,, it is difficult to predict more than a day in advance the exact hours
these quality of supply components will be active. Customers respond to
of supply components using either manual or digital computer techniManual techniques for the tactical control can be used by customers who
on digital logics for normal price variation response, because the magnitude
the quality of supply components can justify heroic response mechanisms
are not easy to preprogram into a digital computer. Customers often
strategic decision plans in advance to determine how they will respond to
quality of supply components. The level of sophistication of such preplanstrategic decisions ranges from elaborate shutdown and rescheduling
for large industries to a residential customer who simply decides to
down the air conditioner, tell the kids to stop watching TV,6 minimize
king, and to turn off all nonessential lights.
to Price-Quantity Transactions

general point of importance relative to customer response to price-quantity

must be discussed. For many types of price-quantity transactions,
utility exercises. its right to limit quantity usage under a price-quantity
only at those times when the quality of supply components of the
spot price d6minate. Thus there is a close relationship between customer
mechanisms to price-only transactions when the quality of supply
cOlnooncem is large, and their response to quantity restriction signals sent by the
ity under a price-quantity transaction. Of course the size of the response may


lmanj such

Tactical Control Service

One basic premise of the spot price based energy marketplace is that the utility
not act like Big Brother and take over customer decision-making prerogaHowever, a utility can provide tactical control hardware and services for
the customers (particularly residential and small commercia\). For example, the
utility could offer to provide digital computers, switches and internal communication systems which will respond to hourly spot prices and other inputs as a
tactical control service. This does not constitute utility Big Brother action
provided the customer specifies the strategic decision-making logics to be used
by the hardware. The utility-provided system simply implements the customer's
desires. The customer would probably want to keep override power realizing
that each override costs money.
A Residential Example

Consider a sophisticated residential customer who sees a 24-hour update or

one-hour update spot price combined with forecasts of future prices.

The residence is equipped with digital logic, internal communicat

metering and control hardware, and a user-friendly human-<omputer inter
(displays, buttons, etc.) Two-way electronic communication exists with
utility. The overall digital display and control system can be viewed as an eXl
system combined with optimization logics. The following functions are perl
med by this system. Note that the customer's use of the functions evolves as ti
gain experience, especially under price-only transactions where they do not hi
to figure out in advance how they will respond.

Help Customer Understand: The customer learns how much electrial enel
is used for the various services it provides and develops an understanding
cost per service associated with use of electric energy.
Learn Customer's Desires: The customer states desired service levels (e.,
ideal temperature of house) as well as cost versus service tradeoffs (e.g., he
many dollars savings are required before they will accept deviations from tl
ideal temperature).
Warn Customer: The customer is warned when the spot price is high enoug
to cause a cost per service that exceeds a customer's prespecified critical valul
e.g., when it is going to cost more than 50 to run the dishwasher. If tb
customer desires, use of specific appliances (e.g., dishwasher, washin
machine, dryer) is automatically inhibited when cost per service gets too higl
Provide Routine Control: Space conditioning, thermal storage, water heating
refrigerator-freezer defrosting, swimming pool pumping and heating, etc. arl
routinely optimized by combining forecasted future spot prices and weatheJ
conditions with models for house thermodynamics, hot water usage patterns
etc., using the customer's desires and cost-service tradeoffs.
Provide Special Control: Implement special customer requests with minimum
costs such as have living-dining room at nOF from 7-11 PM for a party;
having swimming pool clean and heated next Saturday afternoon; run dishwasher and washing machine before 6.30 AM tomorrow; etc.
Suggest Heroic Control: Help the customer decide what to do when faced
with very high spot prices due to quality of supply components. Can be done
beforehand with a simulated high-price scenario (weeks to months) or in
response to an actual high price.
Develop Mathematical Models: Use observed behavior to estimate mathematical models for house thermodynamics, hot water usage patterns, etc. for use
in control logics and to help customers understand how they use electric
Communication with Utility: Inform the utility of the customer's desires,
control logics, etc., so utility can better forecast the impact of spot price
changes on aggregate customer response.

The existence of the energy marketplace can cause the residential customer to
purchase new appliances, etc., that are better able to respond. Examples are

4. Implementation


thermal storage and dual firing (electric and natural gas) space conAs time goes by, appliance manufacturers start to produce appliances
to be able to exploit time-varying prices.
Indlustrial Example

l1S(;USSIOn of an industrial foundry's response was given in Section 1.3. The

tactical control system is similar to the sophisticated residential
O~."t.,.,,,,.c in that it is also an expert system with user-friendly input-output,
ion logics, etc. The major difference is in the nature of devices being
and factors being considered. The industrial expert system looks
further in time to consider production scheduling for the next week or
An additional point is that for such a customer, electricity is one of the
items in its budget. Decisions about rescheduling, etc. will be based on
assessment of net impact on the bottom line.

types of customer planning decisions are

of Transactions
Investments for Energy Control Systems, New Process Devices,
Systems, etc.
of Uncertainty on Choice of Transactions

3 recommended that particular types of price-only transactions be

on the basis of customer classes with customers having the option of
a faster price-only signal or price-quantity transaction. The following
~Hre<;~,E'.<; the question of how customers should decide which, if any, option to


t price

customers were asked what type of electricity rates they would like to see,
answer would usually be that the bill should be low and have no uncertainty,
the rates should be simple. If the desire to minimize uncertainty and
JIJllfJ""''')' were the only criteria, customers would rarely choose to go on a
price-only transaction. However, the potential of a reduction in the bill
increasing the difference between the benefits received and the bill) can
. very strong motivation to choose a more sophisticated transaction.
Relative to the uncertainty issue, many customers are initially appalled when
think about the possibility of having to face very high prices (e.g. 50 cents
one dollar per kilowatt hour) at some times. However, after consideration,
cus;tolnelrs realize that their prime concern with uncertainty is not the hourly
price itself but rather the uncertainty in their monthly and yearly electric
bills.7 A highly fluctuating hourly spot price (due to both normal and
ty of supply components) can result in monthly and yearly energy bills
are quite predictable because of time-averaging effects. In fact, customers
use their capability to respond to reduce the uncertainty in the energy bills.


I. The energy marketplace

Customers who appreciate the potential of the energy marketplace will VV .....V.,"
wide variations in normal spot price behavior, and the times when the q
of supply components become very large. Variations provide the customers
greatest opportunity to save money and reduce their monthly bills
increase the benefits they receive from electric energy.
A Necessary Condition for Capital Investment

Discussions now turn to how customers make capital investments in

presence of the energy marketplace. A necessary condition for customers
make capital investments is that the utility and regulatory commission make
firm commitment to the establishment of a spot price based energy malrketr,la()e
and define the rules under which it will be operated (such as methods of
of supply computation, revenue reconciliation, etc.). Unless such a .. Vi""" ...." ...
is made, customers will be very reluctant to make capital investments unless
payback period is very short (e.g., one to two years).
A commitment to the establishment of the energy marketplace with a particular character does not imply a prespecification of how the spot prices
going to behave in future years. However, it does provide a basis for reasonable
forecasts of future-year price behavior - i.e., guidelines for customer investment decisions.
Types of Capital Investment

Customers will usually find it much more advantageous to make capital investments to exploit the energy marketplace if they are constructing a new plant or
building (or are already planning a major retrofit). However, the key issues are
essentiaIIy the same whether new construction or retrofitting is considered, and
hence the foIIowing discussions apply to both.
Examples of possible capital investment's are


Add Storage Capacity: Addition of thermal, electrical or process storage

enables the customer to better reschedule usage without major effects on
service or overaII production.
Add Fuel Switching Capability: Use gas or oil when the price of electricity is
high, electricity at other times.
Add Cogeneration or Self-Generation.
Add Process Monitoring and Digital Diagnostics: Use sensors and computers
to see where the electric energy is being consumed.
Add Digital computers: For use in either direct load control mode or as an
expert system to aid plant and building operators.

As discussed before, customers need a forecast of future energy marketplace

conditions in order to make such capital decisions. These forecasts can be
provided by the utility or by outside consulting firms. The utility or outside firms

4. Implementation 97

services to customers to aid their capital investment

large industrial or commercial organizations internally allocate the cost

. energy in inappropriate ways. A large industrial customer who
electric energy costs to its various operations within the plant on the
of square foot of building occupied, or number of personnel involved,
the individuals in charge of operations with counterproductive signals
how they should use electric energy. A large commercial or residential
whose owner rents to various tenants and charges the tenants for
usage on the basis of occupied space square footage provides these
with little motivation to worry about electric energy usage.
);<lJ.uLauvllal changes in cost allocation, so that middle-level managers
tenants see realistic energy bills, can be an important step towards
ust()m~~r planning decisions for an energy marketplace.

value of the spot price at time t, given the information available at time r,

E{Pk (tlr)} +. Covariance Term

}: Best guess of ;fk(t) given information at time r

,;: A(t)


t rJu(t)

+ rJQsJ(tlr) + )'R(t) +



simple looking equations have been presented for the components of

5.1). We nqw discuss the problems of actual implementation.
Computation of the true value of p( t) at hour t is not an easy task for several
s. There will probably be disagreement on how some of the components
as quality of supply and revenue reconciliation should be defined and even
the definition of system lambda (e.g., marginal versus incremental). For
implementations, approximations for losses and for the effects of
time coupling (e.g., A(t) depends on conditions at other hours)
be required.
The fact that the true Pk (t) may not be calculated does not destroy the value
implementing a spot price based energy marketplace. The actual value
",a,vUl.a ..,u will be much closer to the true values than the present-day flat or
f-use rates, etc. The goal of implementing the spot price based energy
marketplace is to improve the coupling between the utility and its customers, not
to achieve theoretical optimality.8
The calculation of the individual components of (4.5.1) is now disussed.
System Lambda: A(t)

System lambda has been defined as the partial derivative of generation fuel and

98 I. The energy marketplace


Demand Increment (MWh)

Figure 4.5.1. Change in marginal cost for a range of demand increments.

maintenance with respect to demand at hour t. There are various ways it can be
In the best of all possible worlds, lambda would be obtained as one of the
direct outputs of a unit commitment logic which automatically takes care of all
multiple time period dependence associated with unit startup and shutdown
costs, hydro storage and dispatch, purchases and sales with other utilities, etc.
Unfortunately, in many real-world applic~tions, life will not be so easy since the
existing unit commitment logics may not be appropriately defined and in fact
may not even be computerized. Therefore other approaches, which are not as
elegant, will be discussed.
There is an incremental-decremental approach, wherein the demand in the
existing system dispatch logics is varied at a given hour t to see how the total
costs at hour t change. Because of the shapes of the heat-rate curves, unit
shutdown costs, etc., a plot of the change in costs with respect to size of
increment or decrement of demand will often not be a smooth or even monotonic function as illustrated by Figure 4.5.1. A reasonable approach is to use a range
of incremental and decremental costs and do some averaging to obtain the
quantity to be called A(t).
A planning type production cost model (deterministic or probabilistic) can be
used instead of an on-line program in the control center. This production cost
model could be used, for example, once a day to generate a A vs. demand curve
for use during that day's operation. The inputs to this production cost model
are, of course, based on the generators available that day, the demand characteristics predicted for that day, and predicted purchases and sales from other

4. Implementation


The separate model approach has many advantages during the initial
phases of an energy marketplace.
Many control centers have a computerized economic dispatch logic which
tes a quantity called system lambda every five to ten minutes. This
does not include multiple-period effects but is still a reasonable quantity
use. However, care is needed since the lambda generated by an existing
program may not include, for example, the cost of gas turbines and/or
effects of purchases and sales. Also such economic dispatch codes are usually
set up to predict lambda 24 hours in advance.
Many control centers engage in purchases and sales each hour. Examples of
sales structures are the Florida and California Power Broker
which develop a frequently updated price quote that drives interutility
exchange. This quote could be used as the needed lambda in some cases.
In all cases, the relationship between the computed lambda and the loss
:onl0(me:nt of the hourly spot price has to be considered. For example, a
)Urcn:ise or sale quote to another utility may include losses and so is not the true
at the swing generator.
The conclusion of the above discussion is that there are various ways to get
the quantity we call lambda. The choice for any particular implementation is
dependent on the explicit capability of the existing central control
Usually an adequate approximation is readily calculable .


Quality of Supply: )'Qs(t)



ideal world, the generation quality of supply component is determined by

clearing forces, i.e., it is simply increased during times of generation
shortages until the demand reduces enough to maintain the desired
reserve margin between generation and demand (or reaches the level
custom"ers not on spot price rates face rotating blackouts). Such market
can be done in a predictive mode. For example, if demand is predicted
be too high at 4 PM, a generation quantity of supply component may start
introduced at noon to give customers time to respond and to provide the
with data on the types of response tha t is occurring. However, as discussin Chapter 2, this ideal approach has some negative political and social
In practice, other methods of determining the generation quality of
component may be chosen, especially during an initial implementation


A common implementation for the generation quality of supply component

expected to be based on an annualized cost equation such as (2.4.1). Loss of
probability, LOLPy(t), is not a quantity that is normally computed in real
by present-day economic security functions. One approach is to use a
bilistic production cost model (as discussed under the calculation of
lambda). This could be done either for each hour or once a day by
JonlPtltlnlg a LOLP vs. demand curve. Remember also that in (2.4.1), multi.:VU'''''''''vu of LOLPy(t) by any constant does not change yos(t).


100 I. The energy marketplace

A second approach is to remember that the use of the LOLP measure is really!
an arbitrary (albeit reasonable) choice. Simply to define the generation qualit~
of supply component to be a function of operating reserve margin, where thi~'
function is zero unless the operating reserve margin is below a certain level after,
which the generation quality of supply component rises smoothly.9 This genera"
tion reserve margin could be defined from outputs from the unit commitment
dispatch logics, or could be determined simply as some deterministic functionl
of demand and available generation.
An example of this second approach is to use (2.4.1), i.e.,


A Qs .;. -'---




L (1;,(/)


k[g(/) -

[gcri,.;.(/) -



> gcril.;,(t) - 6g


t1g: A parameter chosen by engineering judgment

k: A constant whose value does not affect YQs(t)

If desired, one could view the ar (t) of (4.5.2) as an approximate LOLP, (/). Note
the relationship between the YQs(t) of (4.5.2) and the aggregate I'/Qs(t) of (2.5.2)
(2.5.3). Initially most utilities will want only a crude model.
Network Losses:

'1L,k (t)

If an on-line load flow is available covering all buses, it can be used to obtain
line flows and their partial derivatives with respect to bus injections, thus
enabling the quantification of the network loss components I'/L,k' However, such
an on-line load flow for all transmission-level buses is not always available and
is rarely if ever available (today) at the distribution network level. Hence
aggregate network models of some type are required. The level of detail needed
depends heavily on how the customer classes (i.e., the k index) have been
Total line losses L for the DC load flow approximation presented in Appendix
D can be expressed as a quadratic function of bus injections y by
L = Lo + y' By.IO The dimension of the bus injection vector y and the corresponding B-matrix can be reduced by aggregating several or many buses
together, by assuming the demand (generation) at each individual bus is a fixed
percentage of the sum of demands (generation) at all of the buses to be
combined. Present-day economic dispatch is often done by use of a B matrix
with each generation bus incorporated, but all demand aggregated into one bus,
A two-bus equivalent can be obtained by aggregating all generation into one

4. Implementation 101

and all load into a second bus (Le., no customer class distinction) to yield
loss model such as

'~"'~'~'J (4.5.3) will probably not yield a satisfactory approximation for

whose geographic generation pattern varies widely with demand level

which are heavily involved with purchases-sales with neighboring utiliOne approach in this case is to hypothesize a loss model with a more
JIJlIVll~'a.",u structural form such as


+ Qsd(t)g(t)

+ a 2 d 2 (t) +
+ Q6 S (t)


+ Q4i(t)

Total demand during hour t

Total generation
Total sales

then to estimate values for the ao, ai' ... coefficients by using regression
. on whatever historical data or load flow planning studies are available.
a structural form such as (4.5.4), it is easy to compute its derivative with
to demand" and hence the network loss component of the hourly spot
Quality of Supply: '1QS.k(t)

ideal word implementation of IJQS.k is the explicit network model (market

or. cost function) approach wherein both generation and demand
. Aft exact implementation requires a load flow combined with a search
Such capabilities exist at the transmission-generation level of some utilitoday, using either computer or computer-aided human search. However,
rarely if ever exist down to distribution level voltages.
An aggregate model based on the capital costs of expanding the network as
(2.5.2) can be used as needed. If network expansion is done for other reasons
new load growth (such as buying and selling), regression models similar to
can be developed, provided sufficient data is available. However, instead
using polynomials as suggested in (4.5.4), hypothesized structures based on
form of (2.5.3) should be used.
Here too, the definition of customer classes (k) has a major impact on the
of sophistication required.
Revenue Reconciliation: YR(t), '1R.k(t)

the constant multiplier approach to revenue reconciliation is used, the only

.jmplementation problem is the computation of the multipliers. Unfortunately
may not be an easy task. Conceptually the mUltipliers can be computed

102 I. The energy marketplace

using a production cost computer program combined with a load flow, if an

explicit network model is being used. Demand response to price effects can
require iterative solutions.
During initial implementation for a particular class of customers, an easier
approach is to choose multiplier values so that spot based transactions are
revenue neutral relative to the present-day rates seen by that class of customers.
This is a nice way to get started on spot pricing. However, it too presents
problems because revenue neutrality for a class does not imply revenue neutrality for individual customers.
As discussed in Chapter 8, the constant multiplier approach is a special case
of more general approaches such as Ramsey pricing and weighted least squares.
If a more general form of Ramsey pricing is to be implemented, there is the
difficult practical problem of determining the customer response parameters,
such as the price elasticity of demand.
Prediction of Future

Thus far the calculation of the components of Pk(t) have been discussed
assuming information at hour t is available. For actual implementation, it is also
necessary to calculate the best guesses of the value at hour t given information
available at hour r, t > r. Thus it is necessary to compute Jc(tlr) = E{A(t)lr},
A unit commitment study done at time r will yield Jc{tlr) directly. If other
procedures are used, best guesses of generation availability and load forecasts
(see Section 4.6) can be used. Such best guesses can also be used to yield the loss
and quality of supply component guesses.
If generation quality of supply is computed using a loss of load probability
type forecast as in (2.4.1), it is theoretically necessary to incorporate, probabilistically, the effects of possible generation outages between time r and hour t.
However, a rigorous treatment of such effects in probably not justified, especially in initial implementation. A heuristic approach is to use (4.5.2) where ~g
increases as t - r increases, but even such modifications may not be justifiable
in practice.
Covariance Term

As discussed in Chapter 2, the covariance term of (4.5.1) is determined by the

correlation between hourly spot price variations and those of the derivative of
demand with respect to price. The data problems associated with its actual
computation can be so difficult to solve that we recommend the covariance term
be ignored except in special cases where there is a strong a priori reason to
believe it is very important. 11

The utility operating functions associated with metering and communication

with customers and calculating the hourly spot price were discussed in Section

4. Implementation




Economic Security

Spot Price
Based Transactions

Spot Prices




4.6.1. Major functions of utility operation.

and 4.5 respectively. Other utility operational issues such as economic

control and the computation of spot price based transactions are now
igure 4.6.1 summarizes the overall operational functions in terms of a
loop, wherein load forecasts drive economic security functions which
then sent back by a price calculation to affect the load. This feedback of price
load is the major distinction between the energy marketplace and present-day
transactions. Appendix B provides background discussions on present-day
Calculate HOllrly Spot Prices box of Figure 4.6.1 was discussed in
4.5. Each of the three other blocks of Figure 4.6.1 are now discussed.

short-term operational needs, the utility needs a load model which can be
to forec~st the hour-by-hour demands for the next day to one week as a
of time of day, expected weather patterns, and expected prices. Presentload models incorporate only weather and time dependence. The modeling
price effects cannot ignore the impact of rescheduling of demand; in other
the demand at hour t depends on both the price at hour t and the prices
demands at other times before and after the hour t.
,', The development of such a price-dependent load model could appear to
a major obstacl~ to the implementation of the energy marketplace,
the necessary data to specify and develop such a model is simply not
ble today. However, this data availability problem will resolve itself
provided the energy marketplace is introduced in a gradual fashion.
particular, for the first few customers seeing spot prices it will not be
to build a price-dependent load model into the utility operations,
I)CI~i:1ll:'C the response of just a few customers will not be sufficient to influence
the spot price significantly. As the degree of penetration of spot price based rates
increases, the need for an approximate price-dependent load model will become
real, but by then sufficient data will have been obtained to develop it. As the
penetration of spot price based transactions increases further, more accurate

104 I. The energy marketplace

load models will be needed but the necessary data to develop them will have
become available. A related phenomenon with similar consequences is that
given customer's response level will increase with time, therefore leaving
time to gather data. Thus the availability of data to develop the needed
model will occur automatically provided energy marketplace transactions
introduced gradually.
Given the availability of data, it is a nontrivial but relatively
exercise to do the statistical manipulation needed to obtain a
load model.
Economic Security Functions

The purpose of the economic security functions of Figure 4.6.1 are to

strategies to control the generation and network, so as to provide electric pn.~rO'\I'AW
at minimum cost, subject to a system security constraint. (I.e. that
hardware is not damaged, blackouts do not occur, etc.) Economic
functions have to take into consideration the existence of uncertainty arising
from equipment outages, errors in weather forecasts, etc. Present-day economic
security functions handle such uncertainty primarily by using an VU'~ll-lVlJ'U
feedback type of philosophy. Best predictions of future behavior are
with deterministic decision logics, which maintain a sufficient reserve margin to
handle unexpected events. This process of prediction combined with deterministic control strategies is repeated periodically and/or after major events have
occurred to provide the necessary feedback (tracking) response to uncertain
events when they occur. An energy marketplace implementation can be quite
similar in basic structure. Neither the criteria nor the basic control philosophy
have to be changed.
To be more specific, very little if any mpdification will have to be done to the
economic dispatch and automatic generation control functions in the energy
marketplace. However, the energy marketplace will have a sizable impact on
unit commitment logics, since an additional feedback-iterative calculation is
required to handle the effects of price on demand and demand on unit commitment. Although the energy marketplace complicates the unit commitment
computations, the feedback has the very positive advantage of reducing the
impacts of future uncertainty on operating costs, and hence provides a more
efficient unit commitment behavior. Also, operating reserve margins can be
reduced if customers carry some or all of the necessary operating reserve.
Emergency state control functions such as load shedding could be extensively
changed by the introduction of the energy marketplace. It all depends on the
character of the chosen transactions.
Evaluate Transactions Values

Given the Pk(tlr) (as discussed in Section 4.5), it is necessary to compute the
values of the spot price based transactions.

4. Implementation


a one-hour or 24-hour update price-only transaction is to be implemented,

themselves are used.
the period definition (see Section 3.3) is longer than one hour, time averagthe Pk(tlt) is required. For example, if a three-hour period is used, then
AM, 8AM, 9AMlt) = "3[p(7It )



be used. (The theoretically present covariance term would probably be

in this case.)
in Section 4.3, there is often a close relationship between the times of active
of supply components, and the times when price-quantity limitations on
are called for by the utility. Therefore estimation of the quality of supply
)rnl)Oflents is especially important for price-quantity transactions.
'ty transactions can complicate the feedback loop of Figure 4.6.1
price-only transactions. As one example, if the price-quantity
olll:'''''~.IUll has a limitation on the number of times a year quantity control can
exercised, the utility operator is obliged to decide whether a control action
be used today or saved until tomorrow when it might be needed more.
another example, some types of price-quantity transactions present pardifficult problems when it comes to predicting customer response.
on System


power systems are operated by highly trained human beings who use
computers tOl)fovide the needed information in a usable form. The price
of Figure 4.6.1 has two different effects on the system operators:

It makes their lives more difficult by introducing a new set of numbers (e.g.
prices) wlilich they have to handle
It makes their lives easier by giving them new control capabilities which
reduce the impact of uncertainty and reduce (eventually eliminate) the trauma
of having to resort to rotating blackouts or other unpleasant control actions



n the

e the

We personally believe that as system operators come to understand the benefits

of price feedback, they will find its rewards make it well worthwhile.
Impact on Power Pool Operation

The discussions thus far have tacitly assumed the implementation is being done
for a single utility which has its own central dispatch and control system. If the
utility is part of a centrally dispatched power pool involving other utilities,
additional implementation issues arise. The actual problems to be solved depend
on the specific nature of the pool. However, some generic issues can at least be
The actual operating costs of a given utility within a pool may be determined
by the own-load dispatch logics (see Appendix B), combined with formulas
which distribute savings of pool operations among pool members.


1. The energy marketplace






Figure 4.7.1. Major functions of utility planning.



The pool lambda may require correction for losses to translate it into a given
utility lambda.
Network and generation quality of supply components may be defined differently for different members of the pool.
Revenue reconiliation components will definitely be different for different
members of the pool.
It should be emphasized that a utility's membership in a power pool does not
prevent it from implementing a spot based energy marketplace even if the other
pool members do not.

Figure 4.7.1 summarizes the planning functions of a utility in terms of a

feedback loop wherein load forecasts drive generation and network supply
models which in turn drive financial models that determine expected rates which
are then fed back to the load forecast demand model. (Appendix C provides
background material on power system planning.)
Figure 4.7.1 is superficially similar to the operating feedback loop of Figure
4.6.1. However, there are some important differences. First, the price feedback
loop for operations in Figure 4.6.1 is a new feature introduced by the energy
marketplace, while the planning feedback loop of prices of Figure 4.7.1 exists
today in most utilities. Second, the operating criteria is relatively simple
(minimize cost subject to security constraints), while planning criteria involve
mUltiple, often conflicting attributes such as desire to minimize costs and
environmental impacts while maximizing reliability. Third, uncertainty is
handled in operations by the use of simple open-loop feedback and the use of
conservative operating reserve margins. Such deterministic approaches to uncertainty have also been used for planning in the past, but today the need for
more sophisticated techniques for dealing with uncertainty is becoming widely
Each box of Figure 4.7.1 is discussed below.
Load Modeling Forecasting

For planning, the utility needs a load model which can be used to forecast future

4. Implementation


patterns and levels many years in advance as a function of time,

and prices. Many, but not all, existing load models have this capability
present-day type rates). Modifications of these existing models will be
to enable them to handle spot price based transactions. We presently
that long-range energy marketplace load modeling will be done almost
in terms of modeling the effects of price-only transactions, since their
are much easier to handle than most price-quantity transactions. '2 If
uantity transactions are to be offered, they will be studied in separate,
long-range and all-encompassing models.
operational load modeling of Section 4.6 had the nice feature that the
data to develop the models can become available at the same time the
themselves become needed. Unfortunately, this same phenomenon
t be expected for the planning problem. A major source of uncertainty is
long-range response of customers to the energy marketplace, e.g. the type of
luit)m(~nt (both control and end use) they will install to exploit the potentials
marketplace. It will take many years after a commitment to an energy
has been made before hard data on such long-term response
available. Planning will have to be done before accurate long-term
models have been developed.
uncertainty associated with long-range customer response would have a
I effect on planning if it dominated the other uncertainties a utility
has to face. Io( turns out, however, that uncertainties in load growth, cost
cost and '\.vailability of different types of fuels, regulatory treatment,
nvironmental impacts and constraints apper to dominate the effects of the
customer'response uncertainty. 13 The damping effect price feedback
on the impacts of uncertainty should make the overall planning process
accurate for an energy marketplace .


to the simplified picture of the planning process presented by Figure

.1, commitment to a spot priced based energy marketplace will not require
major modification of the existing supply models such as production costing and
generation expansion planning computer programs. The type of plans that are
developed will of course change, e.g. there will often be much less reliance on
peaking plants.
) und for

Financial Models
Many aspects of present-day financial models apply directly to an energy
marketplace utility. Additions to existing programs will be required to predict
how future spot prices will behave given a particular planning scenario.


The main operational issues of concern to a regulatory commission are the

specification of the price formulas and the monitoring of the utility's behavior.


r. The energy marketplace

In a spot based energy marketplace, the regulatory commission has to specify

(or accept) a formula for calculating the spot prices rather than specify a
particular numerical value for a particular rate. Fortunately there exist various
precedents for such a procedure. The fuel adjustment clause is one example.
Present-day interruptible rates and many direct load control schemes provide
precedents that are closer in time scales of operation to hourly spot prices. The.
utility decides in real time, using a formula, whether or not to interrupt a;
customer. The regulatory commission has prespecified (or accepted) thar:
formula. Even if the formula is a verbal statement of the conditions for interruption rather than an explicit set of equations with numerical parameter values,
the principles are the same.
T~e revenue reconciliation multiplier is an example of a parameter in the.
equations that has to be reset, for example on a yearly basis. The detailed
calculations for achieving revenue reconciliation change under spot pricing, but
the principles remain the same. The regulatory commission still has to decide
each year what costs can be put into the rate base, what a reasonable rate of
return on investment is, etc.
An important concern is whether implementation of an energy marketplace
requires the regulatory commission to monitor very closely the day-by-day
utility operation. Customers may be concerned that without such close monitoring, the utility would manipulate prices to increase their profits. We believe that
establishment of an energy marketplace will change somewhat the details of how
the regulatory commission monitors the utility's behavior, but that close monitoring of day-by-day utility operation will not be required. Concern over
possible price manipulation can be handled through revenue reconciliation.
If an extensive futures market and/or bidding as part of a price-quantity
transaction eventually becomes operational, the regulatory commission will
have to develop new procedures for monitoring and regulating such operations.
On a planning time scale, the regulatory commission has to specify (or accept) .
a structural form for the equations used to calculate the spot prices. This
requires decisions on which quality of supply approach to use, how much
network detail to include, etc. During an initial implementation phase, the
structure of the spot price equations may change (Le. as experience is gained).
However, the structure should become fixed as soon as possible.
A key regulatory commission planning decision is when to make a firm
commitment to the establishment of a spot price based energy marketplace. This
will not necessarily be an easy decision. For example, customers who are being
cross-subsidized under the present system of rates will tend to complain rather
loudly about changes which result in more equitable sharing of costs.

Implementation of an energy marketplace proceeds through various phases


Tests and demonstrations

4. Implementation


Operation for certain classes of customers 14

Commitment to establishment for all customers
Full-scale implementation
mp,lernetltation of the ..initial tests and demonstrations, and operation for
classes of customers, are relatively straightforward. Full-scale implefor all customers changes the way utilities and their customers plan
operate their lives. A priori, no utility need move beyond phase two.
A pragmatic point of view is essential in implementing a spot price based
marketplace. The goal is to provide greatly improved feedback between
utility and its customers. Actual implementation will vary widely between
utilities depending on their needs and existing facilities. Implementaof an effective marketplace does not require the calculation of the true
of the spot prices. Arguments on the definition of the true spot price can
left to university professors who need to publish academic papers in order to
'n or advance their professional status.

mentioned in the notes for Chapter I, Tabors, Schweppe, Caramanis [1988]

a recent report which summarizes implementation of spot pricing ~nd related
of the ideas of this chapter evolved from our "California Studies" (see
Caramanis, Tabors and Flory [1982], Schweppe, Tabors, and Cara[1984], and Flory [1984]). Portions of section 4.1 are a condensed version
I:lory paper. The Integrated Communications Systems portion of section
is based on ICS [1986] .


. The ideas underlying Table 4.1.1 are taken from Flory [1984] which contains a much more
complete development.
If operating reserves are to be carried by the customer's load, much shorter time intervals are
required, but the ideas extend in a fairly straightforward fashion.
Robert Peddie, originator of the CALMU system and former Chairman of the Southeast
Distribution Board in England, taught us the importance of this point of view.
Any industria!' plant or commercial building which already has a PC and modem in its control
room can implement such a communications system with no hardware cost. Such customers
usually already have recording meters to take care of billing communications.
We are not considering the case where the network quality of supply is negative.
Turning ofT the TV does not reduce usage very much but is a way to get the kids away from the
Fuel adjustment clauses cause today's customers uncertainty in their annual energy bills.
Present-day control centers (see Appendix B) are very sophisticated systems, designed to improve
the economic efficiency of generating electric power (subject to security constraints). However,
their implementation also involves some arbitrary choices and the use of approximation. For
example, the fuel cost of a coal-fired plant that buys from various sources (at different prices)
is not uniquely specified and the mUltiple-period time couplings (e.g., unit commitment and
maintenance scheduling) are rarely if ever handled in the theoretically optimum fashion.
This approach is used in many present-day direct load control and interruptible contract


I. The energy marketplace

10. Lil accounts for losses such as in transformers which occur even if there is no demand.
II. Once a class of customers had been on spot pricing. the utility could examine the data to see
whether the covariance was indeed small.
12. Some present-day load models have become very complex because of a requirement to incor-.
porate the effects of particular types of price-quantity transactions.
13. Utility planners can, after all, initially assume little or no response to spot prices. This will give
them a conservative, worst case, plan since any response to spot prices will be a pleasant surprise.
They can also work directly with large customers to see that new investments are planned to
enhance response.
14. It is important to realize that a small number of customers account for the majority of electricity
use. Exact data is hard to get, but Bohn [1982 page 326] estimates for example that in two
utilities 0.1 % of customers account for about 30% of energy used. These customers are logical
candidates for spot pricing since their energy bills are large, they already have good metering.
and are quite sophisticated (i.e. profit maximizing) in their choices about energy use.


spot price based energy marketplace is designed to operate in a regulated

(regulared private company, or government owned). However, its
,nl"m,f' n1 ation opens a door to deregulation of some or all generation. This
is already being opened slowly in the United States at both the federal and
level t~rough regulations concerning cogeneration, small power producand wheeling. The purpose of this chapter is to look behind that slowly
g door and try to see what might be there.
his chapter only presents a set of basic ideas; it does not analyze their
because such analyses have not yet been done. Since the advantages and
~''''~''''N have not been quantified, we are not advocating deregulation (i.e.
do not know whether there is "a lady or a tiger" behind the door).
This chapter shows how the establishment of a spot price based energy
rKetPlace in a regulated environment (which we do advocate) can evolve
or into a deregulated system. The reader may be surprised to learn that
trip from regulation to deregulation need not be very long (although it may
Many articles have been written on deregulation. In this chapter we only
the approach that evolves naturally from the regulated, spot price
energy marketplace.
The deregulated energy marketplace to be discussed here is summarized in
5.1.1. It has three main participants:


I. The energy marketplace

Bulk Transmission System

Distribution Systems


Market Coordinator


------- r-------







Energy Brokers




Electric Energy Flows

Information. and/or Money Flows

Figure 5.1.1. A deregulated energy marketplace.


A single regulated T and 0 company that controls the transmission and

distribution system, and acts as a middleman in the energy marketplace.
Many independent, private generating companies which sell energy to the T
and 0 company. The amount sold at each moment is the decision of each
generating company.
The users of electricity who buy as much energy as they want from the T and
o company.

Section 5.1 discusses the different elements of Figure 5.1.1. Sections 5.2 and 5.3
discuss how the deregulated marketplace of Figure 5.1.1 fulfills the basic
functions of an electric power system that are summarized in Table 5.1.1 (See
Appendices A, B, and C for background on the functions of Table 5.1.1.).
Section 5.4 concludes the chapter with a scenario that might lead to Figure 5.1.1.

The Regulated T and D Company

The transmission and distribution (T and D) company of Figure 5.1.1 acts as

an intermediary, both physically and financially. It buys all the energy offered
for sale at each moment, while selling all the energy demanded by users.' It also
periodically collects all payments by users and pays alI generating companies for
whatever they produced. The difference covers the T and 0 company's costs of

5. A possible future: deregulation


Operation and Control

Dispatching generation minute by minute

Operating the transmission system during normal conditions; system security, system dynamics,
Controlling the overall system during emergency state conditions such as sudden generation or
transmission outages
Setting prices to customers; choosing which customers will receive electricity during system
Term Operation and Planning

Unit commitment, maintenance scheduling, and fuel purchasing

Investment planning. Choosing what kind, where, how large, and when to build
- Generating units
- Transmission lines
Local distribution systems
Forecasting future conditions
'... " .. """" .... 0

and operating the system and the marketplace. Its specific duties

, maintain, and operate the transmission grid and distribution sy:;tems.

the spot prices and communicate them to independent generators
the stability of the electric power system through a combination of
pricing, direct management of certain physical attributes, and price-quantity
Collect 1Jloney from users and pay money to generators.

on and
o the T
:>f each

T and

lnd 5.3
: basic
.1 (See

T and D company is regulated in a traditional rate-of-return framework (or

owned and operated by some government agency). Revenue reconciliation (to
embedded capital costs, etc.) is done as detailed in Chapter 8, only with
. ,,_.,.,..--- to the T and D system, by modifying the unreconciled spot prices (up
down as needed) or alternatively using a revolving fund or surchargerefunds.
Generating Companies

,Generation represents the bulk of the assets in most power systems. In the

,u".,,"'......ted marketplace of Figure 5.1.1, these assets are owned by a number of

private firms which
acts as
It also
lies for
osts of

Build, maintain, and operate generation and storage units

" Sell electricity to the T and D at the current spot prices
" Meet zoning, environmental, and other restrictions, like any industrial firm
/) Are not subject to regulation by public utility commissions

114 I. The energy marketplace


Are barred by antitrust laws from explicitly cooperating with other generating
companies in their areas, or owning too many units in one region 2
Are motivated by profit
Are forced by competition to act in socially beneficial ways

Today, ownership of generating units is based on geography. One utility owns

almost all the units within its service territory. In contrast, ownership in the
deregulated marketplace of Figure 5.1.1 is based on functional type. Different
generating companies specialize in constructing and operating different plant
types. They are then able to develop expertise, for example, in coal or nuclear
power, instead of maintaining proficiency in all generating systems. This avoids
the current problem of one utility owning and operating a single nuclear plant.
Today more than 200 local monopolies control generation. Even if ownership
shifts to a smaller number of interspersed competitors, the electricity generation
business will be less concentrated than many other U.S. industries.

From a functional point of view, the electric energy users of Figure 5.1.1 behave
in much the same way, independent of whether the energy marketplace is
regulated or deregulated. Large industrial--<:ommercial and sophisticated residential customer see one-hour and 24-hour update spot prices while small
residential customers see billing period updates. However, the deregulated
energy marketplace employs a significant number of price-quantity transactions
involving short-term (seconds to minutes) transactions required for emergency
state control (e.g. to carry operating reserve on the load). Such short-term
actions are also desirable in a regulated marketplace but are not as essential.
Although going from a regulated to a deregulated energy marketplace has
little functional impact on the users, the hope of those who advocate deregulation is that the prices themselves will be lower (in the long run).
Information Consultants and Energy Brokers

Figure 5.1. I shows two other participants: Information Consultants who

forecast future spot prices (short and long term) and Energy Brokers who
arrange side contracts to shift risks. Their roles are discussed further in Sections
5.2 and 5.3.
Decomposition of the Regulated T and D Company

The structure of Figure 5. I. I. combines the bulk transmission and the local
distribution into one company. An alternative structure which has both advantages and disadvantages involves a single bulk transmission company and many
individual local distribution companies, all of whom are regulated. We will only
discuss the combined T and D company since it is somewhat simpler, but almost
all of the ideas apply equally to the decomposed structure.

5. A possible future: deregulation




section discusses how the short-term operating and control functions of

5.1.1 are fulfilled in the deregulated energy marketplace of Figure 5.1.1.
Spot Pricing to coordinate Generation and Load

the deregulated world of Figure 5. I.l, the generators are not centrally dis. Instead the Market Coordinator of Figure 5.1.1 sends each generator
pot price and each generator self-dispatches itself by generating if the spot
paid for electric energy exceeds the plant's marginal operating costs. A
ve reader might say, "Such generator self-dispatch and central utility
are theoretically equivalent." Such a reader would be right.
The Market Coordinator of Figure 5.1.1 keeps supply and demand in balance
continuously adjusting the spot price. On nights when demand is minimal,
spot price declines until only generating units with low operating costs
ain on-line (e.g., nuclear or wind-powered units and perhaps some coal-fired
. As demand increases during the morning, the spot price may rise to the
where owners of electricity storage units, which had purchased and stored
during the night, begin selling back energy.
On the demand side of the market, users reschedule their electricity-intensive
to times of low spot prices, and generally repond to spot prices to
their net benefits from electric power usage.
Ifil sudden tempOra'ry outage of large generating units occurs, the spot price
increases.,Generating units already on-line increase their output to
mum. Users' 'process-control computers automatically delay the on
of air conditioners, furnaces, heaters, pumps and similar equipment.
generators bring their units on-line. Owners of reservoir hydro units open
sluices to,increase production. If the outage is severe, the regulated T and
company exercises some of its operating reserve price-quantity contracts for
first few minutes. These measures gradually reduce the spot price, which
to decline as more units come on-line. If the spot price remains quite
some customers close down portions of their operations to save money.
y, equilibrium is restored at a higher spot price than before the outage.

of Spot Prices

a regulated energy marketplace, the utility's central control system knows the
costs of all the generators, so the marginal fuel cost component of
prices is computed by the formulas of this book. However, under a
structure, the regulated T and 0 company's Market Coordinator
not necessarily know the precise operating cost characteristics of the
generators. Hence, the marginal operating cost component of spot prices
be determined empirically by observing how the generation responds to
t prices at various times. Alternately, the Market Coordinator could ask
private generating firms to furnish their operating cost data (confidentially


U6 I. The energy marketplace

of course) to facilitate system dispatch. We can see little reason for the privat
firms to say no or to try to play games and provide incorrect data.
Since the Market Coordinator knows the details of the transmission an
distribution systems, the network components of the spot price are determine
directly by the equations in the main text of this book.
There is no generation revenue reconciliation term; i.e. in (2.2.1), I'R (I) = O.
The network revenue reconciliation term IJR.k(t) remains.
Conclusion: Spot price evaluation is mostly but not entirely the same for both t
regulated energy marketplace and the deregulated system of Figure 5.1.1.

System Security

Security means avoiding major disruptions. The Market Coordinator of Figure

5.1.1 performs system security functions such as contingency evaluation and
state estimation as is done today.
The Market Coordinator uses the available reactive control inherent in the
network to maintain voltages within limits, as much as possible. However, when
the voltage control capabilities of the network are exceeded, or when line flows
threaten to exceed their thermal or dynamic limits, the participants' generation
and usage patterns are changed by introducing transmission quality of supply
components into the spot price for energy (reactive as well as real energy is
charged/paid for).
The Market Coordinator maintains sufficient reserve to respond to major
unexpected generation losses, dying tie-line support, and other dilemmas using
price-quantity contracts which can be exercised as needed. Either the customer
would be paid up front, or a very substantial payment would be made whenever
the option is exercised. Analogous contracts with generators provide spinning
Conclusion: The security monitoring and control functions look a lot like those of a
regulated energy marketplace.

System Dynamics and AGC

An interconneted system of generators can display oscillating or unstable

behavior due to either small or large (e.g., faults) disturbances. Users and
privately owned generators who aggravate such undesirable dynamic behavior
are penalized (charged extra) by the T and 0 company while those whose
characteristics improve overall dynamic behavior are rewarded (paid).
Appendix F introduces the concept of the F APER (Frequency Adaptive
Power Energy Rescheduler) which is a way for user devices to respond to
frequency derivations in a simple fashion that the user doesn't notice.
If the regulated T and 0 company is interconnected with other T and D
companys, automatic generation control (AGC) systems are needed to keep

5. A possible future: deregulation



close to its desired value (60 or 50 Hz), to keep the sum of tie-line
near the net scheduled interchange levels during normal operation, and to
emergency support when needed. The regulated T and D company pays
generators to accept and respond to AGC signals. The price level is
ned by the marketplace. Of course, it is conceivable for the regulated T
D company to own some limited generation (pumped hydro would be nice)
sole purpose is to accept AGC and emergency response signals.


At long last, the deregulated marketplace has resulted in something new: the
for pricing dynamic behavior. Research) is needed before the details of how to do
are clear, but no major obstacles are foreseen.

Dynamic pricing might also be desirable in a regulated energy marketplace.

example, the dynamic effects of private generation in the U.S.A. operating under the
PA (PL 95-617) might be best handled by pricing rather than trying to specify a set
standards which have to be met.


relaying is done by the regulated T and D company following today's

s. The regulated T and D company establishes standards for the
relaying at the generator-network interface. Internal generator
is specified by the private generation firms.


of Monopoly Power
rll>rl"P ...,five

reader probably started to have second thoughts about deregulated

operation when she/he began to worry that a single generating firm
t have m,pnopoly power in a region and hence be able to drive the spot price
'''''''y ..J (beyond the marginal cost of generation). We have assumed away
a possibility in our formulation of Section 5.1. However, more analysis is
before we are willing to believe the assumption is true. This is the first
many reasons why we do not advocate adoption of the deregulated marketof Figure 5. I.I without further study.




consider the long-term operation and planning functions of Table 5.1.1

how they are done in the deregulated world of Figure 5.1.1.
Long-term operations (hours to years) and planning (one to 20 years) for any
power utility requires forecasts of certain critical variables, such as fuel
and demands for electricity. Decisions are made and implemented on the
of these forecasts. When forecasts can be compared with actual events,
predictions are made, and the decisions are reevaluated, and possibly
Then the process is begun again.
This generic process is the same, whether done by a regulated utility, or in the
energy marketplace of Figure 5.1.1. However, there are major

1i8 I. The energy marketplace

differences in the nature of the variables, as well as the motives of the

Under today's system, the regulated utility is expected to act in the
interests of its customers by minimizing their electricity costs. Under the
regulated structure of Figure 5.1.1 private generator owners try to
their own profits. But because of the market mechanism's "invisible hand,"
also improve the welfare of their customers, assuming that a competitive
structure arises.
Investment by Private Generators in a Deregulated World

Since ownership and investment decisions are decentralized and deregulated,

there is no need for central power studies or central decision making processes.
Instead, private firms build plants if they think they will be profitable.
This is the approach followed in any industry that sells its product in a spot
market. For example, wildcat oil wells, refineries, and new hotels are all built
because investors expect the investment to be profitable. The basic investment
rule is the same as in any other deregulated industry: if net revenues (appropriately discounted and adjusted for taxes) are expected to be appreciably larger
than building costs, go ahead and do it. This rule is, in reality, quite similar in
concept to the optimal capacity expansion process of a utility trying to minimize
their customers' bills under a regulated energy marketplace.
NUMERICAL EXAMPLE. Assume a private firm is considering construction of a
new coal-fired generating unit, which would have a marginal operating costs of
2.05 k Who Thus the unit would generate at full capacity whenever the spot price
is above 2.05 kWh, and shut down otherwise. 4 Whenever the spot price is above
2.05 kWh, the unit will more than cover its operating expense, and it will have
a positive net revenue equal to the difference between the current spot price and
2.05 kWh, times the unit's output. This riet revenue makes up the positive cash
flow which covers profits and the unit's capital cost.
Figure 5.3.1 shows how spot prices would have varied for a Midwestern utility
for two weeks in 1980. 5 During the week of January 7 to 13, because the spot
price rose above 2.0c/kWh each day, the unit would generate daily. Its total net
reveneues for the week are the shaded area, multiplied by the unit's capacity in
megawatts. During the week of October 6 to 12, on the other hand, the unit
would generate for only a few hours on Wednesday, and about 15 hours on
Saturday.6 Looking at the data for the whole year, the unit's total net revenues
would be $69,900 per megawatt of generating capacity, if the unit were available
whenever needed. In fact, a coal unit would be down for maintenance some
weeks. To keep profits as large as possible, all maintenance would be scheduled
during weeks with low anticipated spot prices. Thus, the actual net revenues
would be somewhat below $69,900 per megawatt for the year. If the private firm
invested in a 500-megawatts multiplied by $70,000, or about $35 million per
year. If the tax-adjusted value of this and the corresponding net revenues in later
years are larger, when discounted to the present, than the construction cost of
the unit, then it is a profitable investment.

5. A possible future: deregulation 119






Monday Tuesday Wednesday Thurs. Friday



Shows net revenue under the spot price curve for the week January 7 to 13. Only
one unit's net revenues (based on a marginal cost of 2.0Se) is shown for sake of

Week of October 6 to 12, 1980





0.0 10

~ firm
n per
)st of


30 40

50 60


80 90 100




Shows net I evenue under the spot price curve lor the week October 6 to t 2 Only
one unit's net revenues (based on a marginal cost of 2.0Se) is shown for sake of
clarity. Total revenue for a year is the sum of weekly revenues for each week the unit
is available.

Figure 5.3.1. Spot prices and net revenues: an example.

Economies of Scale: Generation

Economies of scale are a key issue in the argument of whether private firms will
build generators that are socially desirable (economically efficient). Private firms
will tend to build smaller plants that can be put on-line fast, so they can start
to get their money back sooner. If one lives in an ideal world where future costs,

120 I. The energy marketplace

demands, technologies, environmental concerns, regulatory actions, etc. can be

forecasted perfectly, then big customized power plants that take a long time to
build usually appear to be most efficient. In the real world, with all its uncer-.
tainty, smaller, modular power plants that can be constructed rapidly ca
generally be a much better investment. However, more analysis is needed before
explicit answers are available. Many industries like to argue that bigger is better,
and therefore that government policy should encourage concentration in their
industry. We believe that the extent of economies of scale is something that
competing companies can sort out in a deregulated market. For example, there
is now an arguement that smaller and inherently safer nuclear units could be
built, the pebble bed, or High Temperature Gas Cooled reactors. Similarly,
many cogeneration advocates believe that their technologies are not given a fair
chance because of utility dislike. The virtue of deregulation is that advocates 0
different approaches can test their ideas on a level playing field. Of course,
health and safety regulation would still be needed as for any industrial plants.
Investment by Regulated T and D Company

In a regulated energy marketplace, the vertically integrated utility can coordinate its generating and network expansion plans so the network is not overloaded and desirable system dynamics are maintained. One of the potential
weaknesses of the deregulated energy marketplace of Figure 5.1.1 is that such
coordination will not be as close.
When a private generation firm is considering building a new plant, it will pay
the regulated T and 0 company to do load flow and system stability studies to
determine whether the new generation can be expected to see unfavorable
network and/or dynamic quality of supply components of the spot prices due
to weak transmission in the area. If the network is too weak, the regulated T and
D company will decide whether or not it is socially desirable (relative to the
customer's bills) to strengthen the network. Alternatively, the regulated T and
o company might offer to build new lines at the private firm's request and then
charge for their use by using the line-by-line decomposed method of revenue
reconciliation (discussed in Chapter 8) so the private generation firms pay for
the amount of the new lines they use.
A potential obstacle to overall economic efficiency of the deregulated marketplace is that the regulated T and D company will have to forecast what other
types of private generation might be built in a given area in subsequent years.
For example, for a given new generating plant, adding a 130 kV line might be
optimal, but if a second generator were to be added a few years later, building
a 230 kV rather than 130 kV line might be better. This exemplifies the economy
of scale issue for transmission. There is also the potential problem of the
regulated T and D company building lines for a proposed new generation plant
which is subsequently cancelled. These potential problems are still another
reason we do not advocate deregulation as in Figure 5.1.1 until detailed analyses
are done.

5. A possible future: deregulation



in a Deregulated World

tion and planning of any electric power system requires forecasts of future

ill pay
lies to
!s due
:0 the
I then

In today's system (or a regulated energy marketplace), all the demand foreand decisions are done by the central utility. Demand and equipment
ty are the critical variables influencing long-term operation (unit com, plant maintenance, and fuel purchase) decisions. For investment
ng, other variables such as future fuel prices, the cost and availability of
labor conditions, and the possible availability of alternative generation
gies are also forecast.
nder the deregulated energy marketplace of Figure 5.1.1, separate forecasts
made by each private generation firm (or purchased from Information
ts) and by the regulated T and D company. Private generation firms
their long-term operating decision on forecasts of future spot price patterns
the appropriate time span. These spot price forecasts replace the demand
unit availability forecasts used in the regulated case. These firms' investment
are also based on a similar spot price behavior foreast in addition to
variables used in the regulated case (fuel prices, capital prices, and so on).
One advantage of the deregulated case is that many different, independent
and subsequent decisions are made. Some private generation compago bankrupt because of their forecasting errors, but others will realize
profits. 7 Thi~ is more desirable than the centralized case, where a single
's forecasti!lg errors can influence all of the generators in a given
. Moreover, price feedback makes spot price patterns easier to predict.
, today we do not know whether the errors in long-term
of spot price behavior will have a larger or smaller impact on operating
investltlent decisions than the forecasting errors of the central regulated
This'is still another reason we are not advocating establishment of the
energy marketplace of Figure 5.1.1 at the present time.
and Long-Term Contracts

electricity is a risky business. At present, regulatory action or

in response to changes in demands and costs, increases these risks to
regulated utility. Under the deregulated marketplace of Figure 5.1.1, the total
are the same. However, the risks borne by private generators and users are
Long-term fixed-price-fixed-quantity contracts permit private generators to
some of the risks of power generation. 8 Such contracts are purely financial
;"'~t ..".",
, just like commodity futures contracts in agricultural and metals
They may be purchased by producers, customers, or speculators. The
Broker of Figure 5.1.1 can expedite such transactions by bringing buyers
sellers together. These contracts do not affect the efficient operation of the
energy marketplace since they are based on fixed quantities of energy at


I. The energy marketplace

NUMERICAL EXAMPLE. Assume Firm A sells to Firm B a contract for 200 M

of energy at $50 per MWh. Each hour, the two firms close out that hour'
contract by comparing the actual spot price with the contracted forward pric
If the spot price turns out to be $60 per MWh, Firm A pays Firm B $2,0000
(60 - 50) x 200. Owning generation equipment automatically puts any fir
into a long position in electricity. Generators can hedge this long position b
going short with futures contracts, i.e., taking the role of Firm A. Conversely
a user can hedge its implicit short position by going long, i.e., buying contracts
For example, if Firm A owns a 200 megawatt coal-fired power plant, it migh
want to lock in the $50/MWh price to satisfy lenders. When it signs the futu
contract, its net revenues each hour are then composed of two parts. One is a
financial gain or loss from closing out the long-term/futures contract. This gain
or loss is almost completely independent of how much the firm generates. 9 The
other component is its revenues earned by generating electricity, which depend
on the actual spot price paid by the T and 0 company minus its variabk
operating costs. If the spot price for a given hour is too low, the firm will not
generate, but still gains revenues from the futures contract. If the spot price for
a given hour is only 1O$/MWh, B pays A $8,000, or (50 - 10) x 200, regardless of how much A generates or how much B consumes that hour. Generator
A can only sell its output for 10 $/MWh. If this is less than its incremental
generating cost, it will shut down for the hour. Therefore, although it earns
nothing from generating electricity that hour, the $8,000 gain on its contract
hedges A against the change in the electricity price. Conversely, B can buy
energy for only 10 $/MWh for that hour, but it pays out $8,000 in addition to
the value of its actual electricity use. If the spot price is high, Firm A generates
200 megawatts and makes larger operating profits. However, it loses on the
futures contract and in effect, gets to keep only 50 $/MWh generated. The net
effect is that A's decision to generate and B's decision to use electricity depend
only on the spot prices. The T and 0 company need not be concerned that Firms
A and B have a futures contract between them. This is very different from
fixed-price, variable quantity long-term contracts which can disrupt efficient
operation of the power system. IO
A perceptive reader has undoubtedly already observed that these long-termj
futures market transactions in the deregulated world of Figure 5.1.1 look like
the long-term market transations discussed in Chapters 3 and 4 for the regulated
energy marketplace.

Regulation in a Deregulated World

The deregulated energy marketplace of Figure 5.1.1 still has the T and 0
company, which is subject to regulation. 1i This regulation is done pretty much
as in the present system except that only the rates the T and 0 company charges
to transmit and distribute the energy are regulated (e.g., the revenue reconciliation mUltiplier is set to recover network capital costs). The regulatory commission does not directly set the rates that the users payor that the generators

5. A possible future: deregulation


regulatory commission maintains incentives to encourage the regulated

o company to operate and plan efficiently. These management concerns

to regulating today's vertically integrated companies.

important issue remains. What would prevent regulators from reimposprice controls through the back door, by limiting how much the T and 0
is allowed to pay for purchases? Paradoxically, a property unique to
power systems makes it very difficult to impose such limits. Electricity
most other commodities because supply and demand must always be
balanced. A deviation of only a few percentage points for a very short
can cause a total blackout. Thus across-the-board price manipUlation
extremely conspicuous because the T and 0 company is forced to
involuntary power rationing, or do nothing and watch the whole power
black out. Either alternative leads to strong public pressure on regulators
,,"plpl't,vp price discrimination, however, is not as easy to prevent. Suppose
forced the T and D company to offer a price lower than the spot price
private nuclear unit. The unit still keeps generating as long as this price is
its marginal operating cost. This can be avoided by having the T and D
and each private generator sign contracts that rule out price discri. Such most favored nation type contracts are found in other commodWith this protection, the regulators are obliged to bide by the forces
and demand.
any industry, of course, legislators could override contracts and
regulation: If potential investors though such legislative action was a
~possibility, private investment would be retarded or even prevented.


5. fl, 5.2, 5.3 addressed the deregulated energy marketplace of Figure

. In this section, we hypothesize a scenario of events (happenings) wherein
regulated structure can move toward and possibly eventually encompass
complete generation deregulation of Figure 5.1.1. Table 5.4.1 summarizes
four steps of this scenario.
I: The Regulated Energy Marketplace

I of the scenario is the topic of the remainder of this book. We strongly

its implementation. We have highlighted our conclusions that there are
similarities between the regulated energy marketplace of the rest of the
and the deregulated system of Figure 5.1.1. However, important differenexist, such as
In the regulated marketplace, users buy energy generated (or purchased) by
their own utility and pay directly for the generation capital costs via revenue
In the deregulated marketplace of Figure 5.1.1, customers effectively shop
around to buy energy from the cheapest available generator (with network

124 I. The energy marketplace

Table 5.4.1. A Scenario that Might Lead to Deregulation

Step I: Establish the regulated energy marketplace of this book. Users and private generators
spot price based rates which incorporate both generation and network revenue reconciliation.
Step II: Establish mandatory wheeling based on the wheeling rates of Chapter 9 which incorporat
both generation and network revenue reconciliation.
Step III: Allow users and private generators to renounce their obligation to be served.
Step IV: Wait and see what happens.

costs added) without explicitly worrying about generation capital costs. The
regulated T and D company acts like a middleman in the transactions.
Steps II and HI of Table 5.4.1 can remove these differences.
Step II: Mandatory Wheeling

Wheeling is the transmission of electric energy from a buyer to a seller through

transmission and/or distribution lines owned by the wheeling utility. It is a
common practice in the present system when the buyer and seller are other
utilities. Step II makes wheeling mandatory and allows private generators and
users to play buyer and seller roles.
Step II is a major change from present-day wheeling because it involves
wheeling rates based on the spatial variation of the spot prices. The wheeling
rate from point A to point B is the difference between the spot prices at points
A and B. As discussed in Chapter 9, the impact of including the generation
revenue reconciliation component into the spot prices that define the wheeling
rates can be very dramatic. Generation revenue reconciliation can dominate the
wheeling rates most of the time.
At present time we are not ready to advocate the implementation of Step
II. More study is needed to make sure that mandatory wheeling based on spot
prices does not lead to horrendous regulatory problems, prohibitive transactions costs, etc.
Step HI: Renouncement of Obligation to be Served

The wheeling rates of Step II use spot prices which incorporate revenue reconciliation for both generation and the network. As a result, a private user, for
example, would not find it advantageous to buy from a private generator or
other utility solely to escape large generation capital costs of his/her own utility
(say resulting from a new nuclear plant entering the rate base). The underlying
logic is that as long as the user's own utility has an obligation to serve the user,
the user has an obligation to help pay for generation capital costs. A similar
logic applies to private generators from which the utility has an obligation to
buy. (See also discussions in Section 9.5).

5. A possible future: deregulation 125


III of Table 5.4.1 establishes procedures which allow a user or private

to renounce the obligation to be served. Such renouncing would
be accompanied by payments from the user to the utility or vice versa
;>en,C1lflj!; on whether the local utility has over- or under generation capacity .
.....,''<It ... h', a long advance warning (say five to ten years) of the intention to
could be required. When a utility no longer has an obligation to serve
user or private generator, such users or generators become free agents
can buy/sell from other free agents or regulated utilities at wheeling rates
incorporate only the effect of network capital costs; i.e., generation capital
are not included. They can buy from wherever the operating costs are
(subject of course to the network costs due to losses, finite network
and network capital costs).
concept of renouncing an obligation to be served has been neither
out in any detail nor analyzed. Thus at the present time, we advocate
study, not immediate adoption. Renouncing of obligation to be served
a host of interesting questions which may not have nice answers.
, it is much easier to deal with than the fully deregulated marketplace
Figure 5.1.1.
IV: Wait and See What Happens


>n to

result of Step III is a mixed system which combines regulated utilities (with
own customers whom they have to serve) and free agents, users and private
Step IV Ls to wait and see how many users and private generators
to enter the free market and what affect such actions have an overall
s.xstem behavior, etc.
Sometime during this wait-and-see period it could become necessary to introduce the~ynamic pricing discussed in Section 5.2. However, as discussed
, it c~uld also come into existence during Step I, i.e., as part of the
regulated energy marketplace.
If free agent status becomes popular and has desirable effects, the regulatory
commission (or legislature) might take further action such as discouraging (or
preventing) the regulated utility from building more generation. This would
eventually lead to the deregulated marketplace of Figure 5.1.1. The regulatory
commission might even go all the way and have the regulated utility sell off its
existing generation to the highest bidders (which introduces another host of
questions for which answers do not exist today).
The main advantage of the four-step scenario of Table 5.4.1 is that one is able
to play court to deregulation for a long time before committing to a binding
marriage. It is also possible to achieve many of the potential advantages of
deregulation without full implementation, i.e., living with a mixed marriage of
both regulated and deregulated participants.

As stated many times in this chapter, we do not advocate establishment of the

deregulated energy marketplace of Figure 5.LI at the present time. There are

126 I. The energy marketplace

simply too many unanswered questions. We included this chapter in the

because the ideas are fascinating and further analysis may prove them to
Going all the way from today's system to the deregulated energy
of Figure 5.1.1 in one step is a mind-boggling concept. Fortunately for
who favor deregulation, there is a scenario which leads towards deregulation
a reasonable step-by-step process.

There are a host of papers and reports for and against deregulation, with
as many definitions of what the term means. Since deregulation is really
side excursion on the main journey of this book, we choose not to provide
detailed guide to the literature. The ideas of this chapter result from a """''''''''6
of some of our earlier deregulation papers. (See the Annotated Bibl,roo,rll',\hv
section on Deregulation and Wheeling). Many other issues in those papers
not covered here.
The deregulation concept of this chapter is based on a supply and demand
marketplace. Most of the other electricity deregulation literature is oriented
only to the supply side i.e., to deregulating generation without altering the way
users buy electricity. We believe that deregulation which considers only the
supply side of the supply-demand equation is dangerous and could have very
negative results.
A second major difference between this chapter and most of the rest of the
deregulation literature lies in our concern that the economics and physical
security of power systems not be destroyed or compromised. Many other
deregulation scenarios bypass issues such as those discussed in Appendices A,
B, and C. If they examine issues such as plant dispatch at all, they propose that
a central purchasing agent will coordinate the deregulated generators, telling
them when and how much to generate. Long term purchase contracts with
variable quantities and fixed prices are advocated. We believe that such arrangements give all the power to the central purchasing agent, and quickly reduce the
"deregulated" generators back to the status of virtual utilities, with little incentive for innovative behavior. Issues such as responsibility for outages and for
spinning reserves would be very complex in this scenario, and the resulting mess
of contingent long term contracts would have few advantages compared with
present regulation. The new proposal for deregulation in the U.K. may provide
some evidence about this, since it seems to have some of these characteristics.
It is also interesting to compare the situation in electricity with that in natural
gas. Wellhead price deregulation replaced years of strict price regulation. It lead
eventually to considerable deregulation of demand, and to the rise of spot
markets for natural gas.
Our earlier papers on deregulation contain further discussion of these issues
and a discussion of partial deregulation. For example, the present regulatory
system could be maintained but with free entry allowed into generation by
anyone willing to be paid the spot price.


S. A possible future: deregulation


far the most comprehensive reference on non-spot price deregulation

Is is Joskow and Schmalensee [\983]. While their book is often spoken
"anti-deregulation" and we are often called "pro-deregulation", neither
is correct. To quote the last paragraph of their book,
lation is to playa role in helping to improve the efficiency with which electricity
uced and used, it must be introduced as part of a long-term process that also
regulatory and structural reform ... We can begin this process [of experi""",'''''JII in a number of dimensions, including pricing and wholesale power transacnow without resolving what are currently irresolvable debates about the comadvantages of competition and regulation, while preserving the greatest opporassociated with both competition and effective, enlightened regulation.

agree. Experimentation is appropriate now, while drastic change is preWe also agree that reforms of pricing methods (spot pricing) will have
bigger beneficial effect than deregulation could, at least in the next one
two decades.
Finally, we have a substantive disagreement with the implicit assumptions in
Joskow/Schmalensee book, and in much of the rest of the deregulation
. The issues is the willingness of private companies to build power
which would sell into a spot market, without a firm "take or pay
1." This is closely related to economies of scale in generation, since it is
likely for small plants than large ones. We expect that plants costing
a few hundrell million dollars will be quite viable, and not suffer
ISe:O'nomles of scale compared with 1000 MW, billion dollar plants. Although
large plants look good on paper, they are inflexible, take a long time to
(for a variety of reasons), have poorer reliability records, and impose
securi~y costs on the rest of the system. [Ford, 1985] In a full spot
tplace, these liabilities would be reflected in the spot price paid to them.
any industries today are characterized by plants costing several hundred
dollars, which are built despite no guarantees about a market for their
and we expect that electricity could/would be the same.
The real resolution of this issue is an empirical question. It can actually be
today, without much deregulation. State utility commissions have the
power to allow a utility to pay spot prices to any private firm that wants to build
a generator and sell to the utility. We hypothesize that if such offers are made
(with appropriate protection e.g. an enforceable promise that the prices will not
be reregulated once the plant is built), private firms will come forward to do this.
See Bohn et al. [1984] for additional discussion.

1. Sales directly from generators to users are not forbidden. However, when spot prices are
properly calculated, they are irrelevant; customers and generators always do at least as well by
dealing only with the T and D company.
2. Precisely defining how many is too many is a difficult task that depends on overall system


l. The energy marketplace

3. Berger (1983) lays a foundation for such research.

4. This is oversimplified, as it ignores ramp rates, minimum downtimes, startup costs, and
changes in a unit's heat rate at different output levels.
5. A real planning study would of course use forecasted, not historical, values.
6. In fact, the unit might not be committed for this week, just as a central utility would not
7. Bankruptcy will not in and of itself disrupt the operation of a generating unit. If the
it can produce is more valuable than the fuel it uses, it will continue to operate, albeit under
8. Another protection against this is geographic diversification. A single firm can own
units in many regions, thus diversifying the risk of sustained regional price fluct
9. Under perfectly competitive assumptions, or if the generator is very small relative to the
the gain or loss is completely independent. If the plant is large or the market is imperfect,
firm could effect the spot price.
10. As discussed in Appendix B, utilities presently participate in a variety of long-term
among themselves, either as independent utilities or as members of a pool.
long-term contracts are much more like the contracts being discussed here than the hYPri . nriN>,
variable quantity contracts often used for present-day PURPA generators. This should
II. The regulatory commission may also oversee the futures market, and prevent anyone "Pt,,.,,.tir.a
company from obtaining monopoly power in any geographic region.



t pr~sented the basic ideas without burdening the reader with mathematical
Part II presents the explicit concepts and equations which enable
calculatiQn of hourly spot prices and associated energy marketplace transacPart n~tries to prevent mathematical manipulation from obscuring the
concepts. Nevertheless, a lot of complex-appearing equations loaded with
symbols and sub- and superscripts will be found. Some but not all readers
be happy to see such equations replace the handwaving of Part I.
mathematical theory of the most general case could be presented in one
mathematically beautiful step. However, we have chosen a less elegant
,....."'''h which starts with simple cases and proceeds sequentially to more
formulations. This step-by-step approach takes more pages but should
hope) make it easier to understand what is going on. The sequence is as

Chapter 6 considers only generation costs, without revenue reconciliation.

Chapter 7 adds the network costs using an explicit network model where
Kirchoff's laws play an important role.
C"' 8 addresses revenue reconciliation.
Chapter 9 presents the theory for evaluating spot price based transactions.
Chapter 10 addresses investment decisions in a spot price based energy



This $apter assumes a simple model of the overall utility-<::ustomer system



Ignores transmission and distribution network costs

Ignores revenue~reconciliation
Aggregates all individual customers into a single demand
Aggregates all individual generators into a single generator

The model's simplicity makes it easier to understand the basic concepts underlying the hourly spot price.
Section 6.1 starts off with the simplest of all situations, which considers only
fuel and variable maintenance costs assuming there is always enough capacity
available. This leads to the "system lambda" component of the hourly spot
price. Sections 6.2 and 6.3 then address the impact of capacity constraints,
which leads to the generation quality of supply components. Two approaches
are presented, a cost function approach in Section 6.2 and a market clearing
approach in Section 6.3. Section 6.4 presents a side discussion on using prices
to "dispatch generation" as well as demand. Section 6.5 concludes with a very
important discussion on the multiple time-period couplings of supply costs and
demand responses. Since such multiple time-period effects greatly complicate
the notation, we usually do not explicitly include them in the notation of this
book. However, in practice they can rarely be ignored.

132 H. Theory of hourly spot prices


In this section, the hourly spot price is discussed considering only generation
operating costs associated with fuel and variable maintenance and assuming
that there is sufficient generation capacity available so that demand never
exceeds supply. This unrealistic assumption is removed in Sections 6.2 and 6.3.
t: Time index in one-hour steps

1,2, ...

j: Generator Index

1,2, ...

gj.max(t): Maximum possible output of jth generator during hour t (kWh)

git): Actual output of jth generator during hour t (kWh)
gmax(t) = kjgj.max(t): Maximum possible generation during hour t (kWh)
g(t) = Ljgj(t): Actual total generation during hour t (kWh) 0 ~ g(t) ~ gmax (t)
k: Customer index

k = 1,2, ...
dk(t): Demand of kth customer during hour I (kWh)
d(t) = Lkdk(t): Total demand of all customers during hour t (kWh)

These variables are such that

o gj.max and hence gmax (t) are random processes in time t because of generation

equipment outages.
dk(t) and hence d(t) are random processes in time t because of random
phenomena such as weather.
gj(t) j = 1 ... are functions of the and des) s = T 1 t ... T2 as
determined by economic dispatch and unit commitment logics which the
utility uses to minimize its operating costs. T2 - TI is the time interval
considered by the unit commitment logic.

An extremely basic assumption used in this chapter is

The spot price pet) for hour t is specified assuming the values of the random processes
des), gj.max(s), j = I ... and s = T 1 t are known.

The one-hour update spot price discussed in Chapter 3 is specified at the

be,ginning of an hour to hold for all of that hour. Hence, the basic assumption

6. Generation only 133

of this chapter is that predictions for one hour into the future are perfectly
accurate. Chapter 9 discusses the impacts of removing this assumption.
Gj,F[git)]: Fuel cost of jth generator with gj(t) kWh of output during hour t ($)
Gj,M[gj(t)]: Maintenance cost of jth generator with gj(t) kWh of output during
hour t ($)
Gj .FM [gAt)] = Gj.f[gj(t)] + Gj.Mfgj(t)]: Total fuel and maintenance cost of jth
generator ($)
GF[g(t)] = LPj.f[gj(t)]: Total fuel cost of all generators with total output get)
during hour t ($)
GM[g(t)] = LjGj.M[g/t)]: Total maintenance cost during hour t ($)
GfM[g(t)] = Gffg(t)] + GMfg(t): Total fuel and maintenance cost during hour
t ($)

A more precise notation for GFM [get)], etc., would be GfM fgj(t) j = I ... ], but
in this chapter all generators are assumed to be aggregated into a single equivalent generator.
Assume the total cost (ignoring capital) of providing electric energy is given



Theit'"the ~ourly spot price for the kth customer is given by

a(Total Cost)
Pk(t) =

M, (t)

In this chapter only an aggregate demand d(t) is considered, so there is only one
hourly spot price given by
(I(Total Cost)
p( t)


iJd( /)

A key assumption made in this section (and in Sections 6.2 through 6.4) is that
there are no multiple time-period couplings of the costs. Thus g(s) is independent of d(t) t =I s. This yields
(t) == DG FM [get)]



Multiple-period coupling of costs is discussed in Section 6.5.

A constraint that is required by physical laws and therefore imposed throughout this book is:

134 H. Theory of hourly spot prices

Energy Balance Constraint: Total energy generated during hour t has to equal total
amount used.

For the present case, losses are ignored, so the energy balance constraint yields
get) =



Thus (6.1.2) becomes


aG rM [d(t)]



Substituting (6.1.1) into (6.1.4) yields

pet) =




YFCt); Fuel component of hourly spot prices

() _
YF t -



YM (t); Maintenance component of hourly spot price


Fuel and maintenance components are combined into system lambda, where

System Lambda






so that (6.1.5) yields




As the model becomes more realistic in subsequent chapters, additional

components will be added to the equation for the hourly spot price pet).
However, the definition for A(t) (i.e., YF (t) + YM (t will not change.
Optimum Properties of pet)

A first course in basic economic theory often derives the fact that marginal cost
pricing is optimum in a social welfare sense. We will now repeat the derivation
which proves that (6.1.9) is optimum. It is important to understand the principles in this simple case.

6. Generation only


B[d(t)]: Total benefits all customers receive from using d(t) kWh of electrical
energy during hour t.

Social Cost I

GFM [g(t)] - B[d(t)J.


Hourly Spot Price p(I): Rate charged to customers which minimizes the social
cost assuming optimum customer behavior.
Customers exhibit optimum behavior if d(t) is chosen to maximize the difference
between the customers' benefits and their bills, i.e. to maximize



Thus d[p(t)] is determined by the condition 2

= pe,)



Substituting d(l) = d[p(I)] into (6.1.10) yields
Social Cost = C1ML!;(t)] - B[d[fl(t)]]




At~Hs point, it would be easiest to substitute g(l) = d[p(t)] and proceed to

differentiate social cost ,x,ith respect to p(t). However, we will follow a more
roundlfb~utapproach that is more representative of the procedures used in
subsequent chapters. We first form the Lagrangian

/l,(t): Lagrange multiplier introduced for the energy balance constraint d(t) = get)

and then find the optimum values of both g(l) and p(I). Setting


yields, using (6.1.8) and g(t)

11,(1) =







l ( )
c t

OB{d(t)J] Od(t) =






Using (6.1.12),

= A(t)


which is (6.1.9).
The derivation of (6.1.16) differentiates the Lagrangian net) of (6.1.14) with
respect to g(t) and p(t). An alternative approach (that yields the same result) is
to find the optimum get) and d(t) by setting

an(t) =

to yield (6.1.15). The argument that leads to (6.1.12) can thus be viewed as a way
to introduce prices p(t) in a way which causes the customer to choose a value
for del) which satisfies the social optimality conditions for d(t). We mention this
alternative approach because later on, when the problem formulation gets more
complicated, we will sometimes use it. It is best to understand the basic ideas for
simple problems before the notation starts to get messy.
Solution for p(t):

The equation p(t) = A(t) is not a closed-form solution. In general, A(t) is a

function of demand del); i.e.
A(t) =


and demand is a function of price, i.e.



Thus (6.1.9) is really

p(t) =


i.e., it is an implicit equation for pet) which must be solved.


6. Generation only


The solution of (6.1.17) for pet) is usually viewed in the economic literature
as the intersection of the supply and demand curves (see Figure 2.10.1).
Effect of Purchase-Sales From-To Other Utilities



Many utilities purchase and/or sell electric energy from/to neighboring utilities
via transactions ranging from long-term multiyear contracts to hour-by-hour
economy transactions. The cost of purchases and profit of sales are explicitly
included in the GFM [g(t)J cost function. Thus in many situations, },(t) is directly
determined by purchases and/or sales and only indirectly by the utility's own
variable fuel and maintenance costs. This phenomenon is of particular importance for a utility with only hydro generation.
In most subsequent discussions in this book, we talk about A(t) as if it is
determined directly by the utility's own explicit fuel (and other) costs because
it simplifies the discussion. However, at various points, we reintroduce the fact
that purchases and sales cannot be neglected.


Section 6.1 considered only fuel and variable maintenance costs. An additional
term due to quality of supply (i.e. ability of generation to meet demand) is now
add~>to the cost function.
CG1lsider a utility with demand d(t) and maximum available generation gmax (t)
duri~g hour t. If d(t) >~ gmax (t), the system will black out. Therefore a utility will
take'lextreme measures to prevent such a condition from occurring Define
gerlt.l'U): Critical Generation Level: When g(t) = d(t) approaches gerit.i,(t), the
utility takes~measures to prevent g(t) from exceeding gerit,y(t).J
geril.l,(t) = gmax(t) - gres(t).
gres(t): Operating reserve requirements. Can be a function of gj(t) j = 1 ... and

The measures a utility starts to take to prevent get) from exceeding gcrit.y(t) cost
money. Define
GQS[g(t)]: Generation quality of supply costs incurred to provide reliable energy to
customers during hour t; i.e. to prevent g(t) from exceeding gcrit.,(t) ($).

I'Q5(1): Generation quality of supply component of hourly spot price.





Thus for the present case, the hourly spot price p(t) of Section 6.1 expands to




Three possible ways to quantify GQsfg(t)] are



Costs of emergency purchases

Costs of load management unserved energy
Allocation of annualized cost of peaking plant

Costs of Emergency Purchases

A utility that is interconnected with its neighboring utilities may be able to buy
emergency energy over the transmission tie lines. In some cases, a utility can also
buy additional energy from customers with emergency back-up generators or
cogeneration systems. Hence one way to quantify the GQs[d(t)] costs is in terms
of such purchases.
In actual operation, utilities are often buying and selling energy from each
other even ifno one utility is near its own generation capacity. In such cases, A(t)
includes the effects of such transactions. Hence the YQs(t) resulting from emergency purchases can be considered to be part of A(t) also. We choose to take this
point of view in the rest of this book. Thus in all further developments we
assume that
Costs of emergency purchases are included in A(t) and are not part of I'Qs(t).

Costs of Load Management/Unserved Energy

When g(t) > gcrit,y(t), a utility can exercise load management to reduce load
instead of, or in addition to, buying emergency power. Ways to achieve such
load management include

Interruptible contracts
Direct load control
Reduction in utility house load
Rotating blackouts

Such load management introduces direct costs to the utility, and/or the
These load management-unserved energy costs can be quantified by
GQs[d(t)] =


uy(t): Unserved energy during hour t due to generation shortage (kWh)


6. Generation only



u;.(t) =

d(J) - gerit., (t)

d (I) > gcrir.,. (t)



Cost of unserved energy (/kWh)

This yields

OQs;(I)b, (I)





get) > gcrit.;(t)

b,.(t) =






The value of 80s.y(t) in general depends on u)"(t).

If uK(t) in (6.2.4) is replaced by UE(t), the expected unserved energy (see
6.2.16), then (6.2.17) yields
YQs(t): 0os.;(I) LOLP,(I)
LOLPg(t): Loss of Load Probability at hour t


The modeling of the cost of unserved energy is a nontrivial exercise.


Cost of Peaking Plant (Annualized)


co~j,uter studies with human judgment has arrived at a future generation

Assume the utility has gone through its planning process and by combining

conStruction plan. The expenditures associated with generation capacity

addrtions needed to maintain system reliability at some specified level can be
used to calculate

A Qs .y : Annualized cost per kW of the cost minimization generating technology which

must be install~d to meet generating system reliability requirements ($/kW/year). The
equivalent yearly cost of capital depreciation plus interest which depends on interest
rates, inflation rates, and assumed plant lifetime.

The calculation of a value for AQs .y can be done conceptually by use of generation expansion computer programs that look many years into the future. This
approach requires a lot of computation and is subject to the effects of forecasting errors in future load growth, costs, etc. (see Chapter 10). A more pragmatic
approach is to define AQs,y in terms of the annualized capital cost of a peaking
plant such as a gas turbine,
Given AQs,y, it is reasonable to quantify the generation quality of supply costs

L GQS[g(t)]


KQs .)": Generation plant addition made to maintain reliability of generation supply

Denoting by ,1. incremental changes yields

140 II. Theory of hourly spot prices

YQs (I) -

aGQsfg(t)] _

or, since g(t)

YQs(t) =

AQS.;. !l.g(l)


I:!.KQs .;
AQs .; I:!.d(l)

In order to proceed further with (6.2.7), it is necessary to model the effect of

a change in d(t) t = 1 ... 8760 on KQs .y' One general structural form for such a
model is

I:!.KQs .;


L a(l)



Allocation function

Examples of possible allocation functions follow.

Peak Demand
a(l) =

d(l) = dmax


i.e., K Qs .r is tied directly to the maximum demand dmax

A verage of Large Demands

Define dmax ." to be the nth largest demand, n

dmax.N-1 > d(t), all other t


d(l) > dmax N


0,... N - I; dmax > dmax . I '"

This average model can be extended to a weighted average if desired.

Loss of Load Probability


6. Generation only


LOLPy(t): Loss of load probability due to generation, i.e., probability that

d(t) ~ gcril.y(t) during hour t
LOLHy: Expected annual loss of load hours (generation)



Further discussions in this book usually use the loss of load probability
allocation function of (6.2.11), i.e.,

However, it is important to emphasize that there are many other equally

reasonable approaches.
Note that the 'Yos(t) of (6.2.12) does not change if LOLPy(t) is multiplied by
any constant that does not depend on t since the same constant will then
mUltiply the denominator also. Thus it is the shape of LOLP(t) versus time (or
rather get) = d(t that is important, not its absolute magnitude.
Derivation of (6.2.8) to (6.2.12)

The a!location function~' of (6.2.8) through (6.2.12) are reasonable. For those
who til<e mathematical l!1anipulation, we will now present a derivation.
Assllme the generation plant addition Kos.y is chosen to satisfy a constraint of
the form

Cy =

Cy[d(t)'~- KQs,yl ::; Specified Value


1= I

cy[d(t) -

KQs,y]: Function which depends on del) - KQs .y and other variables

Assume the constraint of (6.2.13) is active (binding). Then to maintain it, incremental changes in d(t) and Kos.y must satisfy
t = 1 ... 8760


Because of the special structural form assumed in (6.2.13), (6.2.14) yields

8~ oc;-ld(t) - KQs.;l f::.K


1 .... 8760


KQs., 1

142 II. Theory of hourly spot prices


IlKQs y


aCy[d(t) - KQs .y]



L a(t)


The special cases of (6.2.9), (6.2.10), and (6.2.11) result from (6.2.15) by

Peak Demand: (6.2.13) is


KQs . ~ (Installed Capacity) - (Desired Reserve Margin)

Average of N Largest Peaks: (6.2.13) is

I( N(dma.

dma 1

+ ... dmm

N - I - KQs .,

(Installed Capacity)

- (Desired Reserve Margin)

Loss of Load Probability: (6.2.13) is

AVEy ~ Specified Value
AVEy: Annual Vnserved Energy
AVEy = L~~6? VEy(t)
VEy(t): Expected unserved energy during hour t (kWh)

Assume KQs y is nonzero but small and that K Qs y is perfectly reliable generation
(i.e., is always available when needed) and assume that hourly future demands
are known. s Then

[d(t) _ K .y - gcrit.y(t)]


p[gcrit.y (t)]d[gcrit.1 (t)]

gcrit.x(t): Available generation capacity during hour t
p[gcrit.;.(t)]: Probability density of gcrit.,(t)

The partial derivative of (6.2.16) with respect to d(t) yields

d(t)-KQS .)

a(t) = aUE,.(t)





6. Generation only


Evaluation of LOLPy(t)



The reader should notice that LOLPy(t) is a probability and hence can take
values between 0 and I. If LOLPy(t) is estimated at the beginning of hour t, all
uncertainty affecting the system during the hour is known according to the
assumption of Section 6.1. Hence, if calculated at hour t, the value of LOLPy(t)
as derived above is either 0 or 1, because gent.y(t) g is known and
UEy(t) = del) - gerit.y(l) if del) > gent.y(t), zero otherwise. However, as discussed in Chapter 4, a more pragmatic approach is to use a smoothly varying
LOLP)'(l). Several methods are discussed in Chapter 4, one of which is repeated
here, namely
k[d(t)] - gcnt.;.(t)

LOLP.(t) =


d(t) >

gent.;. (t)



t\g: A value chosen by engineering judgment

k: A constant whose value doesn't effect )'Qs(t)

The assumption that all relevant uncertainty is revealed at the beginning of
each hour is removed in Chapter 9.



6.2, the generation quality of supply component was quantified by

adCting a term to the C'ost function. In this section a different approach is taken.
I;)emand d(1) is assumed to depend on price p(/) and we simply set
pet) =



(6.3.1 )



I'Q8(t): Chosen such that get) = d(t) never exceeds gent.,(t).


This is called the market clearing approach because when the demand del)
approaches the critical level gerit.y(t), the value of I'Qs(t) is raised until the market
clears,6 i.e., until demand reduces enough.
Mathematical Derivation

In order to derive (6.3.1) (6.3.2), the hourly spot price p(/) is defined as in Section

6.2 to be the price which minimizes social cost

Social Cost =

GFM [get)] - B[d(t)]

However the minimization is now subject to both the constraint get)

the additional constraint


d(t) and

144 II. Theory of hourly spot prices


The minimization is done by introducing the Lagrange multiplier /lQS,y(t) to

yield the Lagrangian (which is (6.1.14) plus an extra term)
GFM [g(t)] - B[d(t)]

!l(I) =

+ Ite(t)[d(t)

- g(t)]


ItQs.,(t)[g(t) - gerl,.y(t)]

As in Section 6.1, /le(t) is determined by finding the optimum get) and is given

Ite(t) =


Using d(t)



d[Q(t)] and setting

yields, after using the chain rule,

Ite ()

_ oB[d(t)] =


Assuming optimum customer behavior, i.e. (6.1.12), yields





where the generation quality of supply component I'Qs(t) is the Lagrange multiplier

I'QS (t)

If g(t) < geri,./t), then the constraint (6.3.4) is automatically satisfied, so

I'Qs(t) = O. If the demand tries to exceed gcrit.;,(t), fQS(t) increases to reduce the
Use of Linear Demand Response Model

In order to explicitly evaluate I'Qs(t) of (6.3.8), it is necessary to assume a

demand response model, i.e. a model for how d(t) varies with pet). Define

doer): Demand that would exist if p(t)

ACri,(t) = A(t) when get) = gerit.;(t)

= ACri' (I).


Assume a linear demand response model (other forms are discussed in Appendix

6. Generation only

d(l) =

d,,(I) {I


[P(I) -

Aedt (I)



(J(t) is the demand elasticity parameter (a negative number). If do(t) > derit.;.(t),
then it is necessary to set get) = gerit.:.(t) in (6.3.10). This yields
[gmt/(I) - do(t)]A,,,t(l)



A . (I) [gern.y(l) - do (I)]
do (t)!3(I)




10/kWh = 100$/MWh




Then:'6.3.12 yields


- 7f $!MWh
-1.0 /kWh










which illustrates the major impact of the demand elasticity parameter


146 II. Theory of hourly spot prices


In Section 6.3, prices were used as a mechanism to cause the customer to choose
the socially optimum value of demand d(t) while the generation get) is specified
by the utility to meet the demand. A somewhat different perspective on what is
going on leads to the conclusion that the generators can also be viewed as
"price-takers" who "self-dispatch" themselves based on prices.
To see this, consider what happens when the Lagrangian of (6.3.5) is optimized directly with respect to g(t) and d(t) (as discussed in Section 6.1). This

A(t) - JJ.e(t)




JJ.QS,y(t) = 0

JJ.e(t) = 0

OGFM [get)]



Now assume that


Generators are paid at a rate of Pg(t) for generating g(l).

Customers pay at a rate of pAt) for using d(t).

Assume further that the generators self-dispatch themselves by choosing g(t) to


subject to the constraint that

get) ~ gcrit,y(t)

and the customer (as before) chooses del) to maximize

B[d(t) -



Equation (6.4.3) leads to the result that get) is specified by

A(t) - Pg(t)

JJ.QS,y(t) =


while condition (6.4.4) yields



+ pAt)

= 0


6. Generation only


The results (6.4.5) (6.4.6) yield the optimum conditions (6.4.1) (6.4.2), provided

From this point of view, the customers and the generators are making
independent decisions on the values of d(t) and g(t) based on the same prices
(paid by customers and to generators) while the price is adjusted (somehow)
until the energy demand constraint d(t) = g(t) is met. This concept may not
seem to be overly exiciting for the special case being considered here in Chapter
6. However, it becomes more interesting in later chapters when many different
generators are considered and some of them are owned by the customers.

The discussion of the preceding sections used a model that is over-simplified in

many ways, one of which is the consideration of only a single one-hour period;
i.e., the multiple period (intertemporal) dependence of both generation costs
and customer benefits (demand response) was ignored. The more realistic
multiple period case is now discussed.

For most utilities, the marginal cost of generation during any given hour
depends on the generation levels during previous hours and those expected in
futur't hours. Examples of such multiperiod time couplings are startup and
shutdown costs, hydro::with storage, and various types of fuel contracts.
Ins,tead of using (6.1.8), the multiple period definition of }.(t) is given by




E: Expectation over future uncertainty

Time interval T) to T2 : Range over which time coupling of costs occur.

Possible ranges of T2 - T) are

24 Hours: For a pure thennal system.
One Week: For a thermal system with pumped hydro.
One Year: For a system with long-tenn hydro storage.
The calculation of }.(t) from (6.5.1) is now more complicated but the basic
hourly spot pricing concepts are unchanged.


n. Theory of hourly spot prices

Customer Demand Response

The customer demand response model underlying (6.1.l2) did not have any
multiple-period coupling dependence. In practice, the potential of customers to
reschedule electric power usage from times of high price to times of low price
provides such multiperiod coupling. Examples of residential rescheduling are
thermal storage of either heat (in hot water heaters, buildings, or thermal bricks)
or cold (in buildings or ice). Industrial examples include the variations of the
times at which metal is melted, water is pumped, and very large machines are
As with multiple-period generation costs, multiple-period demand' response
complicates the computations but does not change the basic principles. (Section
9.4 is an example of an analysis with multiple-period coupling).

The random nature of demand, cost, and equipment availability is a basic fact
of life in a spot price based energy marketplace. However, these random
characteristics did not playa major role in this chapter because we assumed the
hourly spot price is specified after the values for demand, costs, etc., for the
given hour are learned - i.e., after the random processes have been realized (or
at least predicted perfectly).
Two of the basic components of the hourly spot price were developed:

+ YM(t): The generation fuel, operation and maintenance operating component.

YQs(t): The generation quality of supply component.
A(t) = YF(t)

The generation quality of supply component was quantified in two ways:


As the derivative of a cost

As the market clearing value

where the market clearing result is based on minimization of a social cost

function. The cost-based approach requires the choice of particular cost
function. The market clearing approach required the use of a customer demand
response model whose structural form and numerical values can be the subject
of extensive debate.
For an ideal world, we recommend implementation of the market clearing
approach, since it yields maximum economic efficiency and is conceptually the
"cleanest." However, we have found in discussions throughout the industry a
lot of concern about the practicality of the market clearing approach and more
acceptance of the cost function approach as the "real-world" way to proceed.
We believe (hope) that eventually the market clearing approach will become
accepted but pragmatically recommend that the cost function approach be used

6. Generation only


at least for inital implementation. Two cost function approaches (cost of load
management/unserved energy and annualized cost of peaking plant) were
presented. It is easier to obtain the necessary input data for the peaking plant
In Chapter 7 we will combine both the cost and market clearing approach to
quality of supply into a single formulation so that the reader can always choose
which one best meets his/her needs.

The basis of this chapter is Caramanis, Bohn and Schweppe [1982], although the
formulation used in this chapter is much simpler to make it more understandable. Outhred and Schweppe [1980] discuss quality of supply components from
a more general point of view, albeit with words, not specific equations.
The basic mathematical approach used here (and in Chapter 7) of minimization of a social cost function subject to constraints is classical economics, which
is used by many authors. Our approach differs from most of the other literature
because of the time scale of concern, which, of course, can make a large
difference in practice.
Most of our approaches in Sections 6.2 and 6.3 to quantifying the quality of
supply component are translations of existing ideas on marginal cost pricing to
the one-hour time scale where events are revealed before prices are calculated.
We '}l1le the term "quaHty of supply component" because it seems to be the most
des1$Eptive term available. Other authors use terms such as "capacity compon~nt" or "reliability component."
Mbre detail on the multiple time-period problem of Section 6.4 can be found
in Caramanis, Bohn, and Schweppe [1982]. See also Kaye and Outhred [1988].




I. Social welfare is the negative of social cost.

2. It is explicitly assumed that d(t) is the sum of many independent customer actions so that no one
customer can have an impact on p(t).
3. Impacts of buying and selling from neighboring utilities are also included in g"".;.(t).
4. To derive this result, modify the social cost of (6.1.10) to be GFM[g(t)] + GQs[g(tl] - B[d(!)].
5. This is the approach taken in hour-by-hour chronological production cost simulation models
used widely by electric utilities for planning studies.
6. Market clearing is related but not necessarily equivalent to the cost of unserved energy as
discussed in Section 6.2.
7. In general, g,nt . . (t) depends on the operating reserve requirement g,,,(t) which can depend on
generation levels and patterns. Such dependence is ignored for the sake of simplification. It can
be included if desired.
8. When p,(t) is assumed to be independent of g(t).


Irt:Chapter 6, network effects and revenue reconciliation were ignored. Here

netw01:k effects are introduced.
ct:i'apter 6 started with a simple case and then added complexity. Here the
procedure is reversed. The general case is presented first and then special cases
are discussed" This change of pace helps keep readers from becoming bored.
Chapter 6 d~veloped its ideas first for the single-hour case, and then in Section
6.5 discussed the effects of multiple time-period coupling of costs and response.
In order to keep the notation from becoming either hideously complex or
completely obscure, only the single time-period case is discussed here. Extensions of the network effects to multiple time periods is conceptually straightforward.
Section 7.1 formulates the overall problem and Section 7.2 then presents the
general results and its derivation. Sections 7.3, 7.4, and 7.5 discuss models for
the three network components of the hourly spot price. Section 7.6 presents a
two-bus example to try to clarify the meaning of the general equations. Section
7.7 discusses the price difference between two ends of a line. Section 7.8 discusses
customer owned generation. Sections 7.1 through 7.8 assume an explicit model
for all the network lines is available. Section 7.9 discusses the use of an aggregated network model which does not consider individual lines. Section 7.10
discusses generalization of the results to the case where reactive power and
voltage are important.



II. Theory of hourly spot prices


The power flows over a given line in a network and the losses in that line are
determined by the generation and loads at all buses, not just the buses at either
end of the line.' The line flows are determined by Kirchoff's Laws and depend
on the network's interconnected structure as well as physical parameters such
as impedances.
In general, the solution of Kirchoff's Laws involves both real and reactive
powers combined with voltage magnitudes and angles. This is called an AC load
flow. In most of this chapter we consider a special case (approximation) that
involves only real power flows called a DC load flow. 2 The full AC load flow case
is discussed in general terms in Section 7.9; see also Appendix A.
Generation-Load Definitions

As in Chapter 6 define
g/t): Output of jth generator during hour t (kWh)
dk(t): Demand of kth customer during hour t (kWh)
Ljg/t): Total generation
d(t) = Lkdk(t): Total demand

(7.1.1 )

g(t): Vector of all generation
~(t): Vector of all demands
A given bus can contain a generator, a load, or both a generator and a load.
Define as in Chapter 6 3
GFM [get): Total Generation Fuel and Maintenance Costs
GFM [g(t)] = L,GFM,j[g;(t)]

GQslg(t)): Generation Quality of Supply Costs
get) ~ gcrit,;,(t): Maximum Total Generation Constraint
gcrit,y(t): Depends on maximum available generation and operating reserve
requirements; see (6,2, I)
g)(t) ~ g),max: Individual Generator Maximum Capacity Constraints
B[~(t)J: Total Customer Benefit

Assume that the kth customer acts optimally, i.e. the demand dk(t) is the
function of price Pk (t) defined by

7. Generation and network


Network Definitions

, are

A well-posed network flow problem results by specifying the gj and dk at all but
one bus, which is called the swing bus. 4 We assume the swing bus contains only
a generator and let * denote the value of j which specifies the swing bus
generator. Define


*(1): Vector of all gj(t) except g*(t), the generator on the swing bus.

Thus the dimension of *(t) is one less than the dimension of (t). Define
Zi(t): Energy flows over line i during hour t (kWh) i = I ...
z(t): Vector of all line flows
y(t): Vector of net bus injections for all buses except the swing bus, i.e. the

- differences of bus generations and demands

If all buses except the swing bus contain both generators and loads,


[*(1) - ~(t)

Using Kirchoff's Laws,

Z;[*(I), ~(I), nel>work parameters, network structure)





The De load flow approximation of Appendix D yields




H: Transfer admittance matrix which depends on the network parameters

- and structure

This yields

OZ;(I) _
ildk(t) -




H;k: Appropriate element of !i


where the negative sign results from the definition of y.




L;[z;(t): Energy loss in line j during hour f (kWh)


L L;[z;(I): Total Line Losses




154 II. Theory of hourly spot prices

When line losses are present, the energy balance constraint used in Chapter
6 becomes
get) =


L gj(t)

L dk(t)


L L;(t)



L[::[,[*(t), q:(t)])


L[::[ K(t)]J

Network lines have maintenance costs which depend partly on the line flows.
NM,i(t): Maintenance cost for line i resulting from flows during hour t ($)
NM(t): Total Maintenance Cost
= LjNM,i[z;(t)] = NM[~(t)]
Any line i has a limit on how much energy flow it can handle. In practice, this
limit can vary with time duration of flow, outside temperature, direction of flow,
and conditions at other buses. However, for the present discussion, assume the
power system has to be operated such that
!Zi(t)! :;:; Z"m ..


The more general case just introduces a lot of extra notation. This line flow
constraint is analogous to the maximum available generation constraints
gil) < g},max (t) of (7.1.3). The physical constraint on line flow is not necessarily
a hard constraint (as in (7.1.12 because a small violation will not instantaneously cause the line to fall down or burst into flames. Instead, overly high
flows cause loss of line life and increase the probability of a line failure immediately or in the future, which decreases the quality of the supply the utility is
furnishing its customers. s Furthermore, line flow constraints may be used to
reflect other system operating constraints such as stability. A cost function
which gets large rapidly as the line flows approach or exceed some level can be
used to capture these effects. Define
NQs,;{t): Quality of supply costs of line i during hour t ($)
NQs(t): Total network quality of supply
= Li NQS,;[Zi (t)] = NQs[~(t)]

(7. 1.I 3)

The difference between line maintenance costs NM,i(t) and quality of supply
costs N Qs )!) is not always sharp. We define them separately because in some
applications, a separate point of view is appropriate.

The purpose of this section is to derive for the hourly spot price the following

7. Generation and network


equation, which incorporates all of the cost functions and constraints of Section
Pk(t) =



A(t): Generation Marginal





and Maintenance



for j = '" which is the swing bus which is located

at a generator that is on the margin
YQs(t): Generation Quality of Supply




J.l.Qs.y: Lagrange multiplier arising from the maximum total

generation constraint (7.1.3)

'IL.k(t): Network Loss


+ YQs(t)] a;IJi)



L oL;(z;(t



'IM,k(f): ,Network Maintenance

oNM [z(t)]

L aNM.;[Z,(t)] oz;(t)



'IQS.k(t): Network Quality of Supply

Od~(t) [NQs[~(t)] + ~ !lQs.~,;(t)Z;(t)J


Odk(t) =


(t)] oz;(t)


H;k: Element of Transfer Admittance Matnx

Lagrange multiplier arising from the maximum line flow constraints (7.1.12)

A more concise way of writing (7.2. I) is


+ tlk(t)
+ YQs(t)



n. Theory of hourly spot prices

'1 L.k (t)

'1M.k (t)

'10S,k (t)

L ~ (t) OZi(t)



OzJt) {y(t)Li[Zi(t)]



/lQs.~,;( t)

The '1k(t) can be positive or negative.

In subsequent sections we discuss the various components in detail and try to
provide some insights into (7.2.1). The rest of this section is just the derivation
of (7.2.1).
Generation Revisited

Two approaches were used in Chapter 6 to quantify the generation quality of

supply: Section 6.2 added a quality of supply cost term Gos (t), while Section 6.3
introduced a Lagrange multiplier f.1.os. y (t) to handle the constraint on maximum
available generation. The derivation for both approaches could have been done
simultaneously by forming the Lagrangian
n(t) =

GFM [get)]

GQsfg(t) - B[d(t)

Ile(t)[d(t) - get)]

/lQs.l'(t)[g(t) - gcrit.y(t)]

and differentiating to find the optimum get) and p(t) where d(t)
would have yielded

YQs(t) =

d[p(t)]. This


which is the sum of the two separate results. Then by assuming GQS = 0 or
= 0, either of the two Chapter 6 approaches can
be obtained (actually in some applications it could be desirable to keep both
types of quality of supply terms).
This concept of using both a cost function and a constraint for the quality of
supply is used in the following.

gcrit.,(t) is very large so J1Qs.y(t)

Derivation of (7.2.1)

The first step is form the Lagrangian

net) =


(Generation Costs, Constraints)


(Network Costs, Constraints)


(Customer Benefit)


L[::(t)) - get)]

(Energy Balance Constraint)



7. Generation and network


(Fuel and Maintenance Costs)

+ Godg(t)]

(Quality of Supply Costs)

11os.,.(t)[g(t) -' gcriq.(t)

(Maximum Total Generation


L f.lmaqJ(t)[g/t)

(Maximum Individual
Generation Constraint)

- gma,/t)

L NM,[z,(t))

(Maintenance Costs)

L Nodzj(t)

+ L, f.lOS.'I.,(t)[Zi(t)


(Quality of Supply Costs)


(lndividual Line Capacity



The second step is to set the derivatives of the Lagrangian of (7.2.2) with respect
j = I ... and Pk (t) (or dk (I)) k = I ... equal to zero. This yields








Using (7.1.5), (7.2.4) yields

() =

Pk t



rid, (t)


rid, (t)



Note that if the hourly spot price at a generator busj is defined to be given




-11 ( t )[3L[Z(t)] - - 3N{z(/)]



such a definition is consistent with (7.2.5) provided the sign of gj(t) is changed
to make it a negative load. Thus (7.2.5) and the basic result (7.2.1) apply to all
buses j and k, and the optimality conditions (7.2.3) (7.2.4) can be neatly
summarized by


n. Theory of hourly spot prices


8gj (t)

adk(t) =


The third step is to evaluate lie (t). Assume without loss of generality that the
generator on the swing bus is on the margin so the * is a value of j that denotes
a marginal generator. Then since [*(t) does not contain g*(t},
8z i (t)


8z i [g*(t), d(t)]

i = I ...


and (7.2.3) yields


t) =

8g* (t)

From the definition of G[[(t)] it follows that


..1.(t) =



8GQs [g(t)]

( )
iJos.y t

8GFM .* [g* (t)]


where the fact that g* (t) is on the margin is used to get IImax.y,* (t)
ing (7.2.10) into (7.2.5) yields
Pk(t) =





O. Substitut-


+ ----""-adk(t)

The final step to get (7.2.1) is to insert the definition of N[z(t)] from (7.2.2)
and collect terms.
Note that the Lagrange multipliers IImax,y,) which were introduced to handle
the gj < gmaxJ constraints do not explicitly appear in (7.2.1). Their role in life is
to help determine which generators are at their maximum values.
Although (7.2.1) is a simple-appearing form, a lot of computations can be
required to find numerical values. First it is necessary to solve the optimum

7. Generation and network


generation dispatch problem to determine the marginal generator (location of

swing bus *). Second, many of the right-hand side terms of (7.2. I) depend on
dk(t)k = I ... which in turn depend on Pk(t).
In the next three sections, we develop explicit models for the three network
components of (7.2.1). Then we consider a special case of particular importance
- the two-bus system.

The network loss component of the hourly spot price is given by (7.2.1) as

'( )
[A t





)] '" oL(z;(t OZi(t)

OZi(t) adk(t)


Loss Model: L(z(t

The DC load flow approximation theory of Appendix D yields for the losses in
the ith line

where R, is a constant that depends on the resistance of the line. More accurate
app~ximations can be, used (such as including the effect of no load losses) but
(7.3.2) is satisfactory in most applications. The total losses
L(t) =

L L,(t)

can thus be written

L(t) =



!i: Positive definite diagonal matrix with main diagonal elements Ri

"T" denotes matrix (vector) transpose
Substituting (7. \.7) into (7.3.3) yields 6



I'/L,k(t) as Function of Line Flows

Using (7.3.1) and (7.3.2),





Rizi(t) odk(t)



II. Theory of hourly spot prices

which expresses '1u(t) in terms of individual line flows.

IIL.k(f) as Function of Bus Injections

Substituting (7.3.4) into (7.3.1) yields

- 2["\(t)




kth row of B


The network maintenance component of the hourly spot price is given by (7.2.1)

__ "oNM.,[z,(t)] oZ,(t)


Maintenance Cost Model: NM.ilzj(t)J

Although maintenance costs associated with a given line depend to some extent
on the energy carried by the line, good mathematical models for this dependence
are not known at the present time (by the authors at least). Therefore we will
just hypothesize a structural form which seems reasonable.
The hypothesized form is

NM,,[z,(I)] =



Constant that does not depend on z,(t)

N~ ,[z,(t)] =

Iz,(t)1 <

Quadratic in Iz,(t)1




The structure of NM.j[zj(t)] is based on the assumptions that


There is a threshold effect so that below some critical loading level, maintenance is independent of line loading.
Maintenance is proportional to heating (losses) when line loading is above the
critical value.

I1M,k(f) as a Function of Line Flows

Substituting (7.4.2) into (7.4.1) yields

I1M.k(t) =

L I1M.k.,(t)


I1M.k., (t)

BM.ilz,(t)1 Odk(t)

Iz;(t)1 >


IZj(t)1 <


7. Generation and network


OM,i: A constant ($jkWh)

It is also possible to express 'lM.k(t) in terms of bus injections, but the line flow
form of (7.4.3) is most natural.
Equation (7.4.3) is simple to use. The problem is to specify the constants, 8M ,;
and Zerit,i' Note that in practice, The Nt.i term will often dominate, so care is
needed if historical data are used to try to specify the needed constants.
In general we believe the effect of the maintenance component is usually not
critical. This is the reason it was not discussed in Part I of this book. It is
included in the general theory because it could be important in some applicaSECTION 7.5. NETWORK QUALITY OF SUPPLY: I1Qs.k(t)

The network quality of supply component of the hourly spot prices is given by
(7.2.1) as
(7.5.1 )

In most applications, both network quality of supply terms will not be used at
the same time.


Quality of Supply Cost Model: NQs;(z;(t))

Assume {lQS.'J.i(t)

o in (7.5.1).

A reasonable cost function is

1 z,(t)1

Quadratic in z,

IZ,(I)I >



where the value of the cost for large line flows may vary with the line flow itself.
Comparison of (7.5.2) with the network maintenance cost (7.4.2) shows that
they are very similar in appearance. We discuss the two terms separately because
they model two different types of phenomena: maintenance is associated with
normal operating conditions, while quality of supply is associated with times
when the system's capacity is being stretched. Thus even if identical-looking
equations were used, '1M.k(t) and '1QS.k(t) could behave numerically in quite
different ways.
Market Clearing Multiplier: I'QS,,,i(t)

Assume NQS.i(t) = 0 in (7.5.1). Assume the flow in one particular line, iO,
exceeds a hard limit, The flow in line iO can be reduced by simultaneously
redispatching the generators (to get a new line flow pattern) and by customer
response to changed prices.
Consider first the effect of customer response. Assume only line iO is overloaded. The hourly spot price for the kth customer is

162 H. Theory of hourly spot prices




The transfer admittances H,fJk k = I, ... are positive or negative depending on

whether a change in the kth customer's demand increases or decreases the flow
in line iO. As the magnitude of J1.Qs.~.;o increases, the customers readjust their
usages until the line overload is removed, i.e., until the market clears. This
market clearing action is similar to the market clearing approach to generation
quality of supply of Section 6.3. However one basic difference is
For generation quality of supply market clearing, all customers see the same YQs(t), while
for network quality of supply, each customer sees a different 'lQ,.k(t) as determined by

Customers a long distance (electrically) from the overloaded line iO have a small
(in magnitude) H jOk and hence will not respond. A second difference is that
network quality of supply can be positive or negative.
A third and perhaps even more basic difference between generation and
network market clearing is
The market clearing approach for generation capacity limit always requires customer
response, but customer response is not necessarily needed for network line capacity

For many (but not all) cases, generation redispatch can be used to remove a line
overload even if no customer responds. 7 Conceptually each generator sees a
price (see (7.2.6,

and redispatches itself. Thus in this case, the market clearing value of J1.QS.~.iO is
determined by the generation costs, not by customer response.
Quality of Supply Costs Versus Market Clearing

The difference between the use of network quality of supply cost functions and
the market clearing approach is not basic. Market clearing involves a hard
constraint, while quality of supply costs are penalty functions.

The general equations of Section 7.2 are now solved for special cases involving
a two-bus system. Hopefully this adds insight to the nature of the equations. A
more general three-bus example is provided in Appendix D.

7. Generation and network

Swin9 Bus



G,"""" 2



Lood, d

Bus 2

Bus 1

Figure 7.6.1. Twobus example.

Problem Formulation

Figure 7.6.1 shows a two-bus system with generators at both buses and a load
at Bus 2. Only a single hour is considered, so the time dependence on t is
dropped from the notation. The numbers are chosen so that Bus 1 is the swing
For this example, assume


No maintenance costs
No network quality of supply costs


quality of supply costs

ThO; the quality of supply components are given by the Lagrange multipliers.
F'\Ir most cases to be discussed, the demand d is assumed to be fixed and
independent of price. This assumption is removed for the last case.
The generator's fuel and maintenance costs are modeled by

Thus the marginal generation costs are constant, indepel)dent of generation

level (the three-bus example of Appendix D is more general).
The losses are modeled by

Since Bus I is the swing bus, it folIows from Figure 7.6.1 that





II. Theory of hourly spot prices

The assumed numerical values are


5 /k Wh =

50 $/MWh

10/kWh =



The value for R corresponds to a 5% transmission loss when


0; i.e.,

when z = d = 1000 MWh



The capacity limits, gl.max, g2,max, and Zmax values are quantified in each particular
case considered.
General Equations

For this example (7.2.3) and (7.2.5) yield











2Rz )fJe


The Bus I equation is much simpler both because it is the swing bus and because
Generator I is always on the margin. Similarly (7.2.6) yields

Case 1: No Generation or Line Limits

Assume gl,maX' g2.maX' gent.y(t) and Zmax are all very large so they impose no
constraints. Thus f1.Qs.y = f1.y,max.2 = f1.Qs.~ = 0 and (7.6.1) becomes




Adding the conditions

7. Generation and network


gives four equations in the four unknowns, gl' g2, Z, and J1.e' If it is assumed that
g2 > 0 then Ily.max.2 = 0 and solving for g2 yields

or numerically


100 - 50
- (2)(5)(10 5)50

- 9000MWh

so the assumption is wrong. s The actual solution yields


5 10- 5 (IOOOi


and (7.6.2) yields


50$/M/Wh =

P2 =50[1


+ 2(510- )1000]

55$/MWh = 5.5/kWh

Thdact that g2 = 0 is no surprise because the peaking ,.1,2 is so much more

than the base loaded AI . However, just the condition that ,.1,2 > AI does
not imply g2 = 0 because of losses. For example if ,.1,2 = 5.1 /kWh instead of
10 /kWh, then



and substituting into (7.6.2)




5.1 /kWh

which is as expected; i.e. the price at each bus is the marginal cost of generation
at each bus.
Case 2: Line Flow Limit Active


gl.max, g2.max,

ft).. max.1 = t1y.max.2




are very large so

However now assume

600 MWh


Thus the Case I solution is not valid as it would imply


For Case 2, it turns out that g, > 0, g2 > 0, so both generators are on the
margin. Thus (7.6.1) becomes


,,1,2 -


~Q'," =



The other equations to be satisfied are



which is four equations in the four unknowns, g" g2, J.lQs," and J.le.
Solving these equations yields

or numerically

100 - [I

2(5(10- 5 600](50) = 47$/MWh

5(10-s)(60W =

780 MWh

where of course
g2 =

1000 - 600 = 400 MWh

For this case, (7.6.2) yields


50$/MWh =


2(5 10- )(600)]

The fact that for this case






7. Generation and network


is not a numerical accident. The presence of an active line flow limit effectively
decouples the prices at the two buses.
Case 3: Generation Limit Active on Base Loaded Unit gl



gen,.)" and zm~x are very large so



However, now assume



500 MWh

The values of gl and g2 for this case are obvious by inspection, i.e.




For this case, (7.6.1) and (7.6.2) are not valid, since g2 not gl is the marginal
generator. For this case
P2 =


Case 4: Generation Limits Active on Both Generators




/los ., =


However, now assume





are large so

168 II. Theory of hourly spot prices

Obviously, if d = 1000 there is no solution, as generation is unavailable to

take care of the losses. Hence it is necessary to include the d2 dependence on
price, i.e. d[P2J satisfies

The actual values depend on the assumed structural form of d[P2J, and the value
of the demand elasticity, etc.

Section 7.6. considered two adjacent buses connected by a single line. Here we
consider two adjacent buses, but they are now embedded in a network. Two
buses are adjacent if there is one line that directly connects them in addition to
the many others interconnecting through the rest of the network. The price
difference between these two adjacent buses (i.e. the price difference across
interconnected line) is discussed in this Section.
For simplicity we ignore the network maintenance component and
network quality of suppply cost component; i.e. assume
NM [Zi(/)]


Then (7.2.1) yields for the kth bus



+ ~ ~,(I)




(Network Quality of Supply)

k =

I, k = 2, denotes two adjacent buses


The Buses I and 2 can actually be generation or load buses or COmUl1Jlt1L1VII~

Using (7.7.1) and (7.7.2),
/).PI2(t) =


'r U/) [OZi(t)



- Od (t)

Equation (7.7.3) shows that, in general, the price differences between

adjacent buses depend on the conditions throughout the whole network. U

7. Generation and network


the DC load flow approximation, the partial derivatives of the line flows with
respect to the bus injections become the constants H21 and H 12 , but they depend
on the rest of the network configuration. However, if we make one further
approximation on the network, (7.7.3) is greatly simplified. The additional
assumption is
Constant Reactance to Resistance Ratios: The ratio of the reactance to the resistance for
every line i in the network is constant, i.e. the same for all lines, i = I ... ,

See Appendix 0 for a discussion on this approximation.

The constant reactance to resistance ratio assumption is often not too bad for
high voltage transmission lines. It is also not always very good. In practice, its
should always be studied before being used for actual detailed price
It can be bad if the network combines both high voltage transand distribution lines.
However, when a constant reactance to resistance ratio is assumed, it is shown
Appendix 0 that (7.7.3) reduces to

: Real power flow over the line (from Bus I towards Bus 2) and the
effective resistance of the line that directly connects Buses I and 2

7.7.4) shows that the effect of the losses have been greatly simplified (the
quality of supply term still involves coupling of prices throughout the
If no network quality of supply terms are present (i.e. 11Qs.",i(t) = 0 all 0. then
.7.4) says
price difference L\PI2(r) between two adjacent buses. I and 2. is proportional to the
I power flow ::1,(1) on the line connecting the two buses.

the DC load flow

(7.7.4) says that (with no network quality of supply)


the price difference across line 12 is proportional to the derivative of the

in the line with respect to line flow. Equation (7.7.5) is of course equito the equations used in Section 7.6 where, since there was only one line,

170 II. Theory of hourly spot prices

the constant reactance to resistance ratio assumption was automatically satisfied.

Bus I is adjacent to Bus 2
Bus 2 is adjacent to Bus 3
The price difference between Buses I and 3 can be computed by summing the
price differences between Buses I and 2 and between Buses 2 and 3. Thus, given
a DC load flow solution for the whole network, i.e. knowledge of all the line
flows, the price difference between any two buses can be easily computed. Even
if the constant reactance to resistance ratio assumption is not good enough for
actual pricing calculations, it can provide a good feeling for how prices will vary
throughout the network.

The formulation thus far has considered the hourly spot price Pk(t) seen by the
kth customer for using dk(t) during hour t where the generation levels get) are
determined by the utility. We now consider the case where some customers have
their own generators and sell energy back to the utility.
The derivation and discussion of the basic result (7.2.1) implied that del) > 0,
but such an assumption was never used. Furthermore the benefit Bddk(t)] could
actually be negative if desired. Therefore if the kth customer owned a generator
with output gt(!) and had a demand of lik(t), the net demand dk(t) is given by

df(t) - g'f(t)


Similarly, if the fuel and maintenance costs of the customer-owned generator are
GFM.dg~(I)l, the net benefits are
Using (7.8.1) and (7.8.2) in (7.2.2) leads to the important conclusion that
The hourly spot price Pk(O at bus k is valid even if the kth customer has a generator.

This conclusion had already been reached using (7.2.6) but is important enough
to repeat.
If the kth customer has only a generator and no load (dk(t) = 0); then the
customer is paid Pk(t) by the utility for the energy gHt) which is fed into the
network during hour t. The customer determines the value of g'k(t) by minimizing

7. Generation and network


subject to g%(t)


~ g~ax.k

i.e., by self-dispatching the generator based on the hourly spot price. Thus gk(t)
is determined by the condition

J.i.max.;'.k (t)

if gHt) = g~ax.Y.k

For most applications, the reader can stop reading this section at this point.
However, we will now discuss some fine points which can be important in some
Nonuniqueness of Pk(t)


Assume for simplicity that the kth customer has only generation and no load.
The use of the Pk(t) of(7.2.1) causes the customers to self-dispatch in a fashion
which minimizes social costs. However, if the generator is at its maximum
output, i.e., gW) = g~ax.k' then any value of Pk(t) which is greater than



will )'~eld the same level of generation. Thus the theory, as presented here, does
not necessarily specify a unique buy-back rate for customer-owned generation.
There is a demand-side analogy to this non uniqueness result. Assume the kth
customer has Qnly demand dk(t) and no generation. If the Pk(t) is low enough
that dk(t) is at its maximum possible level (or at least is completely insensitive
to price), then any value of Pk (1) that does not influence dk (t) can be used.
Throughout this book we assume that the buy and buy-back rates for
customer demand and generation are given by the Pk(t) of (7.2.1) because we
believe this to be the "fair and equitable" approach. We discussed the possibility
of using a lower value (when the generation or demand is at its maximum) only
because it exists.
Role of Generation Quality of Supply

In Section 6.4, the concept of the utility generation being self-dispatched was
discussed. This idea is now extended to the present case.
It follows from the mathematics of Section 7.2 that the optimality condition
for the utility gj(t) are satisfied if each utility generator chooses gj(t) to minimize

subject to




iI. Theory of hourly spot prices


gj(t) :0:;; gma'J(t)



L gj(t)

Comparison of the utility generation self-dispatched criteria of (7.8.6) with

the customer generation self-dispatch criteria of (7.8.4) shows that a basic
difference exists. The utility criteria (7.8.6) explicitly includes the system-wide
generation quality of supply terms GQdg(t)] and the J-lQs.y(t) Lagrange multiplier
for the constraint get) ~ gcrit.y(t). The customer criteria (7.8.4) is only concerned
with fuel and maintenance costs (the Pk(t) seen by customer generator of course
includes YQs(t. This result is "as it should be." It is the utility's responsibility
to worry about generation quality of supply - e.g., to maintain the operating
reserve which motivates the constraint g(t) ~ gcrit.y(t).
In Section 9.2, the use of price-quantity transactions is discussed as a way to
pay customers to help carry the operating reserves.

Thus far in this chapter it has been assumed that all lines are individually
modeled. In practice this is rarely true. Aggregation is almost always required
at the distribution level and often it is desirable at the transmission level.
The modeling of the various network components, losses, maintenance and
quality of supply will be discussed for an aggregated network which does not
model individual lines.
Network Losses IIL.k(t)

Equation (7.3.6) gives 11L.k (t) as a function of bus injections. Thus in some sense,
this equation does not involve the explicit calculation of line flows, i.e., the
explicit modeling of individual lines. However, the computation of the "Bmatrix" of (7.3.4) does require manipulation of individual line parameters.
Hence, use of (7.3.6) can be viewed as disaggregated at a planning level but at
least partially aggregated at the operational level.
The buses used in (7.3.6) can be lumped together in various ways to reduce
the level of detail (i.e., the dimensions of the B-matrix). Some of the potential
problems that arise are discussed in Section 4.5. With enough aggregation of
buses, the network loss component becomes independent of customer locations,
i.e. has no k dependence.
Network Maintenance: IIM.k(f)

As discussed earlier, good data on variable network maintenance costs are not
readily available today. Hence even when network losses and network quality
of supply are handled at a disaggregated line-by-Iine level, an aggregated
approach to network maintenance may be desired.

7. Generation and network


The basic approach is to use regression methods on past maintenance costs

to try to isolated the variable cost component. This approach and its difficulties
are discussed in Section 4.5.
Network Quality of Supply: tlQS.k(t)

If individual line flows are not being modeled, the market clearing approach
cannot be used and it is necessary to resort to the use of cost functions. By
analogy with Section 6.2 on the cost function approach to generation quality of
supply, two approaches are discussed:

Cost of unserved energy

Cost of annualized network investment

Cost of unserved energy

Following the development of Section 6.2, one possible formula for I1Qs(t) is

IIQ5(t) =


0QS.,/(t): Cost of Unserved Energy ($jkWh)

bn(t) =

d(t) > dcril.,,(t)


Th6:problem with implementating (7.9.1) is the difficulty in specifying a reasonablesalue for dent y . One approach is to say that b~(t) = 1 any time any load is
not being served because of a network capacity shortage. This philosophy has,
however, many obvious shortcomings.
Cost of annualiZ'ed network investment

Section 6.2 yielded an equation (6.2.12) for YQs(t). The aggregated network
analogy to (6.2.12) is

A Q5 .n: Annualized cost per kW of incremental network capacity selected by an

optimum network expansion planning exercise ($jkWjyear)

Loss of load probability due to network limitations at hour t





The difficulty associated with using (7.9.2) lies in the fact that practical models
for computing LOLP~(t) are not readily available today (except for special
cases). Therefore, we will follow the usual procedure when one doesn't have a
model: simply assume a reasonable form. The simplest reasonable form is


n, Theory of hourly spot prices

k[d(t) - d ]

d(t) > derit,.



LOLP (t) =

Substituting (7,9,3) into (7,9,2) shows that the choice of a value of k has no effect
on "Qs(t) - a most fortunate occurrence. Thus to use (7,9,2) it is just necessary
to specify a value for derit,q' In some price studies we have used "engineering
judgment" to specify dcrit,q as some percentage of the maximum demand where
the percentage ranged from 80 to 95 percent.
When (7,9,3) is used, (7,9,2) becomes for d(t) > dcri"q
1JQS ( I)

AQs.,Jd(t) - dcr;,.'!]


[del) -" der;,.,,]


where the summation is only over d(t) > dcrit.q' A simple approximation to
(7,9.4) results when d(t) is viewed as a random variable whose probability
density is flat for dmox ~ dcri,.q' This point of view yields
p: Value of probability density for dm.. > del) > der;. .
I 8~ [de )
[dm., - deri, . ]2 p
8760 ,~
t - er;I.. ~

Prob[d(t) > deri, . ]

Substituting (7.9.5) (7.9,6) into (7.9.4) yields

1JQS ( I ) =

2AQs . [d(t) - derit ]

8760[dmax - dcrit . ] Prob [d(t) > derit . ]


Partially Aggregated Models

Thus far in this section we have discussed completely aggregated network

models. In practice, partially aggregated network models will also be used,
One partially aggregated network model contains individual bulk transmission lines and an aggregated distribution system. However, the bulk transmission network itself may be partially aggregated using network reduction
procedures (either mathematical or engineering judgment based). Similarly, the
distribution network may be partially disaggregated into voltage levels and/or
regional levels.
When dealing with certain large industrial customers who have, for example,
a special need for reliable service, the feeder network near them may be explicitly
modeled even if the bulk transmission network is partially aggregated.
The choice of level of network aggregation and disaggregation is very dependent on the particular situation. The use of common sense appears to be the best
general rule.

7. Generation and network



The preceding discussions have considered only real energy. Reactive energy
flows can also be important since they affect both real line losses and voltage
magnitudes (see Appendix A). In practice, it is sometimes desirable to include
a spot price on reactive energy as well as a constraint on the allowable voltage
magnitudes at each bus. The preceding equations can be generalized by viewing
the energy and prices as complex numbers whose real and imaginary parts
correspond to real and reactive energy respectively.9 When this is done, a
nonlinear AC load flow has to be used to solve the network equations instead
of the linear DC load flow approximation used here.

The basic principles used in this chapter are the same ones used in Chapter 6
where only generation was considered. However, the equations are now much
more complicated in appearance because of Kirchoff's laws, which determine
how line flows and losses come into play.
The biggest effect of the network is that spot prices become spatially dependent, i.e. they depend on where the customer is located on the network (unless
an aggregated network model is used). Under normal operating conditions, the
generation components of the hourly spot price usually dominate the network
components. However, when the network is heavily loaded, the network comp~ents can dominate.
;Jhe network components of the hourly spot price should never be ignored
entirely unless explicit computations have proven them to be negligible.



The basic reference for Chapter 7 is Bohn, Caramanis and Schweppe [1984]. See
also Schwejilpe, Bohn, Caramanis [1985] Caramanis, Bohn, and Schweppe
[1986] and Caramanis, Schweppe and Roukos [1988] which address wheeling.
Dobbs [\983] looks at optimal spot pricing and finite line capacity.
Readers with a power system background in economic dispatch and optimal
load flow (with line flow constraints) will find that the equations of this chapter
look familiar. This is to be expected, since such power system theory motivated
their development. See Appendix B for further discussion on this relationship
to power system operation. Ponrajah [1984] is an example of a very close tie. The
work of Luo, Hill, and Lee [1986] dealing with bus incremental costs is related
to spatial spot pricing, but uses only system lambda and losses.
The network causes prices that vary spatially. Several previous results in the
economic literature have studied how public utility prices should vary over
space. Relevant models include Takayama and Judge [1977] (which was not
directed at electricity), Craven [1974], Dansby [1980], Scherer [1976, 1977], and
Schuler and Hobbs [1981]. All of these models are deterministic and most are
static. Agnew [1977] is stochastic but does not deal with electricity.
Scherer has the best model of electricity line losses and line constraints, and
includes T and D investment options. Scherer's approach is to use a mixed


n. Theory of hourly spot prices

integer programming model of an electricity generation and transmission

network. In his model, spatially distinct prices appear as dual variables on
demand at each point in the network. In his numerical case study he found that
prices between different points at the same time varied by up to 30 percent. The
absolute and percentage variations across space changed over time [1977,
p.265]. He does not discuss these results, but presumably they reflect the
different losses resulting from different optimal load flows at each level of total
system demand.
Much of Takayama and Judge [1971] concerns pricing across space. They
consider only competitive markets, but use an explicit optimization method of
finding equilibrium, so their analysis is equally applicable to a welfare maximizing monopolist. They assume a constant transport cost per unit between two
points, no transport capacity limit, and no losses. This makes their models more
appropriate for conventional commodities than for public utility products such
as electricity. They also assume linear demand and supply functions. But their
framework does provide insights into more general spatial and temporal pricing
problems. For example they discuss "no arbitrage" conditions that bound the
price differences between different location (197l, p.405J. Their models do not
include capital, so they provide no insights into optimal investments in transport
I. A nctwork contains transformers, capacitors, etc. as well as lines. The clrects of such devices ar~
assumed to be "lumped into the lines" or treated as "separate lines."
2. The term "DC load flow" has historical origins and does not imply the use of a direct current
transmission distribution network.
3. [n Chapter 6, we expressed GrM[g(I)] in terms of total generation g(l) rather than the vector g(t).
4. If all bus injections were specilled. the energy balance constraint that total generation be equal
to total load plus total losses would almost certainly not be satislled.
5. Very high flows for a long time can cause a line to overheat, and expand so much it sags down
and touches the ground with unpleasant effects. Flows that are too large also do nasty things to
6. We use the symbol B for histoncal reasons. It obviously bears no relation to the customer benefit
which we also denOte by B.
7. Generation redispatch is the only way line overloads can be removed in present-day systems
(except by completely blacking out selected customers).
8. A more elegant derivation of (7.2.1) would have included a constraint that g(l) ~ o.
9. Caramanis, Bohn, and Schweppe [1982] carry out the basic derivation of the chapter for this more
general case.


R~enue reconciliation was ignored in Chapters 6 and 7. We finally face up

to it;many problems in this chapter. Revenue reconciliation is a complex topic
whicb can be approached in many ways. Most but not all of the ideas to follow
are based on existing techniques.
Assume that all customers see hourly spot prices, i.e. the kth customer sees
Pk (t). Thus over some time interval T R ,
Received by Utility
t = ) ... TR


L L Pk(t) dk(t)


(8.1 )

Total Operating and Capital
Costs Incurred by Utility
t = I ... TR


The capital costs are the embedded costs (interest, debt payment and rate of
return of investment) of existing generation plants and the network as allocated
to the time interval t = I ... T R

178 II. Theory of hourly spot prices

The goal of revenue reconciliation is to make A(TR ) close to zero in some sense.
Section 8.1 starts out by considering how to modify the hourly spot prices to
achieve revenue reconciliation which combines generation and network costs.
Two approaches are used in Section 8.1: the conventional one of "Ramsey
pricing" and the simpler to use but less elegant weighted least squares. Section
8.2 contains the "buy-back" version of Section 8.1. Later on, Section 8.5 returns
to the problem of Section 8.1 but with reconciliation decomposed into generation and network components or even all the way to individual transmission
Jines. Sections 8.3 and 8.4 present the surcharge-refund and revolving fund
approaches. Section 8.6 briefly discusses the role of fixed charges. Section 8.7
contains an interesting, but not critical side discussion on nonlinear pricing,
Section 8.8 concludes with a discussion on revenue neutrality.

The first approach to revenue reconciliation discussed here modifies the formula
for computing the hourly spot price so that for some fixed value of TR (say one
year), the expected value of the revenue R(TR ) equals the expected value of the
costs C(TR ) where the expectation is an average over all equipment outages, fuel
costs, demand variations, etc. that occur over the time interval t = I ... TR
CO(TR ): Administration, metering, billing, and other fixed but noncapital costs
over t = I ... TR ($)
C)'G(TR): Embedded capital costs of installed generation over t = I ... TR ($)
C~(TR): Embedded capital costs of instaIled network over t = 1 ... TR ($)
NM (TR): Network Maintenance costs over t = I ... TR ($)
GFM(TR): Fuel and generation maintenance costs over t = I ... TR ($)
GFM(TR ) = r.;~1 r.PFMJ{gj(t)]



{h(t): The hourly spot price developed in Chapter 7 for customer k during hour

t; i.e. from (7.2.1)








The problem formulation is as follows I :

Choose Pk(t) which are "close to" pdt) k == I .. " t = I ... TR and which

{~f Pk(t) dk(t)}


8. Revenue reconciliation


This formulation assumes the formula for computing Pk(t) is a priori specified,
Le. before time t = I.
General Structure of Equation for Pk(t)

There are many ways to modify Pk(t) in order to achieve (8.1.4). However, the
following forms result from' all of the approaches to be discussed:
Multiplicative Form:

Additive Form:

Mathematics leading to explicit equations for

presented shortly. However, as will be seen, the
making of relatively arbitrary assumptions and/or
obtain. Fortunately, it turns out that "reasonable"
theory yield


mk(t) and/or ak(t) will be

mathematical requires the

use of data that is hard to
special cases of the general


(8.1. 7)


i.e., t6'e multiplicative and additive corrections become independent of customer

k and time t at least for all k belonging to some specified customer class. Using
(8.1.7) or (8.1.~) in (8.1.4) yields2
E{C(TR )}




L L E{pdt)d,(I)}



a == E{C(TR )}

2:: L E{p,(I)dk(I)}



Combining (8.1.7) with (8.1.3) yields








'IR.k (I)

YR (I): Generation revenue reconciliation component of the hourly spot price

I'R(t) =




180 II. Theory of hourly spot prices

tlR.k(t): Network revenue reconc'~iation component of the hourly spot price

tlR.k(t) =


+ tlM.k(t) +


Readers who believe that the use of a simple, straightforward, easy-to-explain

approach is more important that fancy mathematical derivations will probably
accept (8. 1.7) or (8.1.8) as the "best" choices. Such readers can skip the rest of
this section and go to Section 8.2. In most subsequent developments we choose
to use the constant mUltiplicative form of (8. 1.7).
Readers who want to understand the mathematics and assumptions underlying (8.1. 7) and (8.1.8) should continue with the rest of this section. In some
applications, it may be desirable to use the more detailed forms of (8.1.5) and
(8.1.6), and the rest of the section provides explicit equations.
Weighted Least Squares Derivation

The weighted least squares approach to deriving (8. I .5) (8.1.6) starts by defining
Measure of closeness _
of fit of Pk(t) to Pk(t) -

{r. L

[Pk(t) - I\(t)f}

Q, (I): Weighting function

all t and k

Minimization of this measure subject to the constraint (8.1.4) is done by forming

the Lagrangian


Setting the derivative of (8.1.15) with respect to Pk (I) equal to zero yields]


which yields4
Pk(t) =





dk(t): Demand of kth customer

Sk(t): Price elasticity of demand of kth customer (a negative number)

Pk (t) ad, (t)

---dk(t) apk(t)

flR: A constant determined by substituting


(8.1.17) into (8.104)

8. Revenue reconciliation


Table 8.1.1. Weighted Least Squares: Effect of Choice of Weighting Function Q,(t)
{ Q,(t) =

{ p, (I) =



Choice 2 yields additive form: a,(I) = I'R[I + ",(I)]
. Choice 4 yields Illultiplicative form: 111,() = I'R[I + ,()]
Value of PR varies with choice

E{C[TRJ} -

E{~ f

E{~ f Qk(t)df(t)[1



T~le 8.1.1 summarizes the equations for Pk(t) that result from (8.1.17) for
four;:aifferent choices of Qk(t).5 The motivation follows from (8.1.14), since the
valu~ of Pk(t) is closest to Pk(t) for customers k and/or times t when Qk(t) is
small~ Thus, for example, Choice 2 of Table 8.1.1 results in hourly spot prices
Pk(t) that are closest to Pk (t) when dk (t) is large so most revenue reconciliation
is done for cus~omers k and/or times t with small demands. 6 Such arguments can
be used to justify Choices I to 4 of Table 8.1.1. Note however that different
personal judgments can yield choices not in Table 8.1.1. For example, one might
argue that Qk (I) = dk (t) should be used so the fit of Pk (t) to Pk (t) is closest when
the demand is small.
There are three important points to note on the role of the elasticity Ck(t) in
Table 8.1.1. First, if ck(l) is a constant, independent of customer k or time I, (i.e.
if ek(l) = c), then the value of C does not have to be known since C always
appears with IlR and (l + c) can be factored out of the denominator of (8.1.19).
Second, if ek(l) = c, then for Choices 4 and 2 of Table 8.7.1,

m = tiR[1


a = tiR[1

+ cl
+ el

where of course the value of IlR changes with the choice of Qk(I). Third, since
the magnitude of Ck(t) is usually small relative to one, the actual value of Ck(t)
is not very important for any of the choices.



Theory of hourly spot prices

When considering Table 8. 1.1, note that, from (8.1.19),


If the utility "underrecovers" by charging Pk (I), then

Ji.R > 0, mk(l) > 0, ak(t) > 0

If the utility "overrecovers" by charging Pk (I)

Ji.R < 0, mk(t) < 0, ak(l) < 0

It is also important to note that the sign of Ji.R can change from year to year
(assuming TR = 1 year). For example Ji.R can go from negative to positive when
a nuclear power plant "comes on-line" and enters the rate base.

Because of the network components, it is theoretically possible for Pk(l) to

become negative for special cases. If this happens, Choice 4 for example has to
be changed to

Throughout this book we assume Pk (I) > 0, since the case of a negative p(t) is
too rare to justify the use of the more complex notation of (8.1.20).
Ramsey Pricing Derivation

"Ramsey pricing" is a term often used in economics to denote the prices Pk(t)
which minimize "social cost" subject to the revenue reconciliation constraint of
(8.1.4). It is also known as "second-best pricing".
In Section 7.2, the following Lagrangian was developed to find the Pk(t) which
minimized social cost (see (7.2.2:
a(t) =


N[~(t)] - B[q:(t)]


L[~(t)] - g(t)]

For the present problem with expectations and a time interval I to TR , define
another Lagrangian by adding the revenue constraint of (8.1.4):

Setting the derivative with respect to Pk(t) equal to zero yields (after manipulation as in Section 7.2 and remembering that the Pk(t) of Chapter 7 is now called



8. Revenue reconciliation


Table 8.1.2. Ramsey Pricing: Effect of Choice of Model For Derivative ad,(I)/apdt)







ad,(t) =
ap, (I)

p,(t) =

ct, (I)

lii,(I) - J.iRd,(t)ct,-I(t)](1

(X, (I)d, (I)

[Pk(t) - 11Rctkl(t)](l

ct,(fl/p, (I)
fl, I












, ~,(I) is a negative quantity which does not depend on tI, (I) or 1', (I),
, For (,hoice 4 ~,(I) = ", (I),
, Choice 4 yields multiplicative form: "', (I) = [\ + i'R + I'Ri", (t)]-',

, J'R varies with the choice,


Thus after some more manipulation 7





The weighted least squares approach required the choice of an arbitrary

function Q, (I). The Ramsey pricing approach appears to be more
elegant, since (8.1.23) looks like a much more specific equation. Unfortunately,
evaluation of the derivative ad, (1)/Op, (I) requires the use of a customer demand
response modf:l, and today the choice of such a model structure is almost as
arbitrarl as the choice of a weighting function. Appendix E presents four
different customer demand response models, each resulting from a different
benefit function. The resulting derivatives are given in the first column of Table
8.1.2. The second column of Table 8.1.2 gives the resulting equations ror Pk(l)
using (8.1.23).
If the C/.k(t) of Table 8.1.2 are assumed to be constant, independent of k and
t, then Choices 4 and 2 yield the constant mUltiplicative and additive forms of
(S.1. 7) and (S.1.8) (the strict additive form also requires J.LR ~ 1).
If the derivative ad,/ap, is assumed to vary with customer k, Ramsey pricing
can yield prices that are very sensitive to its value. A customer with a very small
derivative (in magnitude) would pay a very high price9 when J.LR > 0 and a very
low price when J.LR < O. Thus their prices could swing widely from one year to
the next. The more intuitive but less elegant weighted squares approach is much
less sensitive to the value of the derivative.
The Ramsey pricing formulation of(S.1.21) is "short-run," in that it does not
include the effect of prices on long-run customer decisions associated with

184 H. Theory of hourly spot prices

capital investment etc. Care is needed in using the results since myopic applications of(8.1.23) run the risk of getting outside the range of validity of the model.
To illustrate, consider an industrial customer (the kth customer) who uses
one MWh during hour t at full capacity and receives a benefit of 50 /kWh from
the electric energy. Assume that during hour t, this customer has orders to fill
which require operation at full capacity, i.e. at I MWh. Then this customer will
behave as follows:

IMHwwhenpk(t) < 50/kWh

Hence Odk(t)/OPk(t) = when Pk(t) < 50 /kW. If,uR > 0, the result of(8.1.23)
is that the short-run Ramsey pricing says that this customer should pay 50/
kWh independent of Pk(t) provided Pk(t) < 50/kWh. A long-run model that
looked over many years and had a long-range price elasticity that reflected the
fact that charging 50 /kWh might cause the customer to move would yield more
reasonable results at the expense of even more difficult-to-implement equations.
Some of these "problems" associated with (8.1.23) go away if it is assumed
that the derivative is constant for all k in some customer class.
An interesting possibility results if the derivative is assumed to be constant for
all k (or all k in some customer class) but assumed to vary with aggregate
demand. It can be argued that elasticity goes down (in magnitude) with increasing demand. This yields a time-varying m(t) that increases the variability of the
hourly spot prices.

During the mathematical derivations in Section 8.1, it was implicitly assumed

that the demand dk(t) was positive. Now we allow ddt) to be positive or negative
where a negative dk(t) implies that the customer is feeding energy into the
network. Thus we address "buy-back rates" where the utility is buying energy
from the customer.
Amount Customer
Pays Utility for
Energy ($)
Amount Utility Pays
to "Buy-Back" Customer
Generated Energy ($)

fiL,,1I (I) ddt)

P,.buy(t) ddt)

Thus Ih.scil and are both positive and

If revenue reconciliation is ignored,



ddt) < 0


is the Pk of Section 8.1.


This was discussed in Chapters 6 and 7.

Now consider the Section 8.1 approach of modifying spot prices a priori to

8. Revenue reconciliation



achieve revenue reconciliation. Following the constant multiplier theme, assume


(\ +
(\ +










The following logic on how

earlier discussions.


and mbllY "ought to be" related resulted from

If: mscl\ > 0

Then: Utility is underrecovering if Pk(t) is charged
Therefore: Utility has too much capital invested in capacity
Thus: Utility doesn't need the "capacity associated" with customer generation.
Hence: mOllY < 0
A somewhat different logic chain which leads to the same conclusion is





If: /Ilsdl > 0

Then: Utility is increasing its net revenue
If: mOllY < 0
Then: Utility is also increasing its net revenue





have different signs.

fr one believes that it is important to have simple, easy-to-implement

equations, the "best" way to implement (S.2.6) is to replace (S.2.4) (S.2.S) with



m)Pk (I)




(\ - I1l)Pk(l)


An alternative way to write (S.2.7) is to use the convention that



Pk.,,1\ (I)

(Selling to Customer)


(Buying from Customer)

and then instead of (S.2.7) to use




+ I1lIPk.sel\(t)1
+ I1llih.huy(t)1


where Pk.blly(t) is now a negative number (whose magnitude equals the
of (8.2.7. The form (8.2.8) is more "symmetric" then (8.2.7).
We will use (S.2.7) (S.2.S) in subsequent developments. Thus readers who are

186 II. Theory of hourly spot prices

happy with hand-waving arguments can skip the rest of this section. However,
for those who like to see mathematical gymnastics, we will define some general
fomulas and then show how (8.2.7) (8.2.8) result as a special case. The mathematics is also useful in understanding what is going on and provides some
alternative equations which have potential use in certain applications.
Weighted Least Squares Derivation

The derivation of (8. I. I 7), i.e.


holds whether elk(t) is positive or negative. However, care in the choice of Qk(t)
is required because it is necessary that Qk(t) > O. Thus, for example, Choice 4
of Table 8.1.1, should be replaced by

P. (I)


Qk (I) = Id (/)1

Use of (8.2.10) in (8.2.9) yields

(8.2.11 )


If the assumption is made that lek (t)1 = lei for all k and t, independent of
whether buying or selling, then (8.2.11) yields (8.2.4) (8.2.5) where (because c is
negative for usage and positive for generation)




i-lR (I

If the effect of


lei is ignored,

which is (8.2.7). Similar types of conclusions result from other choices of Q,(t).
Ramsey Pricing Derivation

The Ramsey pricing approach of Section 8.1 (minimization of social cost subject
to reconciliation constraint) yielded (8.1.23), i.e.

This is directly applicable for both buying and selling, since no assumption on
the sign of elk (I) was made.
In order to be more explicit, consider Choice 4 of Table 8. 1.2 which yields

8. Revenue reconciliation







when dk(t) > 0


ek(t) ~ 0

when d, (t) < 0






As discussed in Section 8.1, the short-run character of our Ramsey formulation leads to the need for care in its strict application. A similar situation arises
in the present buy-back case. To illustrate, assume J..tR > 0 and consider a
customer-owned generator with maximum capacity of I MW and fuel costs of
IO/kWh, independent of amount generated. This means the customer will
behave as follows:
Customer generation


when Pk(t) > IO/kWh

Customer generation

when Pk(t) < 1O/kWh

Hence the customer's elasticity ek (t) is "zero" when the buy-back price is greater
than IO/kWh. The result is that (8.2.13) yields 10

. Pk(t)


p,(/) < JO/kWh


p,(t) > IO/kWh

Thu! short-run Ramsey pricing leads to the conclusion that customer-owned

gem;iation should be paid just enough to keep the generator "on-line" whenever
Pk(t~ exceeds the customer's marginal costs. Hence customer-owned generation
chance of ever recovering its capital costs because the social cost function
does not take customer capital costs into consideration.
If the assumption is used that lek(t)1 = lei for all k in some customer class,
then one can Zvrite


6k(l) =

1r.1 sgn (d, (I))

and (8.2.13) yields

pdt) =



+ ~~

sgn (d,(I)))


which is of the form of (8.2.4) (8.2.5) with



[I +
[I +



+ lei

fiR -




This reduces to (8.2.7) if J..tR/lel is small in magnitude relative to I


+ J..tR'

1118 U. Theory of hourly spot prices


In Sections 8.1 and 8.2 we specified a priori (before time = I) formulas for
hourly spot prices which yield reconciliation in an expected value sense over the
intervalt = I ... TR The "surcharge-refund" approach to be discussed here is
a posteriori.
For simplicity assume dk(t) ~ 0. Buy-back rates follow in a similar fashion.
The basic approach is to

Set Pk(t) = Pk(t) t = I ... TR

After time T R, compute L\(TR) = C(TR) - R(TR) from (8.3)
If L\(TR) < 0, give each customer a refund after t = TR to "use up" the excess
If L\(TR) > 0, send each customer a bill (surcharge) after t = TR to "makeup" the revenue shortage

This surcharge-refund approach requires the specification of a formula to

allocate the surcharge or refund among individual customers. Ideally this allocation is independent of a customer's energy usage (see Section 8.7). However, this
criterion is very difficult to achieve in the real world. It is not met in the following
Allocation of Surcharge-Refunds Among Customers

t.k(TR ):

Amount of money to be refunded/surcharged to customer k after time

One "reasonable" way to specify

customer's bill; i.e.

t.k(TR )

is to make it proportional to the kth


Llk(TR )

= m

L dk(t)Pk(t)


L L cik (t)p, (I)


Using this approach,


kth Customer's Total

L d(t)Pk(t)



Bill for t

I ...



L dk(t)Pk(t)(l


TR ($).

+ m)

8. Revenue reconciliation 189


Hence this approach has yielded an effective spot price, (l + m)Pk(t), which is
identical to the multiplicative form of the modified spot price (8.1.7) of Section
8.1, except that in (8.3.2), m is specified a posteriori after time TR using Ll(TR)
while in (8.1.7) m is specified a priori before time 1 using E{Ll(TR)}'
An alternative approach is to make Llk(TR) proportional to the kth customer's
energy usage; i.e.




This yields an effective hourly spot price of Pk(l) + a, which is the additive form
of the modified spot price of Section 8.1, the only difference being the time at
which the constant a is specified.
Analogous ways can be found to allocate Ll(TR) among customers to yield
most of the formulas of Section 8.1 and Section 8.2.
A Priori vs. A Posteriori

If revenue reconciliation is done by a constant multiplier term, should 111 be

spet!ltped a priori (before time I = I) or a posteriori (after time t = TR )?



c)ne possible advantage of the a posteriori surcharge-refund approach is that

durmg t = I ... T R , the customer "sees" Pk (I), which is the social cost minimizing price. In practice, however, customers can try to guess (before t = I) what
misgoingtobe(aftert = TR)andactasiftheir"actual"priceis(1 + m)Pk(t).
Thus this may not be much of an advantage after- all.
Another potential advantage of the a posteriori surcharge-refund is that
reverse reconciliation is achieved exactly at time TR , while the a priori approach
of Section 8.1 and 8.2 only achieves E{Ll(TR )} = O. However, it can be argued
that exact revenue reconciliation reduces the utility's incentive to operate efficiently.
A potential disadvantage of the a posteriori surcharge-refund approach is
that customers who do not attempt to guess what the value of m is going to be
can be hit with a major unexpected bill (if the utility is underrecovering) after
t = T R This would put a strain on utility---customer relationships.
Neither approach completely dominates the other. However, in the real
world, we recommend the a priori approach of Sections 8.1 and 8.2 because it
eliminates customer guessing. We will use it in subsequent discussions.


In Sections 8.1,8.2 and 8.3, approaches to revenue reconciliation were discussed

which attempted to make sure the "books were balanced" at the end of a fixed

190 II. Theory of hourly spot prices

time interval TR , which is usually one year. We now consider the "revolving
fund" approach, which can be viewed as the result of assuming TR is extremely
Set Pk(t) = Pk(t). Define as in (8.3)

C(T) - R(T)


where now T is a running time index with one-year steps. The revolving fund
(referred to with a variety of different terms in the literature) works as follows.

If ~(T) < 0 (i.e. utility is received too much revenue), put the excess into a
"revolving fund" which is invested in the capital market.
If ~(T) > 0 (i.e. utility is receiving too little revenue), remove the funds from
the revolving fund (if it is positive) or if necessary, borrow the money from
the open capital market.

With this approach, the customers see Pk(t) = Pk(t) while simultaneously the
utility neither makes excess profits nor incurs inappropriate financial losses.

RF(T) =

L ~(s)


s= I

RF(T): Revenue accumulated in the revolving fund at year T; can be positive or

negative ($)

In practice, (8.4.2) would be modified to reflect the interest received (paid) when
RF(T) is positive (negative).
The equations of Sections 8.1 through 8.3 keep ~(T) close to zero by modifying the hourly spot price. With the revolving fund approach, the hourly spot
price cost is not modified either a priori or a posteriori. Instead, the utility
generation and network investment policy is used as the mechanism to try to
keep RF(T) close to zero over a period of many years.
The basic fact motivating the revolving fund is
If the utility has an "optimum" generation and network system, revenue reconciliation
is automatic since marginal cost prices exactly cover the operating and capital costs (see
Chapter 10).

This yields the following type of investment rule:


If ~(T) is negative, the utility is overrecovering and should consider investing

revolving fund money to construct new generation and/or transmission facilities.
If ~(T) for some length of time is positive, the utility is underrecovering and
should consider deferring new construction until the demand grows (or old
plants are retired etc.) enough that ~(T) becomes negative.

8. Revenue reconciliation


In practice, of course, the actual investment rules would have to be carefully

designed to consider not just 6.(T) but also RF(T), long-range load and fuel
availability forecasts, etc. Basing expansion investments at year T only on .1(T)
could easily lead to a highly oscillatory behavior.
The big advantage of this revolving fund approach is that customers see the
"optimum" hourly spot prices; i.e. Pk(t) = lh(t). One disadvantage is that it
may not always work. For example, consider a utility which presently has a
major overinvestment in generation that results in major under recovery (i.e.
6.(T) > 0). The interest resulting from borrowing capital to cover the shortage
may grow "faster than the load" so RF(T) "never" returns to zero (at least not
in several lifetimes).12 Even if it works, the revolving fund concept has the
disadvantage of being foreign to many accepted regulatory principals such as
"today's customers should pay today's costs." A revolving fund can look a lot
like CWIP (Construction Work in Progress), which is not acceptable in many
states in the US.
Despite its potential problems, the positive attributes of the revolving fund
concept should at least be evaluated and the revolving fund considered as an
alternative to the ideas of Sections 8.1 through 8.3.
The decomposed revenue reconciliation approach to be discussed in Section
8.5 can be extended to have separate revolving funds for generation and
network specific revenue requirements.

Sectf(;n 8.1 used a single revenue reconciliation constraint that contained all
gener~tion and network costs. An obvious variation is to introduce multiple
revenue reconciliation constraints such as to
Achieve revenue reconciliation separately for generation and the network by using two

Achieve revenue reconciliation separately for each generator and each line by introducing a separate constraint for each generator and each line.

It can be argued that capital costs are "more equitably" distributed among
the customers when such decomposed revenue reconciliation is done for different parts of the system. For example, the capital costs of a major new
generating plant or transmission line would then be borne most by the customer
who makes "most use" of the new facility.
There are, however, disadvantages to the decomposed revenue reconciliation
approach. If the goal of revenue reconciliation is to keep Pk(t) as close as
possible to Pk(t), then the addition of extra constraints helps defeat this goal,
since they will increase (or at least not decrease) the "distance" between Pk(t)
and pdt). Thus there is a tradeoff between equity (customers bear "fair" share
of costs) and economic efficiency (Pk(t) close to Pk(t.13


II. Theory of hourly spot prices

Decomposition of Revenue

Using (8.l.l1), it is reasonable to define

R;,(TR ): Revenue obtained from generation components of the spot price

L [yet)


YR (t)] get)



R,,(TR ): Revenue obtained from network components of the spot price


R,,(TR) =

L L ["k (t)

1 k

"R.k (t)]dk(t) - [y(t)

YR (t)L(t)


because since

L gj(t)


g(t) - Ldk(t)

(Total generation),
(Total Losses),

it follows that

A further decomposition of revenues results by writing


Ry)TR): Revenue associated with generator j


L [yet)

YR (t)] gj(t)




Revenue associated with line i


R",,(TR ) =

L L ["k,;(t)


- L;(t)[y(t)


"R.k.;(t)] dk(t)


YR (t)J

+ "M.k,;(t) +


8. Revenue reconciliation 193

Li(t): losses on line i


+ YQs(t)]



Using such equations, it is possible to compare the revenues "obtained" from
different components of the system to the costs of those components. This can
be useful in helping to guide future investment decisions even if aggregate
revenue reconciliation is done.
Separate Generation and Network Reconciliation

We believe that separate revenue reconciliation for individual generators and

lines is not desirable except for special cases. However, we believe that separation of revenue reconciliation into generation and network, or generation,
transmission and multiple levels of distribution, is desirable for many applications. An example of the use of five distribution levels is all 69 kV lines; 13 kV
rural lines; 13 kV urban lines; below 13 kV rural; below 13 kV urban. We only
present equations for the case of separate generation and network reconciliatioJ1~The multiple distribution level case follows in a similar fashion.
Fitlowing the development of Section 8.1, define

erR ):

Total costs associated with generation
C./TR ): Total costs associated with the network

The problem 1's to define the revenue reconciliation components i'R (t) and YJR.k(t)
of the spot price so that
E{C\.(TR )}

E{R\.(TR )}



E{R,/(TR )}


where R/TR ) and R,/(TR )} are given by (8.5.1) and (8.5.2).

Rather than repeat the full development of Section 8.1, we will simply use the
constant multiplier form to motivate the assumption l4 that

)'Qs(t)] = my)'(t)


so that

194 II. Theory of hourly spot prices

The use of the absolute value of 11k (t) is motivated by the form of the buy or sell
equation of (8.2.8) and the fact that I1k(t) can be positive or negative.
Substituting (8.5.9) into (8.5.7) and (8.5.8) yields the conditions

+ my)






I L E{[lJk(t)



- (I





which can be solved for the desired values of my and

on the value of my.


Note that



Generators Selling to Network Selling to Customers

The preceding equations have an interesting interpretation. A single utility

which generates electric energy and transmits it to its customers can be viewed
as being departmentalized into a "Generating Department" and a "Network
Department" where

The Generating Department sells get) to the Network Department at a price

(I + my) y(t) where my is determined by the Generating Department's
revenue requirement given by (8.5.11).
The Network Department sells ddt) k = 1 ... to its customers at a price pdt)
given by (8.5.10) where 111" is determined by the Network Department's
revenue requirements given by (8.5.12).

This follows by considering the Network Department's net revenues to be

( Revenue

(Amount ReceiVCd)
Amount Paid to
from Customers
Gencrating Department

I I p,(t) ddt)

which using get)




L(t) yields after manipulation

( Right Hand Side of (8.5.12) )
without Expectation Operator
( Revenue =

If the values of fJ.? and m" are not equal, then the use of the aggregate
reconciliation procedure of Section 8.1 can be viewed as causing the Generation
Department to "subsidize" the Network Department or vice versa. Since my and
mil can have different signs, the Generating Department can overrecover while
the Network Department is underrecovering.

8. Revenue reconciliation





The developments of Section 8.1 assumed the total costs C(TR ) of (8.1.2) are to
be recovered (in some sense) by the hourly spot price Pk(t) where C(TR ) included
CO(TR ), the administration, metering, etc. costs. In practice it can be more
desirable to recover some of these COCTR ) costs such as metering and billing in
terms of a fixed charge so that






Special distribution system costs such as an extra transformer for a large

industrial customer could be recovered by doing separate revenue reconciliation
for individual network components. However, in practice, it may be simpler just
to use a fixed charge, either a lump sum or spread out as a monthly charge.




This section provides a broader perspective on revenue reconciliation that leads

to "nonlinear pricing." This section can be skipped since the ideas are not used
in subsequent developments.
In order to reduce the number of symbols to be carried along, the following
simplifying assumptions are made:



o~;No network losses, maintenance or quality of supply (i.e. network ignored).

o No generation quality of supply (i.e. no constraint on maximum available

o No uncertainty (i.e. no expectation or random variables).
This corresponds to the conditions of Section 6.1 except that here we work with
disaggregated demands dk(t) k = 1 .... However the basic ideas to be discussed
apply directly to the more general cases of Section 8.1, 8.2, etc.
A More General Approach to Revenue Reconciliation

Rk(t): Revenue received by utility from customer k using dk(t) during hour t. ($)

In earlier discussions, the revenue was constrained to have the "linear" structural form of



However, now the revenue Rk(t) is not constrained to be of any particular

structural form.

196 II. Theory of hourly spot prices

Following earlier developments, a Lagrangian is formed by


n(TR )

[GfM[g(t)] -

IlR[C(TR) -

Bk[dk(t)] - Ile(t)[g(t) -




t= 1



where iJ.R is the revenue reconciliation constraint multiplier. As earlier, it is

assumed that all customers optimize their own welfare, i.e. choose dk (I) to

so that dk(t) is given by

adk (I)



for this special case yields (as expected)

Ii.(t) =

).(/) = aGFM[g(/)]



The optimum revenue Rk(l) is found by setting

This yields, using (8.7.3) and (8.7.4)


The optimality condition of (8.7.5) can be satisfied by choosing

Rk(/) =

Rk O



8. Revenue reconciliation


A.(t): Assumed to be independent of dk(t) (i.e., many customers)

Rk O: Constants k = I, ... chosen so that J1.R = 0, i.e.


C(TR ) = L,LkRk(t)



Thus the Rk(t) of (8.7.6) is the revenue which minimizes social cost subject to
both energy balance and revenue reconciliation constraints.
The constants R k O in (8.7.6) can be chosen in any way that satisfies (8.7.7)
provided only that they satisfy the condition
t = 1 ... TR


The problem with implementation of (8.7.6) is the difficulty of specifying Rk,o

that are considered (by the customers) to be fair and which satisfy (8.7.8). This
is the identical problem discussed in Section 8.3. The "ideal" surchargesrefunds are independent of dk(t).
Nonlinear Pricing

R(t): Total revenue obtained by utility during hour t ($)
R(~= LkRk(t)

Th;n for the ideal Rk(t) of (8.7.6)


Ro +

A(t) d(t)



where ~ is uniquely specified by the revenue constraint (8.7.7). When the

constant multiplier approach is used for the linear prices of (8.7.1) (for our
simple model, Pk(t) = A(t,
R(t) =


m)A(t) d(t)


Figure (8.7.1) plots R(t) of (8.7.9) and (8.7.10) versus d(t) assuming ~ and mare
greater than zero, i.e. underrecovery in the absence of revenue reconciliation.
Figure (8.7.1) provides motivation for using a "nonlinear" pricing scheme
that yields a R(t) vs. d(t) curve which looks more like the ideal R(t) of (8.7.9).
One approach is to hypothesize a nonlinear polynomial



Ideal Revenue:
Slope = A(t)

Linear Pricing Revenue

Slope = (1 + m) A(t)

Demand, d(t) (kWh)

Figure 8.7.1. Total revenue versus demand.

where the Pk.1 (t), Pk.2(t) etc. are found by repeating the derivation of Section 8.1.
A more explicit nonlinear pricing function motivated by Figure 8.7.1 is



For a fixed value of r:t. (based, for instance, on judgment), a value for Po can be
found. Using (8.7.12), revenue reconciliation impacts more on small demand
levels. Obviously one can also hypothesize nonlinear pricing structures where
revenue reconciliation impacts more on large demand levels, etc. Equations
(8.7.11) and (8.7.12) are only two examples of the many possible nonlinear
pricing structures.
It should be noted that the developments of Section 8.1 and 8.2 started with
the hypothesized "linear" form of (8.7.1), but can yield nonlinear prices. Table
8.1.1 and 8.1.2 showed that some choices of the weighting function QR (t) (for
weighted least squares) and the derivative od(t)/op(t) (for Ramsey) yielded Pk(t)
which are functions of dk(t), i.e., nonlinear pricing structures.
We do not pursue these nonlinear pricing ideas further because they complicate the analysis without any proven real-world advantages. The resulting
formulas generally depend strongly on the hypothesized customer benefit
function Bddk(t). However it is important to emphasize that
The basic principles of hourly spot pricing and the energy marketplace apply to nonlinear
as well as linear pricing structures.

If subsequent research reveals a particular nonlinear pricing structure with

desirable real world properties, it can be incorporated into our basic framework.

8. Revenue reconciliation



The preceding sections have discussed revenue reconciliation for a full implementation of a spot price based energy marketplace. During the initial testing
and implementation phase, it may be expedient to do revenue reconciliation by
the concept of "revenue neutrality" where
A spot price sequence Pk (t), t = 1 ... 8760 is said to be revenue neutral if it recovers the
same annual revenue that would be recovered under a present rale, assuming no change
in the kth customer's behavior, i.e., no response to price charges.

A general approach to this revenue neutrality considers some class of

customers where the k index denotes individual customers. Define
d2(t), t = 1 ... 8760: Last year's hourly usage of customer k (kWh).
Bill k : Annual bill that would be paid by kth customer under a present rate ($).


= 1 ... 8760: Spot price sequence expected next year with no revenue
Pkft:> = (I + m)Pk(t): Revenue neutral spot price sequence.
Pk(t), 1



desired to find the value of m that minimizes in some sense

+ m)R2

Bill, - (I






If there is only one customer of concern, then





and exact revenue neutrality is achieved.

For the more general case of multiple customers, a reasonable approach is to
choose the value m that minimizes


[Billk - (I + m)R2]2

Qk > 0: A weighting parameter


200 . . n. Tneofyof

This approach yields

L (Bill k)R2!Qk


Possible choices for Qk include

An Extension

Because of the cross-subsidies that exist in many present-day rate structures

(even within a given class of customers), the value of m resulting from (S.S.4)
may not yield anything close to revenue neutrality for many individual
customers. This may not be considered desirable for initial testing of spot price
One approach is to specify a different multiplier mk for each customer, but this
has many obvious disadvantages.
Another approach which may help is to replace (I + m) Rlc in (S.S.3) by

L [1

met, a)p,(t)dZ(t)

+ ....


Vector of parameters to be determined by minimizing (8.8.3)

One possibility for m(t,



a" sin



is a Fourier series representation

a2 cos


+ .,.

where the fundamental frequency is 24 hours, one year, or whatever does a good
We must emphasize that revenue neutrality as it is being discussedhere is not
a fundamental approach to revenue reconciliation. It is merely an expedient
which may prove useful during intitial testing of spot pricing.

A variety of different approaches to revenue reconciliation were presented (and

we have by no means covered all possibilities). For an ideal world, we recommend the revolving fund approach, since customers see the unreconciled spot
price that achieves maximum overall efficiency. Unfortunately, this revolving
fund concept is not compatible with much of present-day regulatory practice.
The same economic efficiency could be obtained by the surcharge-refund
approach if the amount of surcharge or refund seen by an individual customer


8. Revenue reconciliation


could be made independent of that customer's energy usage. Unfortunately,

there appears no way to accomplish this that would be practically acceptable.
Hence for the present-day real world, we have to recommend the use of revenue
reconciliation by modifying the hourly spot price as in Section 8.1 or 8.2.
The Ramsey pricing approach to modifying the spot price is theoretically
superior to the weighted least squares approach, but implementation of Ramsey
pricing requires data on customer behavior that are not presently available.
Hence the weighted least squares approach may be preferred on pragmatic
grounds. Fortunately, both the Ramsey and weighted least squares approaches
reduce to the constant multiplier formulas if enough assumptions and/or approximations are used. For present-day, real-world implementation, we recommend the constant multiplier approach where the constant multiplier can be
allowed to vary by customer class. The trade-off between aggregate and separate
generation and network (or multiple level network) revenue reconciliation
involves efficiency versus equity issues. Aggregate reconciliation is of course the
easiest to implement.
Because of the inherent uncertainty faced by power system planners, the sign
of the revenue reconciliation terms can be expected to change over the years (i.e.
overrecovery followed by underrecovery, etc.). We believe it is important for a
particular utility or regulatory commission or government agency (as appropriate) to choose one method and "stick with it." Without a firm commitment
to +W1e method, revenue reconciliation becomes a "political football."
Jtshould be emphasized that the choice of a revenue reconciliation procedure
is riot a problem that is unique to the spot price based energy marketplace. It
als{\ rears its ugly head in present-day types of transactions.
Some key revenue reconciliation issues not discussed here are the questions
of which costs should be covered, what is a fair return on equity, etc. We did
not discuss them here because they are the same whether one is dealing with a
present-day system or a spot price based energy marketplace.
We assumed throughout this chapter that all customers are seeing an hourly
spot price. The basic ideas also apply in an actual energy marketplace where
customers see a variety of spot price based rates including 24-hour updates,
billing period updates, and price-quantity transactions.
One concept not considered in this chapter is the use of a demand charge to
recover eIpbedded capital costs. Demand charges have essentially no role in a
spot price based energy marketplace (see Section 3.7).

Many of the aggregate reconciliation ideas of Sections 8.1 through 8.4 are based
on the extension of existing ideas into the time scale of hourly spot prices. The
weighted least squares concept in Section 8.1 appears to be new. The disaggregated reconciliation ideas of Section 8.5 are based on Schweppe, Bohn and
Caramanis [1985].
The general problem of efficiency constraining prices to meet a budget con-


II. Theory of hourly spot prices

straint has been vigorously debated in the economic literature. Hotelling's [1938]
article considered the problem of financing public works such as bridges where
the marginal cost of crossings is usually trivial. His answer to the pricing
problem was to set prices through taxes which (ostensibly) would not distort
consumption decisions, such as income taxes or inheritance taxes. Coase [1946,
1970] argued that from a broad public policy perspective, user support was an
important market test for efficient allocation of resources, and thus fees should
cover the total cost of the enterprise. He suggested the use of multipart tariffs
(such as declining block rates or a fixed fee plus a commodity charge) as an
alternative to government subsidies. Vickrey [1955] stressed that a misallocation
of resources can result if marginal cost pricing principles are not followed.
Baumol and Bradford [1970] proposed optimal departures from marginal cost
pricing with a generalization of Ramsey's [1927] rule. A much discussed special
result of their analysis is the "inverse elasticity rule":
If the cross elasticities of demand between the commodities in question are zero, then the
percentage deviations in price from marginal costs should vary in inverse proportion with
the own price elasticity of demand.

If cross-elasticities are not zero, a somewhat analogous rule still holds. More
recently, proposals for nonlinear pricing or multipart tariffs (see e.g., Willing
[I978]) have been suggested to be Pareto superior to the Baumol and Bradford
Peddie et al. [\983] address revenue reconciliation with variable elasticities
directly in a spot pricing context (although called dynamic pricing).

I. E( C( T R )} in (8.IA) is assumcd to be a given number because this corresponds to the way revenue

reconciliation is done for most utilities. An alternative approach is to consider thc fuel cost
component GFM (TR ) of C( T R ) to be a variable which depends on the g,(t). which in turn depend
on the Ii, (I). which in turn depend on the (I, (t). This approach leads to a slightly different set
of equations for p, (I). If a reader prefers this approach, it is relatively straightforward to modify
the equations appropriately.
2. Remember that Ii, (I) depends on pdl) so that, in general, (8.1.9) and (8.1.10) are not explicit
3. We ignore cross-elasticities and assume d, (I) docs not depend on p, (T) T #- I. We also ignore
the fact that I', (I) depends on d, (I) which in tUI'll depends on p,(I). The extra terms resulting
from such dependence could be carried along if desired.
4. The expectation operator is used only in evaluating JIlt.
S. If Q, (I) = Q, independent of time. demand, or price, then (8.1.17) and (8.1.19) show that p'(l)
does not depend on Q.
6. This argument is valid if demand response evidence indicates that demand is elastic during peak
periods and nonelastic during off-peak periods. Hence prices deviating from optimal spot prices
during ofT-peak periods do not affect customer behavior appreciably.
7. As in (8.1.16), we ignore cross-elasticities.
8. In the future, when a lot of customer response data become available, this statement may
9. This is the "inverse elasticity" rule that results from Ramsey pricing.

8. Revenue reconciliation


10. A more elegant mathematical argument which leads to the same result would introduce a
Lagrange multiplier to handle the explicit constraint of 1 MW on the customer's generation
11. An interesting academic exercise is to go to the other extreme and assume TR = I hour in
Section 8.1. This effectively destroys any resemblance to marginal cost pricing.
12. It should be noted on the other hand that overcapacity generally implies low spot prices. Hence
if spot prices are left unchanged, they would reinforce load growth and hence increase the chance
that revenues will catch up with revolving fund deficits.
13. Other problems with the separate revenue reconciliation approach applied to individual generators or lines can arise from old equipment whose capital costs are already "written off."
14. The mathematics of weighted least squares can be used to yield (8.5.9) and (8.5.10) in a relatively
straightforward fashion. However, we have not developed a corresponding Ramsey pricing


6 through 8 presented the theory underlying the hourly spot price
defified as follows:
Hourly Spot Price: Marginal value of energy ($/kWh) for the next hour computed at the
beginning of the hour assuming complete knowledge of the operating conditions, costs,
etc., that will ex,ist during the hour.
Here in Chapter 9, we discuss how to compute spot price based rates - i.e.,
transactions which are based on the hourly spot price.
Sections 9.1 through 9.4 discuss transactions with different time scales and
types of contracts between the utility and its customers:

Predetermined Price-Only: A ($/kWh) price determined at the beginning of

some time period (e.g., hour, day, month, or year) which applies until the next
price update at the end of period. A short-term fixed-price-variable quantity
contract (see Section 9.1).
Price-Quantity: A short-term utility-customer contract wherein the customer
allows the utility to exercise certain quantity control between times of updates
of the predetermined price-only transactions (see Section 9.2).
Long-Term Contract: A long-term utility-customer contract (days to many
years) combining fixed-price-fixed-quantity contracts with price-only and/or
price-quantity (see Section 9.3).


II. Theory of hourly spot prices

New Customer Contract: A long-term utility-customer contract (many years)

involving special rates, etc. for new customers based on their long-term price
elasticities (see Section 9.4).

Sections 9.5 and 9.6 return to an hour-by-hour time scale and discuss
wheeling rates, i.e., the theory of how to charge for transmission services.
Section 9.5 discusses spot wheeling rates that two nonadjacent utilities ought to
pay for trading electric power using the transmission networks of other interconnected utilities. These wheeling rates depend on the hourly spot prices
prevailing at the boundaries of each utility. Section 9.6 considers wheeling
between a user and a private generator or a user and another utility. The key
issue here is how revenue reconciliation for generation is handled.
The only ideas developed in Chapter 10 from Chapter 9 come from Section
9.1, and they are not essential to the main theme of Chapter 10. Therefore
readers can take one "giant step forward" to Chapter 10 if the concepts of this
chapter are nol of immediate interest.

The framework used for the discussion is a marketplace where some customers
see hourly spot prices while the remaining see predetermined rates.
Costs and benefits of generating and consuming electric power at a specific
moment in time are assumed to be independent of past and future generation
and consumption levels in order to simplify the derivation (see Section 6.5).
Revenue reconciliation as discussed in Chapter 8 is not included here. It can be
added if desired at the expense of notational complexity. For simplicity, we
discuss only the case of customers buying energy from the utility. However, the
results generalize to the case of customer-owned generation that is being bought
by the utility.
Problem Formulation and Notation

The predetermined rate of concern is defined as

PA(tlr): Rate ($/kWh) to be charged during hour t as calculated given all the information
available at hour r, r < t.

For a 24-hour update (see Chapter 3), r might be 4 PM while t varies 24 times
with one-hour steps starting at 2 AM the next morning. However, to simplify
metering, a given implementation might keep Pk(tlr) constant over some range
of t. A three-hour example is-

In general symbols'
- Pk(tlr)

Constant for all r



9. Spot price based rates


I ct::. I(r): Is mathematical notation for all the hours I belonging to the set t(r).

T(r): Total number of hours in the set t(r)



For the three-hour example, the set t(r) contains 2 AM, 3 AM, and 4 AM and
T(r) = 3.
To simplify the appearance of subsequent equations, the t and t notation is
often suppress and we define



rA = PAUl,): r, is the value of the spot price based rate predetermined at time, and valid
for I ct::. I(r).

For the two different classes of marketplace customers,




Customers Seeing Hourly Spot Price: dk(t) denotes their demand.

Customers Seeing Predetermined Rates: dk(t) denotes their demand.

The results generalize readily to the case where different classes of customers see
predetermined rates with different update times (as determined by r) and period
definitions (as defined by the set t(r.
General Result for Predetermined Rates


Tbihourly spot prices, pdt), and predetermined rates, r k , seen by each of the
tw~classes of marketplace customers are shown later in the section to be given






as defined inc Chapter 7, and



E { {ik(l) -"-.-



I f 1(1)


E {(I(~f'(



( , J...

E: Conditional expectation operator applied at time [ given all information available

at time,

A more accurate (and classy) notation would use E {Pk(t)!r} instead of just
E {Pk (t)}, etc.
When the set 1(1') contains only one hour I, (9.1.2) can be written


Cov { Pk (I), Dd('(t)}

+ ----"--;----:-"--<orA


Cov{a, b}

E{[a - Ea][b - EbJ}

Discussion of General Result (9.1.1), (9.1.2)

The basic result (9.1.2) is complicated by the presence of the derivative which
depends on the choice of a customer response model.
If the derivative iJdf(t)/iJrk is constant in time and deterministic, then (9.1.2)

T(r): Number of hours in t(r)

While (9.1.3) reduces to

In Appendix E, four different customer response models are developed which

Linear: Od(t)/iJr = P(do(t)/ro)
Power: P(d(t)/r)
Exponential: P(d(t)/ro)
Log: P(do(t)/r)


Elasticity Parameter

ro: Nominal Price

do(t): Nominal Demand (when r


Assume P and ro are deterministic and constant in time. Substituting the exponential or power response models into (9.1.2) yields




while the linear and log response models yield a similar result with the nominal
dO.k(t) replacing dk (t).2 When the set t(r) is one hour, (9.1.4) becomes

9. Spot price based rates 209

Thus the predetermined rate rk can be greater than or less than the expected
value E {Pk (I)} because the covariance can be negative or positive for a specific
It should be noted that since dk(/) depends on Pk(/), socially optimal marketplace behavior of hourly spot price customers can be achieved by appropriately
selecting either dk(/) or Pk(t). The same is not necessarily true with df(t) since
the predetermined rate rk can be constant over many hours. This limitation in
the degrees of freedom results in reduced social welfare, which should be
compared to the savings in metering, communications and transactions costs.
Derivation of General Result (9.1.1) and (9.1.2)

The definitions of Chapter 7 are used, except the demands are decomposed into
the two classes of customers: those that see hourly spot prices, and those that
see predetermined rates. Define (in addition to the definitions of Chapter 7)
df(t): Demand of kth customer seeing predetermined rates (kWh)
dP(/) = "f.kdt(/)
dP(/): Vector of all dk(t)
B[~P(t)]: Total Customer Benefit for the customer class seeing predetermined

Asrwme a bus k may have at most one generator and one customer from each
"'" classes. As in Chapter 7, assume for simplicity that the swing bus has
of th~wo
only"!t generator. Thus the individual line flows are given by

The energy balance constraint of Chapter 7 becomes

get) =




The usual assumption of rational behavior of marketplace customers is made,

o Bddk (t)]
odk (t)


We now follow our usual path and form a Lagrangian,

!l =








O(t) =

is the


of Chapter 7 except for the existence of both dCI) and



(Generation Costs, Constraints)

(Network Costs, Constraints)


- B[~(t)J

(Benefits, Spot Price Customers)

- B[~P(t)]

(Benefits, Predetermined Rate Customers)


dP(t) + L[z(t) - get)}

(Energy Baiance)

Since generation dispatch and hourly spot decisions are made on-line whereas
predetermined rates are set in advance, the first order optimality conditions are



Equation (9.1.7) yields, by following the same steps as in Chapter 7,

Pk(t) ==


)le(t) 1 + - a(d


which yields (9.1.1).

Equation (9.1.8) gives


E{[aN[Z(t) - aB[df(t)]

)l (t)

(I +

aL[Z(t)])] Me(t)}

or using (9.1.6) and (9.1.9),

L E { [Pk(t)

- r.) -iJ-




which yields (9.1.2).

Predetermined Price-Only Rates and Rationing Costs

The relationship between predetermined price-only rates and hourly spot prices
has been derived without modeling rationing costs. In practice, predetermined
rate customers will sometimes be rationed by the utility on occasions of generation or network capacity shortfalls.

9. Spot price based rates 2 t 1

The incorporation of these rationing costs for predetermined customers leads

to the conclusion that
Rationing costs for predetermined rate customers place a ceiling on the hourly spot price.

This is a reasonable conclusion. After the hourly spot price reaches a certain
level, it is socially better to drop a predetermined rate customer than to raise
further the price seen by hourly spot price customers.
Derivation of (9.1.11)

Generation capacity shortfalls can be "distributed" between the two classes of

customers by raising hourly spot prices and/or rationing predetermined rate
customers. Define

Actual consumption level of predetermined rate customer k after rationing. If there is no rationing, ~k(l) = df(t)



Rddf(t) - ~k(t: Rationing costs incurred by predetermined rate customer k

at time I if it desires consumption df(l) and is rationed to ~k(l)

df!) -

Rk is defined for simplicity to be a function of the amount rationed,




The amount of rationing is assumed to be determined on-line, that is the utility

decides at each hour t, ~k(t) for all k subject to the constraint

To derive the first-order social cost minimization necessary conditions, the

Lagrangian is modified to become

L 'E{n(t)}

(Generation Costs, Constraints)

net) = G[~{t)l


(Network Costs, Constraints)


(Benefits, Spot Price Customers)


(Benefits, Predetermined Rate Customers)

- g(t)



(Energy Balance)

212 11. Theory of hourly spot prices

+ L Rk(df(t)

- 6 k (t

(Rationing Costs of Predetermined Rate

(Amount Rationed is Constrained to be a
Positive Quantity)

Considering that generation dispatch, hourly spot prices and rationing (if
needed) are on-line decisions while predetermined rate setting is an off-line

OPk (t)

06 k (t)


I~) E {o~:t)}



Equations (9. J. 14) and (9.1.15) yield (9.1.1). Equations (9.1.16) and (9.1.1 7)
yield, after collecting terms using (9. I. 9),

a6 (t) - IlR.k(t) =

~E {[[pdt) +


a6 k (t)

06 k (t) - 1lR.k(t)] odf(t) -




When (9.1.18) is substituted in (9.1.19) we obtain (9.1.10) which yields (9.1.2).

Hence, the inclusion of rationing costs does not change the relationship between
predetermined rates and hourly spot prices.
However, (9.1.18) together with complementary slackness which implies
J1.R.k(t) = 0 when ~k(t) = df(t) (i.e., when there is no rationing) yields

No rationing when


Rationing of predetermined rate customers when hourly spot prices would

otherwise exceed marginal rationing costs to predetermined price customers.
The amount of rationing is determined at the level that achieves

9. Spot price based rates


Thus there is always a ceiling to the hourly spot price when a relatively large
segment of marketplace customers see predetermined rates. The ceiling is the
marginal rationing cost to the predetermined rate customers. When rationing is
done through a nondiscriminating procedure such as rotating blackouts or
brownouts (as opposed to a priority list in order of increasing rationing costs),
the spot price ceiling might be the average customer cost of unserved energy.
In Chapters 3 and 6, several different ways to specify the quality of supply
components of the hourly spot price were discussed, two of which were
Market Clearing: Change the hourly spot price until customers respond enough
so that rationing is not required.
Cost of Unserved Energy: Use quality of supply cost function based on cost of
unserved energy.
The ceiling demonstrates a relationship between the market clearing and the
cost of unserved energy approaches.

transactions of Chapters 6 through 8 and Section 9.1 are fixedpriCe-variable quantity contracts wherein the utility specifies a fixed price (for
thenext hour in Chapters 6 through 8 or for hours, days, months, etc., in Section
9.I}and the customer can buy any amount of energy at the quoted price. In this
section (and in Section 9.3) we explore other types of contracts which involve
some specification of quantities as well as prices.
Motivation for Price-Quantity Transactions

Two potential motivations stated in this book for considering price--quantity

transactions are to

Reduce transactions costs and exploit utility administered load control

Provide the utility with more accurate control of load for system security

Transactions costs can be lowered by reducing the information that needs to

be communicated between the utility and the marketplace participants. Transactions costs can also be lowered by transferring implementation of customerspecified load-control strategies to the utility, whenever this results in reduced
overall costs of control implementation hardware. As discussed in Section 3.4,
this often leads to disguised price-only transactions such as certain practices of
direct load control and interruptible contracts. As a direct load control example,
assume a certain customer's end use has a value that does not warrant purchas-


lCTheory of hourly spot prices

ing electricity at a price exceeding a threshold level known in advance. The

customer may authorize the utility to interrupt service to that end use whenever
the spot price exceeds the threshold. This agreement reduces the price information that must be communicated by the utility to an infrequently transmitted
service interrupt/reinstate signal. In this case, customer billing can be done at a
discounted predetermined price calculated by averaging the spot price over all
instances when it is lower than the threshold level prespecified by the customer.
This idea can be extended to multiple end uses and multiple threshold levels that
result in a priority service system. An example of an interruptible contract that
is a disguised price-only transaction is the common contract provisions that
failure to drop load when requested leads to a financial penalty customers have
to pay. The financial penalty is simply a high spot price.
One situation where price-quantity transactions do have a clear advantage
over price-only transactions is consumer participation in system security
control, where the operating reserve needs of the utility (see Appendix B) are
covered at least partly by the customer's load (or generators), rather than
exclusively by the utility's generators. In this case, the utility needs to have fast
(time scales of seconds to minutes) and accurate control of a relatively small
percentage of the total load. Although the need to exercise system security
control is infrequent, the catastrophic consequences of its failure justify more
careful consideration. The need for transactions characterized by priorities and
quantities agreed upon in advance arises through system security control-related
uncertainty, that reveals itself during the time period that the spot price discussed
in previous chapters is effective. This uncertainty is distinctly different from the
uncertainty considered in Section 9.1 under predetermined price-only transactions, because it is associated with the need for fast and accurate response
required by the engineering characteristics of system security control.
This section discusses both system security control and reduced transactions
costs motivated price-quantity transactions.
System Security Control Motivated Price-Quantity

The formulation of the hourly spot price has so far considered generation and
consumption costs and benefits, transmission line flows, and energy balance,
assuming that all relevant uncertainty is known at the time of price setting.
System security control introduces additional considerations, some of which can
be summarized in the following two issues: 3

Contingency Planning Requirements: Preparatory actions have to be taken in

advance to secure resources (e.g., spinning, cold start and other operating
reserves) which enable system control under a range of possible uncertainty
Security Control Implementation: Actual use of the resources secured
through contingency planning. It takes place after uncertainty has revealed

9. Spot price based rates


The above two issues are systemwide. The only contingency planning requirement in Chapter 6 was an operating reserve requirements constraint. Utilityowned generators were assumed (see discussion in Section 7.8) to be dispatched
subject to the operating reserves constraint while other participant individual
decisions ignored it. This assumption is now removed by treating operating
reserve requirements as an additional commodity. Marketplace participants,
consumers and non-utility-owned generators can also contribute towards
meeting system security planning and control requirements. Therefore their
costs in preparing to supply resources (contingency planning requirements
participation) and their costs in actually supplying them, if needed, should be
distinguished and modeled. For example, a generator that contributes to operating reserve requirements (a common contingency planning method) incurs costs
for maintaining an operating state, and possibly opportunity costs for not
generating at a higher capacity so that it will be capable to make its reserved
capacity available if needed for security control. The same generator will incur
additional fuel and maintenance costs if it is called upon to actually utilize its
spinning reserves for security control implementation. Marketplace participants
are also responsible for the level of resources needed for effective contingency
planning and control implementation. For example, a large generator imposes
a higher burden on operating reserve requirements than a smaller generator.
The major results that follow when security control uncertainty is modeled
are~iscussed next.
Prii;., with security control modeling

The'\planning and implementation issues introduced above are modeled by

introducing four new system quantitites.

= Minimum contingency planning resources that must be secured at the

beginning of the time period (MWh).
S< = Uncertain quantity of resources that are actually needed during the time
period to implement security control. A random variable that can assume
values between 0 and Sp (MWh).
Sk.r' S,.< = Market participant k contribution to Sp and Sc' Sk.c is a random
variable between 0 and Sk.p (MWh)

The units of the above quantities are energy. A more precise characterization
. might be capacity available for a certain duration. For notational simplicity, we
assume that security control implementation requires capacity for a fixed a
priori known duration, and hence energy units are appropriate.
The model used in previous chapters is augmented by the addition of two new



with Lagrange multiplier lip



with Lagrange multiplier 11<


L 5,.,

216 II. Theory of hourly spot prices

and also the inclusion of the relevant costs of participants for supplying Sk.p and
The constraint g(t) ~ gcrit,y(t) imposed in earlier formulations will always
be met if inequality (9.2.1) is satisfied. The Lagrange multiplier /-lQSs (of the
earlier formulation) is zero when enough generating capacity to meet load and
operating reserves is available. The Lagrange mUltiplier /-lp will be zero much less
often since operating reserves must be always secured.
The reader should note that gcrit.y(t) which was defined as a constant in Section
6.2 (i.e., gCrit.y(t) = gmax(t) - gres(t is actually dependent on individual generation levels. Indeed gres(t) in (6.2.1) is equivalent to Sp of this section which
depends on generation levels. For example, gres(t) may be considered to equal
the output of the largest generator at time t. In the simpler formulation of
Chapter 6, gres(t) was assumed to be a constant and hence

for all k was used in deriving (6.3.6). This simplifying assumption is removed
here and Sp is generally considered to depend on the gk levels.
After the usual formation of the Lagrangian and optimization with respect to
Sk.p and Sk,c' the following results are obtained.
For a generator:

For a demand:

Ih(t): Normal operating spot price as in Chapter 7
Pp: Contingency planning spot price; equals /-lp
Pc: Security control implementation spot price; equals /-lc

If Pc > marginal cost to k for providing


To see how security control prices work, consider a customer (load) who
pledges all of his/her demand. Thuss
If Skc


If Skc


Therefore if

dk =

dk =


9. Spot price based rates


BiIl k : Bill at end of pricing period

[Pk(t) - PpjSkp




[- Pp - PcjSkp




The contingency planning spot price Pp (which equals the Lagrange multiplier
J1.p) has an intuitive interpretation when the utility provides alJ required operating reserves by spinning standby generators. In this instance Pp is the incremental cost incurred by the marginal spinning generator k in supplying the last kWh
of operating reserves Skp' In equation form,

a(Spinning cost of generator k)






Depending on the characteristics of the marginal contributor to system operating reserve requirements, Sp, Pp may be quite small and at times may even be
The result (9.2.3) (9.2.4) has a form which is similar to the familiar results of
previous chapters where the various additive components of the optimal spot
Pfice correspond to modeling a specific marginal cost imposed by an increment
the participant's consumption or generation level. In this case, each participant is charged/paid for its incremental impact on system contingency
planning reserve requirements and system security control implementation
requirements. It should be noted that most loads do not affect Sp and Sc. Thus,
except for very large loads,


so (9.2.4) usualJy reduces to

However, for large generation the last two terms of (9.2.3) can be important.
For example, if gk is the largest generator's output at time t, security control
contingency planning could require Sp = gk> which results in


thus implying that the kth generator should be "charged" at the rate of pp.
Similarly, if generator k is suddenly forced to disconnect from the grid causing
the need for Sc = gk> it should be charged at the rate of Pc' The spot price


II. Theory of hourly spot prices

modification of (9.2.3) internalizes security control costs and can thus be viewed
as providing the proper incentives for generators to behave in a way that
minimizes system security related costs.
Implementation of security control prices

One overall procedure for implementing security control is as follows. The

hourly spot price, contingency planning quantities Sk.p, and contingency
planning price Pp are decided at the beginning of the pricing period so as to meet
the energy balance and planning requirements constraint (9.2.1). This could be
done in an iterative bidding process as follows. First the market maker (e.g., the
utility) announces the normal operation spot price Pk, the contingency planning
spot price Pp, and the probability distribution of the security control implementation spot price Pc' Then participants respond by announcing their demand
level db contributing to planning requirements Sk.p, and reservation price h.c
which will be used to prioritize their contributions to security control impJementation. 6 This iterative bidding can be repeated until the constraints are met.
The resulting quantities and prices can then be used to implement the priority
of reserve utilization and credit/charge participants optimally for their contributions.
The security control procedures described above can be extended to cover
multiple quantities and contingency prices set by a single participant and even
have those prices and quantities depend on the specific time in the time period
that quantities will be used for implementing security control, as well as the
duration of their use. At the same time, a range of simpler transactions can be
designed. We next discuss some of these simple transactions which happen to be
particularly applicable when participant costs (reservation prices) are well
known in advance and do not vary substantially with time.
Prepayment or postpayment only

The general security control approach involves a price reduction Pp for pledges
and a postpayment Pc for actual control. The resulting bill for demand is
given by (9.2.5). Two simpler cases are


Prepayment Only:


Prepayment only contingency planning price (i.e. price reduction)

Postpayment Only:

Peo: Postpayment only control implementation price.


9. Spot price based rates 219


Values of Pbo and Pco can be chosen by matching the expected values of the bill
of (9.2.5) to those of (9.2.6) and (9.2.7).
To illustrate how this can be done, define


probability that Skc

= 0,

i.e., that no control is required .

Then the expected bill, E{Billd, for the general case (9.2.5) is given by



E{Bill k } = n[Pk(t) - pplskp

(I - n)[ - Pp - .okclskp

skp[nPk(t) - Pp - (I -




.okC: Expected value of Pc over all values exceeding the reservation price of customer k


For the prepayment only case, the expected bill from (9.2.6) is given by

Equating(9.2.8) and (9.2.9) yields

Pbo =

- [pp

+ (\ - n)/\cl


the postpayment only case, the expected bill from (9.2.7) is given by

so that equating (9.2.11) and (9.2.8) yields


When considering the results (9.2.10) and (9.2.12), remember that for most
situations the probability n of no control will usually be close to one.
Many present-day interruptible contracts are examples of prepayment only,
Interruptible Rate = Pk(t) -


Hence a consistent setting of such rates should follow the procedure outlined
above if the security control philosophy is being used.
A new pricing proposal (Chao and Wilson [\985]) related to post payment
only is interruption insurance, in which the utility pays customers who are
interrupted. One difference is that the proposed insurance payment rates differ
across customers, while in the energy marketplace proposed here, the same Pc

220 II. Theory of hourly spot prices

applies to all customers. The two proposals are alike, however, in that they both
cut off exactly those customers who are most willing to be interrupted.
Reduced Transactions Costs Motivated Price-Quantity

Price-quantity transactions often arise as disguised price-only transactions. The

threshold value of a certain end use example discussed earlier in this section is
a case in point.
Using the notation of Section 9.1, define
pdtlqk): Spot price that will prevail at time t if event qk occurs.

Assuming qk is a binary event, define

= 0: Event resulting in a spot price smaller than the customer's threshold

value of end use.
qk = I: Event reSUlting in a spot price larger than the customer's threshold
value of end use.
n: probability qk = O.

The expected spot price over possible qk events is then given by

Noting that

we define further

Pk(tlqk = 0)


[Ph (t Iqk = I) - Pk (t Iqk == O)Jn

A customer may contract to purchase interruptible power Sk by either agreeing


Drop Fixed Amount: i.e., on utility's signal, the interruptible customer

reduces demand by a prespecified fixed amount Sk'
Drop to Prespecified Level: i.e., on utility'S signal, the interruptible customer
reduces demand dk to a prespecified level lk' Thus Sk = dk - lk'

The utility can provide financial incentives to an interruptible customer by

either a combination of pre- and postpayments or by using one of the two
simplified approaches discussed above, namely:

9. Spot price based rates


Prepayment Only: Charge the interruptible customer lower rates

expected bill is

Postpayment Only: Charge the interruptible customer the expected spot price
rk but offer a payment when interrupting at the rate i'k.l so the expected bill

k O

so the

The pricing structures for the above prepayment only and postpayment only
approaches are identical to the corresponding security control structure of
(9.2.6) and (9.2.7). However, the underlying logic for computing the price is
basically different since in the present case the price-quantity transactions are
really just a way to implement a price-only transaction.
Multiple Levels of Reliability

The basic ideas can be extended in many ways to yield quite sophisticated
utility-customer transactions. An interesting extension involving varying levels
of reliability is the following. A utility could offer its customers a choice in the
level of reliability they want to buy so that customers who pay a lower rate are
lo~r in the priority ordering and thus get interrupted more frequently.
Direct Load Control

Andther disguised price-only transaction is direct load control wherein the

utility provides the customer with financial incentives to be allowed to directly
control individual customer appliances.
As an example, assume the utility offers a special fiat rate ($/kWh) for the
energy used to provide hot water or to heat bricks (or to make ice) used for space
conditioning under the condition that the utility directly controls the heating or
cooling devices. Consider a 24-hour update cycle where
d(t): Energy used by device during hour t.

Assume tnat for a particular day, the utility knows a priori that the customer's
needs can be represented by the constraine

L d(t)



where dreq may depend on day of week and/or outside temperatures (for space
conditioning). Then the utility will obviously control the device's d(t) to meet
the constraint while minimizing


II. Theory of hourly spot prices

where Pk(t) is the utility's marginal cost of providing energy to the customer.
The flat rate the utility offers the customer is based on the expected values of the
Pk(t) and the optimum d(t); t = I ... over say a one-month interval.
Such direct utility control of water heating and/or thermal storage can be
viewed as a utility-provided tactical control service (See Section 4.3).
Discussion of Price-Quantity Transactions

Price-quantity transactions are based on the behavior of the hourly spot prices.
Many price-quantity transactions are very close to or equivalent to special types
of price-only transactions.
In general we question whether there are actually significant transactions
cost reductions from such transactions, compared with just putting the
customer onto a standard price-only spot price, perhaps with only two price
levels. The metering, communications, and customer response costs will be
close, or even lower for the spot price since customers do not have to pre-select
and communicate the quantity information. Hence for large (non-residential)
customers, we will be surprised if transactions cost savings outweight the
reduced benefits of price-quantity transactions.
However, price-quantity transactions based on the utility system's needs for
security control are basically different than price-only transactions, and fill a
unique role of fast and accurate response. A large practical difference between
security control motivated price-quantity transactions and present interruptible
contracts or interruption insurance proposals [Chao and Wilson, 1985] is the
duration of advance commitment required from participants. Most customers
(and generators) have time-varying and stochastic willingness to provide
reserves, as well as stochastic time-varying overall demand. For example, an air
conditioning load can provide more reserves on a hot summer day while it can
provide no reserves during the winter. Industrial consumers such as electric
furnace factories are affected by the overall business cycle, specific orders in
hand, shift changes and the physical production cycles of their equipment. It is
therefore inefficient to force them to precommit to a fixed level of interruption
for a year as is the practice with many interruptible contracts. Hence, the
reserves market should be frequently resolved for the appropriate group of
participants. A reasonable time scale is roughly one order of magnitude longer
than the duration of reserve use. For example, "spinning" type reserves may be
beneficially reauctioned each hour or possibly every few minutes. Of course such
auctions would be conducted computer to computer, using standard response
logic set by the customers to place their bids.

The price-only transactions of Section 9.1 are a fixed-price-variable quantity

9. Spot price based rates


contract between the utility and its customers. The price-quantity transactions
of Section 9.2 combine price- and quantity-based transactions (such as interruptible) where these quantity transactions operate within a price-only update
Here in Section 9.3, we combine price-only transactions with a fixed-pricefixed-quantity contract that works on a slower time scale than the corresponding price-only update cycle. These long-term fixed-price-fixed-quantity
contracts are motivated by customer desires to know in advance what their
energy bills are going to be.
Two basic implementation approaches are
Utility Long-Term Contracts: A fixed-price-fixed-quantity long-term contract
between the utility and one of its customers.
Futures Market: Fixed-price-fixed-quantity futures market contracts between
buyers and sellers as negotiated by an independent energy broker.
A futures market will evolve only if there is sufficient demand for its existence.
Basics of Fixed-Price-Fixed-Quantity Transactions

Assume the fixed-price-fixed-quantity contracts are being combined with onehour update spot prices. Extensions to other predetermined price-only transac,tions such as 24-hour update (and to price-quantity) is conceptually straightfqJ;.ward.
The bill for energy used during hour t by the kth customer with a fixed-pricefixed-quantity long-term contract is
Bill, (I) = Pk (I)[d, (I) - du(t)] + dk.r(t)fk(t)($)


PACt): Hourly spot price of Chapter 8 ($/kWh)

dk(I): Actual energy used during hour I (kWh)
dk/I): Amount of energy specified in the fixed-price-fixed-quantity transaction (kWh)

fk(t): Price of fixed-price-fixed-quantity contract ($jkWh)

The values of dk,r(t) and/k(t) are specified hours, days, weeks, months or years
before hour t .
Equation (9.3.1) illustrates the two key features that result when a price-only
one-hour update is combined with a fixed-price-fixed-quantity contract:

If the actual demand equals the quantity fixed in the contract, i.e., if
dk(t) = dk,r(t), then the bill is independent of Pk(t), i.e. depends only onlk(t).
The customer's decision at hour t on how much energy to use (i.e., the value
of dk(t)) is based solely on the hourly spot price and is independent of either
dk/(t) or Ik(t).

224 H. Theory of hourly spot prices

To illustrate the second feature, let Bk[dk(t)] denote the customer's benefit
function so at hour t the customer chooses dk(t) to maximize
Bddk(t)] - BilIk(t). This yields a dk(t) given by

which is the equation used throughout Chapters 6, 7 and 8.

Equation (9.3.1) can be rewritten in the form

If the fixed-price-fixed-quantity contract is a long-term contract with the utility,

the customer can be viewed as paying the utility

Pk (t)dk(t) for the energy used.

[fk(t) - Pk (t)] dk/t) for the long-term contract. This quantity can be positive

or negative.

The Billk(t) can be negative if, for example, dk(t) = 0 andh(t) < Pk(t).
For the futures market-energy broker implementation, there are various
possible ways the cash could flow. One approach is

Customer pays Broker Billk

Broker pays Utility Pk(t)dk(t)

Another approach is
" Customer pays Utility Pk(t)dk(t)
o Customer pays to or receives from Broker (h(t) Pk(t dk;(t)
The following subsections address the issues of how a customer chooses a
value of dk,f(t) and how the utility or energy broker specifies a value for h(t).
Customer Choice of dk,r(t)

Leth(tlr) be the priceh(t) as specified at the much earlier hour r. The customer
at hour r specifies the value of dk,f(t) given knowledge of h(tlr). Two cases of
interest are
Case I:hUlr) > E{Pk(t)lr}
Case II:h(tlr) < E{Pk(t)lr}
In Case I, where hUlr) is greater then the expected value of the spot price
(given all information available at time r), the customer loses money on the
average. Hence the customer chooses dk,f(t) > 0 only if the customer receives a

9. Spot price based rates 11S

benefit from having a prespecified price - i.e., if the customer is willing to pay
a premium for the insurance provided by the fixed-price-fixed-quantity
In case II, where!k(f\r) is less than E{Pk(t!r)}, the customer makes money on
the average. The expected profit increases as dk.,(t) increases. In this case, the
customer chooses a value for dk./(t) based both on the benefit of a prespecified
price and on the customer's willingness to gamble. For example, if a customer
chooses a very large dk./t) and when hour t finally comes, Pk(t) < };,(t\,), the
customer will have to pay a lot of money. Case II might occur when utility
long-term contracts are affected by revenue reconciliation, as will be discussed
Energy Broker Specification of fk(tl,)

An independent energy broker acts as an intermediary between individuals who

want to buy or sell energy in a futures market. Thus it is the futures marketplace
that really decides the value oflk(t\,). The value of/k(t) determined at times '\
and '2 will usually be different. The energy broker adds a "fee" for his/her
services to cover operating expenses and to make a profit. The energy broker's
fee could be regulated.
Utility Specification of [,.(tl,)


can use various methods to specifY!k(t\r). Four of these are

Act Itke an energy broker

't\1aximize profit
Social welfare revenue reconciliation
Use long-term contracts to achieve revenue reconciliation

Utility as energy broker

The utility can act just like an independent energy broker where fees are either
set at costs or are set higher with the profits used to reduce the bills of regular
Maximize profit

The utility can act as the energy seller who is part of a futures market and simply

where/k(t!,) is chosen to yield dk./t) which maximize

On the average, the utility always makes a profit which can be used to reduce
the bills of its regular customers.

Social welfare revenue reconiliation

The utility can set the socially optimum h(t) by repeating the derivation of
Section 8.1 where the revenue reconciliation constraint of (8.1.4),

is modified to become

Either the weighted least squares or Ramsey pricing approach can be used to
find the optimum Pk(t) and/k(t) simultaneously. The Ramsey approach yields
h (t) that are very sensitive to the benefit function chosen to model the value to
the customer of prespecified prices and the method of modeling the customer's
willingness to gamble - i.e., risk aversion.
Use of long-term contracts to achieve revenue reconciliation

As discussed in Chapter 8, the ideal method of revenue reconciliation does not

modify the unreconciled i\ (I) to get Pk (t). Instead some other mechanism (such
as a revolving fund) which does not depend on dk(t) is used. Long-term fixedprice-fixed-quantity contracts fulfill this criterion because they have no effect on
dk(t) (at least in a short-term operations sense). Thus, for example, the utility
could try to fix theh(t) such that
E{C(TR )} =

E{,~f 1\ (t)d (t)}


E{,~lf (j~(t) -



where Pk (I) is the unreconciled spot price of Chapter 7 which is now seen by all
of the utility's regular customers. This should work (in theory) if the utility
overrecovers when charging Pk(t), since then

(i.e., Case II as discussed above), and one expects that enough customers would
be willing to make a profit (on the average) to enable the condition (9.3.4) to be
met. If the utility underrecovers when charging p(l), the condition (9.3.4) is
achievable only if the benefits of the insurance are large enough to that the
resulting dk./(/) are large enough.
Discussion of Long-Term Contracts

The basic concept of combining price-only transactions with fixed price-fixed

9. Spot price based fates


quantity contract is sound. However we have not yet fully explored the many
issues associated with its implementation, e.g. utility incentives to manipulate
forward prices, and the exact terms of such transactions.

Consider a utility with e)<.cess generation capacity. The utility's existing

customers could welcome new customers onto the system because new
customers help share the capital costs and hence can reduce the prices seen by
the old customers. Therefore, it can be argued that under certain excess capacity
situations, new customers should be offered "special incentive rates" to encourage them to move into the utility's service territory. Old customers who add
new end use devices could also see special rates.
Consider a utility with a shortage of generation capacity. A similar argument
can be made that new customers should be made to pay "special, high rates" to
discourage them from moving in.
Similar arguments can also be applied to customer owned generation, albeit
the signs change.
One of the problems with implementing such special customer contracts is
that after a new customer has moved in (or installed new end use equipment),
the new customer rapidly becomes an old customer. Thus implementation
requires the use oflong-term contracts that specify the manner in which a special
new customer evolves into an old customer.
Tf)ts section uses the Ramsey pricing approach to revenue reconciliation to
derive equations that justify special customer contracts from the social welfare
max~ization point of view. The Ramsey approach to revenue reconciliation
(see Section 8.1) gave rise to the inverse elasticity rule; i.e.
Customers with small elasticities (in magnitude) should pay more in times of underrecovery (over capacity) and less in times of overrecovery (under capacity).

Conceptually, one can define two types of elasticity

Long-Run Elasticity of Capital Stock: Determines the customer's decision
to move into a new area or to buy new end use equipment.
eSR' Short-Run Elasticity of Operation: Determines the customer's hourly operating decisions.


The basic argument for special customer incentive (or disincentive) rates is that
the effect of the long-term capital stock elasticity eLR can dominate the shortterm operating elasticity BS R '
We will now do some mathematical gymnastics which quantify the above
ideas. Only readers with a special interest in the subject should proceed with the
remainder of this section. We do not address the subject again at any later part
of the book.


n. Theory of hourly spot prices

Ramsey Pricing Revisited

I n Section 8.1, the following equation was derived (see (8.1.23):

(9.4.1 )

Pk(t): Hourly spot prices without revenue reconciliation
Pk(t): Spot price with the Ramsey approach to revenue reconciliation
Ji.R: Lagrange multiplier adjusted to yield the desired annual revenue
Ji.R > 0: Implies excess capacity (i.e., underrecovery)
Ji.R < 0: Implies shortage of capacity (i.e., overrecovery)

The derivation of (9.4.1) ignores cross-elasticities; i.e., it was assumed that

t "" r

If this assumption is removed, the generalization of (9.4.1) is


fh: Vector of the Pk(t) t
Vector of the Pk(t) t
dk : Vector of the dk(t) t



I ... TR

I ... TR
I ... TR
Dk = (odk/opd: A TR by TR matrix
T R : Time Tnterval for revenue reconciliation
Derivation of (9.4.2)

To keep the book from having too many equations, the derivation of (9.4.2)
ignores all uncertainties, network effects, capacity constraints, etc. Thus, we
return to the case of Section 6.1 except that the cross-elasticities and Ramsey
revenue reconciliation are included.
The resulting Lagrangian is


G(~) - 1. Bk(c!.k) + ~:[ - 1. c!.kJ


+ Il{ Cr

f efc!.kJ


: Vector of generations get) t = I ...


~e: Vector of energy balance multipliers J1.e(t)

G(): Total generation costs over t

I ... TR

I ... TR

9. Spot price based rates

Bk(0<): Total benefits of customer k over t

C r : Revenue to be collected


1 ... TR

J-IR: Revenue constraint multiplier

The proof follows the by now boring pattern of setting

OQ(TR) =





Assuming optimum customer behavior, i.e.,




which is (9.4.2) with ~

= B,k

A Simple Model for Customer Response

dk(t) =



Kk : Capital stock of customer usage device (kWh)

Uk(t): Utilization factor


Uk(t) ~ I

Further assume


(e PR): Time average of all prices, t :=
TR - T

1 ... T R

Thus the kth customer's decision on what values of Kk to choose depend on all

230 II. Theory of hourly spot prices

of the prices, i.e. Pb over the revenue reconciliation time intervaLS The utilization decision at time t depends only on the price Pk(t) at hour t. There are no
cross-elasticities with respect to utilization.
Assume further that


Uk = TR r:.T!!.k: Time average of utilization factors


Short-run ealsticity

Thus the short-run elasticity is assumed, for simplicity, to be time invariant.

Pk oKk
-(3- : Long-run elasticity
Kk Pk


Using this simple model yields


If -a and -b are column vectors, there is a matrix identity that states

Using this matrix identity on (9.4.11) yields


Then after some manipulation


So finally after substitution into (9.4.6)


customers, both with dk

As an example of how to use (9.4.14), consider two

= 1 and Uk = 0.5, k = I, 2.

9. Spot price based rates


Old Existing Customer: k = I


= - 0.01

New Customer: k = 2

= - 0.5


= -


Old Customer Pays:

New Customer Pays:

which means the new customer "participates" in revenue reconciliation much

less than the old customer. Remember the sign of IlR can be plus or minus.

Wheeling is the transmission of electrical energy from a buyer to a seller,

through transmission or distribution lines owned by one or more other parties.
Wheeling causes a variety of physical and economic effects on the wheelers
(t~se owning the transmission lines). This section discusses theoretically
optimal wheeling rates, which capture the relevant engineering and economic
eff~cts, including the effects of wheeling on


Line losses
Redispatch of generators
Transmission line flow constraints
Other power system security issues 9
Recovery of embedded capital costs; e.g. revenue reconciliation

There are many types of wheeling terms and conditions in use today and
others are being proposed. Wheeling can be a controversial subject especially
when it is associated with terms like "mandatory", "free access", etc. Discussions on its advantages can get very lively and we have been engaged in such
discussions; as one example see Schwepe [1988]. However, the discussions here
neither survey nor compare all the approaches and do not address the pros and
cons of wheeling. Consider only the wheeling rates that result from the theory
of spot price based energy marketplace.
There are various possible types of wheeling such as
Utility to Utility: Utility S is selling to the buying Utility B where part of all of the energy
flows over the wheeling utility's lines (there may be several wheeling utilities).
Utility to Private User: Utility S is selling to a Private User B located within the wheeling
utility's service territory.

II. Theory of hourly spot prices

Private Generator to Utility: Private Generator S located within the wheeling utility's
service territory is selling to Utility B.
Private Generator to Private User: Private Generator S is selling to a Private User B
where both are located in the wheeling utility's service territory.
As will be discussed, revenue reconciliation can cause major differences in the
wheeling rates for these different types of wheeling.
Unreconciled rates: General

Wss(t): Energy seller is selling to buyer during hour t (kWh); i.e., amount of
energy being wheeled.
Cw(t): Costs incurred by wheeling Utility during hour t ($).
wss(t): Spot wheeling rate during hour t if revenue reconciliation is ignored
The spot wheeling rate is the marginal impact of wheeling on the wheeling
utility's costs:

where the derivative is taken subject to the condition


An incremental change in the seller's net generation is matched by an equal

but opposite change in the buyer's net generation.

After going through a lot of manipulations involving Lagrangians etc. as in

Chapters 6 and 7, it can be shown that (9.5.1) becomes

oW + '[,1,()I +





Network Costs and Contraints Of (7.2.2)

Total Network Losses

The details of the derivation can be found in Caramanis, Roukos, and Schweppe
(1988). They are not repeated here because the reader is undoubtedly becoming
tired of such manipulations and because (9.5.2) is a reasonable result. The terms

of (9.5.2) can be interpreted as



Effect of wheeling on wheeling utility's network maintenance costs, network

quality of supply costs, and costs of rescheduling to prevent line overloads.

(I)) ilL[z(I))

il W

Effect of wheeling on wheeling utility's generation costs

resulting from losses.

Under some existing wheeling terms and conditions, the wheeling utility
subtracts its extra losses from the amount of energy delivered to the buyer; in
such a case the second term of (9.5.2) is obviously not appropriate.
It is important to note that (9.5.2) can yield a negative wheeling rate! As a
simple example, assume the wheeling energy flow is in the opposite direction
from the rest of the energy flows on the wheeling utility's network. This yields
a negative oL[z(t)/aw. However, it can be shown that the wheeling utility's net
benefits (wheeling revenue minus change in costs) are never negative.
The basic result (9.5.2) holds for all types of wheeling: utility to utility, utility
to private user, etc. However, for utility to utility wheeling, implementation of
(9.5.2) requires a lot of book-keeping because the partial derivative with respect
to W implies simultaneously changing the net scheduled interchange of Utilities
Band S and then modeling the effects of their AGC, etc. (see Appendix B).
Computer programs to do this book-keeping are available (see Caramanis,
RoU\kos, Schweppe (1988 but the details are not covered in this book. Instead
we w11l go into more interesting issues associated with revenue reconciliation. To
sim'i,lify the algebra, only the special case of bus to bus wheeling is considered.
Bus to Bus Wheeling

Bus to bus wheeling occurs when the seller provides all the energy at bus Sand
the buyer receives all the energy at bus B, where both buses are in the wheeling
utility's service territory. This case applies directly to the Private GeneratorPrivate User type of Wheeling and to the other three types if only one or two
of the wheeling utility's tie lines to external utilities are affected (which could be
a reasonable approximation to many cases). However even if the explicit bus to
bus equations cannot be applied, the basic principles to be discussed are valid.
For bus to bus wheeling (9.5.2) becomes


or equivalently

where the Band S subscripts obviously refer to the buying and selling buses.


n. Theory of hourly spot prices

Three Classes of Wheeling

The method of revenue reconciliation to be used depends on the relationship

between the wheeling utility and the buyer and seller. Three classes of wheeling
to be discussed are
Class N: Wheeling between two parties which the wheeling utility has no
obligation to serve.
Class 0: Wheeling between two parties, both of which the wheeling utility has
an obligation to serve.
Class ON: Wheeling between one party which has no obligation to be served and
one which the wheeling utility must serve.
Examples of these classes are
Class N: Utility to utility.
Class 0: Private generation to private user.
Class ON: Private generator or user to/from external utility

Revenue Reconciliation

After reading Chapter 8, the reader should suspect that there are many ways to
do revenue reconciliation for wheeling. We will only present some of the main
results by using the simple, constant multiplier approach. Define
(Uss(t): Wheeling rate including revenue reconciliation.

Class N: A reasonable approach for Class N is


Class 0: A reasonable approach for Class 0 is to use the form of (9.6.1) with





(I - 111)[;'(1)

+ IJs(I)]
+ 17s(l)]

which yields






Class ON: A reasonable approach for Class ON when the wheeling utility has
no obligation to serve only the buyer is to use the form of (9.5.3) with






IJs(t) - m/lJs(t)/



9. Spot price based rates

which yields

Comparison of Classes

The main difference between Class 0 (9.5.6) and Class ON (9.5.7) is the factor
of 2 which multiplies y(t). The main difference between Class N (9.5.5) and
Classes 0 and ON is that Class N has no yet) term (except as it enters into the
loss components of the I'/(t).
The impact of the generation marginal costs y(t) can be very large. If no
network quality of supply components are present, the network component I'/(t)
contains only loss components, so, for example,

'1e(t) =

yet) adB(t)'

Thus, unless losses are very high,

yet) ~ rfn(t)

; to

Therefore, unless large network quality of supply components are present or

unless Iml is very small, the yet) component of the Class 0 and Class ON
w41eeling rates dominate their numerical values.
On Obligation to Serve



Irt the discussions of Chapter 5 on deregulation, a scenario was presented in

Section 5.4 which involved the concept of a user or private generator renouncing
their obligation to be served. If this happened, they would see Class N rather
than Class 0 wheeling rates. The large difference between Class N and Class 0
wheeling rates underlies the potential importance of this obligation to be served
concept when wheeling is being discussed.
Only an hourly spot wheeling rate was discussed. In practice, the actual
wheeling rate might be predetermined with 24-hour update or biJling period
update, might involve fixed-price-fixed-quantity long term contracts or pricequantity transactions, etc. The ideas of Sections 9.1 through 9.4 apply to
wheeling rates as well as to regular spot price based rates.

The hourly spot prices developed in Chapters 6, 7, and 8 provide the basis for
a wide variety of spot price based transactions. This chapter presented the
resulting theory for predetermined price-only transactions, price-quantity transactions, long-term contracts, special rates for special customers, and wheeling
rates. The techniques used can also be applied in other types of situations such
as assignments of power pool reserves. The wide range of possible spot price


II. Theory of hourly spot prices

based transactions illustrate the power (or it is energy?) of a spot price based
energy marketplace.
Many of the results of this chapter bear on politically sensitive issues (i.e.
disputes) of valuing and compensating various non-standard generation and
load management schemes. For example, one argument against large generators
per se and nuclear units in particular is that their size forces the utility or power
pool to carry more reserves, both total reserves and operating reserves. Pricequantity rates for system security control provide a way to measure this penalty,
if any. Another example is the time pattern of solar and wind generators. The
hourly spot price gives a precise way to measure the actual value of energy
generated. Many similar questions surround cogenerators, thermal storage
systems, etc.
We would be happier to see spot pricing used not just to evaluate the value
of existing generation and load management equipments but to actually change
day to day behavior of the units to increase their social value. One is pie splitting
(at least once the units are built), while the other is pie enlarging. We believe that
many existing units can have their social value (and private profitability)
enhanced in this way.

I. Since revenue reconciliation is ignored here, we could use the notation of Chapter 8, and write
p, (tiT). However, we choose to use the Pk(tlr) notation of Chapter 7. In Chapters 2 and 3, Pk(t)
is associated with a one-hour update spot price. In actual implementation, as discussed in
Chapter 4, a one-hour update is a predetermined rate which is specified, for example, five
minutes before the start of the hour. Thus, to be exactly correct, all price-only transactions are
in the category of predetermined rates.
2. Using (9.1.4), the expected revenues received by the utility are the same for /'k and Pk(t),
assuming dk(/) does not change.
3. These by no means exhaust the issues of "dynamics pricing" which can be argued to include the
dynamics of individual generators. For investigation of such details see Caramanis, Bohn,
Schweppe [1987J.
4. For details of derivation see Caramanis, Bohn, Schweppe [1987J.
5. We assume here for simplicity that if interruption is needed it will last for the whole pricing
period. Extension to fractional period interruptions is, of course, possible.
6. That is, constraint (9.2.2) will be met after the value of S is revealed by setting Sk.c = Sk.p with
k values ranging over the required subset of participants in order of increasing contingency
prices. Once (9.2.2) is met, Pc will be set equal to the highest reservation price reached and all
s'.c corresponding to higher reservation prices will be set to zero.
7. This simple constraint assumes the hot water or thermal storage capacity is very large. In
practice, the logic usually has to be more complicated to account for finite storage and
time-varying (within a day) customer needs.
8. In practice, capital stock decisions would consider prices over several revenue reconciliation
intervals, but this assumption simplifies the manipulation of the equations.
9. Much more detail can be found in Caramanis, Schweppe, and Bohn [1986J and Schweppe, Bohn
and Caramanis [1985J.
10. The term "tie line coefficient" is our own; there seems to be no standard terminology.
II. More detailed discussions can be found in Schweppe, Bohn and Caramanis [1985J.










*1ourly spot prices and spot price based rates were derived in Chapters 6
through 9 by considering short-term issues, i.e., capital stock was treated as
fixed. This chapter examines investment decisions in the spot price based marketplace and their relationship to the statistical behavior of spot prices. Optimal
investment conditions from the point of view of society's cost minimization are
derived and compared to optimal investment conditions from the point of view
of individual marketplace participants. The use of spot price notation leads to
very straightforward mathematical results, and permits us to treat generation,
transmission, and end use technology in a single framework.
This chapter limits discussion of investment conditions to the case of ideal
revenue reconciliation which does not affect participant behavior.
Investment decisions are evaluated on the basis of the price duration curve,
a concept analogous to the load duration curve. Statistical hourly spot price
behavior information is shown to be necessary and sufficient in evaluating
investment conditions in most cases.
Section 10.1 presents the overall problem formulation while Sections 10.2,
10.3 and 10.4 discuss the specification of generation, customer, and network
investments. Section 10.5 discusses revenue reconciliation for an optimum
system. This chapter concludes with a discussion in Section 10.6 on short-run
versus long-run marginal cost pricing.
This chapter concentrates on mathematical conditions for optimality. Real
world power system planning (as discussed in Appendix C, in particular, Section
C.I) involves many concerns, beyond mathematical optimality of social welfare.

238 II. Theory of hourly spot prices


The load, generation, and network related terms defined in Chapters 6 through
9 are also used here. The reader should thus review the definitions of $/t), get),
~(t), ~*(t), ("(t), d(t), cJ.(t), ~P(t), dP(t), (iP(t), ~(t), Lf~(t)], G[~(t)J, N[~(t)J, B[(i(t)J,
B[dP(t)] and other related terms.
Additional quantities are defined here to allow inclusion of investment terms
into the model. It is assumed that the capital stock associated with each
generator, customer or line is represented by a scalar and is not subject to
indivisibilities. Whereas the scalar assumption is made for notational simplicity
only, the indivisibility assumption is needed for a differential calculus based
description of first-order optimality conditions.
Using subscripts to indicate generation, demand and network related quantities, as usual, define

K;.i: Capital stock of generator j which is in place at the beginning of the period
under consideration and expected to be retired at the end of that period.
KD,k: Capital stock of hourly spot price participant (customer) k in place at the
beginning of the period under consideration and expected to be retired at the
end of that period.
Kg.k: Similar to above for predetermined rate participants.
K~J: Similar to above for network line.
K y , K D , Kg, K,,: Vectors with elements the quantities defined above.
I;(K-;} The mvestment cost ($) associated with capital stock KyJ utilized by
generator j.
Ik (KO,k), Ik (Kg,d, ~(K~,;): Defined similarly to the above for hourly spot price
participant k, predetermined rate participant k, and network line i respectively.


[(liD)' [P(o), [(Ii~): Similar summations as above.

Definition of Short- and Long-Term Lagrangians

The short- and long-term Lagrangians are defined next. They represent the net
hourly and whole period costs to society respectively.
The short-term Lagrangian is identical to that defined in Section 9.1 of
Chapter 9.
!let) = Gf(t)]


(Generation Costs, Constraints)

(Network Costs, Constraints)

- B(q,(t)1

(Hourly Spot Price Participant Benefits)

- B(q,P(t)

(Predetermine~ Rate Participant Benefits)


+ dP(t) +

L[~(t)J - g(t)J

(Energy Balance Constraint)


10. Optimal investment conditions 239



The long-term Lagrangian is the investment cost of all participants, generation

and network plus the expected value of constraints and net hourly social costs
summed over all hours in the period representing the capital stock's useful life.
The long-term Lagrangian is thus defined as capital cost plus expected short
term welfare:


I(K y )

+ I(KD ) + I(Kb) +


L E{n(t)}


1 ... 0


T: The retirement time of capital stock invested and put into operation at time
E: The expectation operator over random variables to be realized at time t.
Expectations are evaluated with probability density functions that represent
the knowledge at time zero of the behavior of random variables at time t i.e., a conditional expectation.
Conventional (not spot price based) investment conditions are usually
obtained from a formulation which involves a reliability constraint such as an
allowable loss of load probability (or hours) or expected unserved energy. Our
development here assumes that the generation and network quality of supply
terms of the spot price are calculated by the "market clearing approach." Thus,
th~re is no unserved energy in the sense of anyone being "blacked out";
cystomers simply see very high prices instead and reduce demand as required.
Hence, no explicit reliability constraint is included. Since the market clearing
al'proach is being used, there are no generation or network quality of supply
costs; i.e.,


in the definitions of G(t) and N(t) of (10.1.1). The Gos and Nos terms could be
included, if desired, at the expense of more complicated-looking equations.
Two implicit assumptions in the above long-term Lagrangian formulation are

All capital is put in use and retires at the same time.

The present value of hourly costs equals their simple sum.

These assumptions are made only for notational simplicity. Existing capital
stocks with varying retirement rates can be modeled without major structural
changes in the formulation. Also, discounting to yield a more conventional
calculation of the present value of costs can be introduced in the summation sign
of (l 0.1.2) without changing the problem formulation in any qualitative sense.
Equation (10.i.2) is sufficient to present the basic ideas without a lot of technical
notational complexity. However, the reader must be warned that
In any actual calculation using the ideas of the chapter, the equations must be modified
appropria tely.1

240 H. Theory of hourly spot prices

Appendix C characterizes power system investment planning as a multiple

attribute decision making process which includes other attributes besides
present worth of all present and future costs. However, the ideas presented here
can be modified as appropriate to represent the real world more closely.
First-Order Optimal Investment Conditions

The first-order optimality conditions are obtained in the context of a three-part

optimization procedure:

Short-Term Decisions: Generation levels, consumption levels and hourly spot

prices determined conditional upon given capital stock levels.
Medium-Term Decisions: Predetermined rates determined conditional upon
given capital stock levels and "given procedures", according to which shortterm decisions will be made during the period that predetermined rates are
intended to hold.
Long-Term Decisions: Capital stock level decisions conditional upon "given
procedures", according to which medium, and short-term decisions will be
made during the operating life of the capital stock.

The medium- and short-term decisions are those derived in Chapters 6

through 9. The long-term investment decisions on capital stock levels are
optimal when the following conditions are satisfied conditional upon mediumand short-term decisions conforming to Chapter 6, 7 and 9 determined relationships. First order conditions for 10.1.2 are

an LT

an LT

an LT




for alii


( 10.1.3)

for all k

(Hourly Spot Price Customer)

( 10.1.4)

for all k

(Predetermined Rate Customers)


= 0

for all i


( 10.1.6)

These conditions are analyzed and discussed in the following sections.


Utility Generation

Condition (10.1.3) yields


For optimal short-term system operation as in Chapter 7,

Developing the rest of (\0.2.1) holding g,(t) constant (and remembering that
GQs{g(t)] = 0) yields

- flmax.)./(I)




Equation (10.2.2) states that the optimal investment level of generator j

should be such that the costs of an additional incremental investment equals the
present value of the expected benefits from it. These benefits are the three terms
of the right hand side of (10.2.2), i.e., operating efficiency gains, contributions
to meeting total system demand at those times when total available generation
is binding, and finally contributions to additional output by generator j at those
times when it oper'1tes at its maximum available capacity.
I.n order to better understand (10.2.2), we make a series of assumptions that
ar~~ often reasonable. Assume that the capital stock K yj affects only the
maximum capacity of the jth plant, i.e., gmaxj' Thus

If the jth plant is available at hour t, it can act as operating reserve, so

plant available

plant not available


Finally, assume that plant variable costs are proportional to output



Using (10.2.3) through (10.2.6), (10.2.2) reduces to

L E{(j(t)}


242 II. Theory of hourly spot prices

e(t) =


plant is available



plant not available

If the plant is either turned off or not at peak capacity (gj(t) < gmax), then

If the plant is at full capacity, then the optimality condition (7.2.7), the definition
of Gf(t)] of (7.2.2) (remembering that GQS = 0), and use of (10.2.5) yield


J1max';'j(t) =



Thus the condition (10.2.7) becomes

pj(t) - Aj

g/t) =


IIQs.i(t) = YQs(t)

gj(t) <


plant not available

and plant is available

The condition (10.2.9) has a simple physical interpretation. The costs of an

additional incremental investment should equal the present value of the
expected benefits where the benefits are pj(t) - Ai when the plant is at full
capacity and flQs.y(t) when the plant is available but not at full capacity.
Role of Price Duration Curve

Assume that the variable fuel and maintenance costs of generator j are constant.
Then if the effect of flQs.y(t) = YQs(t) is ignored, a good approximation to
(10.2.9) is
Investment Cost of Incremental kW


* (a;) * (Hours)



Area under hourly spot price duration curve from Aj to infinity as in Figure 10.2.1

a,: Probability that the incremental generating capacity will be available for generation during any future hour
Hours: The total number of hours in the service life of the incremental investment<

This underscores the significance of hourly spot price duration curves in evaluating incremental investments. 3
Quality of Supply and Generation Investment Costs

One of the methods for calculating the generation quality of supply spot rpice
components that was discussed in Section 6.2 was the,"Annualized Cost of

10. Optimal investment conditions 243





FigUre 10.2.1. Price duration curve.

Peaking Plant" method. Another justification of this method is now discussed.

Consider a peaking generator with variable generation costs )'p which
operates only when the spot price p(t) exceeds Ap. Assume that when p(t) ~ Ap
the difference p(t) - Ap is roughly equal to the generation quality of supply
component YQs(t). Further assume that for the peaking plant
/[K/.,l =

K/.,A Qs.;

AQS R : Annualized cost of peaking plant ($jkW)

Then assuming the peaking plant is always available, (10.2.9) yields

A Qs .; =

L E{YQs(t)}

Equation (10.2.11) is satisfied if

A Qs .;, a;, (/)

YQs(t) =




n. Theory of hourly spot prices

for any ay(t), a fact which can be verified by substitution of YQS(t) in (10.2.11).
The ay(t) obtained in Section 6.2 was

Thus the Section 6.2 result is one of many possible ways to quantify YQS (t),
Customer Owned Generation

If a customer-owned generator k sells its output to the grid at the hourly spot
price and self-dispatches itself according to the price Pk(t), the individual generator's optimal investment problem is



- GFMfgk(t)J}

subject to





Assume for simplicity that

Then following the procedures and assumptions that lead to (10.2.5),

L: E[Ck(t)]



Comparing the customer investment condition (10.2.12) with the corresponding utility investment condition (10.2.9) shows that the only difference lies in the
effect of the generation quality of supply benefit !lQS,y(t) seen by the utility. This
makes sense because the utility is responsible for quality of supply.
If the effect of the !lQs)t) = YQs(t) term is neglected, then individual profit
maximizing behavior coincides with the utility's optimal behavior under the
mild assumption that the same future price duration curve information is
available to the private and utility planner. 4 Finally, if system security control
requirements are modeled as commodities to which all market participants can
contribute (see Section 9.2), rather than as a constraint on utility-owned generation, utility- and customer-owned generation investment conditions coincide
completely. This happens because security control or quality of supply requirements are properly internalized in the price system and are accessible to all
marketplace participants alike.

10. Optimal investment conditions 245


Spot Price Participants

Condition (10.1.4) yields


z:: E{ _

an(t) M;Jt) + aBddk(t)]}

aKD .k
oKO .k



The short-term optimization conditions of Section 7.2 yield


OJ[KOk ]

oKD .k



E'h!,lation (10.3.3) states that the optimal investment level is such that incremental investment costs equal the present value of the expected incremental
electricity savings over the life of the investment. Equation (l0.3.3) clearly
coinCides with individual consumer profit maximizing behavior.
Equation (10.3.3) shows that forecasts of the properties of the hourly spot
price are necessary to calculate investment decisions. The price duration curve
information discussed above in conjunction with generating capacity expansion
investments may not be sufficient in describing the necessary information
needed to evaluate incremental investments for customers in all cases.
Consumers with the ability to respond to hourly spot prices using some type of
storage (thermal, product, etc.) may benefit from additional information describing the time series nature of the spot prices. Thus, typical daily spot price
trajectory forecasts may be necessary. See Bohn (1982] for an analysis of how
to calculate the incremental value of additional storage.
The investment conditions for generators, lines and customers given in this
chapter ignore the effect of revenue reconciliation. It can be argued that this is
reasonable for utility investment in new generators and lines. However, in
practice, customers would consider future revenue reconciliation if it modifies
the hourly spot price.
Predetermined Rate Participants

Condition (10.1.5) yields



Theory of hourly spot prices

" E{[aBk[df(t)] _ aN{y(t)] _

Ite t


-It ( t ) -- - + _aB-.::..d!....;df:...;.(t...:.!)]}



( 10.3.4)

Substituting the following relationships derived in Section 9.1 (see (9.1.8) and

Pk(t) =

aN{ y(t)]



aLl yet)]
Pe(t) adf(t)


a/[Kb.d =

"E{[" ( )] adf(t)
Pk t
0 no b.k



Equation (10.3.5) represents society's preferred investment behavior. However,

predetermined rate participant k will not behave according to (10.3.5) since its
independence profit maximizing investment conditions are described by
oJ[Kg,d =

L E{OBddf(t)]}



Comparison of (10.3.5) and (10.3.6) indicates that private profit maximizing

behavior of predetermined rate participants achieves investment levels whose
incremental cost differs from that of socially optimal levels by an amount equal
to the term


Under the reasonable assumption of increasing marginal investment costs, when

the term is positive the predetermined rate participants will tend to overinvest,
and when it is negative to underinvest, compared to socially optimal investment
levels. It should be noted, however, that the term will be close to zero when
predetermined rate updates are selected so as to minimize the covariance
between hourly spot prices and demand response to incremental changes in rk
and Kb.k'

10. Optimal investment conditions



Condition (l0.1.6) yields


~ E {Ile(t) a~~::)l + aN;i:.;t))

"L.., J1QS.~.1 ( t ) [OZI(t)

oK. I


l.max ] }



Equation (10.4.\) states that the network investment in line i should be undertaken to the level that renders the cost of an incremental investment equal to the
present value of the stream of expected benefits through losses reductions,
maintenance cost reductions and alleviation of line flow constraints.
Appendix D on the DC load flow shows that if hourly spot prices are paid to
generators and charged to consumers, the system will receive a net income which
can be considered to be the income of the network. Disregarding NM , the
network income is shown in Appendix D (see Section D.2) to be during hour t

L PQs.".,(t)z,(t)


Cj;}mparison with (10.4.1) indicates that the contribution to society's welfare of

an incremental investment equals the contribution of that incremental investment to network net revenues. Thus, socially optimal and network profit
maximizing investment conditions coincide if market participants' transactions
are based on hourly spot prices.
Evaluation of the investment conditions of (10.4.1) require estimates of the
hourly values of
ilL[ y(t)]
d oz(t)
---an ~



Section D.2 of Appendix D shows how these quantities can be computed under
the DC load flow approximation. The impact on system losses of an incremental
investmentS in line i is shown to be proportional to the square of the power
flowing over line i, while the impact on line flows is proportional to the power
flowing over line i. Appendix D (see also Section 7.7) also shows that the power
flowing over line i is proportional to the difference of spot prices across the ends
of line i (assuming a constant reactance to resistance ratio on a\llines). Thus the
optimal investment condition for line i can be evaluated in terms of the behavior
of spot prices at the ends of that line.


n. Theory of hourly spot prices


In Sections 10.2, 10.3, and lOA, we showed that socially optimal investment
conditions are characterized by the following rule: Investments are made until
the last MW of investment capacity "earns" an expected stream of new income
whose present value equals the incremental cost of investment. Assuming that
there are no economies or diseconomies of scale (i.e., the investment cost per
MW of capacity is invariant of the total installed capacity), optimal investment
conditions quarantee that each unit of capital (generation, network or consumption) will make precisely enough to cover its investments costs ifpaid according
to optimal spot prices. Therefore, revenue reconciliation would be automatically
In real systems, a number of issues may result in imperfect revenue recovery.
Two are

Nonexistence of spot price based revenues/costs and hence historical investment levels that are far from socially optimal levels.

Unexpected future developments that upset the expected revenue streams. Of

course under spot pricing the amplitudes of such upsets are expected to be

It is also necessary to worry about intertemporal effects of present generration-consumption levels on future costs. These effects were assumed nonexistent in Chapters 6 through 8. They can be taken into consideration as discussed
in Caramanis (1982]. However, the intertemporal effects modeled in there are
fairly short-term, extending over the unit commitment time scale (days to
weeks). Longer-term intertemporal coupling extending over periods of years is
very cumbersome to model, since it requires long-term marginal cost models of
the form discussed in the next section.

Arguments on short-run versus long-run marginal cost pricing are often encountered in the electricity pricing literature. This section addresses the shortversus long-run issue in the context of a spot price based marketplace.
In the economic literature, short-run marginal cost usually refers to the cost
of incremental production, including variable costs such as fuel and raw
materials but not including the cost of capital or other fixed (sunk cost) factors
of production. Long-run marginal cost usually refers to the costs of incremental
production over a long period, including both variable and fixed costs.
In most power system costing literature, short-run marginal cost is associated
with a production cost model whereas long-run marginal cost is associated with
incremental revenue requirements calculated by a capacity expansion planning

10. Optimal investment conditions 249

Present-Day Long-Run Marginal Cost Pricing


The usual, present-day practice in calculating long-run marginal costs is to solve

a cost minimization problem of meeting a known future demand at a specified
reliability level. The effect of an incremental change in present demand determines the long-run marginal costs. To determine the avoided costs of an
independent generator, the cost minimization problem is solved twice, with and
without the generator. In engineering terms, present-day long-run marginal cost
pricing practices are characterized by an "open loop" control philosophy.
Four major weaknesses in the present-day long-run marginal cost calculation





Computations are cumbersome since they involve multiple solutions of

usually sizable mathematical programming models.
They use grossly oversimplified demand models.
The marginal cost dependence on the huge uncertainties in the future values
of important inputs such as load growth, fuel costs, cost of capital, construction times, etc, is ignored, e.g., the expectation operator" E" of (10.1.2) is not
Most present-day long-run marginal cost approaches deal only with future
capital and operating costs. They do not include the impact of a change in d(t)
on the operating costs at hour t.





We believe that in the current environment of economic change and future

upcertainties, these shortcomings severely limit the usefulness of present-day
long-run marginal cost estimates.
It is argued by some that present-day long-run marginal cost pricing has the
advantage of yielding rates that are more stable over time and hence are
preferred by customers. This need not be true. For example, the effects of events
such as

A major change in forecasts of future demand

A discovery of a new oil or gas field or a major change in world oil prices
A major change in environmental regulations

can cause a major jump (up or down) in a long-run marginal cost price
computed for a deterministic world which ignores future uncertainty or priceresponsive future demand. History has shown that such events are a fact of life.
Energy Marketplace Long-Run Marginal Cost Pricing

A spot price based energy marketplace can provide long-run marginal cost
prices which are selected to maximize the long-term social welfare Lagrangian
Q LT of (10. \ .2) with the conditional expectation operator reflecting future uncertainty at the time of the selection. Assuming that capital stock is an explicit

250 II. Theory of hourly spot prices

function of future demands and denoting for simplicity different types of capital
by one variable, we have the following necessary maximization condition:

The condition of (10.6. I) is satisfied if all marketplace participants are under

optimal spot prices and long-term marginal costs are set equal to the spot price.
To make the above point more precise, we define, ignoring the k dependence for
PLT(t): Long-run marginal cost based price of electricity which satisfies (10.6. I)
p(t): Hourly spot price developed in Chapters 6 and 7.

We can then show that



under the condition that all marketplace participants are under optimal spot
prices and that socially optimal investment decisions as defined in earlier
sections of this chapter are in place. Indeed, as shown in Chapters 6 and 7,
optimal spot prices are selected such that

Od(t) n(t) =


and optimal investment ensures that


Equations (10.6.2) and (10.6.3) imply the necessary conditions of (10.6.1), and
hence long-run marginal cost coincides with the optimal spot price.
Before discussing the conditions under which PLT(t) and pet) may diverge, the
reader should note the difficulties involved in actually arriving at the optimal
investment conditions which insure that (10.6.3) holds. Optimal investments
require evaluation of future uncertainties as described by the conditional expectation operator of (10.1.2). Although in principle we can write the necessary
conditions, their implementation is hard as is proper accounting of uncertainties
in the context of present-day long-run marginal costs models. However, the
reader ought to note that the adjustment to future uncertainty realizations,
through the selection of closed-loop type optimal spot prices, mitigates the
impact of the uncertainty and hence the impact of shortcomings in modeling it.
Another difficulty in implementing spot price based long-run marginal costs

rests on the unavailability of proper demand models that capture the interdependence of intertemporal demands. In principle one needs to use benefit
functions which depend on a whole time stream of demands rather than demand
at a single point in time; i.e., one needs to evaluate


Similar difficulties arise with the intertemporal dependence of generation costs.

Thus, spot price based calculation of long-run marginal costs is in principle
feasible, but by no means does it overcome all the difficulties which present-day
long-run marginal cost pricing methods face. However, spot prices provide an
estimate of long-run marginal costs in a way which is less cumbersome than
most present-day methods. Equation (10.6.1) provides a consistent framework
for long-run marginal cost calculation despite the associated implementation
difficulties. 6
Reasons for pet), PLT(t) Divergence or Similarity: A Conjecture

Hi~orical investment decisions which have been guided by average cost pricing
con-siderations guarantee that current capital stock levels are not at their socially
optimal levels. In addition, the stated difficulties in dealing with uncertain
intertemporal demand dependencies, noncoincidence of social and private
optimal investments for predetermined price participants, etc., indicate that
future capital stock levels will "almost never" reach and hold their socially
optimal levels. We can thus say that the following relationship holds:


(other terms related to capital stock optimality imbalances) (10.6.4)

The reader should note that pet), the spot price of Chapters 6 through 8, does
provide information about the future which is associated with long-run
marginal costs. For example, when generating capacity is in shortage, spot
prices will tend to be frequently high since capacity constraints will also be
frequently active. This behavior of pet) will signal to consumers that increased
demand will require additional generation investments.
We wish at this point to conjecture that the behavior of PLT(t) is likely to be
very similar to that of pet) and hence the magnitude of "other terms" in (10.6.4)
will be small relative to pet).
The motivation for the conjecture is our belief (neither proven nor tested) that
a change in demand at hour t has very limited impact on future capital and
operating costs (many years in the future) for two reasons:



meory or


Future demands are driven more by uncertain exogenous factors (state of

economy, etc.) than by present demand.
The price-<iemand feedback inherent in the spot price based energy marketplace greatly reduces the impact of dk(t) on future costs.

Hopefully, someone will someday develop a good demand model which can
either verify or disprove this belief.

This chapter provided the equations which yield optimal investment-decisions

for generators, the network, and customers in a spot price based energy marketplace.
Three important qualifications on the equations should be restated;
o Revenue reconciliation was ignored.
o The notation is oversimplified and ignores issues that are important in
practice such as discounting to get present worth, and variable plant retirement dates.
o The equations are incremental in nature.
Inclusion of revenue reconciliation complicates the equations, but one can argue
that revenue reconciliation should not be considered in making utility investment decisions. Conceptually it is easy to remove the second qualification on
notation, although in practice it gets messy. Removal of the third qualification
that new generation facilities, customer, end use devices and transmission lines
are added in lumps is nontrivial. However, the given incremental optimality
conditions provide enough useful information to guide generation, network,
and especially user investment conditions in many actual applications. Of
course, if there is enough motivation (and interest in mathematical manipulation) the derivations can be redone using, for example, an integer programming
type optimization logic.
Implementation of the ideas of this chapter is nontrivial because they involve
the expectation operator "E" over future uncertainties associated with load
growth, cost of capital, capital costs of new technologies, etc. Given today's
technology of investment decision planning, an exact implementation is not
practical in most cases. This, however, is not a shortcoming that is associated
uniquely with the energy marketplace ofthis book. The same uncertainties exist
and should be incorporated in investment planning under the present day
system. See e.g. Ellis [1981].
Spot price based transactions provide a feedback mechanism that tends to
smooth the impact of uncertainties. The hierarchical three-part optimization
procedure that characterizes optimal investment conditions as discussed in
Section 10.1 is based on such a feedback mechanism. Investment levels are
optimized conditional upon future generation and consumption decisions that
are not prespecified but are know to conform to the decision rule described by

10. Optimal investment conditions


the optimal spot price formulas. Thus future uncertainties on load growth, fuel
costs and other important unknown variables are substantially smoothed
through spot price driven short-term reoptimization that cushions the impacts
of investment planning shortfalls or longfalls, changes in environmental
standards, new fuel discoveries, etc.
A general discussion of short-versus long-run marginal cost pricing was also
provided in this chapter. The basic spot price equations developed in Chapters
6 through 9 are primarily (but not entirely) evaluated in a short-run context.
Some of the options for computing the generation and network quality of
supply components involved, use the annualized cost of peaking generation
plants and the marginal cost of expanding the network. These approaches can
be viewed as pragmatic approaches to long-run marginal cost pricing. However,
a spot price energy marketplace could be based on pure long-run marginal costs
if desired, provided the massive uncertainties associated with expansion
planning are included in the long-run marginal costing formulation, valid
demand models are used, etc. - i.e., that the formulation of (10.6.1) is used.
Establishment of an energy marketplace based on present-day approaches to
long-run marginal cost pricing is not recommended.
The reader must be careful not to confuse the mathematical optimality
conditions of this Chapter with real world power system planning as discussed
in Appendix C; particularly Section C.I.

The' basis of Sections 10.1 through 10.4 is Caramanis [1982]. Other material
listed in the annotated bibliography also develops investment conditions for
particular parts of the marketplace. Other authors have developed optimality
conditions for generation, but have not used the spot price formulation (which
gives very clean statements of the optimality conditions). See for example the
discussion at the end of Chapter 3. As usual, an exception is Vickrey, who
provided a lucid discussion of the issues and the optimal investment conditions
under responsive pricing.
There has been very little previous work on optimal conditions for transmission investments, although of course utilities have developed transmission
plans by the method of simulating "System behavior with and without a particular proposed line. (See Appendix C.) To be sure, transmission lines are one
topic where the assumption of continuously variable investment sizes can fail,
since the addition of a new line connecting two widely separated (electrically)
buses can lead to a discontinuous shift in power flows. The results of Section
10.4 are most accurate for the case of strengthening an existing line (or examining the size of a proposed line),
The situation on customer end use technology is somewhat different. Since
utilities have long been interested in buying and installing load management
equipment for cuistomers, there are a number of methods of evaluating the
value of such investments. For example the entire second half of Talukdar and


II. Theory of hourly spot prices

Gellings [1987] is devoted to evaluating load management systems. As discussed

in Section 10.3, use of spot price concepts can simplify such analyses since an
end use investment is socially optimal if it would be profitable when evaluated
at spot prices.
Bohn [1982, Chapter 4] provides a general model of complex industrial
customers under spot prices, using a linear programming formulation. Various
kinds of storage, shutdown, fuel switching, etc. can be captured. The shadow
prices on capacity constraints give the incremental value of investments which
enhance customer responsiveness. For example the value of more storage
capacity is shown to be the difference between peak and trough spot prices.
Daryanian [1986] develops a better algorithm for solving this class of problems,
and generalizes it further.
A number of authors have modeled long run marginal costs, and discussed
their relationship to short run marginal costs. The economic literature was
surveyed at the end of Chapter 3. The discussion in Section 10.6 on long- versus
short-run marginal costs deviates significantly from the existing literature in that
a more accurate (and justifiable) definition of long-run costs is used.
I. Caramanis [1982] contains examples of the more detailed equations.
2. Since in practice discounting is important, (10.2.10) will actually have the form of a summation
over annual hourly spot price duration curves extending over the service life of the investment.
3. For a discussion of how price duration curves can be actually calculated using algorithmic and
computational techniques not too different from those used today for capacity expansion
planning studies, the interested reader is referred to Caramanis [1985].
4. In practice, a utility and a private investor may make different investment decisions because of
differences in cost of capital, risk aversion philosophy, etc. This is normal in any marketplace.
5. Represented as an incremental increase in line i's admittance.
6. Given enough research and time, these difficulties can be overcome.


Jan Paul and Bridger Mitchell, "Evaluating Time-of-Day Electricity Rates for Residential
Customers," in Mitchell & Kleindorfer, eds., pp. 247-273, 1980.
AgQcw, Carson E. "The Theory of Congestion Tolls", Journal of Regional Science, Vol. 17, No.3,
pp. 381-393.
Anderson, Ronald, W. The Industrial Organization of Futures Markets, D.e. Heath. 1984.
Baumol, William J. and Bradford "Optimal Departures from Marginal Cost Pricing", American
Economic Review, Vol. 60, June 1970, pp 265-283.
Bergen, Arthur R. Power Systems Analysis, Prentice Hall, 1986.
Berger, Arthur "Pricing Mechanisms in Coupled Dynamic Systems", PhD thesis, Harvard University, May 1983.
Berger, Arthur W. and F.e. Schweppe, "Real Time Pricing to Assist in Load Frequency Control",
submitted to IEEE Transactions on Power Systems, 1995.
Berrie, T. "Interactive Load Control, Parts 1 to 6", Electric Review, 1981-1982.
Bertsekas, Dimitri, Dynamic Programming: Deterministic and Stochastic Models, Prentice-Hall,
Blackmon, B. Glenn, Jr., "A Futures Market for Electricity - Benefits and Feasibility," Discussion
Paper Series E-85-07, Energfy & Environmental Policy Center, John F. Kennedy School of
Government, Harvard University, July 1985.
Bohn, Roger E., "A Theoretical Analysis of Customer Response to Rapidly Changing Electricity
Prices," August 1980, Revised January 1981, MIT Energy Lab paper MIT-EL-081-001 WP.
Bohn, Roger E., "Spot Pricing of Public Utility Services," PhD dissertation, MIT, 1982. Also
available as MIT Energy Lab Technical Report MIT-EL 82-031.
Bohn, R.E. "Procedure for Electricity Auctions with Optional Spot Pricing", unpublished memo,
November 1985.
Bohn, Roger E., M.e. Caramanis, and F.e. Schweppe, "Optimal Spot Pricing of Electricity:
Theory," Energy Lab Working Paper MITEL-08100SWP, 1981.
Bohn, Roger E., Michael C. Caramanis and Fred C. Schweppe, "Optimal Pricing of Public Utility
Services Sold Through Networks," Working Paper HBS 8331, Graduate School of Business
Administration, Harvard University, corrected January 1983.




Bohn, R,E" B,W, Golub, R.D, Tabors, and F.C. Schweppe, "Deregulating the Generation of
Electricity Through the Creation of Spot Markets for Bulk Power", The Energy Journal, Vol.
5. No, 2,1984 pp 71-91.
Bohn, R,. F. Schweppe, and M, Caramanis, "Using Spot Pricing to Coordinate Deregulated
Utilities, Customers, and Generators", Chapter 13 in Plummer et ai, editors, 1983.
Boiteux, M. "La tarification des demandes en point: application de la theorie de la vente au CaUl
marginal", Revue Generale de /'electricite, Vol. 58, 1949, pp 321-340,
Boiteux, M" "On the Management of Public Monopolies Subject to Budgetary Constraints,"
English translation of his 1956 article, Journal of Economic Theory, 1971, Vol. 3, pp. 219-240,
comment vol. 3, p, 341.
Brown, G. Jr. and M. B. Johnson, "Public Utility Pricing and Output Under Risk", American
Economic Review, Vol. 59, March 1969, pp 119-129,
Caramanis, M. "Analysis of Non-Dispatchable Options in the Generation Expansion Plan", IEEE
Transactions on Power Apparatus and Systems, Vol. PAS-102, No, 7, July 1983 pp. 2098-2103.
Caramanis, M. "Production Costing of Interconnected Electrical Utilities Under Spot Pricing",
Large Scale Systems: Theory and Applications, Vol. 9, 1985, pp 1-18.
Caramanis, Michael, "Investment Decisions and Long-Term Planning Under Electrical Spot
Pricing," IEEE Transactions on Power Apparatlls and Systems, Vol. PAS-IOI, No. 12, Dec. 1982,
pp 4640-4648,
Caramanis, M. C, "Electricity Generation Expansion Planning in the Eighties: Requirements and
Available Analysis Tools," Energy Models and Studies 1983, pp 541-562.
Caramanis, M. C, R. E. Bohn, and F, C Schweppe "Spot Pricing of Electricity: Practice and
Theory", IEEE Transactions on Power Apparatus and Systems, Vol. PAS-IOI, No.9, September
1982, pp 3234-3245.
Caramanis, Michael C, R. E. Bohn, and F. C Schweppe "The Costs of Wheeling and Optimal
Rates", IEEE Transactions on Power Systems, Vol. PWRS-I, No, I, February 1986, pp 63-73.
Caramanis, M. C, R, E. Bohn and F. C Schweppe "System security control and optimal pricing
of electricity" Electricity Power & Energy Systems, Vol. 9 No.4, October 1987, pp 217-224.
Caramanis, M. C, N. Roukos, and F, C Schweppe, "WRATES: A Tool for Evaluating the
Marginal Cost of Wheeling", submitted ot IEEE Transactions on Power Systems, 1988.
Chao, Hung-po, "Peak Load Pricing and Capacity Planning with Demand and Supply Uncertainty," The Beff Journal of Economics, Vol. 14, No. I (Sprin 1983), pp, 179-190.
Chao, Hung-po and Robert Wilson, "Priority Service: Pricing, Investment, and Market Organization," American Economic Review, Vol. 77, No.5, December 1987, pp. 819-916.
Coase, Ronald H., "The Marginal Cost Controversy", Economica Vol. 13, 1946, pp 169-182.
Coase, Ronald H" "The Theory of Public Utility Pricing and Its Application", The Beff Journal of
Economics and Management Science, Vol. 1, spring 1970, pp 123-128,
Coate, Thomas M" "Some Strategies for Reducing Electricity Costs at MIT", MIT bachelor's
thesis, May 1979,
Cohen, Linda, "A Spot Market for Electricity: Preliminary Analysis of the Florida Energy Broker,"
Rand Report N-1817-DOE, February 1982,
Constantopolous, Panos "Computer Assisted Control of Electricity Usage by Consumers", PhD
Thesis, MIT, May 1983,
Constantopoulos, p" R. Larson, and F, Schweppe, "Decision Models for Electric Load Management by Consumers Facing a Variable Price of Electricity", in Energy Models and Studies, B.
Lev (editor), North Holland Publishing Company, 1983.
Constantopoulos, P., F. C Schweppe, and R. C. Larson, "ESTIA: A Real Time Consumer Control
Scheme for Space Conditioning Usage under Spot Electricity Pricing", manuscript, 1987.
Craven, John, "On the Choice of Optimal Time Periods for a Surplus Maximizing Utility SUbject
to Fluctuating Demand," The Bell Journal of Economics and Management SCience, Vol. 2,1971,
pp. 495-502.
Craven, John, "Space-Time Pricing for Public Utilities", The Bell Journal of Economics and
Management Science, Vol. 5, 1974, pp 187-195.
Crew, M. A. and P. R. Kleindorfer, "Peak Load Pricing with a Diverse Technology," The Bell
Journal of Economics, spring 1976, pp. 207-231.
Crew, Michael A. and Paul R. Kleindorfer, Public Utility Economics, N. Y.: St. Martins Press, 1979.
Crew, Michael A. editor, Issues in Public-Utility Pricing and Regulation, D.C. Heath and Company,
Lexington Books, 1980.
Dansby, Robert E. "Spatial Considerations in Public Utility Pricing", in M. Crew, editor, 1980.



Daryanian, Bahman [1986], "The Definition and Application of an Optimal Response Algorithm
for Electricity Consumers and Small Power Producers Subject to Spot Prices", MIT Master's
Dobbs, Charles "Optimal Spot Price Theory and Finite Line Capacity", MIT BA thesis, May 1983.
Elgerd, One I., Basic Electric Power Engineering, Addison Wesley Publishing Company, 1977.
Elgerd, OUe 1.. Electric Energy Systems Theory: An Introduction, Second edition, McGraw-Hill
Book Company. 1982. (TKI45.E487).
Ellis, Randall P., "Electric Utility Pricing and Investment Decisions Under Uncertainty." Ph.D
Dissertation, MIT. 198\.
Enders. Robert H. "Power System Load Modeling for Frequency Mediated Control". MS thesis,
MIT, February 1985.
Faruqui, Ahmad and Robert Maiko, "Load Shape Modifications Associated with Time of Use
Rates for Electricity". in Hotelling, ed. Challenges for Public Utility Regulation for the 1980s,
MSU Press, 1981.
Fitzgerald, A. E., Charles Kingsley Jr.. and Stephen D. Umans, Electric Machinery, fourth edition,
Flory, John E. and O. L. Parker, "Communication and Metering Equipment for Electricity Spot
Pricing", IEEE Transactions on Power Apparatus and Systems. Vol. PAS-104, No.6, June 1985.
pp 1363-1369.
Ford, Andrew "Short-Lead-Time Technologies as a Defense Against Demand Uncertainty", in
Plummer, Oatman, and Gupta, 1985, pp 115-136.
Garcia, E. V. and J. E. Runnels, "The Utility Perspective of Spot Pricing", IEEE Transactions on
Power Apparatus and Systems, Vol. PAS-104, No.6, June 1985, pp 1391-1393.
Gellerson, Mark W. and Shawna P. Grosskopf, "Public Utility Pricing, Investment, and Reliability
under Uncertainty: a Review", Public Finance Quarterly, Vol. 8, 1980, pp 477-492.
Golub. Bennett, and Leonard Hyman "Financial Problems in the Transition to Deregulation". in
Plummer, editor, 1983.
Golub, B., R. Tabors, R. Bohn, F. Schweppe, "Deregulating the Electric Utility Industry: Discussion
and a Proposed Approach," MIT Energy Lab Working Paper MIT-EL 81-043, \98\.
Gol1'!b. Bennett, R. Tabors, R. Bohn. and F. Schweppe "An Approach for Deregulating the
~eneration of Electricity", ch. 5 in Plummer, editor, 1983.
Hasfings, B. F. "The Detroit Edison Second Generation Load Management System", IEEE
J'ransactions on Power Apparatus and Systems, Vol. PAS-104 No.3, March 1985, pp 550-53.
Henney, Alex "Privastise Power: restructuring the electricity supply industry," Centre for Policy
Studies, study no. 83, London, 1987.
Hiaasen, Eric M. "Spot Wheeling Rates for Electricity", MIT MS thesis, August 1984.
Hotelling, H. "The General Welfare in Relation to Problems of Taxation and of Railway and Utility
Rates". Econometrica, Vol. 6, July 1938, pp. 242-269.
Hyman, Leonard S. America's Electric Utilities: Past. Preselll, and Future. second edition, Public
Utilities Reports, Inc., Arlington Virginia, 1985.
ICS. "The ICS Project: TranstexT Participant Development Program on Innovative Pricing for
Electric Utilities", unpublished prospectus, Atlanta, August 1984.
Joskow, Paul and Richard Schmalensee, Marketsfor Power: an Analysis of Electric Utility Deregulation, MIT Press, 1983.
Kaye, R. J. and H. R. Outhred [1988], "A theory of Electricity Tariff Design for Optimal Operation
and Investment", submitted to IEEE Transactions on Power Systems, 1988.
Kelly, K., S. Henderson and P. Nagler, Some Economic Principies for Pricing Wheeled Power,
Report of the National Regulatory Research Institute, Columbus Ohio, 1987.
Kepner, John and M. Reinbergs, "Pricing Policies for Reliability and Investment in Electricity
Supply as an Alternative to Traditional Reserve Margins and Shortage Cost Estimations,"
unpublished manuscript. Approx 1980.
Littlechild, S. C. "Spot Pricing in the Electricity Industry: an appraisal of the MIT model",
presented at a conference in honor of A. Charnes, October 1987.
Littlechild, Stephen c., "A State Preference Approach to Public Utility Pricing and Output Under
Risk", The Bell Journal of Economics and Management Science, Vol. 3, 1972, pp. 340-345.
Luh, Peter B., Yu-Chi Ho, and Ramal Muralidharam, "Load Adaptive Pricing: An Emerging Tool
for Electric Utilities," IEEE Transactions on Automatic Control, April 1982.
Luo, J. S., E. F. Hill, and T. H. Lee "Bus Incremental Costs and Economic Dispatch", IEEE
Transactions on Power Systems, Vol. PWRS-I, No. I, February 1986.



Maiko, J. Robert and Ahmad Faruqui, "Implementing Time-of-Day Pricing of Electricity: Some
Current Challenges and Activities," in Crew, 1980.
Manichaikul, Yongyut, "Industrial Electric Load Modeling", Ph.D. Thesis, MIT, 1978. (Also
Electric Power Systems Engineering Lab Report 54.)
Manichaikul, Y. and F. C. Schweppe, "Physically Based Industrial Electric Load", IEEE Transactions on Power Apparatus and Systems, July/August 1979, pp 1439-1445.
Manichaikul, Y. and F. C. Schweppe, "Physical/Economic Analysis of Industrial Demand," IEEE
Transactions on Power Apparatus and Systems, Vol. PAS-99, No.2, March/April 1980, pp.
Marchand, M. G., "Pricing Power Supplied on an Interruptible Basis," European Economic
Review, July 1974, pp. 263-274.
MIT Energy Laboratory, New Electric Utility Management and Control Systems: Proceedings of
Conference, MIT Center for Energy Policy Research and Electric Power Systems Engineering
Laboratory, Report EL79-024, June 1979.
Moretti, Didier Rene, "Pricing Policies for Power System Security Control," Master's Thesis, MIT,
Munasighe, Mohan, The. Economics of Power System Reliability and Planning, Johns Hopkins
University Press, 1979.
Nguyen, D. T., "The Problem of Peak Loads and Inventories," Bell ]ou/'//al of Economics, Vol. 7,
spring 1976, pp. 197-206.
O'Rourke, Paul "A Physically Based Model of the Space Conditioning Load Under Spot Pricing",
MIT Master's thesis, Operations Research Center, January 1982.
O'Rourke, Paul and Fred C. Schweppe, "Space Conditioning Load Under Spot or Time of Day
Pricing," IEEE Transactions 011 Power Apparatlls alld Systems, Vol. PAS-102, May 1983.
Oren, Shmuel, Stephen Smith, Robert Wilson and Hung-po Chao, "Priority Service: Unbundling
the Quality Attributes of Electric Power", EPRI report EA-4851, Palo Alto, 1986.
Oren, Shmuel, Steven Smith, and Robert Wilson, "Capacity Pricing," Econometrica, Vol. 53, No.
3 (May, 1985), pp. 545-566.
Outhred, Hugh and F. C. Schweppe, "Quality of Supply Pricing for Electric Power Systems",
Conference Paper, IEEE Summer Power Meetings 1980, No A-80-084-4.
Peddie, R. A., G. Frewer, and A. Goulcher "The Application of Economic Theory Utilising New
Technology for the Benefit of the Customer", Discussion Paper, South Eastern Electricity
Board, U.K., 1983.
Plummer, James, Terry Ferrar, and William Hughes, editors, Electric Power Strategic Issues, Public
Utilities Reports, Arlington Virginia, 1983.
Plummer, James L., Eugene Oatman, and Pradeep K. Gupta, Strlltegic Management and Planning
[or lectric Utilities, Prentice Hall, 1985.
Po,irajah, R. A. "Derivation and Applications of Optimum Incremental Costs", M.S. Thesis,
McGill University, 1984.
Ramsey, F., "A Contribution to the Theory of Taxation ", Economic 10u/'//al, Vol. 37, March 1927,
pp 47-61.
RAND, "Real Time Pricing and the Future of the Electricity Industry," Unpublished memo by
Derek McKay, October, 1979.
Ray, D. J. "A Cost Analysis of Wheeling in the Electric Utility Industry", PhD thesis, University
of Wisconsin-Madison, 1987.
Ruane, Michael F., "Physically Based Load Modeling", MIT PhD thesis, June 1980.
Sanghvi, Arun "Economic Costs of Electricity Supply Interruptions", Energy Economics, July 1982,
pp 180-198.
Scherer, Charles R. Estimaling Electric Power System Margillal Costs, North Holland Publishing
Co., 1977.
Schuler, Richard E. and Benjamin Hobbs, "Spatial Competition -- Applications in the Generation
of Electricity", manuscript, April 1981.
Schweppe, F. c., "Power Systems 2000," IEEE Spectrum, July 1978.
Schweppe, F. c., "A Rapid Deployment Demand Side Management Tool for Reduction of Summer
Peaks", unpublished manuscript, November 1987.
Schweppe, F. c., "Mandatory Wheeling: A Framework for Discussion", submitted to IEEE
Transactions on Power Systems, 1988.
Schweppe, F. C., R. E. Bohn, and M. C. Caramanis, "Wheeling Rates: an Engineering-Economic
Foundation", MIT Technical Report TR 85-005, September, 1985,



Schweppe, F. c., R. E. Bohn, R. D. Tabors, and M. C. Caramanis, "Comments on 'Regulation of

Electricity Sales - For Resale and Transmission Services (FERC docket 85-17-000 Phase II)"
October, 1985.
Schweppe, F. c., M. Caramanis, and R. Tabors, "Evaluation of Spot Price Based Rates", IEEE
Transactions on Power Apparatus and Systems, Vol. PAS-104, No.7, July 1985, pp 1644-1655.
Also in Load Management by S. Talukdar and C. Gellings editors, IEEE Press, 1987.
Schweppe, F. c., B. Daryanian; and R. D. Tabors, "Algorithms for a Spot Price Responding
Residential Load Controller", submitted to IEEE Transactions on Power Apparatus and
Svstems, 1988.
Sch~eppe, F. C. and R. D. Tabors, "Electricity Spot Pricing in Developing Countries", National
Development Middle East/Africa, November-December 1984, pp 52-60.
Schweppe, F. c., Richard D. Tabors, James L. Kirtley, Jr., Hugh R. Outhred, Frederick H. Pickel,
and Alan J. Cox, "Homeostatic Utility Control," IEEE Transactions on Power Apparatus and
Systems, Vol. PAS-99, No.3, May/June 1980.
Schweppe, F. c., R. D. Tabors, and J. L. Kirtley, "Homeostatic Control: The Utility/Customer
Marketplace for Electric Power", MIT Energy Lab Report MIT-EL 81-033, revised March
Schweppe, F. C" R. D. Tabors, and J. L. Kirtley, "Homeostatic Control for Electric Power Usage",
IEEE Spectrum, July 1982, pp 44-48.
Secretary of State for Energy, "Privatising Electricity: the Government's proposal for privatisation
of the electricity supply industry in England and Wales", Her Majesty's Stationery office,
London, February, 1988.
Sioshansi, Fereidoon P. [1988), "The Pros and Cons of Spot Pricing: an Electric Utility Perspective",
to appear in Energy Polic)" August 1988.
Steiner, Peter 0., "Peak Loads and Efficient Pricing," Quarterly Joul'Ilal of Economics, Vol. 71, 1957,
pp. 585-610.
Tabors, Richard D" F. C. Schweppe, and M. C. Caramanis, "Utility Experience with Real Time
Rates", accepted for publication and to be presented at IEEE Summer Power Meeting, 1988.
Tal..iiayama, Takashi and George Judge, Spatial and Temporal Price and Allocation Models, North
J101land Publishing Company, 1971.
Tatukdar, Sarosh and Clark Gellings, Load Management, IEEE Press, 1987.
Tsitsiklis, John N., "Linear Optimization Problems with Dynamical Structure," MIT Senior thesis,
'December 1979.
Turvey, Ralph and Dennis Anderson, Electricity Economics, Johns Hopkins University Press, 1977.
Vickrey, William, "Some Implications of Marginal Cost Pricing for Public Utilities", Papers and
Proceedings of the American Economic Association, Vol. 45, 1955, pp 605-620.
Vickrey, William, "Responsive Pricing of Public Utility Services," Bell Joul'llal of Economics and
Management Science, Vol. 2, No. I, Spring 1971, pp. 337-346.
Vickrey, William, "Efficient Pricing Under Regulation: The Case of Responsive Pricing as a
Substitute for Interruptible Power Contracts, "Columbia University, unpublished manuscript,
June 1978.
Vickrey, William, "Responsive Pricing and Homeostatic Control," mime.o, 1979.
Wang, D. "Stochastic Modeling and Parameter Estimation of Residential Electric Loads", MIT
PhD Thesis, 1984.
Wenders, J. T" "Peak Load Pricing in the Electricity Industry," The Bell Joul'llal of Economics, Vol.
7, 1976, pp. 232-241.
Whittle, Peter, Optimization Over Time: Dynamic Programming and Stochastic COlllrol, John Wiley,
Wilber, James G. "A Micro-Processor Based Energy Usage Rescheduler for Electric Utility Load
Leveling", MIT bachelor's thesis, June 198\.
Wood, Allen J. and Bruce F. Wollenberg, Power Generation, Operation, and COlllrol, John Wiley
and Sons, 1984.
Woodard, James B., Electric Load Modeling, Garland Publishing, 1979.


'fhe following is an annotated list of reports and papers written at MIT and
subsequently at Harvard and Boston University on spot pricing and related
isslles. Many of these were discussed in the Historical Notes and References at
the end of each chapter, but this list provides a self-contained summary. Spot
pricing is one part of a larger overall approach to electric power systems called
Homeostatic Control, so there are some references to Homeostatic Control.
The list is separated into three areas:
" Spot Pricing
" Customer Demand Response Modeling
Deregulation and Wheeling


"Power Systems 2000," by Fred Schweppe, IEEE Spectrum, Vol. 15, No.7, July
1978, 6 pages. An informal discussion of how a decentralized state-of-the-art
power system could work using spot pricing.
"New Electric Utility Management and Control Systems: Proceedings of Conference," by the Homeostatic Control Study Group, June 1979, 200 pages,
MIT-EL 79-024. Discussion papers and audience reaction from a conference on
homeostatic control.


Annotated bibliography

"Industrial Response to Spot Electricity Prices: Some Empirical Evidence," by

R. Bohn, February 1980, MIT-EL-80-016WP, 30 pages. An econometric examination of three industrial customers in the U.S. which are on a weak form of
spot pricing. Detailed statistics, but no discussion of what customers did to
allow them to respond.
"Homeostatic Utility Control," by F. Schweppe, R. Tabors, J. Kirtley, H.
Outhred, F. Pickel and A. Cox, IEEE Transactions on Power Apparatus and
Systems, Vol. PAS-99, No.3, May/June 1980,9 pages. A complete and relatively concise presentation of homeostatic control's main elements. A few
equations, but no formal derivations.
"Quality of Supply Pricing for Electric Power Systems," by H. Outhred and
F.C. Schweppe, IEEE Summer Power Meeting, July 1980, paper A80 084 4, 5
pages. An intuitive exploration of the quality of supply aspects of spot pricing.
No equations. The following reference develops the same concepts rigorously.
"Optimal Spot Pricing: Practice and Theory," by M. Caramanis, R. Bohn and
F. Schweppe, IEEE Transactions on Power Apparatus and Systems, Vol. PAS101, No.9, September 1982, 12 pages. Provides the basic methodology framework for the spot price based marketplace covers reactive energy.
"Investment Decisions and Long-Term Planning Under Electricity Spot
Pricing," by M. Caramanis, IEEE PAS, Vol. PAS-IOI, No. 12, December 1982,
9 pages. More details and extension of the above in investment areas.
"Utility Spot Pricing Study: Wisconsin," by M. Caramanis, R. Tabors and R.
Stevenson, MIT Energy Laboratory, June 1982, 200 pages. A case study simulating benefits and their distribution associated with spot pricing of industrial
customers in a Wisconsin, U.S. utility.
"Spot Pricing of Public Utility Services," by R. Bohn, Ph.D. thesis, MIT, Sloan
School of Management, Cambridge, May 1982. Also Technical Report MIT -EL
82-031, 300 pages. A general and comprehensive treatment of economic issues
related to spot pricing of public utility services, including investment and
operational issues, from the utility, societal, and customer perspectives. A
generic customer response model/framework is also developed.
"Utility Spot Pricing: California I," by F. Schweppe, M. Caramanis, R. Tabors,
and J. Flory, MIT Energy Laboratory, December 1982. A case study, sponsored
by Pacific Gas and Electric and Southern California Edison, that explores
implementation issues.
"Utility Spot Pricing: California II," by F. Schweppe, R. Tabors and M.
Caramanis, MIT, Laboratory for Electromagnetic and Electronic Systems,
January 1984. A continuation of the above study that considered other
California utilities and that provided the basis for the methodology
of Schweppe, Caramanis, and Tabors [1985] on evaluating spot price



"Homeostatic Control for Electric Power Usage," by F. Schweppe, R. Tabors

and J. Kirtley, IEEE Spectrum, July 1982, pp. 44-48, 5 pages. An informal
discussion of how Homeostatic Control and spot pricing works.
"Optimal Pricing of Public Utility Services Sold Through Networks," by R.
Bohn, M. Caramanis and F. Schweppe, Graduate School of Business, Harvard
University, HBS 83-21,97 pages. Detailed discussion on impact of transmission
distribution networks on spatial dependence of spot prices. Discusses properties
of optimal investment criteria for transmission.
"Optimal Pricing in Electrical Networks Over Space and Time," by R. Bohn,
M. Caramanis and F. Schweppe, Rand Journal of Economics, Vol. 15, No.3, pp.
360-376, Autumn, 1984.
"Electricity Spot Pricing in Developing Countries," by F. Schweppe and R.
Tabors, National Development, Intercontinental Publications, Inc., Westport,
Connecticut, pp. 52-60, November/December 1984. Survey of concepts from
the developing country's point of view.
"Evaluation of Spot Price Based Rates," by F.e. Schweppe, M. Caramanis, and
R.D. Tabors, IEEE Transactions on Power Apparatus and Systems, Vol. PAS104, No.7, July 1985, pp. 1644-1655. Also published in Talukdar and Gellings,
1987. Practical formulas for calculating spot prices for system planning, with
examples from the California studies above.

Costing ofInterconnected Electrical Utilities Under Spot Pricing,"

by M.e. Caramanis, Large Scale Systems: Theory and Applications, Vol. 9,
198~, pp. 1-18.
"System Security Control and Optimal Pricing of Electricity" by M.e. Caramanis, R.E. Bohn, and F.e. Schweppe, Electrical Power and Energy Systems,
Vol. 9, No.4, October 1987, pp. 217-224. Use of price-quantity transactions for
system security contro\, on a time scale of seconds to minutes. Derives optimal
prices for interruptible rates, spinning reserve pricing, etc. The basic equations
hold for any basic (price-only) rates, not just spot prices.
"Pricing Mechanisms in Coupled Dynamic Systems," by A.W. Berger, Ph.D.
thesis, Harvard University, May 1983. Pricing for very fast dynamics (milliseconds to seconds). On this time scale other issues enter, e.g. physical inertia
of spinning generators.
"Real Time Pricing to Assist in Load Frequency Control," by A.W. Berger, and
F.e. Schweppe, submitted to IEEE Transactions on Power Systems, 1988.
Based on the above thesis.
"Utility Experience with Real Time Rates," by R.D. Tabors, F.e. Schweppe,
and M. Caramanis, to appear in IEEE Transactions on Power Systems, 1989.
Various rates exist in the U.S. and Europe that change prices in real time. These
rates, most of which are not based on spot price concepts, still give information
about customer acceptance, feasibility, etc.


Annotated bibliography


Some of the following do not explicitly address spot pricing but contribute to
the development of a solid foundation for the physically based, end use
modeling. Most papers use a control theory approach in which customers
exogenously specify control rules and objective functions. Others use a different
formulation, based purely on profit maximumization by industrial customers.

Electric Load Modeling, by James Woodard, Garland Press, New York, 1979,
350 pages. Provides a deterministic framework for physicaIly based end use
modeling of electrical demand with emphasis on the residential sector.
"Physically Based Industrial Electric Load," by Y. Manichaikul and F.
Schweppe, IEEE Trans. on Power Apparatus and Systems, Vol. 98, No.4,
July/Aug. 1979, 7 pages. Application of the stochastic framewo~k to the industrial sector by individual case studies of seven specific customers.
"Physical/Economic Analysis of Industrial Demand," by Y. Manichaikul and
F.e. Schweppe, IEEE Transactions on Power Apparatus and Systems, Vol.
PAS-99, No.2, March/April 1980, 7 pages. Models loads as random processes.
Does not explicitly model effects of changes in electric rates, but suggests how
to estimate these effects. Based on seven specific case studies.
"Physically Based Load Modeling," by M. Ruane, MIT Ph.D. Thesis, 1980.
Provides a detailed stochastic framework for modeling residential loads.
"A Theoretical Analysis of Customer Response to Rapidly Changing Electricity
Prices," by R. Bohn, 1980, revised January 1981 MIT-EL 81-00IWP, 150 pages.
A series of models of electricity use, emphasizing the response of profit-maximizing customers to spot prices. Derives the increase in customer profits from
various forms of spot pricing. Some discussion of actual case studies, but no
real-life numerical examples. Later incorporated into Bohn thesis.
"Physically Based Modeling of Cold Load Pickup," by S. Ihara and F.
Schweppe, IEEE Trans. on Power Apparatus and Systems, Vol. PAS-100, No.9,
September 1981,9 pages. Application of stochastic structures to modeling load
level response to short interruptions.
"A Micro-Processor Based Energy Usage, Rescheduler for Electric Utility Load
Leveling," by J. Wilber, Bachelor's thesis, Massachusetts Institute of Technology, June 198 I. Learn hot water heater usage patterns and then control under
spot prices.
"Electric Load Management by Consumers Facing a Variable Price of Electricity" by P. Constantopoulos, R. Larson and F. Schweppe, 1982 Joint National
Meeting, San Diego, California, October 26,1982,30 pages. See also "Decisions
Models for Electric Load Management by Consumers Facing a Variable price
of Electricity," by P. Constantopouios, R. Larson, and F. Schweppe, in Studies
in Management Science and Systems: Energy Models and Studies, edited by



Burton Dean and Benjamin Lev, North Holland Publishing Company, New
York, Vol. 9, pp. 273-292, 1983. General discussion of a theoretical framework
for analyzing customer response to spot prices.
"Computer Assisted Control of Electricity Usage by Consumers," by P. Constantopoulos, MIT Ph.D. Thesis, 1984. Theory of above plus detailed numerical
example of optimization of space conditioning under spot pricing.
"Space Conditioning Load Under Spot or Time of Day Pricing," by P.
O'Rourke and F. Schweppe, IEEE PAS, May 1983, Vol. 102, pp. 1294-1301,
9 pages. Develops simple-to-use formulas to evaluate savings-discomfort
tradeoff for space conditioning under spot pricing.
"Stochastic Modeling and Parameter Estimation of Residential Electric
Loads," by D. Wang, MIT Ph.D. Thesis, 1984. Use of semi-Markov Model with
extensive application to heat pump data from one house.
"Spot Pricing of Public Utility Services," by R. Bohn, Chapters 4 and 5. Models
large industrial customers as profit maximizers with explicit storage and rescheduling options. Solves for optimal behavior under weekly update and 24 hours
update spot prices. Later extended in thesis by B. Daryanian.


NeIther deregulation nor wheeling is a necessary part of spot pricing, and in fact
they are much less important than core applications of spot pricing. However,
both topics are getting a lot of attention (at least in the U.S. and U.K.), and spot
pricing has a lot of relevance to them.
There are two types of deregulation/wheeling proposals in the following
articles. One is incremental: how can small amounts of competition (or
wheeling) be allowed, on a voluntary basis, within a conventional regulatory (or
public ownership) system. The other is complete: how could a completely
deregulated system function. (The analogy for wheeling is mandatory wheeling
to/from customers.) Chapter 5 emphasized the complete approach, while many
of the following look at incremental issues. (E.g. PURPA qualifying facilities
could be allowed to select spot pricing as the basis of avoided cost calculations,
giving them an incentive to self-dispatch.)
"The Costs of Wheeling and Optimal Wheeling Rates," by Michael C. Caramanis, R.E. Bohn, and F.C. Schweppe, IEEE Transactions on Power Systems,
Vol. PWRS-I, No. I, February 1986, pp. 63-73. The theory of wheeling,
involving regulated utilities with simple examples. Non-intuitive nature of some
wheeling rates.
"Wheeling Rates: an Engineering-Economic Foundation", by Fred C.
Schweppe, R.E. Bohn, and M.C. Caramanis, MIT Technical Report TR 85-005,
September, 1985. Major expansion of the above, covering revenue reconcilia-

266 Annotated bibliography

tion for wheeling, and more complex cases such as private generators. Mandatory wheeling and related issues.
"Comments on 'Regulation of Electricity Sales - For Resale and Transmission
Services (FERC docket 85-17-000 Phase II),," by F.C. Schweppe, R.E. Bohn,
R.D. Tabors, and M.e. Caramanis, October, 1985. Comments on how FERC
might regulate bulk utility sales and wheeling to encourage optimal prices (spot
price based).
"WRATES: A Tool for Evaluating the Marginal Cost of Wheeling," by M.C.
Caramanis, N. Roukos, and F.C. Schweppe, submitted to IEEE Transactions on
Power Systems, 1988. Summarizes a PC based model for calculating wheeling
rates in multi-bus systems. WRA TES implements the theory of the preceeding
"Mandatory Wheeling: A Framework for Discussion," by F.e. Schweppe,
submitted to IEEE Transactions PS, \988. Does not emphasize spot pricing.
Instead looks at the many paradoxes and definitional issues surrounding
wheeling, such as "paper wheeling" and the possibility that certain wheeling
transactions can reduce social welfare.
"Marginal Cost Pricing for Mandatory Wheeling," by F.e. Schweppe, unpublished manuscript, 1988. A summary of wheeling issues and equations for a
general audience.
The following papers discuss various forms of deregulation:
"Deregulating the Electric Utility Industry," by R. Bohn, R. Tabors, B. Golub,
and F. Schweppe, MIT Energy Lab Technical Report MIT-EL 82-003, 1982. An
analysis of a deregulated system based on spot pricing. Looks at a number of
possible market structures, with mixtures of regulation and deregulation for
generators, users, and transmission system.
"Deregulating the Generation of Electricity Through the Creation of Spot
Markets for Bulk Power," R. Bohn, B.W. Golub, R.D. Tabors, and F.C.
Schweppe, The Energy Journal, Vol. 5, No.2, 1984, pp. 71-91. A revised version
of the above working paper.
"An Approach for Deregulating the Generation of Electricity," by Bennet
Golub, R. Tabors, R. Bohn, and F. Schweppe, ch. 5 in Plummer, editor, 1983.
Emphasizes complete deregulation.
"Using Spot Pricing to Coordinate Partially Deregulated Utilities, Customers,
and Generators", R. Bohn, F. Schweppe, and M. Caramanis, Chapter 13 in
Plummer et ai, editors, 1983. (Note: the authorship of this article was mistakenly
labeled on the way to the publisher.) Looks at a partial disregulation scenario,
similar to today's system, and why spot pricing should be used as the basis of
transactions by deregulated marketplace participants dealing with regulated
utilities. E.g. why sales by PURPA facilities should be based on spot prices.



Theinain text of this book is supported by eight appendices.

Appendices A, B, and C summarize key factors in the analysis, control,
operation and planning of electric power systems that influence or are influenced
by a spot price based energy marketplace. Readers without a power systems
background are urged to at least scan these three appendices, since they provide
a minimal appreciation of the issues and a starting point for further reading.
Appendix D presents the mathematical details underlying the DC load flow
which is an approximate model for how real power flows propagate throughout
the network. This appendix is essential for readers interested in spatial pricing
Appendix E presents four different customer benefit functions and the resulting customer price response models. This appendix is not fundamental and
should be read only by those interested in technical details.
Appendix F presents the basic ideas underlying the F APER (Frequency
Adaptive Power Energy Rescheduler). This is a device designed to allow
customers to provide operating reserve response without knowing it is happening. The idea has excellent potential but is not essential to the main concepts of
this book. Hence, it is optional reading.
Appendix G presents an example of the computation of 24-hour expected
spot price trajectories.
Appendix H addresses a mathematical "fine point" related to the interchange
of derivative and expectation operations.


--r:his appendix summarizes some of the key ideas in the analysis of network
flows and power system dynamics. It also discusses various local controllers that
are on the power plants and scattered throughout the network. Appendix B
discusses centralized control and operation.
A comment on notation: In the main text of this book, "t" denotes discrete
time, usually with one-hour increments. In this appendix, t denotes a continuous
time variable. Similarly in the main text, real power is measured by energy
(kWh) and denoted by y, d, and g as appropriate. In this appendix, real power
is denoted by p or P with units of kW.

Single-Phase AC Power

The instantaneous electrical power flow pet} at any point in an electrical network
is given by the product of the voltage vet) and the current i(t}. Power system
voltages and currents vary sinusoidally with time (ignoring harmonics). Thus
for some point on a single phase system,

Vma< sin (WI)


Ima> sin (wt




Voltage amplitude


Current amplitude

Angular frequency (rad/sec)


Phase angle between v(t) and i(t) (rad)

The power flow is given by



sin (wt) sin (wt


+ IX)

(A.l.l )

Using some trigonometric identities, (A.l.l) becomes

p(t) =

P[I - cos (2wt)] - Q sin (2wt)

(A. 1.2)

where by definition
P: Real Power

VI cos (IX)

Q: Reactive Power
VI sin (IX)



The real power component of (A. 1.2) (i.e., P[I - cos (2wt)]) has a time-average
value equal to P while the reactive power component (i.e., Q sin (2(01) has a zero
time average.
Three-Phase AC Power

Single-phase power has a pulsating characteristic. For medium and large loads
and generators, pulsating power is undesirable. This is avoided by using a
polyphase system, usually three-phase.
The relationships among voltages and currents at any point on a three-phase,
balanced electric network are
Va (t)

V sin (WI)

ia (t)

I sin (wt


V sin (WI - 21t/3)


+ IX - 21t/3)
V sin (wt + 21t/3)
I sin (WI + IX + 21t/3)


I sin (wt


Appendix A 271

where "a", "b", and "c" denote the three phases. The total power flow at that
point is the sum of the powers due to each phase, so

By using some trigonometric identities, (A.1.3) reduces to

pet) = P = 3VI cos (a)


P: Three-Phase Real Power

which is constant in time.

Similarly define
Q: Three-Phase Reactive Power


(A. 1.5)

The fact that the three-phase pet) of (A. 1.4) has no Q sin (wt) component as in
(A.I.2) does not mean there is no three-phase reactive power. After all, each of
the three lines is carrying a pet) that looks like (A.I.2). Reactive power is
sometimes called "imaginary" power. I
Transmission Line Modeling

Consider a balanced three-phase transmission line between two buses (or nodes)
of a network. Assume at Bus} and Bus k the "a" phase voltages are respectively

+ b)
V. sin (wt + 15 k )

JIj sin (wt

with phases "b" and "c" shifted in place by 120 and 240.
A new variable,
8): Voltage phase angle at Bus j,

has been introduced to denote the fact that the phases of the voltage sine waves
vary with location on the network. The voltage magnitudes and angles vary
continuously between buses.


Real power leaving Bus}

Real power leaving Bus} and flowing towards Bus k (may be

or -)

(A. 1.7)

272 Appendices

where the summation is over all buses connected to Busj. Qj and Qjk are defined
Assume line i connects Bus j to Bus k. The equation that relates ~k to the
voltage magnitudes Vj, /ik and voltage phases bj , bk and the characteristics of line
i is

G,v} -

+ n, V; /Ik



sin (t5j


15 k )
(A. 1.8)


R; +


R;: Resistance of Line i


Reactance of Line i 2

We know of no easy way to motivate this equation without a lot of background

development. 3 We just state (A.l.8) to give the reader a feel for the types of
equations involved (and to serve as a starting point for Appendix D).
Note from (A. 1.8) that, in general, Pl2 ::F P21 Because of losses, the power
entering one end of the line is not equal to the power leaving the other end.
The equation for the reactive power Qjk flowing from Bus} towards Bus k has
a form similar to (A.I.8) except that it includes another physical parameter of
the line called the shunt capacitance. These shunt capacitances do not effect the
~k explicitly but do affect the voltage magnitudes and hence do influence the
values of ~k'
Per Unit System

Network reactances, resistances, real power, reactive power, and voltage are
usually expressed in a per unit system wherein a reference voltage and usually
MWh level are specified and all variables are normalized with respect to these
references. In a per unit system, the voltage magnitude V is usually within the
range 0.95 to 1.05.
AC Load Flow

The purpose of an AC load flow computer program is to calculate the real and
reactive powers flowing through the transmission lines of a given network for
some specified bus conditions, such as real and reactive power or voltage
magnitude and real power. The network's structure and parameters (e.g., resistances and reactances) yield a set of 2Nb - I (Nb = number of buses) simultaneous nonlinear equations which have to be solved, e.g.,

Appendix A


~ == [1i';, ... VNb,Dj, ... DNbf

A set of functions build from equations that look like (A.I.8)

Note: The value of Pb k = .. is not included in u and the value of 15k , k = .. is not
included in ~. The bus .. is called the swing bus (Or reference bus)

The value of x is to be found, given u. The line flows are then easily computed
from x using (A.1.8) (x is sometimescalled the state vector). The real power at
the sWIng bus k = .. is-not specified in order not to overspecify the problem. The
power assigned to the swing bus is the difference between the total load power
and the calculated line losses minus the total power at all the generator buses
except the swing bus. Generally, the angle at the swing bus is arbitrarily set to
zero, since the absolute values of the voltage angles [)j are not important.
AC load flows are solved by iterative techniques. Two solution methods are
"Gauss-Seidel" and "Newton-Raphson."

This method first assumes initial values for the voltages (magnitude and angle)
at each bus. Then the voltage at the first bus is calculated to match the specified
powers at that bus, keeping the voltages at all other buses at their intial values.
Then the voltage at the second bus is calculated, keeping the voltage at Bus I
at the value just calculated. This procedure is repeated for each bus until
convergence. This method does not converge fast, but is very simple and usually

This method makes corrections to the initital voltage values assigned to the
buses while taking into account the interactions with other buses. Define

The problem is to find x such that/ex, u) = O. Using a Taylor's series expansion

about the ith guess ~iand keepin-g only linear terms yields


(If"(.>:, u)



The next guess

{(x,)(:5.. - ~,)

for x =

~+ I

- -X, IS the Jaco

Ian matrIX

is given by solving


274 Appendices

This iteration continues until x j + I ;;:; Xi> which means that !(x j , u) ;;:; O.
Solving the linear system of (A. I. IOns not trivial, since
order can be very
high (e.g., 500 to 2000). Fortunately the Jacobian matrix is very sparse (i.e., it
has a lot of zero entries). Special numerical techniques are used to solve these
sparse matrices.


Decoupled AC Load Flow

An AC load flow works with four basic types of variables: real powers, reactive
powers, voltage magnitudes, and voltage angles. The nature of the physical
system is such that there is strong coupling between
Real power P and voltage angle J

and between
Reactive power Q and voltage magnitude V

with much weaker coupling between, say, voltage angle and reactive power. This
property is often used to decouple the AC load flow logic into two separate
subproblems, which then iterate with each other.
DC Load Flow

An approximation called the DC load flow is used in this book instead of the
full AC load flow. This approximation yields a linear set of equations relating
real power P to voltage angle <5, i.e., it is "one half" of the decoupJed AC load
flow. It is developed in detail In Appendix D.4
Optimum Load Flow

An optimum load flow is a computer program that tries to find the set of bus
power injections, voltage magnitudes, etc., that minimize some criteria subject
to constraints which include the condition (A.1.9). For example, the criteria
might be to minimize losses where in addition to (A.l.9) the real power flow
through line i is not allowed to exceed a prespecified value. Optimum load flows
can be important in system operation relative to economics and security (see
Appendix B).
Time Variations

Thus far we have been considering the magnitudes and phases of the sinusoids
to be constant in time. In practice, concern is usually with the time variations
of the amplitudes and phases of the sine waves. Let V(t), P(t), and (j(t) denote
these time variations. V(t) can be viewed as an "amplitude modulation" of the
sinusoidal voltage, etc.

Appendix A 275

X: location of Fault





FigureA.2.1. Three-bus system used to illustrate protective relaying.


A power system is controlled and operated by a hierarchy of local and central

control systems. Local controllers at the individual power plants and scattered
throughout the transmission network are discussed in this section. Distribution
network local controls are somewhat different. Higher-level central controls are
discussed in Appendix B.
Network Relaying

Relays are extensively used for the protection of transmission lines and transformers. A relay contains the logic that decides to open or close a circuit breaker
if certain locally measured conditions are met. Two commonly used network
relays are overcurrent relays and impedance relays.
To illustrate the sophistication of network relaying, consider the three bus
network of Figure A.2.1. Four circuit breakers Cl to C4 are shown. Each has
an associated impedance (also called distance) relay, R I to R4, which detects the
presence of a "fault" and estimates its location by measuring the voltage and
current at the relay's location. The 'x' on the transmission line of Figure A.2.1
represents a fault, which in this case is a short circuit due to a lightning strike
which established an ionized electrical path for current flow to ground (or
between phases). This path is sustained if the potential difference between the
line and ground is high enough.
The following is one possible sequence of events:


= 0, the short circuit occurs.

= 4 cycles,S each relay determines the location of the fault.

8 cycles, relays R3 and R4 open circuit breakers C3 and C4.

0.5 sec, relays R3 and R4 reclose circuit breakers C3 and C4. 6

276 Appendices

At t = 0.5 sec + 8 cycles, relays R3 and R4 reopen the circuit breakers if the
fault is still there.
The circuit breakers Cl and C2 did not trip because their impedance relays RI
and R2 determined that the fault was not within their zone of protection. In this
example, the fault is in the zone of protection of R3 and R4, but not R I and R2.
However, ifC3 and C4 fail to open for some reason, R2 will trip circuit breaker
C2 after a preset time delay.
Network Controllers

Tap changing transformers, switchable capacitors, synchronous condensers,

etc. may be installed at various points on the network to help control voltage
magnitudes. Many of these operate automatically under their own local controllers which adjust the taps, switch the capacitors, etc. to try to maintain the
voltage magnitudes near some prespecified set points. The set points are
adjusted as needed by the central control system discussed in Appendix B.
Power Plant Relaying

Relays are used to protect the power plants as well as the network. They are set
to "turn off the plant" when they sense a problem which could damage the
Because of the massive capital investment associated with any given power
plant, these relays are set very conservatively; i.e., it is much better to "cry wolf'
than to damage the plant. It is the job of the central controllers discussed in
Appendix B to make sure that the sudden loss of anyone power plant does not
affect the service being provided the customers.
Power Plant Controllers

The power plant operators are the most important controllers. A modern power
plant control room has a vast array of displays and switches for the operators'
use. Digital computer-driven display and diagnostic systems are playing an ever
increasing role.
There are also many automatic control loops within a power plant. To
illustrate, consider a fossil steam power plant.
The voltage regulator controls the excitor in order to maintain output voltage
magnitude at the desired set point.
The governor controls steam flow into the turbine so that the frequency does
not drop "too much" as load increases (as will be discussed in Appendix B, the
local power plant governors do not attempt to maintain exactly 60 Hz).
Boilers have extensive automatic firing control loops to maintain pressure and
temperature within acceptable limits while providing the needed steam flow to
the turbine.
The best way to really appreciate what a power plant is and how it is
controlled is to visit one.

Appendix A 277


Power Plant Dynamics

The number of differential equations used to model the boiler of a steam power
plant and its controllers varies from two to 200. Time constants for boiler
transients range from seconds to 20 minutes.
A turbine is often represented by two to four differential equations whose
time constants range from one to 15 seconds.
The generator is often approximated by a set of two to five differential
equations with time constants from 0.01 to 0.1 seconds.
A key equation of motion for the jth generator is given by Newton's second
law to be
HJ. dJ;(t)

P.mechJ (t) - p.eiecJ (t )


Inertia of the generator rotor and turbine

J;(I): Frequency of generated power which is close to being proportional to the speed

of rotation of the generator




Voltage phase angle at generator bus j

PmeohJU): Mechanical power from the turbine to the generator

PclecJ(I): Electrical power from the generator to the grid

This equation is known as the "swing" equation.

Transmission Network Dynamics

As explained in Section A.I, a transmiSSIOn network in steady state can be

represented by algebraic equations like (A.I.9). The dynamics of the network
are so fast compared to the dynamics of the generators that the steady state
model is assumed to hold during transients; i.e., the network transients are
usually ignored and only the variations in P(t), Q(t), V(t) and <5(t) are
computed. 7
Load Dynamics

Load dynamics are not really understood and so are usually modeled algebraically, i.e., transients are ignored. Typical models are constant impedance,
constant power, frequency sensitive, and voltage sensitive.

Three types of power systems dynamics with different time scales of c('ncern are

278 Appendices

Transient Stability: Very fast; cycles to 10 seconds. Nonlinear.

Dynamic Stability (also called Steady State Stability): Slower; I to 10 seconds.
Long-Term Dynamics (also called Slow-Speed Dynamics): Slowest; seconds
to minutes. Nonlinear.

Transient Stability

If a short circuit occurs on a transmission line, the protective relays "clear" the
fault within cycles, as discussed in Section A.2. During this time, the abnormal
conditions cause mechanical transients in the generators which are governed by
the swing equation (A.3. I). If thejth generator is close to the fault, then as long
as the short circuit exists, its electrical power output is zero (or very small),
because it is trying to supply a load with zero impedance. Since the mechanical
power input to the jth generator cannot change within cycles, the right side of
the swing equation is positive. This implies a positive acceleration, i.e., the jth
generator starts to speed up relative to the rest of the system. If the fault is not
cleared in time, the generator's speed (frequency) increases so much that it
"pulls out of step," i.e., loses synchronization with the rest of the system.
Transient stability is usually studied by numerical integration of the nonlinear
swing equations plus other differential equations of the generator-voltage regulator models. An AC load flow is done at each time step to evaluate the effects
of network coupling between the generators and loads. Boiler and turbine
dynamics are often ignored. For large interconnected systems (say more than
tOo generators), such numerical integrations can run much "slower than real
time" even with powerful digital computers.
Dynamic Stability

Large interconnected power systems with relatively weak transmission links can
exhibit small-amplitude, low-frequency (one- to ten-second period) sustained
oscillations. This is called the dynamic stability problem. It is usually studied
and analyzed by linearizing the nonlinear differential equations such as used in
transient stability studies about an operating point and then doing eigen valueeigen vector analysis. In general, dynamic stability involves slower dynamics
than transient stability. Turbine dynamics may be included while boiler
dynamics are usually ignored.
Long-Term Dynamics

Long-term dynamics looks at transients that are much slower than either
transient or dynamic stability. To illustrate, consider a two-generator system
At t = 0 -, Both generators supply the load
At t = 0, Generator 2 is tripped off due to some fault, then

Appendix A


At t = 0 +, Generator I supplies the load by itself using the inertial energy

stored in the rotating mass of the turbine and generator
This causes the frequency at Generator I to drop.
The mechanical input power of Generator 1 then increases due to turbine
action using thennal energy stored in the boiler of Generator 1 to try to match
the electrical load. The firing rate of the boiler at Generator 1 then increases to
try to reach a level which can meet the electrical load.
If the turbine and/or boiler does not respond fast enough, load-sheddingunder-frequency relays drop some load in order to decrease the rate of frequency drop, giving the turbine and the boiler more time to increase mechanical
power. The dropped loads are energized again one by one. Finally, a new steady
state is reached. If generator I is not large enough to meet all of the load by
itself, other generators have to be started up or a blackout results. Generators
have under- and over-frequency relays which prevent them from operating at
too low or too high a frequency for any period of time, since such operation can
cause vibrations which damage the generator.
Such long-tenn dynamic behavior for a multiple generator system can often
be modeled by a basic swing equation like (A.3.1) except that Hk is replaced by
the sum of the inertias of all the generators, Pmech,k(t) by the sum of the
mechllnical power outputs of all the turbines, Pe1ec,k(t) by the total load plus
losses, andfk(t) by an average (over space and time) system frequency. Turbine
and boiler dynamic modeling is very important for long-term dynamics studies
while the faster transients considered in transient and some dynamic stability
studies are usually ignored. Only a relatively few AC load flows are needed.
Long-term dynamic studies can last from seconds to many minutes. They are
usually done by numerical integration of the nonlinear differential equations.

Many good books cover the modeling of three phase AC networks, load flows,
relaying, generation modeling and transient stability at various levels of detail.
Examples of different types include Fitzgerald, Kingsley and Umans [1983];
Elgerd [1982] and Bergen [1986]. A good source for material on boiler-turbine
modeling, dynamic stability and long-term dynamics (also on load flow, etc.) is
the IEEE Transactions on Power Apparatus and Systems. s The December issue
of each year contains a detailed index of papers published that year.
I. The term "imaginary" is motivated by the common use of complex numbers to represent power
where Q(I) is the imaginary part. Readers who are new to the area should refrain from the mistake
of associating the term "imaginary" 'to "it doesn't exist"
2. These resistances and reactances are equivalent circuit values resulting from the steady state
solution of a partial differential equation.
3. If (A.I.8) is obvious to a reader, that reader is wasting his/her time by reading this appendix.



4. The term "DC load flow" arose because the linear relationship between!:.. and .Q is analogous to
the relationship between current and voltage in a direct current network which contains only
resistors. "DC analog" circuits were used to solve for line flows in the days before large digital
computers were available.
5. A cycle is a measure of time equal to 1/60 second assuming the power system operates at 60 Hz.
In many parts of the world, power systems operate at 50 Hz. However, since the authors live in
the U.S.A., a 60-Hz system is assumed in this book.
6. The ionized path from the line to ground should go away in 0.5 seconds.
7. In switching studies, network dynamics can be important but such issues are not addressed here.
8. Starting in 1986, Power Apparatus and Systems has been replaced by three publications, IEEE
Transactions on Power Delivery, Energy Conversion and Power Systems.


Th1s appendix summarizes the key functions performed by a utility's central

control system.
A generic central control system consists of brains (highly trained operators
with extensive digital computer support), combined with an extensive communications system that uses telephone lines and utility-owned microwave lines to
gather measurements and information from around the system and to send
commands. The communications system is often called the SCADA (Supervisory Control And Data Acquisition) system.
As with power plants, readers with little background should visit a power
system control center to really appreciate what they are like. Modern ones are
very impressive.

A key input to the system economics and security functions (to be discussed in
subsequent sections) is a forecast of what future demand will be, say hour by
hour for the next week, or day by day for next month or year.

Diversity of customer demand is absolutely essential to the operation of today's

electric power systems. Define


282 Appendices

0k(t): Demand for electricity during hour t for the jth usage device (air con-

ditioner, motor, lighting, etc.) of the kth billing entity (customer) (kWh).
m(t): Vector of meteorological variables during hour t such as temperature and

A key assumption is
Elemental Independence: At hour t and for a given value of '!!.(t), the :!Jk(t)'S are
statistically independent over j and k.

There are special cases which violate this assumption, but relative to the present
level of discussion, elemental independence is a reasonable assumption.
The demand of the kth customer is
dk(t) =

I0d t)

and the total demand is

The conditional expected value E {d(t)lm(t)} of the total demand at hour t for
given weather conditions m(t) is givenby the sum of the mean values of its
components; i.e.,

By virtue of the elemental independence assumption, the conditional variance

of d(t) is the sum of the conditional variances of the 0k(t); i.e.,
Var{d(t)I'!!.(t)} =

I I Var{0k(t)I'!!.(t)}

For sake of simplicity only,

E {d(t)I'!!.(t)}

Var{d(t)I'!!.(t)} =

Nd E {0k(t)I'!!.(t}}
Nd Var{0dt}I'!!.(t}}

Nd: Effective total number of devices

The diversity of total demand can be measured by the ratio of the standard
deviation (square root of variance) u{d(t)lm(t)} to the mean value. Assuming
that the order of magnitude of individual conditional means and standard
deviations is close to one,

Appendix 13283







E{d(t)l~(t)} ""




Since Nd for any reasonable sized utility is very large, (8.1.3) says that
accurate short-range load forecasts are possible if one is able to forecast
E {d(t)\m(t)} - i.e., the dependence of the total system demand on weather and
of course time of day, day of week, etc. 1 The key and very important point of
this discussion on diversity is that
The randomness introduced by the independent variations of the individual usage devices
can be ignored when considering total demand behavior.

Weather and Time Dependence

One single model structure for the conditional mean demand during hour t is

= Periodic Component plus Weather

Dependent Component
Periodic Component: Fourier series with fundamental frequency of 24-hour
Weather-Dependent Component: Nonlinear function of outside temperature
~(t) which models both heating and cooling and the saturation effects of very
high or low temperature.
E {d(t)lm(t)}



Real-world complications have the Fourier series coefficients varying with day
of week and season of year and a weather-dependent component that includes
"heat build-up" dynamics, modeled for example by ARMA techniques. 2
Many different types of short-range load forecast models are in use and not
all "lookalike" (B.IA). However (B.IA) illustrates the basic ideas.
In practice, the biggest source of error in short-range load forecasts is the
effect of errors in the weather forecasts.

Power system dynamics as discussed in Section AA cover time scales ranging

from fractions of seconds to many minutes. The system economics functions to
be discussed cover time scales which range from five minutes to many months.
Economics of Thermal Power Plants

Define for a thermal power plant


H = Heat input into the plant (Btu/hr)

C = Fuel cost per unit of energy ($/Btu)
F = (H) (C) ($/hr)
P = Electrical power output (kW)


















Figure B.2. I. Curves relating efficiency and costs.

Figure B.2.l shows typical curves relating these quantities with variations in
output power; B.2.! a shows the input-output relation of the plant, B.2.l b is the
"heat rate" (HIP) where the heat rate is the inverse of efficiency, and B.2.l c is
the incremental heat rate. The horizontal axis in the three curves is the actual
power going into the grid. The total electrical power out of the generator is
about 5% higher than P. This extra 5% is used to run the power plant itself

Appendix B 285

(pumps, fans, etc.).

The incremental (marginal) cost of generation is given by
Incremental Cost =

oP =

oH ($jkWh)


The curves of Figure 8.2.1 are smooth functions. In practice, such curves can
be much less well-behaved. One example is the "valve" point loading" issue
associated with many fossil steam power plants. Incremental heat rate curves for
such plants can look like "saw tooth" functions.
Economic Dispatch

The economic dispatch problem is to find the particular output levels for each
available generator that minimize the total fuel costs while meeting all of the
loads plus line losses. Because of network losses, less efficient generators ($/
kWh) located close to the loads may be used more than more efficient generators
located far from the loads. Typically, economic dispatch optimizations are
recalculated every five to ten minutes with a linear extrapolation (based on a
very short-term load forecast) used in between times. "Raise and lower pulses"
are sent to some generators every two to 20 seconds by the AGC system (see
Section BA).
T\;le equations for economic dispatch are closely related to the spot price
equations developed in the main text of this book.)
Unit Commitment

The unit commitment problem considers longer time scales - for example, hour
by hour for one day or one week. Not all of the generators are needed at certain
times of the day. Unit commitment specifies the daily on/off schedule of generators.
A simple example is illustrated in Figure 8.2.2 for a system with three
generators each rated at 100 MW. If the operating costs of the power plants
increases from Generator # 1 to Generator # 3, the most economical way to
operate the system is as shown in Figure B.2.2.
The simple picture of Figure 8.2.2 becomes more complicated when real-life
constraints regarding each generator are considered, such as
Minimum Up Times: The generator must run for a minimum time.
Minimum Down Times: If shut off, the generator must remain in that state for
a minimum time.
Startup Costs: It takes fuel to heat up a cold boiler.
Ramp Rates: It takes time to go from zero to full load.
Crew Availability: If a plant has two or more generators, the operators may be
able to start only one at a time.

286 Appendices








Plant 1



Time (Hours)

Figure 8.2.2. Simple illustration of unit commitment.

There are also system-wide constraints such as transmission line capacity limits
and the need to carry operating reserves (see Section B.3).
Taking all the real-world constraints into account, unit commitment becomes
a very complicated problem. Note that economic dispatch is a subproblem of
unit commitment; i.e., in theory, for each possible combination of generators
that can supply the load, an economic dispatch must be run.
Two of many approaches to the unit commitment problem are discussed.
Priority Lists

Given a set of power plants and their operating costs, the generators with
cheapest operating costs are first committed as much as possible. The effects of
the constraints are then incorporated. This heuristic method requires a lot of
insight about a particular system. System operators with experience and a good
knowledge of the system can be very effective. Alternatively, heuristic optimization logics are incorporated into a computer program.
Dynamic Programming

Conceptually dynamic programming is tailor-made for the unit commitment

problem as it is an ingenious way of finding the optimal path between two states
given a finite number of possible paths. However, pure dynamic programming
introduces dimensionality problems. In practice, a combination of dynamic
programming and priority list heuristics can yield an excellent computer

Appendix B 287

Fuel Purchases

Most utilities purchase some or all of their fuel (e.g., oil and coal) on the open
market. This leads to another type of optimization problem. Issues related to
determining fuel contracts and purchases include prices of different suppliers,
transportation costs, storage capabilities, purchasing conditions, etc. Fuel
contracts can be signed for' time spans of months and years with possible
provisions for small weekly adjustments. Linear programming can be an effective tool for optimizing fuel purchases.
Hydrothermal Systems

The economic operation of a system with hydroelectric as well as thermal plants

is much more complicated than that of a pure thermal system. Hydroelectric
generation introduces a large number of new technical, economic, and social
constraints, such as
Variation of Water Levels in Reservoirs: A large variation can hurt recreation
facilities that developed in the area and have adverse impacts on lake life.
Rate of Water Flow: Flow rates are constrained to avoid fish kill, erosion of
river banks, to allow irrigation of cultivated areas downstream, to allow
navigation, for sewage control, etc.
Weather Conditions (short-term): If a storm is forecasted, the water level of a
rese.rvoir and the rate of water flow may have to be constrained to prevent
Weather Conditions (long-term): Snowfalls in the mountains and the depth of
the snow pack influence hydro operation many months into the future.
There is no standard optimization logic for hydrothermal scheduling because
each system is different. One approach involves iteration between a pure thermal
optimization (for a fixed hydro schedule) and a pure hydro (for a fixed thermal
unit commitment).
Pumped Storage

Pumped storage hydro plants present special economic scheduling problems. A

typical one-week schedule starts on Monday morning with a full reservoir. Some
of the water is used to help meet Monday's peak demand. On Monday night,
thermal plant energy is used to partially refill the reservoir. This cycle is repeated
throughout the week until by Friday afternoon, the reservoir is at its lowest
allowable level. The reservoir is then completely refilled over the weekend.
Pumped storage decisions can be built into the unit commitment logic.
Pumped storage can be used to reduce the peak demand seen by the thermal
generators, to help compensate for thermal start up costs and limited ramp rates
etc., and for operating reserves.

288 Appendices

Maintenance Scheduling, Nuclear Refueling

Power plants have to be removed from the line for routine maintenance (two to
four weeks per year for a large fossil power plant). This leads to maintenance
scheduling which looks a year in advance to schedule maintenance, taking into
account seasonal demand variations, availability of maintenance crews, etc. A
maintenance schedule determined in January for the rest of the year may be
changed completely in April if a major power plant is forced out in April and
requires extensive emergency maintenance. This can affect maintenance on
other plants (as well as the forced-out plant).
Maintenance scheduling optimization is of the integer programming type.
However, it is usually done using heuristic optimization logics and/or the
judgment of experienced operators.
Maintenance scheduling can be combined with nuclear refueling decision
logics (for a utility with nuclear power plants, that is). Nuclear refueling opti
mization involves complex nonlinear relationships covering long-term fuel cycle
costs (which can span several years). For many nuclear power plants, maximum
capacity is reduced at the end of a fuel cycle to get the most energy out of the

Section A.2 discussed local relay logics which act to avoid damage to the
equipment. This section discusses system level procedures used to protect the
system in the event of a failure. As an example, assume a heavily loaded
transmission line is automatically withdrawn from the system by local protective
devices. If the overall system is not prepared for such an event, other lines may
become overloaded, and their overload relays may trip them out of the system
as well. This could cascade into a blackout.
System Monitoring

Voltage magnitudes, power flows (real and reactive), circuit breaker status (i.e.,
open or closed) are monitored by measuring instruments scattered through the
system. These measurements are sent to the central control system using the
real-time communication system.
The information received is processed by digital computers and presented to
the operators via display monitors. The computer compares the incoming
measurements with previous ones and upper and lower operating limits. It
warns operators in case of irregularities in the data or measurements that lie
outside of safe operating regions.
State Estimation

State estimation is a procedure which converts network measurements into an

estimate of the state variables, i.e., the voltage magnitudes and phase angles at
the buses. Redundant measurements are used to counter the effect of metering
errors and bad data arising from, say, meter failure. State estimation algorithms

Appendix B 289

give the estimate of the state variables which minimize the effects of the measurement and modeling errors.
The most common criterion in state estimation consists of minimizing the
weighted sum of squares of the differences between the measurements themselves and the values of the measurements as computed from the estimated state
variables. A state estimation program can be viewed as a type of AC load flow.
Many similar computational methods are used, e.g., Newton's method, decoupling techniques, sparse matrix methods, etc.
System monitoring can be done using the estimated network conditions
rather than the raw measurements. For example, once the state variables have
been estimated, it is easy to compute all the line flows and to test whether they
and all the voltage magnitudes lie within safe operating regions even if they were
not directly measured.
Corrective Control Actions

Corrective control involves changing the operating conditions of the system if

transmission lines are overloaded or voltage magnitudes are not acceptable.
This is usually done by rescheduling the power generation and/or adjusting
generation voltage magnitudes, tap changing transformers, switchable capacitors etc. Sometimes it is necessary to shed loads.
Such rescheduling of real power generation and voltage magnitude can be
done using an optimum load flow (see Section A.I) which minimizes total
opedting costs subject to the network constraints. However, in practice, it is
often done by combining sensitivity analyses with operator judgment.
The network quality of supply component of the spot price as developed in
the main text of this book can be viewed as a way of accomplishing this
corrective control action.
Contingency Analysis

Contingency analysis addresses "what if' questions concerning potential failure

of an important part or parts of the system, e.g., how would a transmission line
outage affect the rest of the network? In principle, the analysis is simply an AC
load flow program that is run assuming various prespecified, possible transmission line outages. If these simulations show that a line is drastically overloaded by some outage, the operators can apply corrective control before the
fact so that, if the event happens, the rest of the system will not be in danger.
Obviously, there is a tradeoff in doing corrective control just in case something
happens. By doing so, the system is more reliable but, by definition, is no longer
dispatched in the most economical way. Often corrective control is not actually
applied before the fact, since outages are not too frequent. The contingency
analysis simply gives the operators a priori guidelines on how to proceed in case
a key outage does occur.
Contingency analysis has to consider the effect of other interconnected utilities. For example, the changes in line flow within a given utility resulting from

290 Appendices

a line outage within the utility depend on the status of the network
generation patterns of the other interconnected utilities. In the case of
dependence, the neighboring utilities may share real-time information on
state of their respective systems. Otherwise external equivalent models are
(developed from off-line studies or identified from measurements).
Operating Reserves

The term "operating reserves" is used in this book to denote the geller,ato,r
reserve the utility has to maintain to prevent blackouts (or major tr"'rtll,"n"'"
and/or tie-line flow deviations) in case of the sudden loss of some generation
tie-line support.
To understand the operating reserves problem, consider again the three-plant
example of Figure B.2.2. If Generator # I fails at 4 AM, a blackout will occur
since no reserve generation is available to take over. Operating reserves is one
way to avoid this. If Generator # 2 shares the load with Generator # 1, but is
not at its maximum output. Failure of Generator # I at 4AM might not cause
a blackout, if the operating reserve of Generator # 2 is large enough and can
react fast enough (see discussion of long-term dynamics in Section A.4).
Operating reserve is sometimes associated with spinning reserve. However,
the term "spinning reserve" actually refers to generators which are connected
(synchronized) to the network but which are not operated at their maximum
output levels. In practice, utilities may also maintain other types of operatingfast-acting reserve such as fast-start gas turbines, pumped and regular hydro,
etc. that can be brought up to full power in less than ten minutes. Operating
reserve can also be "carried" by loads which can be rapidly interrupted by utility
signals or FAPER (see Appendix F) action.
System Dynamics

System dynamics, as discussed in Section A.4, cannot be ignored. In practice,

however, little in the way of real-time modeling and analysis is done to protect
the system from undesirable dynamics.
Transient stability problems have time constants that are too fast for a central
system to handle (today's technology). Conceptually, transient stability contingency analyses could be done on-line but, in practice, suitable models are not
available. Dynamic stability contingency analyses could conceptually be done
on-line but are rarely implemented. Instead, off-line, planning type studies of
transient and dynamic stability can lead to transmission line flow limits which
are then handled in the various system security functions just as if they were, for
example, thermal line overloading limits.
Long-term dynamics (which determine the required operating reserve) are
usually not evaluated by anyon-line mathematical model either. Operating
reserve requirements are usually based on predetermined rules developed from
engineering judgment and off-line planning type studies.

Appendix B 291


Automatic generation control4 (AGC) provides a bridge between the power

system dynamics or Section A.4 and the economic-security functions of Sections
B.2 and B.3.
Isolated Utility

Consider an isolated utility whose power plants are tied together and to loads
by a network that is not electrically interconnected to other utilities. In this case
the role of the AGC is to
Keep frequency close to the desired 60 Hz (or 50 Hz if appropriate).
As discussed in Section A.2, each power plant has a governors which uses
locally measured electrical frequency to increase energy output when frequency
goes down (i.e., when the mechanical power driving all the turbines is less than
the power delivered to the loads and losses). These governors are built with a
"droop" characteristic. Thus, if frequency is initially 60 Hz, then after an
increase in load, generation increases to meet the new demand but the resulting
frequency is less than 60 Hz. This droop characteristic is needed to prevent the
local, independent governors from fighting each other.
For an isolated utility, the AGC readjusts the set points of the local governors
to l'lring frequency back to the desired level. Raise and lower pulses may be sent
to the local power plant governors6 ever two to ten seconds.
Choice of a particular power plant's share in any needed total energy output
change is determined by the economic dispatch logic of Section B.2. Thus, in
addition to maintaining frequency, the AGC also tries to keep the generation
levels as close to the optimum economic dispatch as possible.


Interconnected Operation


Life gets more interesting when several independent utilities are electrically
interconnected. Because of short-term economy transactions and longer-term
contracts (see Section B.5), each utility specifies its net scheduled interchange;
which is the total amount of power that is to flow out (in) along the tie lines
connecting the utility to its neighbors.
Consider two Utilities, A and B. The role of Utility A's AGC system is




In normal conditions, to maintain the sum of the power flowing out (in) over
all of Utility A's tie lines close to Utility A's scheduled net interchange. Thus
if Utility A's generation is greater than its total load plus losses plus its
scheduled net interchage, Utility A's AGC reduces Utility A's total generation.
To maintan frequency close to the desired 60 (or 50) Hz.
Under emergency conditions when Utility B has lost a major power plant(s)
due to local relay actions, to increase Utility A's generation to provide



emergency support by increasing power flow into Utility B. This energy is

paid back by Utility B later on.
Utility B's AGC system works the same way.
This seemingly difficult control task is accomplished by having each utility
compute its own area control error (ACE) given by

B(f(t) -/oj

Pn(t) -


Pn (t)


Sum of all tie-line power flows (m<:asured and communicated to the

central control system in real time)
Net scheduled interchange
Locally measured frequency


Desired frequency

Frequency bias setting.

Each utility raises or lowers its overall generation proportional to the time
integral of its own ACE(t). We will not go through the analysis here, but the
overall result is the desired behavior.
The beauty of this control logic is that each interconnected utility's AGC
system uses only measurements made on its own system. The only overall
interconnected system coordination needed is to make sure each utility'S net
scheduled interchange is correct - Le., that they all sum to zero.
The AGC systems make no attempt to control individual tie-line flows (unless
there is only one tie line). An AGC system controls only the sum of the tie-line
Control of Time

The AGC systems keep the overall system frequency close to nominal but time,
as measured by the integral of frequency, can drift. Thus one utility is assigned
the task of comparing the integral of frequency to a time standard and sending
time correction signals to the other utilities say once or twice a day. In normal
operation, USA utilities try to keep the difference between the integral of
frequency and true time to within 3 seconds (usually it is much closer but in rare
cases it can be much worse). In general, electric clocks tend to run a little slow
during the day and a little fast at night. 7

Many electric utilities are operated as part of an interconnected grid to


Allow purchases and sales which are beneficial to all.

Provide mutual support during emergency conditions.

Appendix B 293

There are various degrees of interconnected cooperation. Two extreme cases are
discussed; independent operation and power pools.
Independent Operation

Consider a bevy of independent but interconnected utilities. A wide variety of

economic transactions can occur between them.

If Utility A's marginal operating cost AA ($/kWh) is greater than Utility B's AB
for the next hour, Utility A may purchase energy from Utility B instead of
generating the energy itself. The price is often based on a split-the-difference
rule; i.e., the sale price is (AA + AB)/2. Such economy transactions take place
each hour and are made by telephone calls between the system operators. Utility
A may be buying from Utility B while simultaneously selling to Utility C.


A wide variety of longer-term purchase and sale contracts are negotiated

between utilities. Examples are firm contracts for a fixed amount of energy for
the next day; contracts for the right to buy energy for the next day; and contracts
for the percentage of the output of a given power plant for the next year.



IfUtility A wants to buy from Utility B but the energy has to flow at least partly
through Utility C, then Utility C may charge a wheeling rate. In today's system,
wheeling rates often bear very little relationship to the actual marginal costs of
wheeling as discussed in Sections 9.5.





Power Pool

A simple power pool uses a single central control system that determines how
energy is to be dispatched from all the utility members' generators to minimize
the total operating cost of all the utilities in the pool. This enables centralized
economic dispatch, unit commitment, and maintenance scheduling. It also
allows for central control of operating reserves and system security.
Power pool operation requires the use of some mechanism to balance the
books, i.e. to transfer funds between the pool members so they payor are paid
for energy obtained from or sent to other pool members. One approach involves
variations on the split-the-difference formulas used for independent operation.
A more sophisticated approach uses the concept of an "own load dispatch."
With this approach, the power pool central office determines, say each week,
how much each utility ought to receive or pay for the energy transactions
performed during the week by evaluating

The cost of running Utility A in a way that it would meet its own load without
purchases from or sales to the pool.
The actual costs Utility A has incurred.

294 Appendices

These numbers determine the amount of money that Utility A receives or pays.
Power pool operation does not stop members from having separate long-term
contract arrangements among themselves. For example, Utility A may agree to
sell Utility B the output from a given plant for a period of one year. Then the
two utilities simply inform the power pool office of the arrangement, so that the
capacity of A is decreased and the capacity of B is increased by the same
amount. System operation is not affected.s
Power pools often have free-flowing tie lines. In this case, the utility members
do not have individual AGC systems and there is one AGC system for the entire
Power pools engage in purchases and sales with other power pools or independent utilities just as if the power pool was a single utility. Utilities within
the pool may also make purchases and sales agreements with utilities outside the

The material discussed in this Appendix (like that of Appendix A) can be found
in many places. One of many good books is Wood and Wollenberg [1985]. The
IEEE Transactions on Power Apparatus and Systems is a good source for
detailed papers.9
I. Actually other exogenous variables such as an industrial strike, a World Series baseball game, etc.
can also effect total demand, but we restrict discussion to weather and time effects.
2. ARMA means auto regressive, moving average.
3. Economic dispatch programs provide values of system lambda A. However, some care is needed
in translating the economic dispatch A into the Aof this book as the economic dispath Amay not
include purchase and sales, effects of running gas turbines, and unit commitment effects, all of
which are included in the A of this book.
4. Also called load frequency control (LFC).
5. Governor action may not exist on large base loaded units.
6. In practice, only certain power plants are usually under AGC. Base load units may not see AGC
signals. Neither do most gas turbine peaking plants.
7. This makes the working hours for most people a little longer than they should be; unless they use
their own watches.
8. Such long-term contracts between pool members (and also between independent utilities) provide
motivation for the marketplace long-term fixed price-fixed quantity contracts discussed in the
main text of this book. They serve a useful long-term purpose without affecting the efficiency of
system operation.
9. As of 1986. IEEE PAS is replaced by these three publications: IEEE Transactions on Power
Delivery, Energy Conversion. and Power Systems.


This appendix, like Appendices A and B, provides power system background

for those who want it. In this appendix, long-term planning issues associated
with building new power plants, transmission lines, etc. are discussed.
Section C. I discusses the overall planning problem in general terms. Sections
C.2 to C. 7 then discuss some of the major tools and techniques used in the utility
planning process.


The title of this section can be viewed as a succinct .summary of what power
system planning is all about. Independent of whether the utility is trying to plan
for future generation, network changes, or load management, there are always
two dominant issues:

There is no unique scalar criterion to define what is the optimum (best) thing
to do - i.e., multiple attributes.
There is massive uncertainty about what the future will hold.

Often, there also exists a third important issue:


There are multiple decision makers - e.g., utility, regulator, environmental,

and customer personnel may all be involved.




Multiple Attributes

An example of the multiple attribute (criteria) problem is the desire to simultaneously minimize all of the following:

Present worth of all future costs (capital and operating)

Rate shock
Cash flow deficiencies
Probability of not being able to meet demand
Environmental impacts; air, water, and land
Social impacts of new construction

In many cases, none of the above criteria are consistent in the sense that a plan
which minimizes one does not minimize any of the others.
Some approaches to this multiple attribute decision making problem are now
discussed, considering both single and mUltiple decision makers.
Constrained Optimization

With this approach, one criterion (attribute) is chosen to be minimized subject

to constraints on the allowable values for the rest. For example, one could find
the future generation, network or load management plan that minimized
present worth of all future costs subject to constraints on rates, cash flows,
reliability, and environmental and social impacts. The weakness of this
approach is, of course, the need to specify the constraints on allowable values
of the rest of the attributes. It is difficult when there is only one decision maker
and usually "impossible" if there are multiple decision makers.
Use a Utility (Preference) Function

The idea of simply minimizing a weighted, linear sum of the individual attributes
is usually not satisfactory. However, for the case of a single decision maker,
there are procedures to develop a nonlinear utility or preference function (the
word utility here is not electrical in character) by talking with the individual to
learn and then model his/her preferences. The resulting utility function is then
the desired scalar cost function which can be minimized. This approach is not
directly applicable to the multiple decision maker case because utility functions
of different individuals are usually different. For example, it would be surprising
if a utility chief executive officer, the head of the regulatory commission and a
customer completely agreed on anyone utility function. Various fancy multiple
decision making methodologies based on questioning, voting, etc., have been
published, but their applicability to most power system planning problems is not
Sensitivity Analysis

The first two approaches used minimization of a scalar cost function with
possible constraints. Given the resulting, optimum plan, sensitivity analysis can

Appendix C




0 0




loss of load
= Inferior Plan

Figure CI.I. Attributes for ten different plans.

be ""done by perturbing the optimum plan in various directions to see how the
various attributes change. If a particular perturbation yields a small change in
the scalar cost function with a major improvement in one or several of the other
attributes, the perturbed plan might be decreed to be better than the optimum
plan. Alternatively, sensitivity analysis can be done on the numerical values of
the constraints to see how the optimum plan changes. Even the definition of the
scalar criteria can be perturbed.
Tradeoff Analysis

Tradeoff analysis is an extension of sensitivity analysis into a more formal

setting. The basic idea is to work with all of the attributes as originally defined.
Instead of trying to find an optimum plan, the procedure is to find plans that
are inferior and then throw them out. A plan is inferior if there exists some other
plan, all of whose attributes are better. As an example, consider a two-attribute
problem with present worth of costs and loss of load probability as the two
attributes. Assume there are ten different plans being considered. Figure C.l.l
shows what might happen if the attributes resulting from the ten plans are
plotted in the cost-loss-of-load plane. It is clear that Plans 4,6, 7,8 and 10 are
inferior and should be discounted. This leaves five plans to be considered. A
curve drawn through the attributes for Plans 2, 1, 9, 3, 5 is called a tradeoff
curve. Everyone would agree (independent of their personal preference for costs
versus reliability) that only plans on the tradeoff curve should be considered. It

298 Appendices

can be argued further than Plans 2 and 5 should be discarded because, for
example, Plan 2 has much higher costs with only a small improvement in
reliability. A curve drawn through Plans I, 9, and 3 is called the knee of the
tradeoff curve. The tradeoff curve approach does not yield a single answer; e.g.,
the choice between Plans I, 9, and 3 still has to be made. With this tradeoff
analysis approach, the final decision (e.g., the choice between Plans 1,9 and 3)
is left to the final decision maker or makers. After all, they are being paid to
make such decisions. Tradeoff analysis makes it possible for the utimate
decision maker(s) to concentrate on the real issues.
Uncertainty and Risk

Utility planning tries to look many years (10 to 30) into the future. Unfortunately, it is very difficult to accurately predict future events over such time scales.
Examples of major sources of uncertainty include

Load growth
Cost of capital
Capital cost of new equipment
Development of new technologies
Cost, availability of fuel
Treatment by regulatory commission
Environmental standards
Salaries of university professors committed to power system research

Utility system planners can make forecasts of the future behavior of all of the
above, but the general rule is
The Forecast Is Always Wrong!

Hence, the uncertainty associated with the forecasts has to be factored into the
decision making process.
Uncertainty implies risk and the existence of risk introduces new attributes.
For example, assume a constrained optimization approach is chosen and the
present worth of future costs (with constraints) is to be minimized. One plan
may minimize the expected value of the cost but may have very high costs for
certain possible futures. A decision maker who is risk adverse could choose a
plan with a small spread in costs even though its expected value is larger.
Some general approaches for dealing with uncertainty are now discussed.
Use Probability Measures

It is conceptually possible to assign probabilities or probability distributions to

all of the uncertainties (load growth, etc) and then propagate these probabilities
throughout the analysis to yield the probability distributions of the attributes.

Appendix C 299

Unfortunately, the specification of the input probabilities is often a very difficult task.
Use Bounds

An alternative to the use of probabilities is to simply use upper and lower

bounds on the uncertainties (e.g., load growth will be somewhere between - 1%
and + 4% per year). This is easier to handle but is not as theoretically elegant
as the use of probabilities.
Use Utility Function

The development of a utility function (as discussed above) by talking with a

particular decision maker can be done in a way which factors in risks and their
effects. Thus a different utility function would be developed for a risk averse
decision maker than for one who was willing to gamble. Such utility function
methodologies usually assume the existence of probability measures on all the
Sensitivity Analysis

Sensitivity analysis can be used to evaluate the effect of perturbations on input

parameters (such as load growth) as well as plans. The results can be used to
guide the final choice of a plan, i.e., to choose plans whose attributes are not
overly sensitive to input uncertainties. This is the most common approach used
Tradeoff Analysis

As when dealing with multiple attributes, tradeoff analysis can be viewed as the
extension of sensitivity analysis for uncertainties into a more formal framework.
One basic idea is to discard plans which are inferior to another plan for all values
of the uncertainty (if only bounds are used) or with some high probability (if
probability measures are used). The concept of the tradeoff curve and the knee
of a tradeoff curve can still be used, although the simple picture of Figure C.l.l
is lost.



Three different characteristics of future loads are often forecasted on a year-byyear basis:



Total energy (for the year)

Peak demand (over the year)
Load shape

The load shape is represented by either chronological hour-by-hour data (often

just for representative days) or a load duration curve which is equivalent to the
probability distribution of the load levels over the year. Multiple subperiod load
duration curves may be used.
Three approaches to long range load forecasting/modeling are discussed.

300 Appendices

Exponential Models

A very simple modeling procedure is to assume a constant percentage growth

(% per year) and to evaluate the rate of growth by fitting a straight line to

historical data plotted on log paper. Different growth rates are usually obtained
for total energy and peak demand. The load shape is either assumed to be
constant or adjusted to match both the energy and peak demand forecasts.
Econometric Models

The exponential growth approach ignores the many exogenous variables that
affect load growth such as the state of the economy, demographic trends, etc.
These factors can be introduced by assuming, for example,
d(t): peak demand during year t l (MW)
d(t) = kiai/; [ei(t)]
ei(t): Exogenous variables such as state of economy, etc. in year t
/; [ej(t)]: Pre specified function of dt)
a j : Coefficients which are estimated from historical data

In practice, a more complex structure than illustrated by (C.2.1) is often chosen,

but the basic ideas remains the same.
A disadvantage of an econometric model such as (C.2.I) is that it is now
necessary to develop forecasts for the future values of the exogenous variables
e,(t). Often these forecasts are obtained from nonutility sources (which often
obtain their forecasts from other econometric models for broader sections of the
The big advantage of an econometric model such as (C.2.I) is that it is
possible to do "what if' studies (i.e., sensitivity or tradeoff analysis), to see what
effect, for example, a change in the economic health of a region will have on load
End Use Models

Section 8.1 of Appendix B used an end use modeling framework based on the
demands of individual appliances of individual customers to discuss diversity.
Such an end use framework can be extended into the long-range forecasting
arena to yield a model for the residential sector that depends on

Customer appliance holdings

Appliance usage patterns

so that hour-by-hour load shapes can be developed for any future year. Econometric models can then by used to see how electric rates, gas rates, customer
income, etc., effect the customers' (aggregated) appliance holdings. Related end
use models for industrial and commercial customers can also be developed
(although they usually have a different structure).

Appendix C


The big disadvantage of these end use models is that they are hard to develop.
Their big advantage is that they make it possible to ask much more detailed
"what if' questions. For example, the effect of introducing more efficient
refrigerators can be studied.
Comparison of Load Forecasting Models

Exponential modeling is easy. Econometric modeling is nontrivial. End use

modeling is a very major undertaking. This, however, does not imply that end
use models yield the most accurate forecasts. Econometric and end use models
require inputs whose forecasting errors can yield a load forecast with even larger
errors than an exponential model. The principle that the forecast is always
wrong holds for all three approaches.
The advantage of the econometric and end use models is their ability to
address "what if' questions; i.e., they enable the sensitivity studies, tradeoff
analysis, etc., that are basic to the long-range planning decision making process
as discussed in Section C.l.

A production cost model is the heart of all planning studies that involve
generation and/or loads. An annual production cost model looks at a given
future year, with a given load shape, and mix of generator plants, and then

E~aluates the fuel consumption and costs needed to meet the load under a
cost minimizing dispatch strategy.
Sometimes evaluates some measure of the generators' ability to meet the load
at all times; e.g. loss of load probability or expected unserved energy.

One key function that distinguishes different types of production cost models
is the approach used to model the effect of generation outages (forced and
maintenance). Another distinguishing feature is whether the load is modeled
chronologically (e.g., hour by hour) or by a load duration curve (number of
hours per year, for example, that the load exceeds a given.levelV
Five types of production models are summarized in Table C.3.1. There are
many variations on these basic ideas and only the main points of each are
discussed. 3
Chronological Simulation

A basic approach starts with hourly demand data for 8760 hours (i.e. a year).
Maintenance scheduling is then simulated to determine which plants are available on a week-by-week basis. Then for each day (or week), a unit commitment
logic is simulated to see what the fuel costs are. Random forced outages of the
generators are handled by derating. For example, a 1000 MW plant with a
probability of 0.1 of being forced out is treated as a perfectly reliable 1000
(0.9) = 900 MW plant.

Table C.3.1. Five Types of Production Cost Models

Types of
Cost Model
Monte Carlo
Merit Order
Proba bilistic
DAM (Discrete
Approximation Method)

Method of

Type of Load




Load Duration Curve


Load Duration Curve

(typical days)

This approach does not yield useful information on the probability of being
able to meet the load at all times.
Variations on this basic approach work with representative days instead of all
365. For example, a typical weekday, a typical weekend, a peak day, a peak
weekend, for four seasons might be chosen to yield 4 x 4 = 16 representative
Monte Carlo Simulation

This approach is a chronological simulation where random numbers are used

instead of derating to model generation forced outages. For example at the start
of a particular day's simulation, a 1000 MW plant with 10% forced outage rate
is viewed as a 1000 MW plant 90% of the time and 0 MW plant 10% of the time.
Multiple random numbers are used until enough "data" has been obtained to
estimate (statistically) the probability distribution of the fuel costs and the
probability of being able to meet the demand. Representative days may be used
with associated probabilities of occurrence.
Merit Order

A simple version of this approach combines an annual load duration curve with
derating of the generators for both maintenance and forced outages. To illustrate the idea, assume there are three generators whose characteristics are

Generator A
Generator B
Generator C


Outage Rate
and Forced)



Appendix C


Demand (MW)

1275 + 100

= 1375

800 + 475 = 1275



Generator A




Figure C.3.1. Merit order production costing.

Figure C.3.1 shows how these three plants would be dispatched against an
annual load duration curve with a peak demand of 1400MW. The total fuel
costs are computed by calculating the areas under the curve, multiplied by the
operating costs.
In practice, subperiod load duration curves are often used instead of an
annual load duration curve so, for example, the effects of maintenance scheduling can be handled more directly than by simple derating. These subperiods are
sometimes developed as daily load duration curves for representative days (as
in the chronological and Monte Carlo simulations).
This merit order approach does not yield useful information on the probability of being able to meet demand. The area in the small upper shaded triangle
(from 1375 to 1400) of Figure C.3.1 is not a meaningful measure of how much
demand will not be served.
Probabilistic Simulation

This approach is a generalization of the merit order approach which works with
load duration curves but explicitly handles the random nature of forced outages.
The basic idea relative to Figure C.3.1 is to start with Generator A and to
dispatch it first, assuming it has a 90% probability of being available. Probability convolution is then used to compute a new equivalent load duration curve
(which accounts for the 10% probability of Generator A's outage) that is seen
by Generators Band C. The convolution process is then repeated for Generator





Oemand (MW)





Demand (MW)





. ..







Figure C.3.2. Graphic illustration of convolution. (a) Loading of a 500 MW plant with 90%
availability. Expected energy = capacity x availability x hours = 75600 MWHjweek; (b) If
the first plant fails, the second plant sees the original customer curve. This event has a
probability of 0.10; (c) If the first plant operates the second plant does not see the first 500 MW
of customer demand. This event has a probability of 0.90; (d) The equivalent load curve for the
second plant is the sum of the two curves weighted by their respective probabilities.

B, etc. This approach yields useful infonnation on both fuel costs and the loss
of load probability and/or the expected unserved energy. Figure C.3.2 elaborates the process of using equivalent load duration curves to model forced outages
of previously loaded units.
There are many variations on the method (other than going to subperiods).
Frequency and duration models can be used to incorporate the length of time
a forced outage can be expected to last. Numerical techniques (such as the
method of cumulants) can be used to reduce the computer time requirements
associated with multiple convolutions.

Appendix C


DAM (Discrete Approximation Method)

This approach returns to the use of chronological models of representative days.

It handles generator forced outages by approximating the many different
possible outage states by the probabilities of a much smaller number of effective
outage states. For example, a system with ten generators has 2 10 possible
combinations of available generators. By "careful" choice, the system might be
represented by five to ten equivalent outage states, each with an associated
probability. The actual number needed depends on the characteristics of the
individual generators. 4
The total fuel costs and measures of failure to meet demands are computed
by doing a chronological simulation for each representative day with each
effective outage state and then mUltiplying by the appropriate probabilities.
Discussion and Comparison

Some production cost models allow the incorporation of the effects of nondispatchable generation (e.g. wind and solar), load management, time-of-use rates,
etc. This can be done with varying degrees of success with all of the five
approaches, but the details of how are beyond the scope of this discussion.
An obvious question is: Which approach is best? Unfortunately, the answer
is: It depends on the nature of the system, the issues of concern, and the amount
of computer time available. For example, when doing a generation expansion
study for 30 years (as to be discussed in Section C.S), computer time may be a
dominant factor and accuracy of less concern because of the many other
uncertainties influencing long-term decision making. Alternately, when comparing, say, a particular load management scheme with adding a particular
combined cycle gas turbine, the accuracy of the production cost model may
decide the answer. A Monte Carlo simulation has the potential to yield the most
accurate answers if enough computer time is available.
It is an interesting and frustrating experience to try to compare the results of
two different production cost programs run for the same utility.

A financial model computes a utility's cash flows, i.e., input revenue and output
costs for fuel, maintenance, personnel, loans, taxes, etc. It also worries about the
stock holders (if a privately owned utility) and indices like the debt-to-equity
For a given future expansion plan, the financial model is run each year to see
how financially healthy the utility would be. For example, building a large
nuclear power plant might cause a utility major cash flow problems during
construction and result in a large rate shock (Le. jump in rates) when the plant
comes on-line and enters the rate base. Furthermore, the amount of utility debt
can influence the interest rate that has to be paid on loans (i.e. the cost of capital)
which, in turn, influences the choice of future construction decisions.
Financial models are, conceptually, just simple bookkeeping. However, they

306 Appendices

can get very complex in practice. Because of the different ways taxes, depreciation, etc. are handled, a financial model developed for one utility may not be
applicable to a different utility.

Consider the basic multiple attribute decision making under uncertainty

problem described in Section C.l. In today's world, a utility worrying about its
future generation needs wants to consider plans involving options such as

Build new plants (various types and sizes)

Extend life of existing plant
Encourage or discourage customer generation
Load management, conservation
Buy from or sell to neighboring utility

For each plan of interest (a particular combination of options), the utility uses
production cost, financial, environmental, etc., models to evaluate the diverse
Unfortunately, the number of possible future plans can get extremely large.
For example, there are many possible types of new plants; they come in different
sizes, and they can be built in different combinations in different years. Combining all these options with the many possible load management options, etc.,
leads to an almost uncountable number of possible plans.
Generation expansion programs are tools which allow the computer to search
over at least some of the many possible futures to reduce the number of plans
that have to be considered in detail. A typical generation expansion program is
designed to choose automatically between plans involving new plant types, sizes,
and timing of construction.
Several difficult approaches are discussed.
Target Mix, Heuristic Search

One standard approach has the user specify a desired Target Mix of generation
plants at some future time, say 20 years from now. A possible Target Mix might
be 20% nuclear, 40% base load coal, 30% combined cycle, and 10% peaking
plant. The computer program then starts with the existing generation mix and
builds new plants each year (taking into account construction time, etc.) as
needed to meet demand with some specified measure of reliability. Heuristic
logics are used to move the generation mix towards the Target Mix. A production cost model is used each year to evaluate the fuel costs and reliability levels.
A financial model might be built into the code or run separately. At the end of
the study period, the various attributes of the reSUlting plan are summarized, i.e.
present worth of all capital and fuel, tons of coal burned, etc. Relative to the
discussion of Section C.I, the user-specified Target Mix can be one of the
options to be considered. The program would be rerun many times for various

Appendix C


Target Mixes and of course for different values of the input assumptions on
costs, load growth, etc.
Optimization Logic

Many generation expansion programs make use of optimization logics which

yield the particular sequence of new plant types, sizes and timings that minimize
some scalar criterion subject to constraints. The scalar criterion is often the
present worth of capital and fuel costs with reliability as a constraint, but there
are many variations. For example, the cost of unserved energy can be added to
the cost function, the numbers of particular types of plants can be constrained,
etc. Popular types of optimization logics are dynamic programming and linear
programming. Decomposition techniques of various types can also be used.
There is no best optimization logic; the choice depends on the specific nature
of the utility and the issues of concern. Relative to the concepts of Section C.l,
the constraints can be viewed as options. The program is rerun many times for
different values of the constraints and, of course, assumptions on the input
costs, load growth etc.
Optimization logics that minimize costs over the whole planning horizon are
especially useful when intertemporal changes in the relative economics of candidate investment options are important.

EGE,(S (Electric Generation Expansion Analysis System) is an example of a

powerful generation expansion program 5 developed for and distributed by
EPRI (Electric Power Research Institute). It can handle a variety of plant types,
including pumped storage and renewable generation (solar, wind). Production
costing is done by probabilistic simulation (two versions are available as
options). The user can choose between different optimization logics: dynamic
programming, linear programming, and Benders decomposition. The program
can also be used to build along a prespecified pathway and evaluate the resulting
user specified expansion plan. Automatic sensitivity analysis logics are provided
to facilitate making the many runs needed to address the fundamental issues
discussed in Section C.l.
Screening Curves

Thus far we have discussed sophisticated generation expansion programs which

use a lot of fancy mathematics. However, we would be remiss if we failed to
discuss a very simple, approximate technique which can sometimes yield results
using the "back of an envelope." This technique is often called a screening
The basic idea is to look at some future year and to determine what would
be the optimal generation mix for that year, ignoring the present mix of plant.



j: Index of possible plants (including type and capacity).

Cj: Annualized capital cost of J th plant ($jyear).
A/ Fuel cost of jth plant ($/kWh).
10: Maximum capacity of jth plant (kW).
hj: Number of hours that thejth plant runs.
TCj : Total cost for year of Jth plant ($).


Dividing (C.S.I) by Kj yields the key equation


To illustrate the use of (C.S.2), assume there are three possible plant types and
it is desired to determine how much of each to build.


Gas Turbine






Figure plots TC)Ky of (C.5.2) versus h to give three straight lines. The
intersection of these lines are then projected down onto the load duration curve
of Figure C.S.l b and the desired capacity of each plant type is determined. The
total fuel costs are computed from the areas under the curve as discussed in the
merit order approach to production costing. In practice, derating to handle
outages is used to modify the Ky.
The number of approximations inherent in this screening curve logic is large.
However, the idea can save a lot of effort if used before going into a big
computer program. Sometimes in this world, one hour of human thinking is
worth more than many hours of main frame computer number crunching. For
example, if the nuclear-coal intersection in Figure occurred for
h > 8760, the nuclear plant should not even be considered.

Attempts have been made to develop network expansion computer programs

that are similar in operation to the generation expansion programs; i.e., choose
the particular network expansion plan that minimizes present worth of all costs
subject to constraints. However, the nature of the network expansion problem

Appendix C




8760 hours. h

Figure C.S.1. Screening curve methodology.

is such that the usual procedure involves using engineering judgments to

hypothesize potential new line additions and then doing multiple load flows,
stability studies, etc., to determine the attributes (cost, environmental impact,
etc.) of a given plan.

We have discussed some of the basic models and procedures used in power
system planning. In practice, there are feedback coupling which have to be
considered, as illustrated by Figure C.7.1. For example, load growth drives
generation plans, which drive network plans, both of which drive financial



Load Management, Conservation

Load Model





Financial Model

Rate Model


Figure C.7,1, Feedback coupling in power system planning.

issues, which determine rates, which effect load growth. In practice, the complexity of the overall problem places a premium on breaking it up into pieces
so load experts can work on load modeling, financial experts can work on
financial modeling, etc. The trick is then to put it all back together to get a
reasonable plan for the future.
Combining the feedback coupling of Figure C. 7.1 with the multiple attributes
and massive uncertainties discussed in Section C.I shows why
Power System Planning is an Art, not a Science!


As with Appendices A and B, there exists a huge literature on power system

planning. IEEE Transactions on Power Apparatus and Systems is a good
starting place for detailed papers. A sampling of more specific references is as
follows. Many of the ideas in this section (and in several other appendices) are

Appendix C


discussed in more detail in an MIT Summer course, which has a more complete
Merrill and Schweppe [l984J discuss initial developments in the tradeoff
approach to multiple attribute decision making under uncertainty. Caramanis
[1983] discusses generation expansion logics with emphasis on EGEAS. Munasinghe [1979] is a good book on generation expansion logics. Hyman [1985] is
a good starting point for financial issues. These references just scratch the
surface of the economic literature on power system planning. We assume that
most readers of this book will already be somewhat familiar with that literature.
A good source is Crew and Kleindorfer [1979] but many others exist.
I. A change in notation, since usually in this book I is an hourly index.
2. A turned-around probability distribution.
3. The names associated with these different model types can be confusing. We try to follow the most
common usage although there are variations in the literature.
4. The difficulty of specifying these equivalent outage states makes this approach the least popular
of all five.
5. We are biased towards EGEAS since the first three authors of this book were instrumental in its
development [Caramanis, Tabors and Schweppe 1982].


This appendix presents the DC load flow model, which provides approximate
but simple relationships between generation and demand levels at the buses and
real power flows through the lines. These relationships yield explicit formulas
giving the marginal impact on network losses or specific line flows from incremental changes in demand or generation at some bus of the network.
The DC load flow provides an approximate solution for a network carrying
AC (alternating current) power. The term "DC" comes from an old method of
computing a solution using an "analog computer" built out of resistors and
batteries where direct currents were measured.

Real Power Flowing Through a Line

From (A.I.8) of Appendix A, the exact equation for


Real power flowing from Bus I toward Bus 2 along line i

is given by


R; + X/

314 Appendices


fli =


+ x/

R i : Line resistance

X;: Line reactance

The equation (0.1.1) can be greatly simplified by making a series of assumptions which are often valid. Assume (>I - >2) is small in magnitude so
cos (<5 1
sin (<5 1

<5 2 )

<5 2 )

(<5 1

<5 2 )

and assume that in a per unit system (as discussed in Appendix A) VI

I/; ~ I. Then (0.1.1) reduces to

I and

Note that using (D.I.3)

i.e., the approximation of (0.1.3) yields a lossless line since the power flowing
into one end ZI2 equals the power flowing out the other end - Z21.
Losses on a Line

Define L; to be the real power losses on line i. By definition

L, =




(D. J .3) cannot be used since it is a lossless line model. Therefore, we return to
(D.I.l), and substitute into (0.1.4) to yield

Assume again that 5 1 - 5 2 is small in magnitude but instead of using (0.1.2),

the second order term is also included. Hence

This yields, assuming V;

I, I/;


Now make a further approximation by substituting the

(0.1.6) to yield


of (0.1.3) into

Appendix D





(D. I. 7)

where we have changed notation so





Finally, assuming


(D. 1.8)

Line Flows as a Function of Network Characteristics and Bus Injections


Nb: Number of buses.

NL : Number of lines.
y: Nb - I Vector of bus injections (generation minus demand) at all buses
- except the swing bus.
y*: Injection at the swing bus (J = *); assumed to be generation only for
0, R: NL x NL Diagonal matrices of the OJ and resistances R j respectively.
Ii: NL x (Nb - I) Reduced network incidence matrix with 0, I, - I elements
- corresponding to network interconnections. Section 0.3 illustrates this
matrix for a three-bus example.
z: NL Vector of line flows.
~: Nb - I vector of voltage angles at each bus, except at swing bus where <5 = O.
L: Total line losses: L = LjLj.
A <5: Vector of angle differences across lines.
Column vector of all ones.
~: Column vector of all zeros except for I in the ith line.


Since the sum of all power entering a bus is zero,


The matrix form of (0.1.3) is

z = nAb


Combining (0.1.9) and (0.1.10) yields


316 Appendices

Solving for ~ yields

~ =

(;iT!;i) -I E.

Substituting into (0.1.10) yields



!;i(t!:D- 1

H is called the transfer admittance matrix.

Network Losses

Using (D.I.8), the total line losses L become


Substituting ~ from (D.1.12) yields



(;iT !;i) - 1;iT! B. !;i(;iT! ;i)-I


Most of the approximations introduced so far are quite good for transmission
lines. We now introduce another approximation which is not as good but which
yields equations that are much simpler. This new assumption is
A constant X to R ratio (reactance to resistance ratio) on all lines; i.e., R; =


all i.

Substituting into (0.1.1) yields

Assuming ~ ~ R;, so
fl, =


~ I, the constant X to R ratio assumption yields


or in matrix form,

Substitution into (0.1.14) yields


and (0.1.12) becomes

Appendix D



Discussion on the Approximations and Assumptions

The assumptions made through (0.1.14) are often used in actual utility operations and planning for transmission lines. The additional assumption postulating a constant resistance-reactance ratio for all transmission lines may lead to
larger approximation errors than are tolerable. However, the results thus
obtained are useful in increasing one's intuitive understanding of spatial
network pricing and can be probably used in long-term planning studies where
general behavior modeling is important but high accuracy is not required.
All of the OC load flow assumptions can become more inaccurate for lower
voltage, sub-transmission and distribution lines and as the line loading increases. Thus when dealing with certain actual operational conditions, it may be
necessary to use more accurate approximations or the full AC load flow
The choice of the swing bus may have a nonzero (although small) impact on
the numerical results. This is due to the approximations resulting in (0.1.3),
which effectively assume a uniform power flow (and hence no losses) at every
point of the line joining Buses I and 2. Depending on where the swing bus is
placed, the flows on anyone line mayor may not include losses on other lines
thus resulting in slightly different numerical results. Although these differences
are likely to be negligible in most real applications, the reader should be aware
of them and not be surprised when they arise in simple textbook type examples.

The OC load approximation of Section 0.1 yields many useful results of various
types which are exploited in the main text. Some of these results are derived
Effect of Swing Bus

Consider (0.1.12) and (0.1.14). Since J:. does not contain the swing bus, i.e. does
not contain y*, it follows that



Difference of Spot Prices at the Ends of a Line


318 Appendices

p: Nb - I vector of spot prices at all buses except the swing bus

e*: Spot price at swing bus

Disregarding NQS and GQs terms, the Chapter 7 spot price equations yield


aL) + -aaz
(-e - -a
2:: -





where !:!:.QS,,/ is the column vector of Lagrange multipliers. Using (0.1.12) and

Using the additional approximation of a constant resistance-reactance ratio

that gives (D. I .15) and substituting,!:: using (0.1.11) yields

Define further

IIp: Vector of price differences across lines

lle = :ie.

Since A e = Q, (0.2.5) becomes


Using (0.1 II) and (0.1.12) yields:i~

g-I~, so (0.2.6) becomes


From (0.2.6) and (0.2.7) it is seen that the price difference across each line,
disregarding !:!:..s.,1' is proportional to the voltage angle difference across each line
or the real power flow through the line.
Impact on Incremental Line Admittance Changes on Overall Network Losses and Flows

The first-order optimality conditions for optimal transmission investment (see

Chapter 10) contains the term fJLjfJK".h where K,/J is a proxy of the capital stock
of line i. Assume K~ is the admittance Q i of line i. Then we want to evaluate


an; - --


Appendix D 319

Use of (0.1.14) yields a complicated equation. However, using the further

approximation of (0.1.15) where!!. = o:(~TQ~)-1 yields

Substituting (0.1.11) yields


where 0, - O2 is the voltage angle difference across line i.

Another important term for transmission investment decisions is
Using (0.1.12),



or using (0.1.11)





<5 2 )



This can be rewritten as


Since C and n; are exogenous and Zj is readily calculable, the iJzjiJn j term is easily
quantifiable for practical studies.
Net Revenues of Transmission and Distribution Network


s: Total revenue received by T and 0 network assuming all users pay the T and 0
network, and all generators are paid by the T and 0 network at the optimal spot price.
Since p is the vector of prices and y is the vector of all injections at all buses
except-the swing bus, it follows that

320 Appendices

Bus 1

Bus 3

Figure D.3.1. Three-bus example.


where elements of yare negative for net consumption of electricity and positive
for net generation at any bus.
Using (D.2.S), (D.2.13) becomes

Using (D. I. 12), the energy balance ~Tl

L = IT!!.l yields

+ y*

= L, and the loss equation


which implies that S > O. When line flows are not limiting, i.e., !!:.Qs.~ = 0, then
the T and D net revenue is proportional to total network losses.

Consider a three-bus utility with generators at all three buses, demand at two
buses only and interconnected with a single tie line through which it inputs
WMWh as in Figure D.3.!. Assume demands, dl and d2 , are completely
inelastic, i.e. insensitive to the price of electricity. Assume that the marginal cost

Appendix D 321

of fuel and maintenance at each generator j, j = 1, 2, 3, is monotonically

increasing with generation level gj and is described by the expression

where a{a~, with $jMWh units and aj with MWh units are specified parameters.
If Pj(t) is the optimal spot price at time t seen by generator j, then optimal spot
price theory and economic dispatch imply that the generation level g; will be
selected to satisfy

subject to 0 ::;; gj ::;; gl.max' Assuming for simplicity that gj.max is not binding, we
have for j = I, 2, 3
aHin [Pi - aj] - In an

if PI > aj




Assume for simplicity that NMl = Nos., = O. Following Chapter 7 we have the
optimal spot price relationships for j = ), 2, 3:


= y(l) + y(l) ~(I) +



L J!Qs.~.,(I) iJg, (I)



In addition to the six equations specified by (0.3.)) and (0.3.2), we have the
energy balance equation accounting for imports W, net injections gj - dj and
losses L,

L [g;

- djl

+ W


i= 1

and the line overload constraint for i

I, 2, 3,

The above ten relationships must be solved simultaneously to yield values for
Pi' gj' 1', !lOS.I,.; at time t for j = 1,2,3 and i = 1,2,3. To do that, however, the
quantities Land Z; must be expressed in terms of gj' ~, tie line power flows W,
and the characteristics of the transmission network. Here is where the DC load
flow relationships are used. The constant network characteristics are included
in the reduced network incidence matrix A and the individual line resistances
and reactances.
To construct i we select arbitrarily positive directions of transmission line

322 Appendices

power flows as indicated in Figure 0.3.1. Thus, for example, if Z3 equals

530 MWh it means that 530 MWh of energy flows over Line 3 during the hour
t from Bus 3 to Bus I, whereas if Z3 equals - 530 MWh it means that 530 MWh
flows over Line 3 from Bus I to Bus 3.
Let A denote the network incidence matrix before reduction. Then A is
constructed as a matrix with as many rows as lines and as many columns as
buses, with each row consisting of zero entries except for a (+ I) entry at the
column that corresponds to the bus where the positive power flow originates and
a ( - I) entry at the column that corresponds to the bus where the positive power
flow terminates. For the three-bus network with Lines i = I, 2, 3 and Buses
j = I, 2, 3, we have line bus entries where positive flows originate A II
An = A33 = I and entries where positive flows terminate A 12 = A22 = A31 =
- I, which yields

To model the swing bus (or equivalently to remove the singularity which is
introduced because the overall energy balance relationship is included in the bus
specific power flow conservation relationships), we form the reduced network
incidence matrix A by dropping from A the column that corresponds to the
swing bus. Selecting Bus 3 as the swing bus, we obtain the reduced network
incidence matrix

1 -I]



Let Ri denote the resistance of Line i and ri

1/ R i , its inverse; then we have


We now assume a constant resistance to reactance ratio for all lines so the loss
matrix B and the transfer admittance matrix H are given by (0.1.15) and
(D.I.16f After the necessary matrix multiplication and inversion operations,

Appendix D 323



(I' I' + I'

I 2


+ I' 2I') )








The Band H matrices are now completely determined in terms of the network
interconnections, the swing bus location and the line resistances. Therefore the
terms o:::,(t)/og,(t) in (0.3.2) are now quantifiable as the corresponding element
of the H matrix, H;k, and the oL/ogj terms can be obtained by explicitly
considering the partial derivatives of total losses given by yT B Y or ZT R z. Indeed,
noting that l is the vector of power injected at each bus except-for theswing bus,

If we denote by B,j the elements of Ii, the losses are given by

and the derivatives become


= 2B12 (gl


2B I2 (g, - d,)

In order to proceed with the numerical example, assume that the line resistances all equal RI = R2 = R) = 4 X 10-- 5 MWh- l , so 1'1 = /'2 = r) = 25000
MWh, which yields



; !i





- 333

To illustrate the impact of line overloads on spot prices, we first assume that
the line power carrying capability is unlimited so the constraints 0.3.4 are not
binding. This implies tha J1QS.,/.i = 0 for i = 1, 2, 3 and that we have to solve
simultaneously only seven equations (0.3.1), (0.3.2) for j = 1,2,3 and (0.3.3)
to determine generation levels and spot prices at each bus and y(t). Because of
the swing bus selection at Bus 3 (0.3.2), '13(t) = yet).
The reader can see by inspection that the resulting simultaneous equations
can be solved by iterating on y(t) and calculating generation and spot price levels



Table D.3.1. Results in the Absence of Line Flow Constraints

Spot Price

81.54mills/kWh (note 111 '" 1.]4)

88.04mills/kWh (note 112 = 7.64)
S0.40 mills/kWh (note 11) = 0.00)


p) = 1'(1)



Total Losses

99.3 MWh

Line Flows







until the energy balance equation is satisfied within some tolerance. Assume
price inelastic demands of 200 MWh at Bus I and 2200 MWh at Bus 2 and
imports through the tie line at Bus 3 of 200 MW. Consider further that the
exponential generation marginal cost relationships have parameters value.








(most expensive)





(least expensive)

The iterative solution yields the results shown in Table 0.3.1. The reader can
verify the above results by substituting in the appropriate relationships. For

500 [In (81.54 - 50) - In 4) = 1032.5

Table D.3.2. Results with a Flow Constraint on Line 2
Spot Prices


PJ = )'(1)

90.23 mills/kWh (note Iii = 36.93)

130.31 mills/kWh (note '12 = 77.01)
536.30 mills/kWh (note IIJ = 0)

Value of Ilos ..,.2(1)



221.S MWh

Total Losses


Line Flows





Appendix D

L =

10- 5 (2.666(1032.5 - 200)2

+ 2.666( -

2(1.333)(1032.5 - 200)( -2200) =




We now remove the assumption that line flow capacities are not constraining
and impose a maximum flow limitation of 1000 MWh on Line 2 but maintain
the high capacity assumption for Lines I and 3. Since Z2 = 1189 MWh for
fJ.QS,q.2(t) = 0 it follows that we now have to determine the value of fJ.QS.q.2(t)
which results in Z2 = 1000 MWh. This implies an iteration over both y(t) and
fJ.QS,II.2(t) until the energy balance equation (0.3.3) as well as the power flow
constraint 22 = 1000 MWh are met. Carrying out this exercise we obtain radically different spot prices which reflect the nonzero network quality of supply
component. The results are reported in Table 0.3.2.

l. The G, and nJ symbols here are obviously not related to the costs G" etc., of the main text. In
(A.l.8) the real power flow on the line was denoted by Pjk' Here we revert to the notation of the
main text and use z to denote line flows.


Various chapters of the main text have used models for customer benefits and
price response. Some general structural forms for benefit functions and price
response are presented here. A completely different approach, designed to help
individual large customers optimize their response to spot prices, is presented in
Bohn [1982, Chapter 4].

Define the individual customer benefit to be

Bk[dk(t)]: Benefit kth customer receives during hour t from the use of dk(t) kWh of electric
(E. \.l)
energy ($).

Define the aggregate customer benefit function to be

B[d(t)] = Total benefit al\ customers receive during hour t from use of d(/) kWh
of electric energy
d(1) = Lkdk(t)

Since Bk(dk(t is usual\y a nonlinear function in dk(t), the use of (E. 1.2) is an
approximation in that the sum of benefits E[d(t)] depends on the individual dk(t)
k = 1,2, ... , not just d(t).
It is often assumed that customer demand dk(t) depends on the price or rate
the customer pays for electricity. Define

328 Appendices

rk (t): Rate kth customer pays for electricity during hour t ($jkWh).

(E. 1.3)

It is reasonable to assume (albeit not necessarily true) that customers behave in

a rational fashion. One type of rational behavior is characterized by
Customer Optimum Behavior Assumption: Customer k chooses demand level dk(t) to
(E. 1.4)
maximize benefits minus costs; i.e., to maximize BAdk(t)] - dk(th(t).

Given (E.I.4), it follows that dk(t) is determined by

(E. 1.5)

where (E.I.5) assumes that the rate rk(t) is exogenously specified or that the
number of individual customers is sufficiently large that the value of anyone
dk(t) does not affect rk (I). The aggregate model analogy to (E. 1.5) is
GB[d(t) _

r t


The benefit function most often used in the main text (for an aggregate model)
is the
Quadratic Benefit Function:

B[d(t)] =


ro(t)[d(t) - do(t)]

{I + d;th(~do~Ot~t)}

(E. 1.7)

do(t): Nominal demand level (kWh)

Bo(t): Benefit when d(t) = do(t) ($)
ro(t): Nominal rate level ($/kWh)

pet): Elasticity parameter

From (E. I. 7)
GB[d(t) =

r (t)

{I +

d(t) - do(t)}


GB[d(t)] =


when d(t) = do(t)

Combining the customer optimum behavior assumption that leads to (E.l.6)

with (E.1.8) yields the

Appendix E 329

Linear Response Function:

d[r(t)] =


{I +

p(t)[r(t) - ro(t)]}

(E. 1.9)

so that
d[r(t)] =


when ret) = ro(t)

From (E.I.9),


ro(t) ad[r(t)]
do (t) or( t)

(E. I. 10)

which means
pet) =


when ret) = ro(t) and d(t) = do(t)

C:(l): price elasticity of aggregate demand at time I

c: I)

r(l) od[r(l)]
del) or(l)


Other Structural Forms

Obviously there are many other possible structural forms for customer benefits.
Table E.l.l summarizes four such structures: linear response (as discussed
above); power; exponential; and log. It is important to note that
A second-order Taylor Series expansion of the power, log and exponential benefit
functions B[d(t)] about d(t) = do(t) yields the quadratic benefit function,

A first-order Taylor Series expansion of the response function d[r(t)] resulting from the
power, log, and exponential functions yields the linear response functions obtained from
the quadratic benefit function.

Hence if d(t) ~ do(t), all four structures are equivalent. However, in actual
applications, the assumption d(t) ~ do(t) may not be valid, in which case a
choice of explicit structures is required. For the power structure {J(t) = S(l) for
all ret) and d(f). Hence the power structure is often called the "constant elasticity" structure. I

Table E.I.I. Four Benefit and Response Functions







Bo(1) +
ro(1)[d(t) - d o(1)]



d(l) - do(1)}



+ P(t)do(t)

{I +

P(I)I,(I) - '0 (I)} }





{( d(t)

+ P 1(1)

Y-1(n _I}



P(1) [In



( d(l)





[d(t) - do(t) ]

[ Tlln

I I [I

{ fl

exp {p(I)lr(l) - r(t)}}

{I + In.L}



+ P(t)

n do(t)







ex [d(l) - do(t)]






{ 1+ P(/) [r(l) - ro(l)]}



[I + In J




Appendix E


The "best" structure in Table E.l.l is not known. However, it should be noted
that since {3(t) is negative, d(t) can go negative for the Linear Response and Log
Response models when ret) ~ ro(t). Thus, special care in their use is required.
The values assumed by do(t), ro(t), Bo(t), and {3(t) will often vary depending
on the context of the discussions. As discussed before, d(t) is viewed as a random
process. One set of assump(ions used in the main text is:
"0(1): Deterministic function of time

do(I): Random process (non stationary)

fJ(t): Deterministic function of t and possibly do(t) (could be constant)

Bo(t): Deterministic function of t and possibly do(t)

Disaggregated Customers

The benefit function and response function have been discussed for aggregate
customer demand d(t). Definition of structures for individual customer benefit
and response functions follow in a completely analogous function. For example,
for the kth customer, the linear response function of (E.I.9) becomes

which implies that all of the variables that determine the kth customer demand
for electric energy during hour I can vary with the customer index k.

The structures of Section E.I considered only a single period of time; i.e.,
demand at hour t depended only on price at hour t. In practice, demand
rescheduling from hours of high price to hours of low price will occur. Thus a
mUltiple time period model structure is also required.
We now assume that demand at time t depends on pricE:s at times 1, 2, ... t,
... Twhere I to T defines an appropriate cycle. We consider a 24 hour cycle, i.e.,
T = 24. Extensions to different cycle lengths shorter or longer are of course
possible. The multiple period linear and power response models are derived
below. Multiple period exponential and log models can be developed similarly
by the interested reader.
Linear Demand Model

The linear demand model is a linear relationship between demand at time i and
prices at times 1,2, ... 24. It is obtained from the self and cross price elasticities
of demand at the nominal (original) prices and demands. We define:
d(i): Input nominal demand at time i; i = \,2, ... 24

332 Appendices


(I): Input nominal price at time i; i = I, 2, ... 24

d(i): Demand at time i; i = I, 2, ... 24

p(i): Price at time i; i = I, 2, ... 24

Cross price elasticity of demand at the nominal price

and demand levels. E[i, j] is negative when i = j and
positive when i " j. It is provided as an input

The above definition together with the linearity assumption, that is

ad (i)
ap(j) =

constant for

, ,

= 1,2",,24

imply the following linear relationshp between prices and demands

~(i) +





E[i,j] ---0--(1,) [P(j) - p0(j)]



P )


The benefit function consistent with the above linear demand relationship is a
quadratic function which is obtained by integrating the demand relationship,
Indeed, the vector form of the linear demand relationship can be written as
~ =


+ f[p - l]


where d, dO, p, pO are the corresponding 24 x I vectors of demands and prices

and CIS 24
24 matrix with elements

a x

EO[' ,]_1_


Solving the above for f!.. and integrating between ~o and ~ we get:
[Change in Consumer Surplus when prices change from original (nominal)


where prime denotes transpose,2

Power Demand Model

The power model is derived from constant price elasticities of demand (own and
cross) and nominal price and demand levels.

Appendix E


The basic assumption is that the elasticity (rather than the slope as in the
linear model) is constant. That is
. .
ad(i) p(J)
E[I,}] = op(J) d(i) = Constant for i,j, = \,2, ... 24


which implies

where d(i), d(i), p0(i), p(j) are defined as in the linear model.
Equilibrium Demand

Given a simple multiple period demand response model as described above and
a supply price relationship which assigns a spot price to a given generation level,
it is possible to find the equilibrium price and demand levels. This requires,
however, the simultaneous solution of a system of equations since demand levels
at each time period are coupled. The task of obtaining this solution is made
more difficult by the fact that we normally do not have an analytic representation of the supply price relationship. A simple iterative approach for finding the
equilibrium price and demand trajectories over all time periods in the cycle is
described below.
Step 1: Given input vectors pO, dO set do Jd = do;.pCq = po
Step 2: Using supply price reJatio-nship for each element of do Jd find corresponding vector of supply spot prices p' .
Step 3: Set d OCW = d oJd + a(d - dOld ) where the elements of the d vector are
obtained from (E-:-2.1) or (E.2.S)setting the p(j) equal to elements of vector
p' determined in Step 2 above. The parameter a is a scalar input between 0
and I appropriately selected to avoid divergence. Typical values of a that
work in most cases are between 0.3 and 0.6.
Step 4: Check for convergence by comparing d OCW and do 1d If sufficiently
different, set do 1d = d OCW and return to Step 2. When convergence is reached
i:..Of:W and e.cqUil
f!..' and finish.
set dequil
I. Since (E.l.6) defines demand as a function of rate, i.e. d[r(t), it follows that

iJ 2 B[d(tJr'
iJ 2 d(t)

This result is not used in this book, but it is sort of interesting.

2. Consumer surplus is defined as the difference between the use value of electricity to consumers
minus the cost of electricity (i.e. its price) to consumers summed over all consumers.


Throughout most of this book, the shortest time interval of interest has been
one hour. The Chapter 9 discussions on price-quantity transactions for security
control and operating reserve considered shorter time intervals down to
minutes, and perhaps seconds. This Appendix discusses a different approach to
the use of loads as fast-acting (seconds to minutes) operating reserves. This
approach is based on the microprocessor controller called a Frequency
Adaptive Power Energy Rescheduler or F APER. A F APER provides fast
responses without any signaling from the utility and without the customer
realizing what is happening.

A FAPER's operation combines on two separate concepts:


Under Frequency Relaying: A major loss of generation or tie-line support can

cause a sizable drop in local frequency for a short period of time. Hence a
common defense mechanism employed by utilities is the use of under frequency relays to drop loads (usually at a substation level) when frequency gets
too low (for too long).
Energy vs. Power Loads: Energy loads such as space conditioning, water
heating, pumping and metal melting require a certain amount of energy over
time; the power level at a given time is not important. Power type loads such
as lights, computers, and television require power at specific times.

_ _ _~!l4(,",,"~....


A F APER users a local frequency measurement as a signal to reschedule the

power usage of an energy type load.
To be more specific, consider electric heating of a building with a simple
thermostatic on-off control. The conventional control (without a F APER)

Turns the heater off when the temperature T(l) at time I (continuous time) is
greater than the specified upper limit T max , i.e. off when T(l) > Tmax.
Turns the heater on when T(l) is less than a specific lower limit Tmin , i.e., on
when T(l) < T min .

A F APER continues to exercise the above two conventional control laws; i.e.,

Turns off when T(l) > Tmax.

Turns on when T(l) < Tmin .

However, in addition a FAPER


If heater is on, turns it off if frequency is less than desired and T min < T(t).
If heater is off, turns it on if frequency is greater than desired and
T(t) < Tmax

The FAPER control logic has two important factors:


Since the control logics of the conventional control dominate, the customer
does not "know" that control action is taking place.
Energy consumption (over the long term) is not reduced when frequency
drops. The power that has been used is simply rescheduled to a later time,
where the time shift depends where the temperature T(l) lies between Tmin and

The FAPER concept makes sense only if many FAPERS are placed on many
energy type loads. For a single heater, no control action results if the heater is
off when frequency drops. Similarly, for a single heater, the length of time before
it comes back on is "random." Hence the utility cannot know in advance what
effect a single FAPER will have. However, when dealing with a large number
of F APER controlled loads, diversity and probability theory conceptually
enable one to predict the overall effect of the FAPER control. This diversity
issue is the same one discussed in Section B.l. A utility can forecast total
demand over short intervals even though the utility cannot forecast the behavior
of an individual customer.

The basic FAPER concept of Section F.l can be implemented in a variety of

ways. Two approaches are described to illustrate the possibilities.

Appendix F






___. 1. v~----





Figure F.2.1. Control functions for a practical on-off FAPER.

An On-Off F APER

Consider an on-off FAPER of the type discussed in Section F.I. Assume again
a heater is being used to maintain some desired temperature range. One practical
FAPER control logic is

Turn heater off when T(l) is above Tupp[f(t)]; i.e., T(l) > Tupp[.f(l)]
Turn heater on when T(l) is less than T;ow[f(t)]; i.e., T(t) < T;ow[f(t)]

where the T;ow and Tupp control functions are given in Figure F.2.1. This gives
a smoother, more desirable response than the simple logic discussed in Section
The "slopes" of the T;ow and Tupp functions are design parameters. Specification of the reference frequency Jo(t) will be discussed later.
A Continuous F APER

Not all energy type load controllers are of the on-off type. Some employ
continuous controllers. For example, assume (again for a heater) that the
conventional controller acts as follows:
u(t): Power to heater at time t (kW)




Tma ,

K [Tm., - T(l) ]
Tm .. - T min

T min

< T(t) < Tm.,



Then a F APER version of a continuous control is

338 Appendices


K [Tm .. - T(t)]
Tm - Tmin

T(t) > Tm

B[f(t) - /o(t)]

Tmin < T(t) < Tma.

Tmin > T(t)

where B is a design parameter and fo(t) is the reference frequency.

Specification of Reference Frequency /o(t)

Both the on-off and continuous F APER designs depend on a reference freIn a simple-minded FAPER,!o(t) = 60 Hz (or 50 Hz depending on
the country). However, as discussed in Section B.4, the system AGC does not
keep the system frequency J(t) exactly at 60 Hz all the time, even under normal
operating conditions. It tends to be a little bit below 60 Hz during the day and
above 60 Hz at night. Therefore a more sophisticated design has a reference
frequency!o(t) that tracks the normal frequency variation. This can be done by
putting the measured J(t) through a low pass filter (time constant of hours) to
get !oCt).


There are many possible variations on the basic concepts. Two are discussed as
Operations to Reduce Normal Power Plant Control

The F APER has been discussed as a method of carrying operating reserve on

the loads by responding to major changes in frequency following loss of generation. Theoretically a FAPER could be made sensitive enough to respond 1.0 the
much smaller normal frequency changes. Thus a FAPER might be useful in
helping reduce the AGC-governor "noise" signals to which individual power
plants respond.
We have built a FAPER suitable for operating reserve control. We don't
know whether a more sensitive version is practical.
Central Utility Control

The F APER is based on the use of locally measured frequency to exploit the
short-term rescheduling potential associated with energy type loads. It is
possible to replace the locally measured frequency with a more sophisticated
and coordinated control signal generated at the utility's central control center.
Such a signal could be transmitted for example, by radio.
Such central utility control has the advantage of enabling direct coordination
of load control with other utility control functions. The concept of rescheduling
the power flows to energy type loads (and such that the customer doesn't know
it is happening) can be maintained.
The potential disadvantage of such central utility control is that the commun-

Appendix F


ication costs might override the benefits of the central action. Another potential
disadvantage lies in the inherent time delays associated with long-distance

As discussed in Section F.I, F APERs make sense only when there are many of
them on diverse loads. A key question is
What are the statistical-probabilistic properties of the aggregate responses of many
F APERs in a closed-loop system which includes the effect of load change on frequency? .

Unfortunately we don't have an answer to this question.

We have done some simulations, and they behaved "reasonably" and about
as expected; i.e., power used by the loads dropped for a while and then came
back up. We tried direct analysis and found that we could not get explicit results
for the key question of closed loop behavior. The system is basically nonlinear
and we could not find tricks which yielded "nice answers." We now believe
detailed, multiple simulations are the way to getting the needed answers.

A FAPER is located "behind the meter" directly on or within an individual

appliance or end use device. Customers can be given incentives to install such
devices in various ways such as
" Law
Lump sum payment
" Payment for response

F APERs can be installed on appliances at the factory by law. This seems

unlikely to occur in practice.
Customers can be given lump sum cash incentives to install F APERs. This
has the disadvantage that the utility won't know when devices fail, and the
customers have no incentive to maintain them.
A conceptually cleaner approach is to develop a frequency-sensitive meter
that records the customer's total load response to frequency change. Favorable
response is rewarded with a cash payment or a reduction in the monthly bill.
Such meters can be built, but their cost might override the benefit of F APER
action. A logical extension of the ideas of this book is to introduce a frequencydependent component to an "instaneous spot price."
More study is needed before a practical approach can be recommended.

The F APER concept has a lot of potential value and could become important
in the power system of the future. However, the F APER concept is like the

340 Appendices

deregulated generation systems discussed in Chapter 5. More studies are needed

before any recommendation for actual implementation can be made. The two
areas particularly needing more studies are the nature of aggregate multiple
FAPERs response and cost-effective mechanisms to get customers to install and
maintain F APERs.



A utility contemplating the introduction of spot price based rates may benefit
from a quick estimation of how the spot prices are expected to behave, now or
in the future. This appendix presents a typical approach to such a study.
A typical study consists of estimates of 24-hour trajectories of the utility's
marginal costs for selected days, complemented by an aggregated characterization of the statistical behavior of hourly costs over large periods of time (a
month or a year). The 24-hour trajectory information is useful when obtained
for typical and peak load days in each season. The statistical behavior of hourly
costs (or hourly spot prices) can be summarized by a cost duration curve that
is analogous to the familiar load duration curve. This appendix presents an
example of the necessary data and how they can be translated to variable cost
trajectories and duration curves. The reader should note that the level of
complexity of the study presented here is intentionally simple in order to allow
implementation on a personal computer using widely available spread sheet
accounting software.
Discussions in the remainder of this appendix will consider a simplified form
of the optimal spot price which ignores network maintenance costs. In particular, the spot price at hour t without revenue reconciliation is considered to


342 Appendices

[Generation Fuel and Variable Maintenance)






[Generation Quality of Supply]



[Network Losses)
[Network Quality of Supply]

).(1) = system lambda at hour f
d(f) = demand level at hour f

= Network losses coefficient consistent with a two-bus model with all generation and load located at each of the two buses





adjustment for revenue reconciliation yields


or alternatively decomposed reconciliation can be done as discussed in Section


In this section we will discuss how a 24-hour trajectory of spot prices can be
estimated. In practice spot price trajectories will vary even if the demand
trajectory stays constant. However, a useful description of all possible price
trajectories is provided by their "conditional expectation," that is, the expected
value of hourly spot prices conditional upon the demand level and other
seasonal parameters (e.g., planned generator maintenance). Estimation of the
conditional expectation of the various spot price components discussed in
Section G.I can be obtained as described below. For notational simplicity we
denote by l(t) a function of d(f), i.e., l[d(l)] is denoted as simply f(t).
System Lambda let)

There are many ways to calculate the conditional expectation of system lambda.
Many available generation planning software tools provide marginal cost information that includes the expected value of the incremental variable cost of
meeting the last kWh of a given load level. This is obtained by finding the
probability that each generator will be the marginal generator given that
demand is at a known level, and then obtaining the average variable cost of all
generators weighted by these probabilities. A plot of expected system lambda
versus load can be obtained by repeating the above exercise for various load
levels. The plot should exhibit a monotonically increasing relationship.
A simpler way to obtain the system lambda versus load curve is the following:

Arrange all generating units in merit order, that is in order of increasing

variable cost.
Derate the capacity of each generating unit by multiplying its rated capacity

Appendix G



















[jIPODIll OIDID 0l1li



Din l1li





Load "'~walts

Figure G.2.1. System lambda vs. load.

by availability. If the unit is on planned maintenance during all or part of the

relevant season, reflect planned maintenance in the derating operation.
o Find the cumulative derated capacity of generating units starting with the first
unit in the merit order and including one more unit at a time. If the system
includes hydro units, place them in the merit order according to the loading
order that results in a capacity factor consistent with their energy availability,
and assign them a variable cost equal to that of the thermal unit preceding
them in the loading order.
" The cumulative derated capacity versus generating unit variable cost will
provide a step function relationship between generation level get) and load
level d(t). Noting that g(r) = d(t) + Bd(t)2 the load levels d(t) corresponding to generation levels get) can be calculated and the system lambda versus
load curve established.
Figure G.2.1. is an example of a system lambda versus load relationship. This
relationship may vary significantly from utility to utility as it depends very
strongly on the utility's generation mix. The minimum input required for its
determination is generating unit capabilities, availabilities and variable costs.
Network Losses Bd(t)

Network losses Bd 2 (t) and marginal network losses 2Bd(t) can be obtained from
the loss coefficient B. The B coefficient can be calibrated from information on
losses at peak load or average losses. For example if losses at peak are known
to be 11 % and peak load is dp then B can be obtained from the relationship

344 Appendices








Load (WW)

Figure G.2.2. LOLP vs. load.


Bd~l =


Alternatively, if average annual losses are given at 6% then B can be obtained

from the relationship

Load duration curve information can be used in place of the 8760 hourly load
data by appropriate frequency weighted treatment of the various load duration
curve load levels.
Generation Quality of Supply I'Qs(t)

Estimation of the "lQs(t) term using say (6.2.5) or (6.2.12) requires input on
generation-induced loss of load probability (LOLP (/)J at various load levels.
This information is usually available at utility planning departments and can be
obtained from generation outage tables or convolution of generating unit
outage probabilities. It can be often fitted analytically with an exponential
relationship. Figure G.2.2 presents an example of an LOLP1,(t) versus load d(t)
curve. Given the LOLP,.(t) relationship, we have from (6.2.5) or (6.2.12)

where C, is either the average cost to consumers per kWh of unmet demand or

Appendix G


alternatively the annualized cost per kW of peaking load capacity divided by the
annual expected loss of load hours due to generation shortfalls (LOLH,).
Typical values of customer cost of unserved energy are $l/kWh, and of annualized capacity costs $30-80/kW.
Network Quality of Supply '1Qs(t)

Estimation of the 'lQs(t) term can be performed similarly to that of YQs(t).

However, in practice, network-induced loss of load probabilities as a function
of load level are rarely available. A reasonable approximation in the absence of
better network-induced LOLP~(t) data is to use (7.9.3)
if d(t) > de
LOLP,,(t) =


and associated equations such as (7.9.7)

Revenue Reconciliation Multiplier (s)

Calibration of the revenue reconciliation components can take place in an

iterative fashion. Values for m or, if separate reconciliation is desired, my and m~
are adjusted until generation and network capital revenue rate base requirements are met. The requisite data is capital revenue requirements and 8760
hourly load data or load duration curve information.

The price duration curve conveys aggregate information on the statistical

behavior of the spot price much like the load duration curve does for load.
Price duration curves may be generated by applying the methods of Section
G.2 on a hypothetical trajectory which is in fact the load duration curve.
Alternatively, however, the methodology of probabilistic production costing
can be used to yield price duration curves. When the probabilistic production
costing framework is employed, a different conditional expectation is used.
Rather than conditioning on the hourly load level and considering conditional
expectations of system lambda, conditional expectations of demand are
employed for given system lambda levels. Considering that uncertainties in
demand are in the short run more predictable than uncertainties in generation
availability, conditioning on system lambda may give better results.

Customer response will undoubtedly have a significant impact on price trajectories and price duration curves. Although the exact behavior of customer
response may be hard to model, the direction of the impact can be investigated
using the multiple period demand response models of Appendix E. The following iterative procedure can be employed.

346 Appendices

Step 0: Input an original demand trajectory which yields a duration curve with
the same shape as the annual load duration curve.
Step 1: Derive spot price trajectories as described in Section G.2 or G.3 using
the demand trajectory of step 0 or step 5.
Step 2: Derive revenue reconciliation multipliers using the spot price trajectory
from step I.
Step 3: Derive a spot price versus generation level relationshp using revenue
reconciliation multipliers of Step 2.
Step 4: Using the supply curve of Step 3 and demand response models of
Appendix E.2, derive the demand trajectory that incorporates customer
Step 5: Update demand trajectory and return to Step I. Repeat until updated
demand trajectory converges to a constant trajectory.


The differentiation of an expression (usually a Lagrangian) involving expectations has arisen a number of times in Chapters 9 and 10, as for example in the
context of predetermined rates or investment optimality conditions. In every
occasion, we interchanged the order of the expectation and the derivative
operators while manipulating the necessary optimality conditions to meaningful
forms. No justification for this interchange was given in the text since it holds
for rather general conditions. The purpose of this appendix is to elaborate on
these conditions.
Consider a function g of two variables x and 11' where 11' is a random variable
with probability density functionfw. It follows that

if/;'(\\') does not depend on x directly (through its arguments) or indirectly

through the limits of integration. To show that (R.I) is true, we represent the
expectation operator explicitly. so
i1E{g(x. II')}



= -;(IX


g(x. 11')j;, (H') dw

Using Leibnitz's rule, as long as the functionfw(w) does not depend on x and

348 Appendices

the set Rw does not depend on x either, we can interchange the order of
integration and differentiation to obtain

f (g(x, w)/w(w

dw =

f (L g(x, W)j~(W) dw

By definition, the expression above is


ox g(x, w)

which yields (H.I).

In the context of the spot pricing derivations, the key assumptions can be
interpreted as follows.
Uncertainty, as regards participant costs, capacity availabilities, and the like,
make spot prices, generation, and demand levels random variables when viewed
in advance. However, this uncertainty is introduced through the dependence of
these variables on exogenous random variables such as weather, product requirements, generation forced outages, etc. The expectation operator thus
applied over the exogeneous random variables and differentiation with respect
to endogenous random variables can be moved to the right of the expectation
The above assumption about exogenous and endogenous random variables is
fairly general. It is possible to think of extreme situations where some exogenous
random variable depends on endogenous variable levels. For example, the
probability distribution of generation forced outages may depend on generation
levels and hence customer load levels. For practical purposes, however, such
dependence is weak and concerns a negligible portion of the relevant exogenous
random variables. It can therefore be safely disregarded and (H.l) assumed to
The issue of derivative-expectation operator interchange as well as the autoand cross-correlation of endogenous and exogenous random variables arises
again in a more subtle manner when intertemporal dependence of the type
discussed in Section 10.6 is modeled. The appropriate framework for dealing
with such models is that of Markovich decisions or stochastic dynamic programming. The decision problem at hand can always be cast in a Markovian
decision framework with appropriate definition of state variables. The interested reader is referred to Whittle [1982] or Bertsekas [1987].


California. xiv
Capacity component, 34, 149
Capacity credits, 7S
CAPITAL, xviii
Capital stock, (See also Revenue reconciliation) 78, 238
Generation capital, 238
Transmission capital, 16.238
Peaking plant capital cost method of
setting quality of supply price. 39.
41-42 (example). 139ff
Chronological simulation, 301-302, 305
Closed loop vs open loop feedback, 103, 106,
249, 250, 252. 309
Communication to customer. 4, 16, 17-18,
84-86, 90, 112-113. 288. 238, 339
Competition, 114, 117ff, 175
Computer, uses of, 92. 109 (footnote)
Conditional expectation, 102, 207, 250. 282.
342, 345
Construction Work in Progress, 191
Contingency analysis, 116, 289, 290
Contingency planning, 113,214-215,289,290
Control: see Power system control
Corrective control action, 289
COSTS, 22 .
Annualized capital costs, 41-42 (example),

AC load flow, 152, 175, 272, 274, 279, 289,

decoupled, 274, 289
AC power, 269, 270
Three phase, 270, 271
Single phase, 269
Advance warning of spot price, 12-13, 59
(example), 100, 172-174
Aggregation across customers, 50
Aggregation over time, 72
Area Control Error, 292
Automatic generation control, 104, 116,233,
285, 291-292

B matrix for losses, 100, 176 (footnote), 316,

317, 318, 322-323 (example)
Bankruptcy, 121, 128 (footnote)
Bidding, 86, 108
Blackout, 19, 123,279,288
Bus. swing,

Cable TV, 84, 86




Line shunt capacitance, 272

Linear demand response, 143-145 (example),
Linear programming, 287, 307
Load duration curve, 48, 87-89, 237, 299, 301,
302, 303, 304, 341, 345
Load dynamics, 277
Load, energy VS. power, 335
Load flow: See AC load flow
Load forecasts, (See also Forecasts), 107,281283, 299-301
Load management, 305, 306
Load shedding, 104,279,289
Load frequency control, 294
Long term contract for revenue reconciliation
Lond term contracts, II, 17,66-67, 121-122
(example), 128 (footnote)
Long term dynamics, 278, 279
Loss of load probability, 39, 41-42 (example),
99, 139, 140-143, 172-174, 297,
Losses (See also: Network losses), 36, 37-38
(example), 51,100,175,285,291

Maintenance scheduling, 47, 118,288,293

Marginal generator, 48, 158
Energy marketplace, xviii, 5, 6, 9-11, 21,
23,26, 311f, 51, 55,73,88,89, 108,
237, 248
Futures market, 66-67, 67, 108, 223-225
Market clearing, 5, 8, 39 (example), 53
footnote, 64, 99
Market information flows, 14-15
Marketplace operation, 82-87
Spot market, 8, 237, 248
Time scale of market resolution, 222
Market coordinator, 115, 116
Metering, 5, 16, 83, 85, 288, 339
Merit order of generating units, 302, 303
Microelectronics, xviii, 4, 13,26
Microshedding: See FAPER
MIT, xv
Monopoly, 5, 114, 117
Monte Carlo simulation, 47-49, 302, 305
MUltiple attributes, 295-298, 311
Multiple periods, 147-148, 149,299

Net interchange, 116,233,291,292

Net revenues, (See also: Revenue reconciliation), 118,248,319-320

NETWORK or transmission system, 112,

113, 124-125, 151
Network construction/investment, 112113, 120,247,308,309,318,319
Network flow, 160-161. 161-162, 165-167
(example), 1681f, 175,269-274,289,
309, 313-320, 320-325 (example)
Network losses (transmission loss), 36, 3738 (example), 41-42 (example), 153,
154, 157, 159-160, 175, 232, 235,
247, 285, 3131f, 3201f, 342, 343
Network planning, 120, 237, 308, 309
Network quality of supply, 34, 40
(example), 109 (footnote), 154, 161,
172-174, 325, 345
Network revenue reconciliation, 32
Newton-Raphson, 273
Nuclear power, 35-36 (example), 114, 119,
Number of price levels, 59, 114

Obligation to serve, 124-125

One hour update: see Prices, hourly
Operating reserves, 16, 22, 38, 63, 64, 77, 104,
109 (footnote), 114, 137, 1391f, 149
(footnote), \71,215-217,286,287,
290, 293, 338
Optimization, constrained, 210, 285, 287, 288,
296, 298
Optimization transactions, 68
Options, II, 67
Own load dispatch, 105

Per unit system, 53 footnote

Period definition, 9-11, 59, 62,78, 105
Planning, xviii, 95-97, 1171f
PLANT, 39, 119, 127
Availability, 36, 64, 78, 98, 121, 126,241,
285, 301
Heat rate, 98, 285
Outages, 47, 60, 63, 64, 115, 126, 30 I, 303305
Plant value, 118,235-236,241-244
Plant marginal cost, 162-168 (example),
Pool, 105, 235, 293-294
Real, 152, 2701f, 313, 314
Reactive, 52, 116, 152, 175, 2701f, 279

Index 353

Control, 22, 269, 274ff, 281, 289, 292,338
Dispatch: see Dispatch, central economic
Dynamics, 80, 269, 277, 280, 290, 291
Operations, 22, 57, 103,275,281-294
Planning, 23,106,113, 139ff,290, 295-311,
Power system security control: See
Security control
Predetermined price participant, (See also Rationing), 237, 240, 244, 245
Prepayment, 218, 221
Price discrimination, 123, 187
Price duration curve, 47-49, 237, 242, 243,
244, 345
Price forecasts: See Forecasts, price
Price trajectories, 7, 14-15, 24, 45ff, 47-49,
24 hours update, 12, 12-13, 18,58,63, 114,
Contingency prices: See Prices, security
Dynamic pricing, 52, 116, 202, 236
Fixed-price-fixed-quantity, 11,66-67,223225
Hourly, 7, 10, 51, 58,63, 80
Mandatory vs. optional prices, 68, 90
Marginal cost prices: See Costs, marginal
Nonlinear prices, 44, 73, 195-198,202
Predetermined price, 206-210, 240, 245
Price only, 10, 58
Price-quantity, xvi, 10, 17,63,63-66,80,
114, 213-222
Real time prices, 27, 52
Security control price, 215-219
Market clearing price, 39 (example)
Spatial price, 35, 168-170
Prices, non uniqueness of, 171
Prices, Predictability of, 7,10,331-333,341345
Priority lists, 80, 286
Probabilistic simulation, 47-49, 303-304
Production cost model, 98, 149 (footnote),
Pumped storage, 287, 290, 307
Purchase/sale, 101,293,294
PURPA (See also: Costs, avoided), 75
Quality of supply, (See also Spot Price Components), 38ff 40 (example), 51, 9293, 137, I 39ff, 149, 239, 242, 289,
344, 345

Ramp rate, 35, 285, 287

Ramsey pricing, 42-44, 102, 182-184, 186187, 202, 226, 228
Ratchet clause, 70
Rate base, 108, 305, 345
Rate shock, 305
RATES, (See also: Prices)
Block rates, 71, 73
Buy-back rates, xvi, 10, 44-45, 75, 170171, 184-187
Flat rates, 4, 8, 10, 20, 71, 72
Incentive rates, 227
Interruptable rates, 8, 63, 65, 71, 219, 220,
236 footnote
Life line, 26, 57, 71, 74
Time of use rates, 4, 8, 10, 20, 22, 23, 71,
78, 305
Rationing, 23, 41, 80, 99, 210-213
Reactance to resistance ratio, 169, 246, 247,
Regulation xviii, 5,42-44, 82, III, 112-113,
119, 121-122 (example), 122, 123
Regulatory commission, 26, 107-108, 122,
127, 128 (footnote), 298
Retrofitting, 96
Relaying, 112-113, 117,275-276,335
Reserves, Reserve margin, (See also: Operating reserves), 100, 106, 114, 116,
126, 137, I 39ff, 236, 286, 287
Residential customers, (See also customers,
residential), 13, 18, 49ff, 80, 94
Responsive pricing, 27, 51
Revenue neutrality, 102, 199-200
Ramsey pricing), 7, 51, 72, 75, 102,
108, 177-203, 228
Additive form, 179
Decomposition, -191-194,342
Generation revenue reconciliation, 194
Ideal revenue reconciliation, 237, 248
Modify spot prices, 76,178-187,201
Multiplicative form, 179, 229, 342, 345
Network revenue reconciliation, 193
Revenue stability, 24, 182, 183, 20 I
Revolving fund, 42-45, 189-191, 200, 226
Surcharge-refund, 42-44, 188-189
Risk,77, 121-122 (example), 298-299
Screening curves, 307, 308-309 (example)
Security control, (See also: Power system
control), 22, 66, I 13, 116, 126, 213,
214-218, 275, 276, 281, 288-290,



Security systems, home, 86

Self dispatch, 115, 146-147, 170-171,
Sensitivity analysis, 289, 296-297, 299, 307
Shunt capacitance, 272
Single phase AC power: See AC power
Slow speed dynamics, 278-279
Small power producer, (See also Deregulation. Rates, Buy-back), 125, 170171
Social welfare/social cost, 33. 143-145, 148,
Social optimality conditions, 237-247
Spinning reserves, (See also Operating reserves). 290, 217
Split the difference, 293
Spot price behavior. 45ff, 47-49,7, 14-15, 115,
Quality of supply, System lambda)
Basic formulas, 34, 32, 9-11, 152, 155,
162-168 (example)
Calculation of, 97-99
Capacity components, 149
Covariance term, 59, 59-62 (example), 68,
69,73, 102, 109 (footnote)
Generation marginal fuel
Generation marginal maintenance
Generation quality of supply, 39 (example), 38ff, 99, 137, 143-145, 148,
Generation revenue reconciliation, 44,
179, 124
Network loss component, 36, 37-38
(example), 41-42 (example). 155,
159-160,152-168 (example). I 68ff,
Network quality of supply, 40 (example),
41-42 (example), 42-44, IO}, 155,
Network revenue reconciliation. 44
Network marginal maintenance, 155, 160161,172-174
Reliability components: See Network
quality of supply; Generation
quality of supply
System lambda, 34, 35, 36-36 (example),
155, 162-168 (example), 294. 342,
343, 345
Spot price, types of, (See Prices; Rates)
Start up costs, 79, 147-148,285
State estimation, 116, 288, 289
Status quo, 4, 20. 21, 127
Steady state stability, 277, 278, 279

Storage (See also: Pumped storage), 287, 236

(footnote), 48, 65, 79, 91
Supervisory control and data acquisition, 281
Surcharge-refund: See Revenue reconciliation, 42--44
Swing bus. 153, 155,273,315,317.322-323
Swing equation, 277, 278, 279
System lambda, formulas for, (See also: Spot
price components), 34, 294, 342,
343, 345, 35, 35-36 (example), 9799
System monitoring, 288, 289

Target mix, 306

Target revenue, 43
Time of use rate, (see Rates)
Time, control of, 292
Time period linkage: See intertemporal effects
Time period for spot price, 148
Tradeoffs and tradeoff analysis, 297-298, 299,
56, 62, 65, 77
Prices, types ot), xvii 77
Custom tailored, 76.227-231
Futures market transactions
Long term contract, 10. II, 17.55,66-67,
121-122 (example),
(footnote), 222-227
Price only, 10, 16,55,58,62,82,92-93
Price-quantity transactions, 10, 222, 17,
55, 63-66, 109 (footnote)
Transactions types, choice among, 87-89, 95.
Transfer admittance matrix. 153, 155,316.322
Transient stability, 277, 278, 279, 290
Transmission lines, 271, 286
Transmission system (See Network), 112-113
Transtext, xiv
T&D company as regulated middleman. 112,
(example). 122

Uncertainty, 24, 25, 61, 95,106,119, 148,250.

295, 298-299, 348
Under frequency relays, 335. 279
Unit commitment, 35-36 (example), 98, 102,
104, 285-286 (example). 287. 293,
University professors, xv, 109 (footnote), 298


Unserved energy, 53 footnote, 139, 139fT, 140143,301. 307

Update cycle, 7, 58,66-67, 72, 78, 80, 85
Utility: see Power system,
Utility-customer interface, 94
Utility function (of consumers), 296, 299
Utility to utility sales, 35-36 (example), 20, 99,

Valve point loading. 285

Vickrey. William, 52, 27, 78. 202, 253
Voltage, 269-272


Voltage magnitude or voltage angle, 152,288,

315,319,277,289, 175,269-272
Voltage regulation, 116,289,276,278

Weather or Weather forecasts, 50, 283, 61, 103

Weighted least squares, 43. 180-181, 289
WHEELING, III, 175,293,231-235
Mandatory wheeling, 124
Bus to bus wheeling, 76
Types of wheeling, 231-232, 234
Win win scenarios, xviii, 236