Professional Documents
Culture Documents
Handbook of Networks in Power Systems II
Handbook of Networks in Power Systems II
Series Editor:
Panos M. Pardalos, University of Florida, USA
Editors
Handbook of Networks
in Power Systems II
Editors
Alexey Sorokin Steffen Rebennack
University of Florida Colorado School of Mines
Industrial and Systems Division of Economics and Business
Engineering Engineering Hall
Weil Hall 303 15th Street 816
32611 Gainesville Florida 80401 Golden Colorado
USA USA
sorokin@ufl.edu srebennack@mines.edu
v
vi Handbook of Networks in Power Systems
Alexey Sorokin
Steffen Rebennack
Panos M. Pardalos
Niko A. Iliadis
Mario V.F. Pereira
Contents
vii
viii Contents
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
.
Part I
Gas Network
Implementation of a Scheduling and Pricing
Model for Natural Gas
Abstract Since 1999, the Australian state of Victoria has operated a natural gas
spot market to both determine daily prices for natural gas and develop an optimal
schedule for the market based on an LP (Linear Programming) approximation to the
underlying inter-temporal nonlinear aspects of the gas flow optimization problem.
This market employs a dispatch optimization model and a related market clearing
model. Here we present the model employed for both the operational scheduling
and price determination. The basic dispatch optimization formulation covers the
key physical relationships between pressure, flow, storage, with flow controlled by
valves, and assisted by compressors, where flow and storage are measured with
respect to energy rather than in terms of mass. But we also discuss a range of
sophisticated mathematical techniques which have had to be employed to create a
practical dispatch tool, including iterating between piecewise and successive line-
arization; iterating between barrier and simplex algorithms to manage numerical
accuracy and solution speed issues, and special methods developed to deal with
scheduling flexibility. The market clearing model is a variation on the dispatch
optimization model which replaces the gas network with an infinite storage tank
with unlimited transport capacity. We address the performance of the model
including accuracy and run time.
The authors wish to thank the Australian Energy Market Operator (AEMO) for providing
information used in this chapter and for review of the content.
W. Pepper
ICF International, Fairfax, Virginia, USA
e-mail: wpepper@icfi.com
B.J. Ring
Market Reform, Melbourne, Australia
e-mail: brendan.ring@marketreform.com
E.G. Read (*) • S.R. Starkey
University of Canterbury, Christchurch, New Zealand
e-mail: grant.read@canterbury.ac.nz; egread@attglobal.net; stephen.starkey@pg.canterbury.ac.nz
A. Sorokin et al. (eds.), Handbook of Networks in Power Systems II, Energy Systems, 3
DOI 10.1007/978-3-642-23406-4_1, # Springer-Verlag Berlin Heidelberg 2012
4 W. Pepper et al.
1 Introduction
Since 1999 a gas market has been operating in the Australian state of Victoria,
based on a “common carriage”1 model, similar to that employed in many electricity
markets. The motivation for introduction of that market concept has been discussed
by Read et al. [1], who demonstrate that, conceptually, such a market can be based
on an LP formulation analogous to that employed in electricity markets. That paper
develops and interprets an LP formulation based on a textbook representation of the
gas flow equations, and does not deal fully with some real-life complexities, such as
the potential non-convexity of some functional relationships. Thus that formulation
is not the one actually employed for dispatch purposes in the Victorian gas market.
Further, while the gas market concept developed by Read et al., like the electricity
market concepts on which it was based, would produce different prices for each
system injection/extraction point in each trading interval, market participants in
Victoria have preferred to trade on a much simpler basis, which captures some of
the inter-temporal variation, but treats all gas, anywhere in the system, as inter-
changeable, within each day. While Read et al. [1] describes a conceptual model
this paper describes the model implemented in the working market and operational
dispatch models as previously presented by Pepper and Lo [20].
The Victorian market produces both the physical dispatch schedules and the
market price are determined in accordance with the Market Clearing Logic
described by Ruff [2] which called for optimization of the physical gas dispatch
in the context of market driven supply and demand to form a realistic constrained
schedule of hourly injections and withdrawals of natural gas into/from the pipeline
system, alongside calculation of another hypothetical unconstrained schedule to
determine the daily market prices and schedules used in the settlement of the
market. These features of the Market Clearing Logic were implemented within
the Market Clearing Engine (MCE) developed by ICF International [3].
The MCE can be run to produce a physical gas dispatch, the so called ‘Opera-
tional Model’, or to determine market prices and trading schedules, the so called
‘Market Model’. Both of these models are run by the gas system operator, originally
VenCorp, now absorbed into the Australian Energy Market Operator (AEMO). The
Operational Model includes a detailed representation of the physical gas system and
can optimize supply, based on supplier injection bids, purchaser withdrawal bids,
1
The use of common carriage is due to the similarities with electricity markets. Victoria actually
use the term “market carriage” to represent its pool based market, contrasting with “contract
carriage” which relates to more traditional scheduling under bilateral contracts for access to each
individual transmission pipelines.
Implementation of a Scheduling and Pricing Model for Natural Gas 5
runs from 10 a.m. for 20 h. Both models are run again, with a new schedule and a
new ex ante price determined. This price is used to settle differences between actual
gas flows and scheduled gas flows from 6 a.m. to 10 a.m., and settles all changes in
scheduled gas flows for the 20 h from 10 a.m.. In effect, the 6 a.m. schedule
provides a form of forward market for the 10 a.m. schedule. Similar processes are
repeated at 2 p.m., 6 p.m. and 10 p.m., the last schedule covering the remaining
hours from 10 p.m. to 6 a.m. the next day. The Operational Model can also be run at
other times if required.
The MCE seems unique in many ways. Aspects of its function are captured in
other models, however. Zhu et al. [4] note that ‘the literature on gas pipeline control
is rather sparse’, yet they do make note of two commercial software packages for
dynamic gas pipeline simulation, the Hyprotech PYPESYS and the Gregg Engi-
neering WinTran model. Johnson et al. [5] discuss the use of WinTran model for the
Tennessee Gas Pipeline which consists of over 15,000 miles of pipeline and 70
compressor stations. These models determine the (operational) gas flows, and
WinTran optimizes compressor fuel use, yet they have no explicit concept of
costs or economics. ICF International has two other models of gas pipeline markets
that have a much different focus than the MCE. The ICF Gas Market Model [6]
represents the North American Gas Market focusing on finding a market equilib-
rium and uses costs and economics but the daily pipeline capacities are fixed.
Similarly, the ICF RIAMS model provides detailed regional modeling of pipeline
capacity and determines a market equilibrium using costs and economics but again
uses fixed estimates of daily pipeline capacity. The Scheduling and Pricing Engine
(SPE) employed in AEMO’s Short Term Trading Model (STTM) [7] for gas
trading at hubs in Adelaide and Sydney (Australia) can be thought of as a form
of Market Model which determines a single price and schedule for the day, but
only represents a simple capacity limit on the ability of transmission pipelines to
supply a hub.
Reviews of recent literature may be found in Read et al. [1] and Zheng et al. [8].
The latter note optimization models being applied to gas production, and gas
pipeline network development and operations, including optimization of gas
compressor operation to minimize fuel use, which may be seen as a sub-problem
of the application discussed here. Both Peretti and Toth [19] and Rios-Mercado
[21] use Dynamic Programming to optimize fuel use. In particular, Wu et al. [9]
discuss the issues in considerable detail, although they take a different
convexification approach than we do. Zheng et al also discuss formulation of the
“least cost gas purchasing problem”, which deals with the same physical flow
requirements as our problem. De Wolf and Smeers [10] suggest a solution strategy
broadly similar to our own, based on successive application and refinement of a
piece-wise linearization approach. But their approach to piece-wise linearization
[11] is different from ours. Several more recent papers describe practical models
developed for the Norwegian Gas Industry. Somewhat like Read et al. [1],
Tomasgard et al. [12] and Midthun et al. [13] both discuss linearization strategies
to deal with non-linear pipe flow losses using Taylor’s expansions, and note that
this strategy “makes analyses computationally feasible even for large networks”.
Implementation of a Scheduling and Pricing Model for Natural Gas 7
But the linearization approach we have adopted here uses the “lambda”, or “convex
combination” method, as discussed by Martin et al. [14]. For a recent PhD thesis on
gas flow linearization techniques see Van der Hoeven [15].
The current paper differs from all of the above in that it describes a Market
Clearing Engine that has been specifically implemented for the purpose of determin-
ing operational schedules, and clearing short term gas market trading on a routine
basis. After describing the physical gas system involved, we present an outline of the
Operational Model formulation. This model was developed on a different and more
pragmatic basis than the formulation of Read et al. [1], partly because it was unclear
how easy it would be to tune the more theoretical formulation of Read et al to produce
accurate results in an acceptable computation time. Thus the priority for the MCE
was that it had to be developed commercially in a limited timeframe to meet specific
performance requirements. But the MCE, as implemented in its “Operational Model”
form, is capable of producing hourly nodal prices, as discussed by Read et al. On the
other hand, we later describe the “Market Model” form of the MCE, which
determines the much more aggregated prices actually used for trading by solving a
highly simplified special case of the Operational Model.
The Victorian Gas Market is a spot market developed to facilitate the trade of gas
between privately owned wholesale suppliers of gas and privately owned retailers
and industrial users of gas across the Principal Transmission System (PTS), the core
gas transmission network.2 As shown in Fig. 1, this is a meshed network rather than
a single long pipeline. The major market participants are both suppliers of gas and
purchasers of gas so are only exposed to spot prices to the extent that their supply
and usage are out of balance.
At its commencement, the market was characterized by a large demand around
Melbourne supplied by a single gas production facility, Longford (represented by
ESSO at the Pre-Longford node), approximately 160 km from Melbourne. The
capacity of the Longford to Melbourne pipeline is small enough that the pipe’s
throughput could be constrained within a day. If this occurred, typically during peak
demand winter days, then the only alternative supply source was from an LNG
facility, with limited storage, at the Dandenong City Gate (DCG Inlet) within
Melbourne. Since market commencement, the PTS has become interconnected
with South Australia (SEAGas Pipeline connected at Pre-Iona), New South Wales
(EAPL Pipeline connected at Pre-Culcairn) and, via the VicHub (at the Pre-
Longford node), with pipelines (not shown) linking to Tasmania (Tasmanian Gas
Pipeline) and New South Wales (Eastern Gas Pipeline). An underground storage
2
As of the Victorian gas market recently coming under Australia’ National Gas Rules the PTS is
now called the DTS, or Determined Transmission System.
8 W. Pepper et al.
Clonbinane
C Bendigo Junc
BBP Reg
Brooklyn – Melb
Outlet Inlet
2B
) Compressor
Kilmore
BL
eg
P
R
lyn
R
eg
ok
Bro Wollert
Brooklyn GL
Outlet Inlet
Melbourne Zone
Ballarat Junc
DCG Inlet
Pakenham
Gooding
Ballarat Zone
Ballarat
field (WUGS connected at pre-Iona) which takes in gas at times of low prices and
releases it at times of high price has also been established. Gas fields off the south
cost of Victoria have also connected (Ottway at Pre-Iona and BassGas at
Pakenham) to the PTS. The major supply node point is still Pre-Longford, though.
Major compressor stations, which effectively pump gas between regions of the
PTS, are installed at Brooklyn, Wollert, Springhurst, and Gooding. These
compressors are powered by a small quantity of the gas withdrawn from the
network. According to Wu et al. [9] “compressor stations typically consume
about 3–5% of the transported gas”. In this case, though, total compressor gas
usage is relatively small. For a 1,200 TJ peak demand scenario with all compressor
stations running no more than 3 TJ, or 0.25% of daily demand, is consumed by
compressors.
Since the commencement of the market, the demand for gas has increased,
particularly in the form of exports to surrounding jurisdictions and increased use
of gas fired power generation. Gas powered generators (GPGs) are located at
Traralgon (Jeerlang and Valley Power GPGs), Brooklyn-Melbourne (Newport
GPG), Brooklyn-Geelong (Laverton GPG) and Wollerton Outlet (AGL Somerton
GPG). With all generators running, the GPG demand is about 5–10% of daily
demand, and 20% of demand during the critical evening period, on a winter peak
demand day.3 AEMO is both a system and market operator. It runs the market and
3
Based on information supplied by AEMO for winters 2007 to 2009 inclusive
Implementation of a Scheduling and Pricing Model for Natural Gas 9
has the duty to manage the system so as to maintain secure operation of the gas
system. Principally this entails managing gas stored within the PTS so as to ensure
that supply and demand match. The core issues which AEMO variously deal with
are forecasting demand, dealing with the characteristics of various supply sub-
systems, managing the use of the transmission system, and determining an appro-
priate end of day storage pattern.
Participants in the Victorian Gas Market are typically retailers who have
contracts with physical suppliers of gas. When scheduled in the Victorian Gas
Market, these participants must nominate gas inflow levels from physical suppliers
in accordance with their contracts. Operators of gas processing plants and
interconnected pipelines typically like to keep gas flows relatively constant over
the day, but may allow gas flow rates to be changed several times during the day.
This means that AEMO must match relatively inflexible supply rates with diurnally
varying gas demand in Melbourne. The gas day begins at 6 a.m. with supply
exceeding demand, causing the pressures in the PTS to rise through the morning.
During the late afternoon, demand typically outstrips supply, and remains higher
than supply rates until late in the evening, causing pressures to drop. Hence AEMO
must manage the system to ensure that enough gas is supplied early enough to meet
demand, and to keep pressures above minimum pressure levels especially during
the evening peak period.
Demand in the Victorian Gas System is quite variable. During the summer
months the peak demand can be as low as 400 TJ per day, unless gas fired power
generators are operating. At the peak of winter, demand can be in the region of
1,200 TJ per day. A sudden and significant weather change can produce large
changes in demand. If enough gas has not been scheduled to flow to Melbourne
from the various supply sources in advance of the change, then the time may not
be available to schedule new supplies from producers or interconnected
pipelines before minimum pipeline pressures are reached. Such ‘surprise’ events
can force AEMO to call on LNG to meet demand. While typical gas prices are in
the region of $3/GJ – $4/GJ, LNG may cost $10/GJ or as much as $800/GJ,
which is the maximum market price. Hence the use of LNG imposes significant
costs on the market. If LNG becomes exhausted, or cannot be vaporized fast
enough,4 then demand curtailment of industrial users must be employed. If LNG
runs out then the (constrained) market may be in shortage situations on peak
days, and curtailment is priced at $800, hence the highest prices for LNG tend to
approach $800. AEMO forecasts the demand for gas (aided by knowledge of
participant forecasts, for which participants are financially accountable) at the
start of the day, and updates that forecast throughout the day as the situation
changes.
4
Storage of LNG is limited, and it is possible to use a significant proportion of the storage quite
quickly if LNG is used too freely. Peak LNG vaporization for 1 h can take about 1 day to replace at
the (emergency) maximum rate of LNG production, though more typically takes several days to
replace.
10 W. Pepper et al.
Sections of the PTS can become constrained depending upon the diurnal sched-
ule and location of injections and withdrawals. These constraints are typically
pressure related and can reduce both the ability to supply consumers and the ability
to take injections from suppliers. The minimum pressure of a pipeline reflects the
pressure associated with minimum linepack storage. These limits cannot readily be
breached – in 1998 an explosion at the Longford gas processing plant required the
system to be shut down for several weeks to protect the system’s minimum
pressures. The maximum pressure of a pipeline is related to the physical capabilities
of the pipeline, though lower limits may be set for broader system operation reasons.
Gas flows are primarily driven by the pressure differences across a pipeline,
where a pipeline may be tens or hundreds of kilometers long. If gas is to flow in the
desired direction, at the desired rate, then a specific pressure difference must be
achieved. If demand is too great, relative to the scheduled supply, and no operator
actions are taken, then consumption can literally suck gas from the pipeline at a rate
which lowers the outlet pressure to the point where the outlet pressure hits its
minimum level. If demand is too low, relative to the specific pressure difference,
then the pressure at the demand end will rise and the pressure difference will drop.
This will reduce the ability for gas to flow away from the injection point and may
limit the ability to inject gas into the PTS. This can make it difficult to move enough
gas through the system in time to meet increased consumption later.
AEMO use gas powered compressor stations to pump gas around the network.
The PTS is divided into zones (see Fig. 1), and compressors allow gas to be
transferred from low to high pressure zones, effectively making gas flow against
the natural direction of flow. They can also be used to ‘pump up’ a zone during off-
peak periods so that gas will flow from it to Melbourne during peak periods.
Pipelines are often configured with other useful fittings, such as regulator valves
which are designed to either restrict flow, or adjust pressure differentials. Check
valves may also be used in a pipe to restrict the gas flow to a single direction.
At the commencement of the market, transmission constraints were rare,
impacting only a few days per year and creating relatively small costs. The
incidence and severity of constraints grew as demand grew, particularly from gas
fired generation. More recently, network augmentation has increased the storage
capacity of the PTS, thus increasing the ability of the system to meet demand at
peak times, and significantly reduced the incidence and severity of constraints.
A major consideration for AEMO in scheduling the system is the end of day
storage. There is a relatively small range of end storage levels within which AEMO
seeks to finish the day. A minimum storage must be maintained for system security
reasons. Too high a storage level equates to high pressures around Melbourne
and this may limit the ability of gas to be supplied from locations like Longford
on the next day.
In summary, then, AEMO must forecast demand and manage the scheduling of
gas to manage both the daily peak demand and the end of day conditions so as to
minimize the risk of surprise events requiring the use of LNG and to manage the
timing of gas delivery and the positioning of stored gas to minimize transmission
congestion.
Implementation of a Scheduling and Pricing Model for Natural Gas 11
5
In this formulation, variables are generally upper case, and constants lower case, whereas Read
et al. [1] use lower case for variables, and upper case for constants. Upper and lower limits are still
represented by over and under bars, respectively. Unless otherwise stated, all variables in this
formulation are positive.
12 W. Pepper et al.
of the pipe segment l, identified by o(l), to the corresponding destination node d(l).
We denote the flow into and out of pipe segment l, in period t, as FIlt and FOtl
respectively. Stm is gas scheduled to be supplied according to injection offer m, in
period t while Dtc is gas scheduled to be withdrawn according to withdrawal bid c.
Hence the operational MCE defines the pipe network mass balance as6:
X X X X
0¼ Stm Dtc UDtn þ FOtl FIlt þ slack : 8n; t (1)
m:nðmÞ¼n c:nðcÞ¼n l:dðlÞ¼n l:oðlÞ¼n
Here the expressions n(m) and n(c) respectively indicate the node associated
with injection offer m and withdrawal bid c. Given each pipe segment has an origin
node and a destination node, the MCE can model pipe segments as allowing flow to
only move from the origin to the destination; the MCE can also allow for reversible
flows by allowing the origin and destination pressures to be reversed. Although not
evident in (1), the formulation also differs from that in Read et al. in that it is
expressed in terms of energies not gas masses.
Extra variables, UD, are added to represent a base load of “uncontrollable”
demand. This accounts for a large proportion of total market demand. Here, and
elsewhere in the formulation, the “slack” term indicates high priced violation
variables, which ensure that the optimization can still solve, if all constraints on
the problem cannot be satisfied.7 This allows the model to produce useful solutions
6
slack is a variable, yet it is also a vector, and as such it is stated in lower case.
7
Although this terminology is common in such models these are not really “slack” variables in the
traditional sense. They do indicate how far the final solution point is from the constraint, but it lies
outside the feasible region, not inside. Thus they have the form of generalized slack variables,
which are positive when the constraint is slack, but negative when it is violated.
Implementation of a Scheduling and Pricing Model for Natural Gas 13
in that situation, giving the dispatch schedule that most nearly meets all require-
ments, and indicating the location and extent of potential violation.
Read et al. introduce specific gas flow relationships for segments within a
pipeline, describing pressure/flow relationships, and also the way in which those
relationships will change over time. It can be seen that, through a series of
substitutions, flows along any length of pipeline will be related to the change in
pressure across the length of the pipe. For the operational formulation; the flow into
pipe segment l, in period t, is represented as a nonlinear function, Qo, of the origin
and destination pressures, PBtoðlÞ and PBtdðlÞ respectively, for that pipe segment.
Thus we implicitly define the pipe inflow as:
The flow out of pipe segment l is defined in a similar fashion, being dependent
again on the origin and destination pressures. The function Qd differs from Qo in that
it also accounts for compressor fuel usage for pipe segments that have compressors,
and the net change in useable linepack in the pipeline from period t-1 to t. If we let
KNlt:t¼0 be the linepack in pipe segment l measured (or extrapolated) at the start of the
optimization, we define flow out of the pipe segment via:
FOtl will tend to differ from FIlt because of changes in linepack stored within the
pipe segment or, in the case of compressors, because of fuel used in the segment. FIlt
and FOtl will be non-negative for directional pipe segments, but may be negative for
reversible pipe segments. Useable linepack in pipe segment l can also be expressed
as a nonlinear function of the origin and destination pressures. If a pipe segment has
a pressure regulator, then the origin pressure is modified by a throttle variable, TRtl .
Regulators are modeled such that they are situated between an origin node and the
corresponding pipe segment l, and are subject to a minimum regulator outlet
pressure limit, opl . If the pressure at the node is PBtoðlÞ then this is the inlet pressure
to a regulator, but the outlet pressure of the regulator, and hence the inlet pressure of
the attached pipeline is PBtoðlÞ TRtl . With TRtl set to zero for pipe segments without
regulators the general equation for linepack on a pipe segment is
In addition, if “reg” denotes the set of pipe segments with regulators then for
those pipe segments that have regulators the following limit applies.
including an aggregate system linepack limit for the whole system, Zone 0. We can
add “slack” variables to allow targeted end of day minimum linepack constraints
for a given zone, mz , to be violated at a high penalty cost, and also a variable MLz
that discourages, but does not prevent, end-of-day linepack exceeding targeted
levels – it is non-negative and has a small penalty cost, ps, applied to it. So we have:
X X
mz ¼ KNlT þ ðFIlt FOtl Þ MLz þ slack : 8z (6)
l2z t;l2z
Read et al. [1] discuss a number of further constraints that may be imposed on
pressures in pipeline segments, or at the end of segments, or on pressure differences
(and hence flow velocity) across segments. Here we simply represent these
constraints generically, in terms of the underlying pressure variables, as:
Note that while Read et al. discuss the possibility of trading end-of-day linepack,
it is not explicitly represented in this objective because it was decided that such
trading was not appropriate. There is, in fact, only a small range of end-of-day
linepack levels that will leave the system in a state from which the next day’s supply
requirements can reliably be met. Thus linepack requirements are specified by the
System Operator, via constraint (6).
Further, the offer/bid terms here have been expressed in a very generic way, not
necessarily tied to particular hours as in Read et al. In fact, when run to determine
market prices, using the simplified infinite tank assumption as in Sect. 8, the same
8
Market participants submit supply offers to the market, these ‘bids’ are composed of up to 10
price and quantity tranches, where we call each individual tranche a ‘step’.
Implementation of a Scheduling and Pricing Model for Natural Gas 15
offers and bids are assumed to persist over the entire remaining gas day, and
corresponding (pseudo-)dispatch schedules will be produced. Conversely, in real-
ity, injections and withdrawals at particular points in the PTS may be constrained by
a number of constraints not discussed by Read et al. These include physical
constraints, such as maximum hourly quantities, ramp limits, and restrictions on
the frequency and timing of changes to gas flow rates, and constraints implied by
market rules. For example, where possible, the rules require pro-rating the
schedules of tied offers or tied bids.9
Thus we describe supplier offers and consumer bids by their aggregated injec-
tion or withdrawal price and quantity step combinations. But these must also be
related to the hourly injection/withdrawal variables employed in the gas flow
modeling representation above. Rather than detail those relationships here, we
will express them in a very general way, using a set of generic “Market Constraints”
(MC), defining relationships between uncontrollable demand in each period, UDtn ,
the supply and demand scheduled to be dispatched in each period from offers and
bids, Stm and Dtc , and the aggregate quantities cleared from those bids and offers, Sms
and Dcd 10:
MCðUDtn 8n; t; Stm 8m; t; Dtc 8c; t; Dcd 8c; d; Sms 8m; s; slackÞ ¼ 0 (9)
The Operational Model is built around a physical gas flow model, as was the
conceptual formulation developed by Read et al. [1]. But Read et al. basically only
stated that the nonlinear gas flow/linepack relationships should be linearized, without
specifying how that should be done in practice. Read and Whaley [16] did explore a
linearization approach based on analytical differentiation of the theoretical gas flow
equations, but the implemented MCE described here adopted a very different, and
more pragmatic, approach. In this section we discuss linearization of the basic gas
flow model without considering specialized equipment like compressors. We begin
by developing the relationship between flow rate and the pressure change across a
pipe segment, a relationship which is then employed within both piece-wise and
successive linear representations of the gas flow problem.
The pipeline system is divided into a number of pipe segments where the
physical characteristics of the pipe segments and constraints on the pipe segment
are relatively uniform within the pipe segment. This includes the pipe
9
The “slack” variables applied to such constraints, have very small penalties, so as to encourage,
but not force, tie-breaking.
10
Parameters associated with a number of these constraints limits/bounds may also be scaled in a
pre-processing step so as to resolve conflicts between quantities of gas previously scheduled and
quntities actually observed, for example.
16 W. Pepper et al.
segment diameter and minimum and maximum constraints on pressure in the pipe
segment. The flow of gas along a pipe segment is a direct function of the average
pressure of the gas in the pipe segment and the pressure decline from one end of the
pipe segment to the other, and hence of the inlet and outlet pressures, as indicated
by Eq. 2 and 3 for FI and FO above. But that relationship is nonlinear, and
approximately related to the square root of the average pressure times the pressure
decline.
The allowable pressure drop along a pipe is actually nonlinear, but our aim is to
develop an LP based formulation. For a short enough pipe segments, and time
intervals, an approximately linear loss function could be considered accurate
enough, as suggested by Read et al. [1], and this strategy is employed by iterative
pipeline simulation models such as Dorin and Toma-Leonida [17]. But our model
of the Victorian Gas System use a piece-wise linearization of performance over
longer pipe segment lengths, and has been calibrated to within acceptable
tolerances against a more general physical gas flow model. Similar experience
was recently reported by Midthun et al. [13], whose linearised approximation, a
Taylor series expansion about multiple-points, was deemed acceptable for work
undertaken for North Sea gas system users and its operator, Gassco.
The gas flows within the pipe segments are derived assuming isothermal
conditions, uniform pipeline diameter and surface conditions, and a constant
altitude (sea level). While not described in this paper, the MCE includes pre-
processing steps to translate measured pressures and pipeline pressure limits at
pipeline altitude to sea-level equivalents. It scales flow and linepack equations to
reflect the higher density of gas at sea level, and then converts pressures derived
from the optimization back to be applicable at the relevant altitude. The Appendix
contains the detailed equations used to estimate flow rates and linepack, given
specified values of the (sea-level) pressure at both ends of a pipe segment. The
MCE went through a rigorous calibration exercise during its development, and the
model results for pressures closely match measured pressures from actual gas days
given the initial conditions on the pipeline system, hourly injections and
withdrawals, and final linepack on a given gas day.
Pipelines, valves, and compressors control the flow of gas between injection/
withdrawal nodes. Additionally, a pipe segment stores gas, so a key parameter is the
linepack stored at the end of each hour and the change in linepack over the hour,
which impacts the flow from the pipe segment. The gas storage in a pipe segment is
primarily a function of the pressures in the pipe segment and its length and diameter
(See Appendix).
For pipe segments in which the flow direction is specified, this pressure delta
must be non-negative and less than a maximum pressure delta. But, for most pipe
segments, the pressure decline required to produce maximum allowable flow rates
is significantly less than the difference between the maximum allowed inlet pres-
sure and the minimum allowed outlet pressure. So the MCE allows for a user
specified maximum pressure decline on each pipe segment set, so that the estimated
flow rate corresponding to that decline will be no less than the maximum flow rate
expected at any time on the pipe segment. Limiting the allowable pressure decline
Implementation of a Scheduling and Pricing Model for Natural Gas 17
in this way allows the MCE to provide a more accurate linear representation of the
nonlinear pressure flow relationship, by linearizing over a narrower pressure range.
For bi-directional pipe segments the same logic applies, but the roles of the origin
and destination nodes are expanded to allow for the possibility that the pressure
conditions will be reversed.
k (11)
P
where aji 0 8j; i and aji ¼ 1
i
But note that other characteristics, such as the destination pressure, the steady
state gas flow (FO and FI), and the linepack stored in the pipe (KN) can be
determined as functions of PO and DP, as described by the Appendix. Thus,
while the location of each point shown in Fig. 3 is defined with respect to only
two variables (Pj(o), DPj), it may have associated with it a vector giving the values
for any number of characteristics that can be calculated as a function of those two
variables. We denote these characteristics by xkj for k ¼ 1,. . .K. We can then create
an effective piece-wise linear representation of any one of these characteristics by
applying the same weights used to represent (PO(j), DPj) in terms of the underlying
grid point values.
X
xkj ¼ aji xkji 8j; k (12)
i
Thus the generalized functions stated in Eqs. 2, 3 and 4 for flow in and out of a
pipe segment, and linepack in that segment, are actually defined by essentially
empirical value determined for a set of grid points. And the fundamental LP
variables in this representation are not actually flows, pressures etc, but the
11
PS indicates the number of “pressure states”, although actually there are PS + 1, including
state 0.
12
For a 2 dimensional representation like this, every point can actually be defined as a weighted
sum of three particular grid points. But, in order to define a convex feasible region, we need to
leave the model to determine which grid points it prefers to use.
13
Together, these two constraints actually mean all weights must lie in the range [0,1].
Implementation of a Scheduling and Pricing Model for Natural Gas 19
weighting variables, aji. This allowed the model to be implemented quickly, and
readily tuned to match observed empirical relationships. Subsequently the same
approach was used to improve modeling of losses on the HVDC inter-island link in
the New Zealand electricity market clearing engine, and to model losses on all
transmission lines in the Singapore market clearing engine. While it is not immedi-
ately obvious whether this type of formulation will increase or decrease computa-
tional times, it has proven at least competitive with more traditional piece-wise
linearization in the latter context. And here it has the advantage of increasing the
length of pipeline segments, and hence reducing their number, relative to using
explicitly defined gas flow equations of the type proposed by Read et al.14
14
Another potential advantage of the convex combination approach to piece-wise linearization is
that, if and when convexity issues arise, they could be dealt with by employing an integer type
formulation, employing Special Ordered Sets (SOS2) as in Martin et al. [14], to force the model to
apply weights only to adjacent grid points. A similar strategy is mentioned by DeWolf and Smeers
(2000a), but dismissed on computational efficiency grounds. Still, that approach has been applied
successfully in modeling HVDC link losses in the New Zealand electricity market model.
20 W. Pepper et al.
the system earlier in the day so that it is available as linepack during the
constrained period. Or they can occur when the system is reaching minimum
pressures at one point on the system but the pressure/flow equations imply that
upstream gas will be diverted to another direction due to pressure differentials in
that part of the system.
Cases where the MCE will want to solve in the shaded area or on the bottom line
are not uncommon. They can occur when the pipeline system will be constrained
during an hour later in the day and it would be beneficial to send gas to some part of
the system earlier in the day so that it is available as linepack during the constrained
period. Or they can occur when the system is reaching minimum pressures at
one point on the system but the pressure/flow equations imply that upstream gas
will be diverted to another direction due to pressure differentials in that part of the
system.
There can also be an issue when negative prices arise, but this is not common.
They occur when either the bid or pipeline constraints force too much gas into the
system, or where the system is constrained during the day, perhaps during the
evening peak. If the bid and pipeline constraints are treated as being inflexible, they
can end up forcing more gas into the PTS during the night, thus producing end-of-
day linepack greater than AEMO would want, and has specified. Compressor fuel
use can seem desirable to the optimization during this period, simply to use up
negatively priced gas, but the impact is usually limited by constraints on pressures
and flows.
Whatever the cause may be, the MCE implements a successive iteration process
to correct the problem whenever the LP optimization recommends solutions that are
not physically feasible. At the first iteration, the problem is solved using piece-wise
linearization, with no special features added. This might produce the iteration one
solution shown in Fig. 4, where the modeled flow is ‘Not Realistic’, and less than
Implementation of a Scheduling and Pricing Model for Natural Gas 21
the flow that would actually occur, given that pressure delta. In successive iteration
mode, the MCE will then do three things:
• Restrict the inlet pressures used in the model from PS4 to PS3 based upon the
three inlet pressures that are closest to the solution
• Restrict the pressure deltas included in the model from PS5 to PS3 based upon the
three pressure deltas closest to the solution
• Create a slope line that restricts the flow as a function of the pressure
All of these are illustrated in Fig. 5 below, where the choice of three inlet
pressures and three pressure deltas allows flexibility of the model to adjust if one
of the constraints (let’s say pressure delta) on one pipeline segment causes the inlet
pressure required on another pressure delta to adjust up or down, possibly making
the constraints on that pipeline segment infeasible. Small deviations from the
constraint line are allowed without penalty. But larger deviations are subject to a
small cost penalty. Additional system wide constraints limit the total deviation
allowed across all pipe segments. These constraints can be relaxed at subsequent
iterations if found to be overly restrictive. The stopping condition is that the
deviation is small enough, or that a maximum of 14 iterations are completed.15
Also, in each successive iteration, the three inlet pressures and three pressure
deltas are reviewed to see if they are still appropriate, and alternatives chosen if
required.
15
Early experience with the model showed that most cases solved within 14 iterations. Also, cases
which did not solve showed negligible improvement in convergence after 14 successive
approximations.
22 W. Pepper et al.
5 Modeling Compressors
Pipelines move gas from node to node, and through the process of transporting the
gas there are pressure losses to the gas. Unlike gas pipeline segments, compressors
move gas from a lower pressure state to a higher pressure state, while consuming
gas in the process. Depending on the inlet gas pressure, a compressor can be run
over a range of speeds (rpm) to achieve various combinations of gas flow and
pressurization (pressure delta), and hence of outlet pressures and gas consumption.
We can think of the feasible region as being defined by three primary (LP)
variables: Flow, Inlet Pressure, and Outlet Pressure, or equivalently, Pressure Delta.
Figure 6 simplifies this representation by projecting the feasible region down on
to only two dimensions, Flow and Pressure Delta, with Inlet Pressure and/or Outlet
pressure being only implicit. This representation allows us to define the maximal
feasible region over which the compressor could operate, in terms of the extreme
low/delta combinations that can be achieved at various rpm, if the pressure (and
hence the density) of gas flowing through the compressor were also allowed to vary
all the way from its maximum (typically limited by the maximum output pressure
pout ) down to its minimum (typically limited by the minimum input pressure pin ).
Our approximation methodology requires identifying performance at a number of
key points, both within the (3-dimensional) range of variations, and at its corner
points. Figure 6 shows some of the key characteristics of each compressor, in terms
of their projection onto this 2-dimensional representation:
16
In practice, it is usual for the axes to be reversed to match compressor manufactures data relating
to efficiency curves, however we present them in this orientation for consistency
Implementation of a Scheduling and Pricing Model for Natural Gas 23
17
There are actually two null operational states one for the minimum inlet pressure, and the other
for the maximum outlet pressure, although they appear as one point in Fig. 6
18
In theory, operating states between zero and the minimum operating level could actually be
achieved, on average, by operating efficiently for only part of the time. If so, the convex
approximation to performance in that region could actually be valid. Hydro generators, for
example, can be validly operated, and represented, in that way. In this case, though, the savings
in operational efficiency would probably be outweighed by increased startup/shutdown costs.
24 W. Pepper et al.
ol ol
ð1 NS p in NSp out Þ2 (13)
cp cp
In the LP, the compressor operating level (i.e. its power requirement and hence
its gas consumption), is determined by linear interpolation between the values
determined by the set of points for which it has been pre-computed. Once the LP
solution is known, though, the operating level can be better approximated using the
quadratic formula:
ð1 NS p in NSpout Þ2 cp (15)
19
This lower limit here defines the minimum running boundary of the feasible region if we force
the compressor to be “on”, thus eliminating the null states. As shown, this boundary is actually
non-convex, but it can reasonably be approximated by a straight line, at the cost of eliminating a
relatively small set of operating states that are seldom utilized, in practice.
Implementation of a Scheduling and Pricing Model for Natural Gas 25
The gas pipeline system in Victoria uses pressure regulation valves at a number of
key points in the system, and check valves to regulate flow and optimize the daily
operation of the system.
A pressure regulator controls the decline in pressure across it, while maintaining
the rate of flow consistent with that for the corresponding supplying and receiving
pipe segments.20 A pressure regulator often has a maximum pressure at its outlet,
which can be pre-set at the valve, and in most cases this value is also the desired
pressure to be achieved. The MCE models the pressure regulator and the pipe
segment at the outlet of the regulator as a single entity with both a maximum and
minimum pressure at the outlet of the regulator. The pipe segment inlet pressure is
allowed to be less than the inlet node pressure, but must be between the maximum
and minimum outlet pressure of the regulator.
Check valves generally constrain flow across them to a single direction, their
operation depends on their being a positive pressure difference between
corresponding inlet and outlet pipeline segment pressures. Check valves are often
installed with by-pass lines, or can be manually and temporarily by-passed, the
model includes such by-passes to allow reverse flow when the pressure difference is
in the opposite direction. Although such an abnormal-operation attracts a high
penalty cost. The modeling of a check valve is more complex. We model this by
determining the flow and pressure delta combinations that occur at each combina-
tion of maximum and minimum inlet and outlet pressures and corresponding flow,
or lack of, across the check valve. This results in there being five potential limiting
operational points, as shown in Table 1. The four points labeled F allow the MCE to
cover the range of possible forward flows and pressure combinations, while the
three points labeled C allow the MCE to represent zero flow at any combination of
pressures where the inlet pressure is less than the outlet pressure (that is, where the
flow is “checked”).
The convex combination approach to piece-wise linearization is then applied, as
above. The problem in modeling check valves is that some linear combinations of
the weights applied to pressure flow states can produce physically infeasible results,
20
In other words, the pressure can decline significantly without implying the increase in flow rate
one would otherwise expect for such a decline.
26 W. Pepper et al.
as for the pipe segment flow equations discussed earlier, but much more severe. We
address this problem, if necessary, using the successive iteration logic discussed
earlier.
Participants in the Victorian Gas Market can make injection bids to supply gas, or
withdrawal bids to buy gas, or specify uncontrollable withdrawal, being gas which
will be purchased at any price. Injection bids are associated with supply points, and
withdrawal bids and uncontrollable withdrawal are associated with withdrawal
points, represented as “nodes” in the network. AEMO can over-ride the aggregate
uncontrollable withdrawal if the cumulative participant forecast differs signifi-
cantly from its own. AEMO also controls how the aggregate uncontrollable with-
drawal is profiled across the network and across the day. These profiles can be
varied during a day as weather conditions or observed demand patterns change. In
combination, the following features allow the Operational Model to more closely
represent how gas is physically injected into the system over the gas day. Collec-
tively they define the constraint set referred to as MC in Eq. 9.
Bids include ten price-quantity steps as well as a response time, an expiration
time, and ramp limits. While participants can freely submit revised price-quantity
steps between daily scheduling intervals, these other parameters must be approved
by AEMO and are only changed infrequently. There may be multiple participants
trading at a supply point. For instance, there are multiple participants bidding to
supply gas to the market at the Longford supply point. This supply point represents
a physical gas production facility. A set of constraints can be imposed on aggregate
supply or demand at a point, and these may over-ride constraints on individual bids
at those points (with logic included to resolve conflicts). Supply and demand point
constraints can impose minimum hourly flow rates, maximum hourly flow rates,
minimum daily flow rates, maximum daily flow rates, response times, expiration
times, and ramp limits.
Ramp limits restrict the rate of increase and decrease in schedules between
hours. A response time can be set for each hour during the gas day and indicates
how long before that time a schedule must be issued for the participant to match that
schedule. Thus if response times are 2 h for all hours after 10 a.m., and a schedule is
issued at 6 a.m., but then revised at 10 a.m., the 10 a.m. schedule will still match the
schedule issued at 6 a.m. for the first 2 h after 10 a.m., with the schedule only
changing to match the 10 a.m. market-clearing dispatch schedule from noon on.
The expiration time is a time during the day after which the participant will no
longer respond to re-schedules. A participant with an expiration time of 8 a.m.
could be scheduled for the day at 6 a.m., but would then retain that same
schedule for the remainder of the day, irrespective of how prices, and other parties’
dispatch schedules, may change when the market is subsequently re-cleared during
the day.
Implementation of a Scheduling and Pricing Model for Natural Gas 27
The Victorian Gas Market also has a form of capacity right called Authorized
Maximum Daily Quantity (AMDQ). This reflects an amount of gas that a partici-
pant can supply at one point, or receive at one point, without being deemed to have
“caused” any constraints.21 Withdrawal bids or injection bids within a participant’s
AMDQ limits are given priority in scheduling if tied with withdrawal bids or
injection bids with the same price that are not covered by AMDQ. The market
model also allows for directional flow constraints. These can be imposed across a
supply point and withdrawal point so as to constrain the net flow. This can be useful
in modeling gas storage fields which can only either take in gas or release gas each
day. The directional flow constraint allows limits to be imposed on the net supply
from a group of points. Minimum hourly net flow rates, maximum hourly net flow
rates, minimum daily net flow rates, and maximum daily net flow rates can be
implemented.
A problem that existed in the early history of the market was that the Operational
Model was not fully reflecting how gas facility operators managed the flow of gas.
They typically manage gas flow at a constant hourly rate, changing rates several
times during the day. The Operational Model imposes small costs on ramping to
encourage flat scheduling, but this feature failed to account for the specific times at
which schedules could be changed. These could be over-ridden when strong
economic incentives existed to encourage more varied scheduling. While it might
be thought desirable to have more variable flow rates, the operators of gas produc-
tion and supply facilities are not generally participants in the market, and ignoring
their physical operating characteristics in scheduling has the potential to create
costs for participants which are beyond their control (e.g. for not following their
schedule). Thus the MCE could produce schedules which were not implementable
in the real-world, and in extreme cases could create system security risks.
This problem has been largely resolved in recent years by allowing bids and
supply and demand point constraints to have defined flexibility limits. If flow at a
bid or pipeline point is specified to be completely inflexible, then the optimization
will determine an hourly flow rate which is constant over the time window for
which inflexibility applies. If an aggregate supply point flow is inflexible, but the
injection bids there are flexible, then the total supply through that pipeline point
must maintain a constant rate, but the participant schedules at that pipeline point
can vary, while maintaining the same total flow.
Inflexibility restrictions are relaxed for the times at which major constraints,
such as minimum and maximum hourly flow limits or directional flow constraints,
change, and for a sufficient period after that, so as to allow prior schedules to be
followed within the response time, and then for the dispatch schedule to be ramped
to minimum or maximum hourly schedule quantities, given the applicable ramp
limits. In practice, most ramp rates are quite fast, so these intervals of flexibility can
be quite short.
21
This being important because it determines who faces penalties and/or receives constrained-on/
off payments when the actual dispatch schedule differs from the market trading schedule.
28 W. Pepper et al.
Each time it is run, the Market Model determines the single market price to be
applied to trades (including deviations from trades cleared earlier) over the remain-
der of the gas day. It also determines the set of injection and off-take schedules that
would occur if there were no limits on the ability of the gas transmission system to
move gas to, or from, any node at any time within the optimization horizon. The
differences between these unconstrained schedules and the schedules produced by
the Operational Model runs, over the same optimization horizon, are used to
determine compensation for the costs created by the constraints.
The Market Model LP is structured similarly to that for the operating schedule
with a few rather significant exceptions:
• All injections and withdrawals are treated as if they were at a single node so that
there are no constraints modeled on the ability to flow gas between locations.
Consequently there is no representation of pipelines, compressors, check valves or
regulators, although pipeline point constraints and directional flow constraints –
which are applied at injection and off-take points, not pipe segments – still apply.
• With no pipeline pressure-flow relationships modeled, successive iterations is
not required.
• As pipe segements are not modeled A system wide total initial linepack is used
rather than pipe segment specific linepack
• The market schedule imposes no upper or lower limits on linepack levels during
the day, other than an end-of-day minimum linepack constraint.
• The end-of-day minimum linepack constraint differs slightly from that used in
the operational schedule formulation. Compressors are modelled in the Opera-
tional Model, and consume gas, but are not modeled in the Market Model. To
ensure that each model has the same change in linepack over the day, and hence
the same pricing in an unconstrained case, the compressor fuel usage determined
in the Operational Model is subtracted from the minimum linepack limits used in
the Operational Model but not from the minimum linpeack limits in the Market
Model. As both models include cost penalties to discourage linepack above
minimum, these minimum levels are effectively target end-of-day linepack
levels. When the Operational Model schedules gas supplies run compressors
then it must supply correspondingly less gas to achieve the end-of-day linepack
target, ensuring that total supply over the day matches that in the Market Model.
9 Solution Methods
The MCE is currently executed as a set of LP problems solved with CPLEX version 9.1.
Early versions of the MCE developed in the late 1990s solved the problem using the
Simplex method, as this is well suited to reliably and accurately determining prices
in markets. The MCE solved the problem up to 14 times in executing success
Implementation of a Scheduling and Pricing Model for Natural Gas 29
iterations for the Operational Model. Even using an advanced basis for each itera-
tion, the Simplex method was found to perform too slowly on some problems, taking
several hours using the technology of the day, so a switch was made to using the
barrier method.
The problem then encountered was that the barrier method had difficulty
maintaining adequate numerical accuracy to cope with the range of numerical
values used in a typical MCE problem. This is because the MCE employs a
hierarchy of penalties; which have been adjusted over the years to minimise
inappropriate interactions between them. Within this hierarchy there are many
very small penalties in implementing tie-breaking, penalizing infeasible flow
rates during successive iterations, and encouraging some other desirable outcomes.
Since the highest market price is $800/GJ, all infeasibility penalties are set greater
than this. Penalties on operator configurable constraints, which can be modified if
infeasible, typically have penalties in the region of $3,000/GJ and $4,000/GJ.There
are also some very large penalties to address physical infeasibilities. For example,
the mass balance constraint, Eq. 8, has a violation penalty value of $9,999.9/GJ. As
at 2010, the smallest penalty used is $2 106/GJ while the largest used is
$10,099.9/GJ, a variation of nine orders of magnitude.
To resolve the precision problem, while using the barrier method, an additional
three iteration phases were introduced to both the Operational Model and the
Unconstrained Market Model. Broadly, the first phase involves solving the problem
with only the large penalties activated, and all small adders removed. This
determines physically feasible aggregate quantities of gas to schedule over the
day. The second phase fixes the aggregate schedules over the day, thus not requiring
large penalties, but uses small penalties to allocate the gas flows optimally across
time. Both phases 1 and 2 are performed for each successive iteration. Finally,
Phase 3 is only performed at the completion of the problem – it involves effectively
fixing the problem to a tight region around the optimal solution, with some minor
penalty costs set to zero, and re-solving the problem using the simplex method to
determine nodal prices (for the Operational Model) and the single market clearing
price (for the Unconstrained Market Model).
The market timelines require that the software be able to reliably produce a
schedule in the time between the window for bids and offers closing and the
deadline for publishing schedules. While this time frame is about 1 h, both the
Operational Model and the Unconstrained Market Model must be solved, and it
may be necessary to solve multiple schedules in that time to allow the operators of
the system to modify compressor commitments and other constraints under their
control to correct issues seen in prior solutions. At the commencement of the market
the Operational Model could be solved, including data input and output processing,
in approximately 15 min using the simplex approach and earlier versions of
CPLEX. The Unconstrained Market Model, which was then solved after the
Operational Model only took seconds to solve.
Since 2007 the Unconstrained Market Model has been solved before running the
Operational Model, providing the operators with insights about the schedule before
needing to determine compressor commitments to be used in the Operational
30 W. Pepper et al.
Model. Today, using CPLEX 9.1 on faster workstations, and despite a significantly
more complicated PTS than existed in the late 1990s, the Operational Model can
be solved within 3 min while the Unconstrained Market Model solves well within
1 min.
10 Conclusions
The Market Clearing Engine described here has been used in the Victorian Gas
Market since 1999, and demonstrates the practical use of optimization techniques to
schedule dispatch, and determine prices, for a complex gas market. In principle it
could be used to support market trading based on determine hourly nodal prices, as
originally proposed by Read and Whaley [16], and explained by Read et al. [1]. In
practice this has not eventuated and the market has evolved along somewhat
different lines. This has occurred for various reasons, but it should be clear that it
is not because it proved impossible to determine hourly nodal prices, since the MCE
model does, in fact, determine such prices. Nor is it because the prices determined
by the MCE model are always the same, everywhere in the network, as may be seen
from the example discussed by Read et al. Thus while, in Victoria, the full nodal
version of the MCE has proved most successful as a dispatch optimization tool, a
model of this type could also be used to clear markets and support trading, and this
paradigm may well prove more beneficial elsewhere, where congestion is more
prevalent, and there is greater economic value at stake.
In this Appendix we present the key equations for deriving natural gas flows and
pressures in the Victorian pipe network. The derivation is based on six initial
equations described in Eqs. 16, 17, 18, 19, 20, 21) below.
Pl ¼ rl RTz (16)
The ideal gas law equation 1 describes pressure, Pl at a point l along a pipeline of
length 0 < l < L. R is the ideal gas constant in units of kPa*m3/( K kg), T is the
temperature of the pipeline in K, and z is the supercompressibility of gas, while rl
is the density in units of kg/m3.
q l ¼ rl v l A (17)
Implementation of a Scheduling and Pricing Model for Natural Gas 31
The gas flow rate ql at a point l, in kg/s, is described in (17). Where A is the pipe
cross-sectional area in units of m2, nl is the velocity of gas in m/s in the pipe at point l,
at the point l measured at sea level.
dP f
1000 ¼ r v2 (18)
dl 2D l l
rl vl D
Re ¼ (19)
m
The Reynolds Number, Re, is defined by (19). In this equation m is the viscosity
of gas measured in kg/ms.
!0:25
0:316 1 fturb 4
f ¼ þ (20)
Re 0:316
The Fanning friction factor formulation in (20) utilizes the Blasius formulation
[18], where Z is the pipeline efficiency (a fraction). It describes friction as gas flows
along a pipe. This makes use of a turbulent friction factor, fturb.
e 0:35
fturb ¼ 0:0053 þ 0:1662 (21)
1000D
Smooth pipelines have no turbulence while rough pipelines have more turbu-
lence. Turbulence becomes more important as the Reynolds Number increases.
Given e, a measure of the roughness of the pipeline in mm, then the relative
roughness of a pipeline can be defined with respect to its diameter as (e/D).
Using empirical data the form of the turbulent friction factor shown in (21) was
developed by gas system engineers working on the development of the MCE. Now,
using Eq. 17 to substitute for nl in (18) we get:
dP f q2
¼ 2 l 1000 (22)
dx D A rl 2
dP 1
¼ 0:5Gf q2l (23)
dx Pl
32 W. Pepper et al.
Where:
fRTz
Gf ¼ (24)
1,000DA2
By integrating (23) with respect to l and observing that for l ¼ 0, Pl ¼ Po, the
origin pressure of the pipeline gives:
0:5
Pl ¼ P20 Gf q2l l (25)
Using (25) to define the value of Pl where l ¼ L, i.e. the destination pressure,
and assuming a constant flow (ql ¼ q), friction factor, and supercompressibility,
then the steady state flow rate can be derived as:
2 0:5
P0 P2L
q¼ (26)
Gf L
dPl
Pl ¼ 0:5G q1:75
l (27)
dl
Where:
0:316m0:25 RTz
G= (28)
1,000ZDia1:25 A1:75
Integrating (27) with the assumption that the flow is constant along the pipe
segment and the requirement that for l ¼ 0, Pl ¼ Po (the origin pressure) we get:
0:5
Pl ¼ P20 Gq1:75 l (29)
Using (29) to define the value of Pl where l ¼ L, i.e. destination pressure, then
for a non-constant friction factor and constant supercompressibility the steady state
flow rate can be derived as:
2 ð1=1:75Þ
P0 P2L
ql ¼ (30)
GL
Implementation of a Scheduling and Pricing Model for Natural Gas 33
ðL ðL
Pl A
I ¼ rl Adl ¼ dl (31)
RTz
0 0
ðL
A
I¼ Pl dl (32)
RTz
0
We also have22:
dFðlÞ ¼ Pl dl (33)
A
I¼ fFðLÞ Fð0Þg (34)
RTz
Substituting the expression for Pl from (25), which was based on a constant
fraction factor, into (33), then integrating over l, we get, for a non-zero flow rate:
0:5
dFðlÞ ¼ Pl dl ¼ P20 Gf q21 l dl (35)
Given the rule (f(x))n differentiated by x gives n df/dx f(x)n-1 then we can
integrate dF(l) by the reverse transformation to give:
1:5
P20 Gf q2 l
FðlÞ ¼ (36)
1:5Gf q2
Evaluation of (36) for F(0) and for F(L), and substituting these into (34) gives:
A P30 P3L
I¼ (37)
RTz 1:5Gf q2
22
Energy ¼ Work ¼ Force Displacement and Pressure ¼ ForceArea; so combining these
two results in Pressure Length ¼ Force DisplacementArea which then equals the flow of
energy past a point, which is WorkArea ¼ EnergyArea
34 W. Pepper et al.
Further, substituting for q from (26) gives linepack for non-zero flow of:
3
AL P0 P3L
I¼ (38)
1.5RTz P20 P2L
Where the flow rate is zero, then Pl ¼ P0 for all l, and (16) implies
AL AL P0
I ¼ P0 ¼ (39)
RTz RT z
AL Pa
I¼ (40)
RT z
Hence
2 P30 P3L
Pa ¼ 2 : if P0 > P L (41)
3 P0 P2L
Or:
Pa ¼ P0 : if P0 ¼ P L (42)
Given user defined values of typical low and high pressures in the system, Plow
and Phigh, and corresponding supercompressibility values zlow and zhigh it is possible
to define an average supercompressability za as:
ðPa Plow Þ
za ðPa Þ ¼ zlow þ ðzhigh zlow Þ (43)
ðPhigh Plow Þ
References
1. Read EG, Ring BJ, Starkey SR, Pepper W (2012) An LP formulation for a natural gas market.
A. Sorokin et al. (eds.), Handbook of Networks in Power Systems II, Energy Systems,
Springer-Verlag Berlin Heidelberg
2. Ruff LE (1997) Victorian Gas Market Clearing Logic Version MCL 1.2. Released by Energy
Projects Division, Department of Treasury and Finance, 30 October.
3. Pepper W (1999) AEMO market clearing engine: mathematical formulation. Version 1.20.4
Draft Report by ICF Consulting to VENCorp (now AEMO)
4. Zhu G, Henson MA, Megan L (2001) Dynamic modeling and linear model predictive control
of gas pipeline networks. J Process Control 11(2):129–148
5. Johnson AT, Marquart BC, Istre ML, Walloopillai RK (2000) Integrating an expert system and
pipeline simulation to enhance gas pipeline operation, profitability and safety. www.psig.org/
papers/2000/0007.pdf. Accessed 23 Nov 2010
6. Hogan WW (1992) Contract networks for electric power transmission. J Reg Econ 4
(3):211–242. ICFI (2010) ICFI gas market model. www.icfi.com/markets/energy/doc_files/
nangasweb.pdf. Accessed 28 Oct 2010
7. AEMO (2010) National gas rules, Part 20 at http://www.aemc.gov.au/Gas/National-Gas-
Rules/Current-Rules.html. Accessed 2 Sep 2010
8. Zheng QP, Rebennack S, Iliadis N, Pardalos PM(2010) Optimization models in the natural gas
industry. In: Rebennack IS, Pardalos PM, Pereira MVF, Iliadis NA (eds) Handbook of power
systems I. Springer-Verlag Berlin Heidelberg, pp 121–148
9. Wu S, Rios-Mercado RZ, Boyd EA, Scott LR (2000) Model relaxations for the fuel cost
minimization of steady-state gas pipeline networks. Math Comput Model 31:197–220
10. De Wolf D, Smeers Y (2000a) The gas transmission problem solved by an extension of the
simplex algorithm. Manag Sci 46:1454–1465
11. De Wolf D, Smeers Y (2000b) The simplex algorithm extended to piecewise-linearly
constrained problems. http://citeseerx.ist.psu.edu/viewdoc/summary?. doi¼10.1.1.14.8090.
Accessed 23 Nov 2010
12. Tomasgard A, Rømo F, Fodstad M, Midthun KT (2007) Optimization models for the natural
gas value chain, In: Hasle G, Lie KA, Quak E (eds), Geometric modelling, numerical simula-
tion, and optimization: applied mathematics at SINTEF, Springer-Verlag Berlin Heidelberg,
pp 521–558
13. Midthun KT, Bjørndal M, Tomasgard A (2009) Modeling optimal economic dispatch and
system effects in natural gas networks. Energy J 30(4):155–180
14. Martin A, Moller M, Moritz S (2006) Mixed integer models for the stationary case of gas
network optimization. Math Program Ser B 105:563–582
15. Van der Hoeven, T. (2004), Math in gas and the art of linearization, PhD thesis International
Business School and Research Center for Natural Gas, Groningen, The Netherlands
16. Read EG, Whaley R (1997) A gas market model for Victoria: dispatch/pricing formulation.
Report by Putnam, Hayes & Bartlett – Asia Pacific Ltd. to VENCorp (now AEMO)
17. Dorin B, Toma-Leonida D (2008) On modelling and simulating natural gas transmission
systems (part I). J Contr Eng Appl Infor 10(3):27
18. Hager W (2003) Blasius: a life in research and education. Exp Fluids 34(5):566–571
19. Peretti A, Toth P (1982) Optimization of a pipeline for the natural gas transport. Eur J Oper Res
11:247–254
20. Pepper W, Lo G (1999). Application of linear program to gas market and pipeline model in the
state of Victoria, Australia. In: INFORMS conference, Cincinnati, May 1999
21. Rios-Mercado RZ (2002) Natural gas pipeline optimization. In: Pardalos PM, Resende MGC
(eds) Handbook of applied optimization. Oxford University Press, New York, pp 813–826
Long-Term Pressure-Stage Comprehensive
Planning of Natural Gas Networks
Michael H€
ubner and Hans-J€
urgen Haubrich
A. Sorokin et al. (eds.), Handbook of Networks in Power Systems II, Energy Systems, 37
DOI 10.1007/978-3-642-23406-4_2, # Springer-Verlag Berlin Heidelberg 2012
38 M. H€
ubner and H.-J. Haubrich
network structures with regard to all technical and economic boundary conditions and
is also used by the German Federal Network Agency for calculating reference
networks with minimum costs for given supply tasks. Exemplary applications dem-
onstrate the method’s capability and the advantages through applying this method for
long-term planning of natural gas networks.
The cost pressure on distribution companies has increased because of the upcoming
regulatory framework in the European power and gas markets. Therefore, in
particular gas network operators, who focused mainly on the compliance of the
technical safety proposals by national technical and scientific associations for gas
and water networks, need to reveal potentials for a cut down of costs in the field of
network planning.
The reduction of directly controllable costs in the short-term, e.g. a decline in
maintenance work or an abandonment of investments, provides a limited lowering
of costs, if the high reliability and safety standards of existing networks should be
maintained. Another option for lowering network costs is the operational optimiza-
tion. The aim of this method is to supply all customers via the existing network at
minimal costs with regard to contractual obligations [1, 2]. Nevertheless, due to the
narrow time planning horizon, advancements in efficiency are constricted to short-
term manipulable expense factors.
A sensible approach seems to be the refinement of the network planning process
in order to determine efficient networks, which provide for an equivalent level of
reliability and technical safety compared to the existing network and at the same
time lower costs. Therefore, the application of a computer-based network-planning
instrument for the objective determination of long-term cost-efficient networks is
necessary. At present, there exists no instrument for the integrated planning of
natural gas distribution networks with different pressure stages. Hence, the primary
objective of this paper is to introduce the status quo of current developments of a
computer-based technique for the integrated planning of natural gas distribution
networks with variable pressure stages.
In case of the long-term planning of electrical networks, computer-based
procedures are deployed that optimize the network structure as well as the dimen-
sioning of necessary operating facilities with regard to the given supply task. All
relevant technical and company internal planning criteria need to be considered.
Numerous studies have proven the overall functionality of the aforementioned
procedures for realistic planning tasks [3]. On the other hand, in the past, the
planning of natural gas networks has been vastly driven by the company’s success
Long-Term Pressure-Stage Comprehensive Planning of Natural Gas Networks 39
Due to the complexity of the network planning process, it seems both necessary and
beneficial to subdivide it into two separate planning stages [5], see also Fig. 1.
Regarding a planning horizon of several decades, the existing network structure and
operating facilities may be ignored. The long-term planning is based on the
aforementioned assumption that is often referred to as Greenfield development.
Therefore, the long-term planning will identify cost-efficient network structures
neglecting the existing facilities.
Demand/Supply
Uncertainties
Scenarios
Possible
Planning Projects
target networks based on the outcomes of the long-term planning [6]. Uncertainties
concerning, e.g. load growth, changes in interest rates, etc., may be considered in
this planning stage.
As the long-term planning resembles in many respects the determination of
reference networks and is therefore attached with special consideration in the
regulatory context, this article will focus exclusively on the long-term planning of
natural gas networks.
Technical Boundaries/
Degrees of Freedom Supply Task Degrees of Freedom
Pressure Stages Equipment
Station X Station Y
Node
Pressure Stage: A
Trench
Admissible Types
Regulator A/B
Admissible Types
Regulator B/C
Admissible Types
Customer (B)
Node
Demand,
Pressure Stage: C
Min./max.Pressure
Customer (C)
Demand,
Min./max.Pressure
The model allows for a definition of multiple pressure stages connected via pressure
regulators within a station. Each node represents a pressure stage in each station.
The closed analysis of possible dislocations of network customers between pressure
stages as well as the fundamental determination of efficient vertical network
configurations, realization and positioning of pressure regulators, is made possible
by the presented network model.
42 M. H€
ubner and H.-J. Haubrich
As the pressure level of each pressure stage is depending on the location and
discharge pressure of the used pressure regulators that is determined within the
optimization procedure, the resulting operating pressure is an outcome of the
optimization. Therefore, it is possible to determine the optimal combination of
pressure stages objectively. Nodes are used to offer both an assignment of operating
equipment to dedicated pressure levels and furthermore a junction of pipelines as
well as pressure regulators. Trenches are composed of a combination of pipelines
with potentially differing technical properties and pressure stage assignments.
3 Methodology
All relevant exact optimization proceedings are based on the iterative resolution
and modification of a relaxation – broken down simplified problem – with a
solution set containing all solutions of the initial problem. The branch-and-bound
search and the cutting plane method apply a relaxation into a linear optimization
problem and disregard all integrity requirements in the first instance. Since in
principle, a greater number of solutions are accepted in the relaxation than in the
initial problem, the optimal value of the relaxed problem is at least as good as the
optimal value for the initial problem, hence providing an upper or lower bound.
The value of each admissible integral solution defines just as well a boundary for
the optimal solution of the initial problem as per definition. The comparison of both
upper and lower boundary yields the maximum distance between a found solution
and the optimum one, even though it is unknown.
below 1% [3] and therefore negligible for realistic problem formulations. Genetic
algorithms are, especially for combinatorial optimization problems, widely adopted
heuristic optimization approach that will be applied for the introduced optimization
problem. They refer to the natural evolution of animals over multiple generations.
They additionally offer the possibility to exploit their population based design with
a parallel execution to achieve further computing time reductions [17].
Initial Population
New Population
Crossover
Selection
Break?
no
yes
Result
decision variables and therefore make a decoding of the problem for a validation
and assessment of the individual possible. Subsequently, an upfront-predefined
number of individuals will be chosen from the search space stochastically or in
combination with already known solutions. These individuals compose the initial
population of the procedure.
Afterwards, the quality of all individuals within the present population will be
assessed with the help of a fitness function. The fitness function includes all relevant
parameters as investment cost input and technical specifications so that a concrete
roundup of each individual’s quality is possible. The termination criterion will be
checked at the end of every generation. For example, a termination criterion may be
a maximum number of iterations or a stagnant convergence over several iterations.
If no termination criterion is met, new individuals are created by means of genetic
operators that constitute the new population. Selection, crossover, and mutation are
counted among the genetic operators. Figure 5 shows an exemplary application to
illustrate the general approach of genetic operators.
The genetic operator selection is chosen to arrange a gene pool from individuals
of a population. All individuals are considered with a probability according to their
fitness values. Therefore, well-adapted individuals receive a higher share in the
gene pool than individuals with a poor fitness value do. Thus, the main functions of
selection operators within genetic algorithms are:
• Selection of individuals participating in the crossover
• Selection of a predefined number of individuals that form a successive popula-
tion from a set of individuals
• Selection of a predefined number of individuals from the set of individuals
resulting from a crossover that shall be further regarded
46 M. H€
ubner and H.-J. Haubrich
Following the selection, at least two individuals are extracted from the gene pool
and recombined with a definable crossover operator to form new individuals. The
main goal of this recombination is ideally to transfer the positive attributes of the
different parent individuals so that the succession unifies these attributes.
The concept of a genetic operator mutation has been introduced to genetic
algorithms in order to replicate the naturally occurring stochastical modification
of genes. For this purpose, individuals from the actual population will be selected
and their genestrings altered at sparse positions. In comparison to the crossover
operator, the mutation operator should be applied the more infrequent the more
advanced the genetic algorithms get as otherwise individuals in the near of a
potential optimum will be altered too excessively. Depending on the chosen
mutation function, one ore more genes of a genestring will be mutated.
Finally, the population generated by the genetic operators and tested for the
termination criterion will replace the actual population. The genetic algorithm does
not necessarily need to replace the old population entirely but can also take over the
best individuals from the old population. If no termination criterion is fulfilled, all
genetic operators will be applied to the new population. Otherwise, a predefined
number of the best individuals will be exported.
Optimization Procedure
Pressure Stage x
Technical Verification, Repair Functions
Technical Assessment
Operational Assessment
Substantial operational requirements for the planning of natural gas networks are
limits for flow velocities in pipelines, flow rates in pipelines and through pressure
regulators and as obligatory limit maximum and minimum pressure levels for all
nodes. The compliance of these limits is likewise checked with the results of a flow
calculation.
Required maximum flow rates are depending on the type of each used pressure
regulator. If the actual flow rate exceeds the maximum admissible value, a poten-
tially more expensive pressure regulator with broader limits or a pressure regulator
with a lower outlet pressure may be inserted. Should the aforementioned possibility
be unfeasible, pressure regulators at different positions within the pressure stage
may be amplified to relieve the pressure regulator under consideration.
Long-Term Pressure-Stage Comprehensive Planning of Natural Gas Networks 49
4 Exemplary Results
The considered supply area (see Fig. 7) exhibits a predominant urban character. The
total area is approximately 100 km2 in size. Two connection points to the
overlaying 16 bar regional distribution grid provide the opportunity to feed natural
gas into the supply area. With a number of 30,000 inhabitants, the total power
output adds up to 150.8 MW. A share of 68% of the power output falls upon the
low-pressure household customers and the remaining output falls upon business
customers in the medium pressure stage. In a prognosis for a long-range planning
horizon, stagnation on today’s level of the demand situation has been determined
(cp. Table 4). Because of the general decline of demand for natural gas due to a
replacement of natural gas for alternative energy sources and a reducing population,
a scenario of a halving of the total demand will be analyzed additionally.
Three potential pressure stages will be considered for the exemplary results: one
high-pressure stage operated at four bar (HP4), one medium-pressure stage (MP),
and one low-pressure stage (LP).
the small geographical dimension of the supply area, the number of pressure
stages will be limited to these three pressure stages.
• Each combination of pipelines of all possible pressure stages may be placed in
every trench connecting two stations.
• The positioning of pressure regulators will be restricted to adequate areas
according to the place requirements (see Fig. 7). Overall, the optimization
procedure may choose from 64 pressure regulators connecting the three pressure
stages.
• Additionally, there are two possible transfer stations connecting to the
overlaying 16 bar transmission network that bypasses the urban area (see Fig. 7).
• In accordance to the planning practice, the number of selectable equipment
categories will be limited. For instance, pipeline types may be chosen from six
categories representing diameters between DN 80 and DN 500 (cp. Table 3).
Furthermore, the following technical requirements are considered:
• Customers are located in the low and medium-pressure stage. The assignment of
connection points is evident in Fig. 7.
• The requirements concerning the pressure level in each pressure stage result
from maximum infeed pressure and minimum pressure at customer as well as
pressure regulator connections (cp. Table 4).
52 M. H€
ubner and H.-J. Haubrich
HP4 MP LP
• In order to restrict noise disturbances for residents in the urban area, the flow
velocity will be limited to 20 m/s.
• The distributed natural gas is a H-gas with caloric value of 11.3 kWh/mN3 and a
medium temperature of 5 C for the relevant peak load case.
Figure 8 illustrates the pressure stage design as well as the network structure for
each pressure stage of the existing network, featuring two connection points with
the overlaying 16 bar high-pressure network along with distributing the natural gas
over three pressure stages to the end customers. While the high pressure stage
(HP4) exhibits a wide-meshed network structure, the medium and low pressure
stages (MP and LP) exhibit a fine-meshed network structure.
Besides the absolute minimum pressures, Table 4 equally provides bandwidths
of the admissible pressure within each pressure stage as the zone between the
feed-in pressure and the minimum outlet pressure for customers or pressure
Long-Term Pressure-Stage Comprehensive Planning of Natural Gas Networks 53
HP4 MP LP
regulators feeding lower pressure stages. In the existing network, the resulting total
bandwidths are utilized only to a low extent indicating that taking into account the
technical requirements, a substantial reduction of overall pipeline length and
equipment dimensioning could be achieved.
In Fig. 11, the annual network costs of the existing networks (differentiated
between investment and operating costs) for pressure regulators, pipelines, and
trenches are depicted for each pressure stage. The total annual network costs
aggregate to 3.2 Mio. € for an interest rate of 9.5% and are mainly influenced by
the necessity to connect household and business customers to the medium and low
pressure stage respectively leading to big pipeline lengths in these stages.
HP4 MP LP
The target network for the scenario halved demand exhibits less isolated
networks in comparison to the target network for scenario of stagnating demand.
This is due to a decreased need for pressure regulators. Furthermore, all target
networks have significantly lower annual pipeline costs compared to the existing
network. On the other hand, costs for pressure regulators are only slightly reduced.
At the same time, the results demonstrate that the existing network is over dimen-
sioned for the fulfillment of the supply task.
The following results are achieved:
• Positions of pressure regulators for a connection of two pressure stages are, due
to the integrated optimization approach, optimally adjusted. This yields drasti-
cally reduced costs for pipelines with simultaneously maintaining a nearly equal
number of pressure regulators.
• The northern pressure regulator feeding in from the 16 bar grid is not necessary
for an economic connection of the considered network area.
• With exception of the medium-pressure stage in the scenario halved demand, the
nodal pressures in all pressure stages max out the admissible bandwidth of
pressures. Consequently, both target networks do not possess spare transfer
capacities beyond the peak demand case.
• The medium-pressure stage in the scenario halved demand is used almost
exclusively for the supply of network customers. Therefore, an investigation
on altered pressure connections of medium and low-pressure customers seems
sensible and has been included in the economic evaluation.
Figure 11 illustrates the comparison of all executed evaluations and the existing
network in terms of annual investment and operating costs.
The majority of annual network costs of the existing as well as of the target
networks fall upon the distribution level of household customers. In contrast to the
Long-Term Pressure-Stage Comprehensive Planning of Natural Gas Networks 55
3,0 HD4
–40.2 % –46.1 % –30.7 % –58.7 % –46.3 %
2,5
MD
2,0
1,5
1,0
ND
0,5
0,0
Existing Stagnating Halved Demand Abandonment Abandonment
Network Demand LP MP
low-pressure stage that is mainly influenced by the given points of supply, advanced
degrees of freedom exist in overlaying pressure stages leading to significant
differences in their costs. The medium and high-pressure stage may take over
different functionalities such as a wide-ranging transportation or localized supply
of natural gas.
The theoretical cost cutting potential of all target networks are in-between 30,7%
with an abandonment of the low-pressure stage and the assumption that each then
required household pressure regulator will incur expenses of a further 500€ and
58,7% if the expenses for household pressure regulators are neglected. Because of
the low number of medium-pressure customers, additional expenses for pressure
regulators are neglectable concerning the target network with an abandonment of
the medium-pressure stage. The optimal annual network costs in the scenario
halved demand are reduced just slightly by approximately 6% compared with the
scenario stagnating demand.
If the target networks for a reduced demand are analyzed it is obvious that the
expenses for three pressure stages is almost equal to the expenses with regard to an
abandonment of the medium-pressure stage. Nevertheless, three pressure stages
offer major flexibilities for a deviation from the projected demand. An increase of
the connection pressure for household customers to medium-pressure leads to
considerably increased expenses due to the necessary household pressure
regulators. As the results have shown, the proposed optimization procedure is
capable of identifying inefficiencies in the existing network and recommends
measures for correction.
2,8
Fitness [M /a]
2,6
2,4
2,2
1,8
0 20 40 60 80 100
Iteration
and consequently on the computing time. The iteration number and population size
have a direct influence on the computing time as well as the quality of the
achievable results. The iteration number specifies how many populations are
generated and evaluated. With choosing a higher iteration number, in general, a
greater part of the solution space will be explored. In order to analyze the influence
of the iteration number in detail, an exemplary optimization for the identification of
the cost-efficient network structure for the already described planning task is
executed.
Figure 12 depicts the bandwidth of individuals’ fitness values evolving with
increasing iterations with a population of 1,000 individuals.
As shown in the diagram, it is apparent that due to the pure stochastic nature of
the generation scheme at the beginning of the optimization and therefore high
distance to the optimum, very fast advancements in the solution quality are
achieved. After approximately 50 iterations, only minor improvements are neces-
sary for the present case. The required number of iterations mainly depends on the
number of degrees of freedom of each planning task. Application of the optimiza-
tion procedure on different real supply tasks demonstrated that typically an iteration
number in the same magnitude as the degrees of freedom provide good solutions
even for huge problems. By using additional supporting local search algorithms, the
iteration number can be decreased relative to comparable genetic algorithms with-
out local search.
The population size specifies how many different individuals have to be created
and evaluated during a single iteration of the genetic algorithm. An increasing
population size leads to a greater diversity amongst the individuals and helps in
preventing a premature convergence to local minima. For a detailed evaluation of
this correlation, population sizes between 50 and 2,000 for 100 iterations are
chosen. Figure 13 illustrates the resulting means and standard deviations of the
annual network costs of the most cost-efficient network structures after 100
iterations for the specified population sizes. With an increasing population size,
Long-Term Pressure-Stage Comprehensive Planning of Natural Gas Networks 57
0,5 0,04
0 0
50 100 250 500 750 1000 2000
Population Size
an initial drastic reduction of the annual network costs is achieved, but eventually
converging to a nearly global optimum.
12
80%
Computing Time
10
60%
Speedup
8
40% 6
4
20%
2
0% 0
1 10 30 50 70 90
Number of Processors
directly dependent on the population size, the number of utilized processors has to
be adapted accordingly.
5 Conclusions
Due to the forthcoming regulation schemes throughout Europe, new challenges for
natural gas network operators arise. In the past natural gas networks have been
planned without the specification of long-term objectives and therefore often exhibit
oversized dimensioning of the underlying network structure as well as the utilized
operating facilities. This in consequence leads to corresponding higher network costs.
Computer-aided optimization procedures provide important assistance for the
retrieval of potentials for future cost reductions. While the development and
application of such optimization procedures for the planning of electricity networks
is widely adopted and state-of-the-art, the planning of natural gas networks has
previously been accomplished by predominantly experience-based example
investigations and thus without the necessary objective future orientation.
The presented approach enables the long-term planning of natural gas networks
under consideration of all relevant technical and furthermore company internal
guidelines. It yields multiple alternatives as well as cost-efficient network structures
with considerable low computational time.
With the aid of sensitivity analyses, possible effects of varying boundary
conditions on structure and costs of cost-efficient networks may be quantified.
These sensitivity analyses allow for a fundamental determination of optimal
planning principles and are therefore able to explain minimal network costs for a
given set of boundary conditions.
Long-Term Pressure-Stage Comprehensive Planning of Natural Gas Networks 59
References
1. Ehrhardt K, Steinbach MC (2005) In: Bock HG, Kostina E, Pu HX, Rannacher R (eds)
Nonlinear optimization in gas networks. Modeling, simulation and optimization of complex
processes. Springer, Berlin/Heidelberg/New York
2. H€ubner M, Haubrich HJ (2008) Long-term planning of natural gas networks. In: IEEE 5th
international conference on the European electricity market, Lissabon, 2008
3. Paulun T, Haubrich H-J, Maurer Ch (2008) Calculating the efficiency of electricity and natural
gas networks in regulated energy markets. In: IEEE 5th international conference on the
European electricity market, Lissabon, Portugal, 2008
4. Hofbauer M, Anders S, Sigrist R, Weßing W (2003) Modell einer kostenoptimierten
Gasverteilung. In: GWF-Gas/Erdgas 144, No. 5, 2003
5. H€ubner M (2009) Druckebenen€ ubergreifende Grundsatzplanung von Gasverteilungsnetzen,
vol 127. Aachener Beitr€age zur Energieversorgung, Klinkenberg/Aachen
6. Paulun T (2006) Strategic expansion planning for electrical networks considering
uncertainties. Eur Trans Electrical Power 16(6):661–671
7. German technical and scientific association for gas and water e.V. (DVGW). Code of practice
G 2000, “Gasbeschaffenheit”, Bonn, 2006
8. Domschke W, Drexel A (2007) Einf€ uhrung in operations research, 7th edn. Springer-Verlag,
Berlin
9. Rothfarb B, Frank H, Rosenbaum DM, Steiglitz K, Kleitman DJ (1970) Optimal design of
offshore natural-gas pipeline systems. Oper Res 18:992–1020
10. Osiadacz AJ, Górecki M (1995) Optimization of pipe sizes for distribution gas network design.
In: Proceedings of the 27th PSIG Annual Meeting, Albuquerque, Pipeline Simulation Interest
Group, 1995
11. Hoeven Tvd (2004) Math in gas and the art of linearization. Dissertation, Groningen: Energy
Delta Institute, Rijksuniversiteit Groningen, 2004
12. Handschin E, Waniek D, Martin A, Mahlke E, Zelmer E (2007) Gekoppelte optimale
Auslegung von Strom-, Gas- und W€armenetzen. In: VDI-Berichte Nr. 2018, Leverkusen, 2007
13. Boyd ID, Surry PD, Radcliffe NJ (1994) Constrained gas network pipe sizing with genetic
algorithms. In: EPCC-TR94-11, 1994
14. Cunha MC, Ribeiro L (2004) Tabu search algorithms for water network optimization. Eur J
Oper Res 3(175):746–758
15. Duarte H, Goldbarg EFG, Goldbarg MC (2006) A tabu search algorithm for optimization of
gas distribution networks. Evol Comput Comb Optimization 3906:37–48
16. Castillo L, Gonzáles A (1998) Distribution network optimization: finding the most economic
solution by using genetic algorithms. EurJ Oper Res 108(3):527–537
17. Rajan SD, Nguyen DT (2004) Design optimization of discrete structural systems using MPI-
enabled genetic algorithm. Struct Multidisciplinary Optimization 28(5):340–348
18. Gottlieb J (2000) Evolutionary algorithms for constrained optimization problems, 1st edn.
Shaker Verlag, Aachen
Optimal Location of Gas Supply
Units in Natural Gas System Network
Abstract Natural gas industry has been confronted with big challenges: great
growth in demand, investments on new GSUs – gas supply units, and efficient
technical system management.
The right number of GSUs, their best location on networks and the optimal
allocation to loads is a decision problem that can be formulated as a combinatorial
programming problem, with the objective of minimizing system expenses.
Our emphasis is on the formulation, interpretation and development of a solution
algorithm that will analyze the trade-off between infrastructure investment expen-
diture and operating system costs. The location model was applied to a 12 node
natural gas network, and its effectiveness was tested in five different operating
scenarios.
1 Introduction
The constant increases in oil prices, the governments’ commitments to the Kyoto
Protocol, and the social and political concerns with the adoption of sustainable
development strategies, have led to the increasing search of new technologies and
alternative energies. Natural gas fits in those specifications, because it’s a clean
energy and provides an important contribution to the security of supply.
The European energy policies encourage the production and consumption of
natural gas. The natural gas market growth implies a significant reinforcement on
A. Sorokin et al. (eds.), Handbook of Networks in Power Systems II, Energy Systems, 61
DOI 10.1007/978-3-642-23406-4_3, # Springer-Verlag Berlin Heidelberg 2012
62 T. Nogueira and Z. Vale
gas supply units – GSUs – and on transport networks, in order to properly meet the
gas load demands. On one hand, the investment in GSUs is significant, and on the
other hand, pipelines also require a large investment whereas they have to face
serious physical and technical constraints.
Thus, the investment decision in GSUs is of major importance: how many of
these facilities should be placed and where, in order to meet gas demand, at the
lowest possible costs? This question is addressed in this article by using an optimi-
zation technique – specifically, using a location problem and model.
A simple location problem can be formulated as a linear programming problem
in which some of the variables are constrained to assume only integer values. Such
problems are called integer linear programming problems. The added complexity in
these problems results from the fact that we usually are interested not only on
location, but also on the optimal allocation of GSUs to gas loads. This led us to
choose an optimal combinatorial problems’ approach [1]. In the case of small
dimension natural gas networks, location problems can be supported by optimiza-
tion solver tools, which are commercially available.
The analytical location methodology is heavily dependent on the system net-
work size [2]. In literature [3] we can find different location methodology
approaches. For large networks, Beasly [4] describes very effective heuristics
dealing with location problems. To deal with the realistic dimension of a gas
system, the mathematical models need to be extended, observing regulatory and
technical restrictions.
To better understand the general concepts of the location problem methodology,
we will apply it to small networks, as a basis for further comprehension.
Natural gas is a gift from nature, and finds its origin in high-pressure conversion of
biomass into coal and hydrocarbons. Part of this million-of-years process is due to
intensive underground dynamics, which brought the natural gas through porous
layers deeper and deeper, until it became captured in an impermeable layer.
Widely used all over the word, natural gas is an attractive fuel, because it is easy
to transport (in its liquefied form), cheaper and clean. It has the highest hydrogen
carbon ratio of all fossil fuels, and when burned or used as feedstock the exhaust
gases hold less CO2 compared to the conversion of oil and coil.
Natural gas incorporates highly important national interests, in terms of econ-
omy and social welfare. In gas producing countries, natural gas reserves are
considered a national asset and have been exploited for the national benefit. For
gas consuming countries, access to gas reserves is also of crucial importance. The
development of natural gas markets in Europe has so far been guided by national
policies that take into account each country’s access position to gas reserves.
Optimal Location of Gas Supply Units in Natural Gas System Network 63
We can distinguish two different models for the gas market development: a
public property focused model and a public utility focused model [5]. The first
model is usually found in countries with voluminous gas fields, and the second
model, in countries without gas fields.
Only a few European countries have gas reserves voluminous enough to allow
for both domestic consumption and export. In these countries gas reserves are
considered as national assets and have been exploited for the national benefit,
with the ambition to capitalize economic value. This became a guiding principle
in the exploitation of gas resources. The organization of markets based on a public
property model focuses in the upstream activities – production and transmission,
controlling access to a limited number of players, and the ownership structure is
public dominant. This model assures maximization of state revenues and pursues
the national welfare and prosperity.
Net importing countries – those without gas reserves – have concentrated on gas
consumption and gas demand. Policy for the public utility model draws on the idea
that natural gas is a vital national energy resource that should be available nation-
wide. In this regard, the state takes an active role in developing the utility-oriented
gas market for economic and political reasons. State control and monopoly regula-
tion was meant to reduce investment risk and to ascertain long-term domestic
consumption by developing a domestic consumer market. The public utility
model is therefore consumer oriented and heavily relies on specific regulation.
Unlike the public property model, it focuses in the downstream activities, also
controlling the access to a limited number of actors. As a consequence, the model
pursues reasonable consumer tariffs and selective services.
In addition to these two models, there is what one could name a commodity
focused model, referring to a gas market where natural gas is perceived as a freely
tradable commodity – reflecting the ideal of a competition-based gas market. This
commodity-oriented model is now challenging all European countries due to the
EU gas Directive (see Sect. 2.3).
Because of its import dependency and expected strong growth in demand,
continental Europe gas industry faces important challenges and opportunities [6].
The growth of the European natural gas market will depend, to a large extent, of gas
price competitiveness as compared to other energy sources and technologies.
Some differences on the national responses to the emerging EU gas market liberal-
ization can be found. A point of departure is the obvious difference among
European countries with respect to access to gas resources.
Liberalization has affected the image and perception of natural gas, which can
no longer be considered only as a public property or a public utility, as the
ownership structure is open to private dominance. So far, the commodity-based
model can be found, where some competition requirements are met. To be able to
perform, a commodity based model requires both international and national
matured upstream and downstream gas systems.
In an attempt to harmonize the differences, European markets are oriented by
Directive 2003/55/EC [9] of the European Parliament and of the European Council,
concerning common rules for the internal market of natural gas. The new current
law facilitates competition by introducing third party access of transmission and
distribution pipelines and financially unbundling transport and trade activities. In
order to ensure efficient and non-discriminatory network access it is appropriate
that the transmission and distribution systems are operated through legally separate
entities where vertically integrated undertakings exist. It is also appropriate that the
transmission and distribution system operators have effective decision making
rights with respect to assets necessary to maintain, operate and develop networks
when the assets in question are owned and operated by vertically integrated
undertakings.
National regulatory authorities should be able to fix or approve tariffs, or the
methodologies underlying the calculation of the tariffs, on the basis of a proposal by
the transmission system operator or distribution system operator(s) or LNG system
operator, or on the basis of a proposal agreed between these operator(s) and the
users of the network. In carrying out these tasks, national regulatory authorities
should ensure that transmission and distribution tariffs are non-discriminatory and
cost-reflective, and should take account of the long-term, marginal, avoided net-
work costs from demand-side management measures.
In the liberalized natural gas market, gas customers should be able to freely
choose their supplier. Nonetheless, a phased approach should be taken to complet-
ing the internal market for gas, coupled with a specific deadline, to enable industry
to adjust and ensure that adequate measures and systems are in place to protect the
interests of customers and ensure they have a real and effective right to choose their
supplier.
The progressive opening of markets towards full competition is almost complete
in European countries with the implementation of appropriate measures to achieve
the Directive’s objectives. Such measures include adequate economic incentives,
using, where appropriate, all existing national and Community tools, for the
maintenance and construction of necessary network infrastructure, including inter-
connection capacity and investment on new gas supply units – GSUs [10].
66 T. Nogueira and Z. Vale
These arguments reinforce the need of GSUs investment, to ensure the security
of supply, and the existence of new natural gas players within a liberalized market.
Having the GSUs economic feasibility in mind, we’ll study the GSUs’ location
problem, which plays a key role in supporting the sustainable development of the
natural gas industry.
A major determinant of the level of effective natural gas supply is the ease to feed
customers, with the correct balancing between the infrastructure investment and the
ongoing transportation costs through the gas network. The number of GSUs and
their optimal location in a gas network is a decision problem that can be formulated
in order to minimize a cost function [11].
In practice, because of economic constraints, if we want to improve a gas system
supply, we must try and find the best location sites for a fixed number of GSUs. We
often optimize an objective function subject to the constraint that we should locate a
given number P of facilities, reaching a solution that is usually different from the
one we find in the case where such a constraint does not exist. Both situations will
be considered in this study: no P value constraint and P value constraint. In the first
case, P is an output of the problem; in the second case, P is an input.
If the GSUs are marine terminals or other kind of aerial storage facilities, we can
consider that they have an unlimited capacity. This means that we can consider all
the supply capacity we need to feed properly gas demand. In case of underground
storage facilities, because of physical issues, maximum specified capacities of
infrastructures are important, so GSUs capacitated fixed charges are given as
constraints of the location problem.
For geological reasons, some GSUs infrastructures constructions are not suitable
or feasible in certain locations. Similar GSUs structures can have different fixed
implantation costs in different locations. To include this particularity in this study,
we will consider an explicit cost, fi, of locating a GSU at each candidate location i.
In this study we consider that these i candidates location are nodes on network.
According Hakimi [12], the optimal places to locate facilities on network can be
found at nodes. This author proved that by relaxing the problem to allow facility
locations on network arcs, total costs would not be reduced.
Location problems include not only the study of the best place implementation in
the network, but also the optimal assignment from sources to loads. The transpor-
tation costs – Cij, are the costs of serving demands at each node j from the sources i,
they are treated as demand-weighted distances. We can understand the importance
of accounting for these costs, by observing Fig. 2.
The long dashed curve shows the routing cost as a function of the number of
facilities that are to be located. The long dashed dot-dot curve reflects the
Optimal Location of Gas Supply Units in Natural Gas System Network 67
accumulated fixed cost of locating GSUs. The fixed facility cost is added to the
routing cost to obtain the total cost, as shown by the solid curve [13].
The total cost initially declines as the reduction in routing cost that results from
the addition of more gas sources more than offsets the additional facility location
costs. At some point, the cost of additional GSUs exceeds the savings in routing
costs and the total cost increases as we add more facilities.
Our goal is to determine the turning point, which correspond to the optimal
number of the natural gas sources to be placed in the gas system in order to
minimize overall costs.
The model developed in this study is versatile and accounts for the referred
natural gas technical and economic constraints. In Sects. 3.3, 3.4 and 3.5, we will
analyze five operating scenarios:
1. No P imposed, capacitated GSUs
2. No P imposed, relax capacitated GSUs
3. No P imposed, unlimited GSUs
4. No P imposed, increase unitary cost
5. Imposed P, capacitated GSUs
that GSUs will be implanted only on the gas network’s nodes, our model considers a
network with m GSUs potential sites and n demand points. The m value can be
equal to n, if we assume that all network nodes are potential sites lo locate GSUs. In
this case, we have m location decision variables and m x n transport decision
variables, totalizing m + m x n unknown variables or decision variables, being Yi
binary location variables and Xij transportation variables.
The problem can be formulated as minimizing the sum of the facility locating
cost, fi and the routing or transportation costs, Cij. These transportation costs are the
total demand-weighted distance (dij. bj) multiplied by a, the cost per unit distance
per unit demand, Cij ¼ a.dij.bj. Being, ai the maximum capacity of a GSU facility at
candidate node i and bj the demand at demand node j, the problem can be
formalized as a combinatorial problem, stated as follows:
Minimize
X
m m X
X n
f i :Yi þ a:dij :bj :Xij (1)
i¼1 i¼1 j¼1
Subject to
!
X
n X
n X
n
Xij Yi b j b0 Xij bai :Yi (2)
j¼1 j¼1 j¼1
X
m
Xij ¼bj (3)
i¼j
X
n
Xij = p (4)
i¼1
Xij r0 (5)
Yi 2 f0; 1g (6)
The objective function (1) minimizes the total cost which is the sum of the fixed
GSUs costs and the routing costs. Offer constraint (2) means that if Yi ¼ 0, the
GSU is not located in i, if Yi ¼ 1, the gas quantity from i source is limited by total
demand. This is applied to unconstrained capacity. The term in parenthesis, applied
for fixed charge problems, requires that the demand for the source i will not go
beyond its maximum capacity and no client will be allocated to an unavailable
natural gas source. Demand constraint (3) stipulates that each client should be
completely served and constraint (4) states that P facilities will be located (for P
imposed cases). If we remove this constraint from the problem, P number will be
calculated as an output. Constraint (5) states that the decision variables are positive
Optimal Location of Gas Supply Units in Natural Gas System Network 69
quantities and constraint (6) is the binary integer condition: the variable Yi is 1 if a
GSU is located in i or 0, if not. The optimal solution is the one that minimizes the
objective function (1).
As binary variables, Yi will assume only the value 1 or 0, which means a GSU is
located at i node or not, respectively. The transportation variables Xij will give us
the information about the gas quantity that should be supplied from i source (if it
was installed a GSU) to j load consumer.
In this section the location model is applied to a 12-node gas network (Fig. 3). Gas
network nodes are marked from 1 to 12. The triangular boxes are the gas demands –
bj, in million cubic meters unit (Mm3).
The distances between nodes – dij, stated on the branches in kilometers, are the
pipelines in which the natural gas is transported. Assuming that the network is
known, we can build the distance matrix (Table 1), an important input of the
location problem.
Each node can be a potential GSU location site, it has its explicit cost location fi
and it could be imposed its maximum supply capacity ai (as seen in Table 2).
Another important assumption to the problem is the a value, which is the cost per
unit distance per unit demand. As an input, its value will have a crucial impact on
the problem result. In our operating scenarios, we consider two different costs, for
a ¼ 0.009 and 0.027.
The problem we want to solve is to find the number and the optimal GSUs
location-allocation sites on the 12-node network so that the total cost is minimized
(Eq. 1). For this size gas network, the problem can be solved by optimization
software, such as the Risk Solver Platform. In this minimization problem we have
156 total decision variables, 12 unknown location variables, Y1 to Y12, and
12 12 ¼ 144 unknown transportation variables – Xij. The results depend on
the constraint assumptions, as seen in the next sections.
Table 1 Network distance matrix, dij
j
i 1 2 3 4 5 6 7 8 9 10 11 12
1 0 150 370 550 240 600 180 330 480 400 580 670
2 150 0 220 400 380 520 330 480 420 550 610 610
3 370 220 0 180 160 300 410 280 200 580 390 390
4 550 400 180 0 340 120 590 460 240 620 430 340
5 240 380 160 340 0 360 250 120 240 470 370 430
6 600 520 300 120 360 0 570 430 120 500 310 220
7 180 330 410 590 250 570 0 150 450 220 400 610
8 330 480 280 460 120 420 150 0 300 370 250 460
9 480 420 200 240 240 120 450 300 0 380 190 190
10 400 550 580 620 470 500 220 370 380 0 190 400
11 580 610 390 430 370 310 400 250 190 190 0 210
12 670 610 390 340 430 220 610 460 190 400 210 0
3.4.1 Scenario 1
We begin by considering that there is a maximum capacity stated for each potential
GSU installed (Eq. 2 in parenthesis), as well as its local implantation cost (Table 2).
In this scenario, P is an output (Eq. 4 not included), the GSUs have fixed charge
capacities and a ¼ 0.009. Results are shown in Table 3.
In Table 3, we have the decision variables’ results for the location variables Yi
and transportation variables Xij. The program optimized the result installing 5
GSUs (P ¼ 5), located at nodes 1, 4, 7, 9 and 11. With this choice, fixed costs
are: 1,000 + 1,500 + 1,900 + 1,650 + 1,250 ¼ 7,300 M€ (see Table 4, third
column).
When location variables Yi have 0 value, it means that node i won’t have a GSU
installed and therefore the respective transportation variables Xij are null as well.
For example, let’s consider node 2, Y2 ¼ 0; if there isn’t any gas source located on
node 2, no gas quantity is transported from node 2 to any other consumer on the
network. As a result, the transportation variables X21 to X2.12 are all zero. This node
2 is not a source point, but it is a demand node and it is supplied by the GSU located
in node 1, as seen in Table 2: X12 ¼ 10 Mm3, which is exactly its demand.
The GSU located in node 1 will supply its own demand plus the node 2 demand,
which is exactly the maximum quantity it can give, 15 + 10 ¼ 25 Mm3, as seen in
the first line, fourth column, in Table 2.
However, to respect the maximum capacities, some gas loads need to be
supplied by two different sources, which is the case of nodes 6 and 10. The demand
node 6 is supplied by GSU 4 with 10 Mm3 (X46) and by GSU 9 with 14 Mm3 (X96),
totalizing 24 Mm3, which is exactly the node 6 load demand. The GSU 4 also
supplies nodes 3 and 4, whose demands are 12 and 18 Mm3, respectively. These
demands plus the node 6 total demand will put GSU 4 out of its maximum capacity.
So, the optimal solution suggests that GSU 4 will supply 12 + 18 + 10 ¼ 40 Mm3,
which is its fixed charge capacity.
Transportation costs are the costs of the optimal transport solution from those
GSUs installed to the assigned gas loads. As an example, GSU 1 supplies its own
demand (X11 ¼ 15) and the demand in node 2 (X12 ¼ 10). There is no transporta-
tion cost to supply one node’s own demand, of course, but for GSU 1 to supply load
2 it costs: 0.009 150 10 10 ¼ 135 M€, as presented in Table 4. Let’s take
now GSU 4: this unit supplies its own demand, as well as nodes 3 and 6 demand.
Transportation costs from GSU 4 to node 3 are: 0.009 180 12 12
¼ 233.28; transportation costs from GSU 4 to node 6 are: 0.009 120 24
10 ¼ 259.2. Total transportation costs from GSU 4 to loads are 492.48 M€,
as shown in Table 4.
The location-allocation results can be summarized in Table 4. The objective
function is 10237.24 M€, which are the optimized system total costs.
As we can notice in Table 4, all gas demands are completely satisfied by open
natural gas sources and the maximum GSUs’ capacities were not exceeded.
3.4.2 Scenario 2
If we increase maximum GSUs capacities, the location program has more flexibility
and shows different optimal results. Let’s now consider that maximum GSUs
capacities were relaxed to 60 Mm3, and the same for each potential node. This
means that all values of the fourth column in Table 2 have the same ai value. The
location-allocation results can be shown in Table 5.
Optimal Location of Gas Supply Units in Natural Gas System Network 73
In this scenario, the optimal solution is to locate 4 GSUs on nodes 1, 3, 6 and 11,
and total costs decrease. This is due to lower fixed location costs, but instead
transportation costs increase. We have less GSUs, so gas need to “travel” more
kilometers, which increases the transportation distances, and consequently the
transportation costs.
3.4.3 Scenario 3
3.4.4 Scenario 4
In the previous section it was assumed that the number P of GSUs installed on the
gas network is endogenously calculated, therefore an output of the problem. Now
we’ll consider that P is taken as a proxy of the number of GSUs, thus constituting an
input imposed by user. In this case, the Eq. 4 is included as a problem constraint.
We assume that P is known, but the location problem will solve its optimal
locations on the network.
3.5.1 Scenario 5
4 Conclusions
The natural gas industry in Europe is currently facing great challenges: European
harmonization, integrated operations of pipeline system and studying optimal
economic locations of gas supply units.
Optimal Location of Gas Supply Units in Natural Gas System Network 75
References
1. Dresner Z (1995) Facility location: a survey of applications and methods, Series in operations
research. Springer, New York
2. Nogueira T, Vale Z et al (2006) Advanced techniques for facility location problem in natural
gas network. ICKEDS’06, Lisbon, pp 347–351
3. Lorena L, Narciso M (1996) Relaxation heuristics for a generalized assignment problem. Eur
J Oper Res 91:600–610
4. Beasly E (1993) Lagrangean heuristics for location problems. Eur J Oper Res 65:383–399
5. Arentsen M, Kunneke R (2003) National reforms in European gas. Elsevier, Oxford
6. Nogueira T, Vale Z et al (2005) Natural gas market in Europe: developments and trends. In:
CAIP2005 – congresso interamericano de computation aplicada a la industria de processos,
Vila Real, Portugal, pp 115–118
7. Jensen J (2003) The LNG revolution. Energy Journal 24(2): 1–37
8. Flanigan O (1995) Underground gas storage facilities. Gulf Publishing Company, Houston,
pp 40–53
9. Directive 2003/55/EC of the European Parliament and of the Council of 26 June 2003
concerning common rules for the internal market in natural gas and repealing Directive
98/30/EC
10. Nogueira T, Mendes R, Vale Z, Cardoso J (2006) Heuristic model for Iberian natural gas
source location. J WSEAS Trans, 1:1343–1349
11. Nogueira T, Mendes R, Vale Z, Cardoso J (2007) A heuristic approach for optimal location of
gas supply units in transportation system. In: 22nd International scientific meeting of gas
experts, vol 1, Opatija, pp 303–311
12. Hakimi S (1964) Optimum locations of switching centers and the absolute centers and medians
of a graph. Oper Res 12:450–459
13. Daskin M (1995) Network and discrete location: models algorithms and applications. Wiley,
New York
14. Nogueira T, Vale Z (2008) Natural gas system operation: Lagrangean optimization techniques.
IGRC – International gas union research conference, ID 147, Category: Transmission, IGRC
Foundation
An LP Based Market Design for Natural Gas
Abstract Many electricity markets are now cleared using Linear Programming
(LP) formulations that simultaneously determine an optimal dispatch and
corresponding nodal prices, for each market dispatch interval. Although natural
gas markets have traditionally operated in a very different fashion, the same basic
concept can be applied. Since 1999, the Australian state of Victoria has operated a
gas market based on an LP approximation to the underlying gas flow optimization
problem. Here we discuss market design issues, using a formulation derived from
the key gas flow equations. Dual variables on key constraints imply prices which
vary by location, as for electricity markets, but also by time. But gas is both delayed
and stored within the transportation system itself. This raises a number of opera-
tional, pricing, and hedging issues which could be ignored in the case of electricity,
but become important when operating this kind of market for gas, or other
commodities, such as water, in a supply network where there are delays and storage.
The authors wish to thank AEMO for information used in this chapter, Ray Whaley for assistance
with the initial formulation, and Peter Jackson for his input.
E.G. Read (*) • S.R. Starkey
University of Canterbury, Christchurch, New Zealand
e-mail: grant.read@canterbury.ac.nz; stephen.starkey@pg.canterbury.ac.nz
B.J. Ring
Market Reform, Australia
e-mail: brendan.ring@marketreform.com
W. Pepper
ICF International, Virginia, USA
e-mail: wpepper@icfi.com
A. Sorokin et al. (eds.), Handbook of Networks in Power Systems II, Energy Systems, 77
DOI 10.1007/978-3-642-23406-4_4, # Springer-Verlag Berlin Heidelberg 2012
78 E.G. Read et al.
1 Introduction
Many electricity markets are now cleared using an LP-based “Market Clearing
Engine” (MCE) to optimize the value of trade, as determined by participant bids
and offers. An MCE formulation typically optimizes power system operations at
quite a detailed level, using a detailed representation of the transmission system,
and simultaneously optimizing dispatch of generation, transmission power flows,
and often ancillary services. This level of detail also enables the MCE to simulta-
neously determine corresponding prices for energy at each “node” in the transmis-
sion system, and often for each ancillary service, in each market dispatch interval.
In 1996, the first market of this type was developed in New Zealand [2]. The
Australian “National Electricity Market” (NEM) followed soon after, in 1997, using
a re-developed version of the same software, NEM1. That development was also
partly built upon the Victorian electricity market, which had been successfully
operating in the State of Victoria since 1994, with interconnected trade between
Victoria and New South Wales commencing mid 1997 [39]. Broadly similar elec-
tricity markets were being discussed or developed in many parts of the world [14],
and are now widespread [36]. Thus an extensive literature was developing on topics
such as the design of such markets, how market participants might behave in that
environment, and how they might hedge their risks over time and space. Hogan [15]
had also proposed an auction based model for allocation of gas transport capacity,
while McCabe et al. [43] had performed simulations of simple market structures.
These developments lead the Victorian State Government to consider the desir-
ability of developing a similar kind of market for trading natural gas, in that State [42].
The gas system operates in a broadly analogous way to electricity, with a variety of
suppliers and consumers, simultaneously injecting gas into, or withdrawing gas from,
various points in an interconnected “transmission network”. As reported by DPI [10],
the Victorian gas system currently has six suppliers, three major retail buyers, and
multiple wholesale traders. As shown in Fig. 2, the gas transmission system is a
meshed network with multiple inter-regional connectors and active underground gas
storages. In 1996, though, there was only one supplier and buyer, operating under a
single long term contract, with a simpler network, and no interconnections. The market
design discussed here was developed to allow that monopolistic arrangement to be
decomposed, and to support evolution towards a more dynamic trading environment.
Convergence towards a consistent market framework also seemed desirable in
view of the way in which natural gas interacts with electricity, both as a fuel for
generation and a competing supplier of end demand. Thus, one goal of the initial
design was to try and create a gas market framework which drew on experience
with electricity markets, aligned with the Australian electricity market, and could
develop towards greater integration over time. The key question was whether the
basic concepts developed for electricity markets could also be applied to gas, in the
sense that an analogous “nodal” market-clearing formulation could be developed,
and form a practical basis for trading. Further, if that was possible, under what
circumstances might it be worthwhile to do so?
An LP Based Market Design for Natural Gas 79
This paper focuses on the conceptual market design issues, using an LP formu-
lation analogous to those employed in the electricity sector, based on a representa-
tion of the gas system dispatch problem in nonlinear form. In reality, from its
inception in March 1999 [1], the Victorian gas market has been dispatched using an
MCE based on an LP model developed by ICF International. As described in
Pepper et al. [26], the MCE incorporates a number of advanced features to deal
with various physical and computational issues, and takes a different approach to
linearizing the gas flow equations. That model has proved accurate and reliable, but
its fundamental variables are a set of convex weights, which does not make for any
intuitive discussion of the forces driving pricing effects in a way likely to prove
meaningful to potential developers of gas markets. (In the same way, meaningful
discussion of pricing effects in electricity markets must be based on a representation
of the fundamental electrical relationships, in a full nodal model, not on the kind of
implicit representation some markets use in practice).
The reader is referred to Pepper et al. for a description of the implemented
model. Since our goal in this paper is really to discuss the implications of adopting
this kind of approach to gas market trading, we abstract away from some of the
detail and base our exposition on the original formulation originally developed by
Read and Whaley [32], as part of the Putnam Hayes and Bartlett (PHB) team
responsible for the market design. Even though this simplified formulation was
not implemented in practice, it provides a more accessible introduction to the
concepts, starting from a standard “textbook” representation of the gas flow
equations, and produces a dual that is more readily interpreted. It will be seen
that the resultant formulation is really no more complex than some of the
formulations that have been discussed or applied to form electricity markets,
particularly if AC power flow equations are modeled [16], and/or ancillary services
co-optimized [31], and/or inter-temporal unit commitment constraints represented
[17]. Thus this kind of market development does not seem infeasible on grounds of
complexity.
The major complication is that the gas flow equations imply that gas is both
delayed and stored within the transportation system itself. This raises a number of
operational, pricing, and hedging issues which could be ignored in the case of
electricity, but become important when operating this kind of market in a gas supply
network. One major motivation for discussing these issues, at this time, is that similar
issues are likely to be important in markets for other commodities, such as water [28],
where delays and storage also occur within the “transportation system” over which
the market operates. On the other hand, while Pepper et al. [26] describes a market
dispatch process that has proceeded down the path of increasing sophistication and
precision, commercial gas market trading is actually based on a highly simplified
version of the formulation. In fact, the initial market clearing logic of Ruff [35] only
involved a daily clearing of a market in daily gas delivery, while also accounting for
overnight storage in the system. For reasons discussed later, the market has still only
moved forward to the point of re-clearing to determine prices for the remainder of the
day at four-hourly intervals. The studies reported by Frontier Economics [11] did
80 E.G. Read et al.
2 Market Concepts
Most gas pipelines in the world operate under a contract carriage model. Historical
overviews of the emergence of competition in the US may be found in Vany and
Walls [37] and Doanne and Spulber [8]. Under this model, the pipeline operator
funds its pipeline by selling access to shippers of gas with varying levels of priority.
Those with the greatest priority have firm access to the pipeline and tend to pay the
most for their access. Those with less priority pay less, but only get access to the
pipeline to the extent that it has otherwise unused capacity. Open access regimes
may be imposed by regulators who require some transparency to these
arrangements, but the basic access arrangement is still via bilateral contracts.
Markets for gas around the world generally operate at hubs between pipelines.
Gas can be delivered to these hubs in accordance with pipeline usage contracts,
traded at the hub, and hauled away on other pipelines or consumed at the hub. In the
US an unregulated natural gas market trades over the New York Mercantile
An LP Based Market Design for Natural Gas 81
Exchange (NYME). Futures contracts are traded relative to the principal Henry
Hub, in Louisiana. Two equivalent virtual trading point markets are the National
Balancing Point (NBP) system in the UK, and the Title Transfer Facility (TTF) in
the Netherlands. A Short Term Trading Market (STTM) for gas along these lines
operates in Australia at hubs in Adelaide and Sydney, and soon in Brisbane.
The Victorian Gas Market is different in that it operates instead on the concept of
“market carriage”. The Primary Transmission System (PTS) is funded through
Transmission Use of System (TUOS) charges, rather than under bilateral
contracting arrangements. This makes it very similar to how electricity transmis-
sion systems are funded. With the network funded in this manner, the Victorian Gas
Market can operate a “commodity only” market for the trade of gas, and has done so
since 1999. Further, with AEMO operating the network, the trade of gas can be used
to determine day-to-day transmission access, and (in principle) point-to-point
transport charges. Given the obvious analogies between the gas and electricity
systems it seems natural to ask whether the concepts that have been applied to
design electricity markets might also be applied to design gas markets.
Ignoring a simplified marginal loss adjustment, the Australian electricity market
differentiates spot prices by region, so there is only one spot electricity price for all
of Victoria. Another point of reference was the New Zealand electricity market,
which determines spot prices for each node at which physical injection or off-take
occurs [2]. Both operate on the basis of prices for half-hourly trading intervals.
Buying and selling (wholesale) electricity is done through a “pool”, where electric-
ity generators offer electricity to the marketplace for dispatch through the electric-
ity transmission network. A central market coordinator receives generation offers
(and potentially load bids), determines which of those should be accepted (i.e.,
“clears the market”), implements the optimal dispatch, and announces the
corresponding spot prices, all in real time. Thus the goal here was to develop an
analogous market design for gas, in which a “market-clearing solution” is deter-
mined by an LP optimization model that simultaneously determines:
• A “dispatch” schedule for all gas “injections” and “off-takes” that is optimal, in
the sense that it maximize the “value of trade” defined as the benefits delivered
to loads (as determined by their bids), minus the costs incurred by suppliers (as
determined by their offers).
• A matching set of “nodal” spot prices, varying over time and across network
locations, defined by the marginal cost of meeting a (possibly hypothetical) load
at each time and location, and applied symmetrically to buy gas from suppliers,
and sell gas to consumers.
In principle, we can divide pipelines up into arbitrarily small cells, and it becomes
somewhat arbitrary as to whether primary variables are defined at cell midpoints, or
cell boundaries. Our basic nonlinear formulation is based on the initial market
82 E.G. Read et al.
design formulation developed by Read and Whaley [32]. They developed their
formulation in terms of average midpoint cell values, assuming velocity and
pressure changes to be implicitly defined at cell boundaries. That approach was
designed to be applied to a fairly discretized representation of a uniform pipeline,
though. Pepper et al. [26] model a more general network, approximating over quite
long pipeline segments, and modeling nodes where multiple pipes may connect,
and various pipe fittings that may induce step changes in velocity and pressure.
We briefly touch on such issues in an Appendix, but here take an intermediate
approach, developing a formulation for a single pipeline, with variables primarily
defined at cell boundaries, which can be thought of as nodes. The major complica-
tion is that natural gas is compressible, unlike some other piped fluids such as water,
so its density (and hence pressure) varies, with gas flows being driven by pressure
differences. We start with general equations, which allow pressures etc. to differ on
each side of each junction, but later assume a single pressure variable at each
junction, so as to develop a simplified conceptual formulation. Pepper et al. provide
a more detailed development of the actual formulation employed.
The pipeline is split into segments as shown above in Fig. 1. A variable at the
centre point of the subscripted nth element represents the average value. We discuss
flow reversals later but, for simplicity, assume flow is from cell n1, to n, to n + 1
etc. For simplicity we also assume time periods to be of unit length. The key
variables are gas pressure and flow, at the beginning of each pipe segment (or
cell), n ¼ 1,. . .,N and at, or from, the beginning of time period, t ¼ 1,. . .,T. Other
variables are derived from these, as necessary. Notationally, constants are
represented using normal fonts (e.g. R or H below), whereas variables and indices
are represented using italic fonts (e.g. qi, mn etc.).
Here the nth pipeline cell has diameter Dn, cross-sectional area An, and length Ln. H
is the gas temperature , assumed to be constant, and R is known as the “specific gas
constant” for the particular gas composition in the pipeline. The rate of gas (mass) flow
at any point (e.g. qtn ) is determined by the pressure of the gas, and its velocity. We will
apply this relationship to the midpoint flow/pressure/velocity values, as in (4). This
allows us to state the mass conservation equation for each cell, as in (6). Note that, in
this equation, injection (ytn ) is treated interchangeably with mass (mtn ), and mass is a
midpoint value, reflecting average pressure across the cell. Thus injection is implicitly
treated as if it were occurring at the midpoint of a cell, increasing pressures at both
ends. This is not likely to happen, in practice, but the cell in which injection is assumed
to occur can be made arbitrarily short, or represented by a “node” as in Pepper et al.
mtþ1
n ¼ mtn þ qtin qto
n þ yn
t
(6)
Modisette and Modisette [22] discuss fluid forces in pipes, initially defining the
forces for a single element. By applying the conservation of momentum, they then
sum forces and flows across all time periods and pipe elements. We use their results
to present what is known as the Bernoulli equation, in its more general form for
unsteady flows, which we state for the midpoint velocity/pressure pair:1
1
In this formulation, the pipe is assumed to be horizontal, thus eliminating the gravitational term
‘g siny’, for an elevation angle of y. The effect of this force is negligible because natural gas is
nearly twice as light as air, at standard conditions. Superscripts i and o are dropped because this
equation applies at any point.
84 E.G. Read et al.
vtn Þ2 ¼ RHrptn p~tn
Dvtn þ v~tn rvtn þ Bn ð~ (7)
Bn = f = 2Dn (8)
Dvtn ¼ vtþ1
n vn
t1
(10)
rvtn ¼ ðvto
n vn Þ=Ln
ti
(11)
rptn ¼ ðpto
n pn Þ=Ln
ti
(12)
Each pipe section will have a working pressure range, and may also have
maximum velocity limits. Since flow in a network, especially on loops, can be bi-
directional the physical lower bound is likely to have a negative value. Later, we set
this lower bound to zero, so as to keep solutions within the range where solutions
are convex.2 In reality, both bounds are more likely to bind at the inlet end of a cell,
but for now we impose them on midpoint flows, because this simplifies the dual of
our formulation, and discuss variations later:
t
Ptn p~tin Pn (13)
t
V tn v~tin Vn (14)
Equations 1, 2, 4, 5, 6, 7 and 9, 10, 11, 12, 13, 14 describe the key physical gas
flow relationships, and consequently feature prominently in the optimization model,
with the associated dual variables generating the information required to create
consistent market trading prices. Apart from this, an initial pressure (mass) and/
flow profile must be assumed, and a target (range) specified for the final period. But
the above model is incomplete, because we have not specified how the input variables
2
Direction of flow is relative to a conventional direction, which for simplicity we define as being
from i (inlet) to o (outlet). As with electricity networks this can be generalized by defining a
conventional direction for all arcs, then allowing the flow to take a +/ value in that direction.
An LP Based Market Design for Natural Gas 85
for each pipe cell link to the outlet variables for the upstream cell. It should be clear
that the flow out from cell n1 to cell n equals the flow in to cell n from cell n1, in
mass terms. Both pressure and velocity may change, though, if there is some kind of
fitting, or compressor, or just a change in diameter at the junction of two pipelines. A
reasonably simple formulation can be produced by assuming that that there will be a
proportional change in pressure at such a junction, at least locally around some likely
solution level with velocity adjusting to match. But we will simplify further, by
assuming that there are no special fittings, or pipeline diameter changes, at junctions.
Thus not only mass flow, but pressure and velocity have the same value immediately
upstream and downstream from each junction. This means that we can drop the
distinction between inlet and outlet variables, with ptin etc. just becoming ptn etc. and
pto t
n etc. just becoming pnþ1 etc. Thus, for example:
We can now state a formulation for a single pipeline, over a gas trading day. We
assume that flow always occurs in a uniform direction, in the direction from cell
n1 to n to n + 1, and use the endpoint formulation, assuming pipelines of uniform
cross-section, with no abrupt pressure changes due to fittings, as discussed above.
Thus Gn becomes simply G from here on. An appendix discusses modeling of
complications such as compressors, fittings, and junctions. Initial linepack in the
pipeline is inherited from the previous day’s end condition, as specified in (20).3
The end state of the system, at time T + 1, must be set to ensure sufficient linepack
carryover; in the required sections of pipeline, to meet next trading day’s
requirements. In the limit we could try to force final linepack in each cell to a
specific value, representing a desired pressure/flow profile for the start of the next
trading day, between defined limits. But this is too restrictive, and could lead to
extreme price impacts, if not infeasibility, as the system will struggle to meet any
exact profile. Still, we may group pipe segments into “zones”, z 2 Z, for which the
aggregate end of day linepack must be between defined limits, as in (21).4
3
We could also specify initial flows to get a more accurate representation of the nonlinear
equations involved. But that increased accuracy would come at the cost of increasing the
likelihood that the LP could not actually find any feasible initial flow/pressure pattern to exactly
match the specified parameters.
4
A combination of upper and lower bounds may suffice to ensure that pressure differentials are
also large enough to create sufficient flow in the next trading day. But pressure differential
constraints may also be added to ensure this directly, and independently of aggregate linepack
levels.
86 E.G. Read et al.
The formulation seeks to maximize the value from allocating gas across time t,
and space n, as expressed by Eq. 16. To the extent possible, the optimization
balances trade between time periods from t ¼ 1,. . .,T, and over all nodes from
n ¼ 1,. . .,N. Participants submit bids and offers as price (Bidtn and Offertn ) and
quantity (xtdi and xtsi ) combinations. Individual participant involvement depends on
the physical configuration of the system, at each specific location. Many pipe cells
will only have extractions, while many others only injection, but most will have
neither. For each individual consumer or supplier spot market bidding is
represented by Eqs. 16 and 17, where the index subscript i indicates a single bid
or offer tranche from a demand (d) or supply (s) side participant.5 The aggregate net
injection of gas into the system into cell n, at time t, is given by Eq. 18, and upper
and lower limits are imposed on this via Eq. 19. Combining all these market bid/
offer curves with the gas flow and pressure equations, and ignoring linepack
bidding, generates the following nonlinear dispatch formulation.
Maximise !
XX XX XX
x; y; m; p; p~; q; Bidtdi xtdi Offertsi xtsi (16)
q~; rp; v; v~; rv t n d2Dn i s2Sn i
Subject to:6
Mass, pressure, flow and velocity relationships, within a cell:
mtþ1
n ¼ mtn þ qtn qtnþ1 þ ytn : mtn (6a)
Bernoulli energy conservation equation with substitution from (10), over cell:
5
The notation i,d ∈ n means i or d is located at node n. i ∈ D(n) means i is a bid from a demand
side participant at n. i ∈ d means i is a bid/offer from d, etc.
6
All constraints are 8 n ¼ 1,. . .,N and t ¼ 1,. . .,T unless otherwise stated. Greek symbols
associated with equation numbers indicate the key dual variables which will be significant in
later discussion of pricing relationships.
An LP Based Market Design for Natural Gas 87
v~tþ1
n v~t1
n þv vtn Þ2 þ RHrptn p~tn ¼ 0 :
~tn rvtn þ Bn ð~ btn (7a)
Pressure bounds:
t :
Ptn p~tn P t
ftn ; f (13a)
n n
Velocity bounds:
t
Vtn v~tn V n : wtn ; wtn (14a)
dtn ;dtn
t
Ytn ytn Yn : (19)
XX
þ PackBidTþ1 Tþ1
zi xzi (22)
z2Z i
The total system linepack at the end of the period T, across all nodes n in z,
would then have to match the linepack purchased for the final period:
X X
mTþ1
z ¼ mTþ1
n ¼ xTþ1
zi (23)
n2z i2z
If we are to clear the market using LP software, we must first linearize the nonlinear
constraints in the formulation of Sect. 4. Read and Whaley [32] originally proposed
to substitute the mass/pressure and flow/velocity relationship in Eqs. 1 and 4 into
Eq. 6a, then linearize the resultant nonlinear flow/ pressure equation directly. They
also showed how to re-arrange the Bernoulli equation (7a) into two expressions, one
involving only velocities, and the other only pressures, and apply Taylor’s expan-
sion separately to each.
The relevant derivatives can certainly be formed, but they will not be discussed
here, partly because that approach was not actually adopted in practice, and partly
because it creates a dual formulation from which it is not particularly easy to deduce
pricing relationships. But there were also concerns with respect to the accuracy and
convexity of the implied approximation. One proposal was to employ an iterative
successive linearization scheme, with each iteration solving an LP linearized
around the solution from the previous iteration. A coarse discretization can yield
a poor result in this kind of modeling, because small errors can propagate and
compound though the equation set. The goal of the representation introduced in
Fig. 1 was to re-express the underlying nonlinear differential equations by a set of
linear difference equations, with both time and space discretized on a sufficiently
fine grid to make the linear assumption reasonable, and to refine the grid further,
around a proposed solution, if the results were deemed to be too inaccurate. But this
approach is similar to employing Euler’s method to solve the underlying differen-
tial equations. In practice, that first order method is known to have stability issues,
and a fine discretization may be required. Thus higher order methods are generally
applied, as described by Dorin and Toma-Leonida [9], for example.
Thus piece-wise linearization within the LP seemed preferable to successive
linearization. It was proposed that a piece-wise linear model could be produced
using Taylor’s expansions, as above, to create “supporting hyper-planes” around a
set of points spanning the feasible region. A critical issue, though, was whether the
piece-wise linearization so produced would actually form a convex LP feasible
region. It can be shown that the flow equations are not actually convex if flows are
allowed to reverse, but Read and Whaley argued that an acceptable convex
An LP Based Market Design for Natural Gas 89
approximation could be found in the vicinity of any likely optimum. There are
actually two possible issues here, and a later section discusses why piece-wise
linearization may break down in situations where gas, or gas flow, turn out to have
negative value. But, concerns about the potential non-convexity of the feasible
region itself focused on the first two terms in the Bernoulli equation, relating to
velocity changes and kinetic energy, and to the possibility of flow reversal.
Read and Whaley wished to retain these terms in the Bernoulli equation because,
at the time, it was unclear whether they would have any significant pricing
implications. But almost all other authors, including Pepper et al. [26], have
considered those terms small enough to be ignored, on the grounds that a gas,
being very light, has little kinetic energy or momentum. Thus it is reasonable to
assume that changes to gas injection or withdrawal rates will primarily be reflected
in changes to pressure/flow relationships. Friction losses will slow the process, but
velocities will quickly respond without significant expenditure of energy to reflect
this new “steady state”, which then evolves over a longer time frame in accordance
with the flow and mass balance equations. Thus most authors use a steady state
version of the Bernoulli equation, in which the first (time derivative) term is
dropped, and most authors also drop the kinetic energy term. Since the pressure
in (7a) in the average pressure over that whole cell, which is proportional to the
mass in the cell, Eq. (7a) can be simplified to:7
hpffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
q~tn ¼ An =Bn Ln rptn mtn : btn (7b)
This form of the equation clearly encloses a convex feasible region.8 In fact it
forms a convex cone, being linear in mtn along any ray where mtn =rptn is
constant. Although we can now drop (11a), defining rvtn , because it no longer
appears in (7b), v~tn itself is still defined by (4) and appears in the velocity bounds
(14a). But v~tn can be eliminated from the formulation, along with (4) and (9a), by
substituting (4) into (14a) and re-arranging to express those bounds as constraints
on flow, as a function of pressure.9
7
A substitution for velocity in terms of flow and pressure is made from (4) and the equation is re-
arranged in terms of q with coefficients grouped. This steady sate equation assumes that the mass
flow rate is uniform across cell n, and period t. While this approximation is commonly employed, it
is not quite consistent with (4), which allows the mass in the cell to change, implying different flow
rates at each end.
mn is always positive and (12a) ensures that rptn is positive, since we are excluding solutions
8 t
where flows reverse. Thus the RHS is just a constant times their geometric mean which is known to
enclose a convex set [3].
9
Note that, if the lower velocity limit is only used to prevent flow reversal, Vtn will be set to zero,
so the lower bound on mass flow is also set to zero, irrespective of pressure.
90 E.G. Read et al.
If we use (1) to substitute for m in terms of endpoint pressures, ptn and ptnþ1 , while
also substituting for rptn from (12a) (which can then be dropped from the formula-
tion) we get:
h ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii
q~tn ¼ An = 2Bn RHLn ðptn Þ2 ðptnþ1 Þ2 : btn (7c)
This is just the Weymouth equation employed by many other authors, in various
forms. If we let s be the ratio of the downstream and upstream pressures, then
substituting sptnþ1 for ptn makes it clear that this expression is linear in upstream
pressure (or downstream pressure) along any ray from the origin, in the upstream/
downstream pressure plane, over the range of interest. (That is for 0 < s < 1, since
otherwise downstream pressure would be higher than upstream pressure, causing
flow to reverse.) Over that range, this expression also forms part of a convex cone,
as discussed by Tomasgard et al. [38] and Midthun et al. [21].
Zhou and Adewumi [41] take a different route. They form a steady state version
of (7) by dropping the first term with its time derivative, but show how to obtain
analytic expressions for flow which account for the kinetic energy term. Re-
arranging their equation (12), for a horizontal pipe, and re-expressing it in our
notation gives a modified form of Eq. (7c).10
2 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 3 rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
t 2
p
q~tn ¼ 4An = 2Bn RHLn RH ln nþ1 5 ðp t Þ2 ðpt
n nþ1 Þ 2
: btn (7d)
ptn
This approximation can not hold for outlet pressures (and hence outlet/inlet
pressure ratios) near zero, because then the logarithm in the divisor tends to infinity,
and the predicted flow falls rapidly to zero. For more realistic outlet/inlet pressure
ratios, closer to unity, the logarithmic term is close to zero, and this equation creates
only a modest adjustment to the Weymouth formula, implying a slightly greater
resistance to flow. Substituting sptnþ1 for ptn shows that this, too, is linear in pressure
along any ray with constant outlet/inlet pressure ratio, and letting that ratio range
from a small value up to 1 also forms part of a convex cone.
Although (7d) provides a convex formulation which accounts for the kinetic
energy term, we will use the much more common Weymouth type equation in (7c).
Martin et al. [19] discuss a piece-wise linearization that could be applied to either
equation, using “convex combinations”, in the context of a mixed integer formula-
tion. In Victoria a hybrid approach was adopted, using convex combinations to
form a piece-wise linear formulation most of the time, but reverting to successive
linearization when required to deal with “convexity issues”. The detail may be
10
Assuming a compressibility of Z ¼ 1, using the Specific Gas Constant rather than the Universal
Gas Constant, and assessing the pressure drop over a cells length Ln.
An LP Based Market Design for Natural Gas 91
found in Pepper et al. [26]. But Tomasgard et al. [38] and Midthun et al. [21] discuss
piece-wise linearization using hyper-planes created around a set of points, as
suggested by Read and Whaley [32]. Using Maple# (see www.maplesoft.com),
and re-arranging provides (7e) as the linearization of (7c) around a point denoted by
superscript *. Here constant terms are enclosed in square brackets, and the simpli-
fication of Ft
n reflects the fact that the square root term on the top line is just a
constant times the expression for q~t t t
n , the flow corresponding to (pn ,pnþ1 ), while the
( ) in the divisor is just the square of the same term.
t t t
q~tn ¼ Ft
n n pn pnþ1 pnþ1 :
pt btn (7e)
Where
2 r ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3
A ðp t Þ2 ðpt Þ2
6 n n nþ1 7 A2n
Ft ¼ 6 7¼
n 4pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 5 2Bn RHLn q~t
2Bn RHLn ðpt n Þ ðpnþ1 Þ
t n
Thus the final simplified LP formulation consists of equations (2), (5) (6a), (7f),
(13a), (14b), and (16, 17, 18, 19, 20, 21).
6 Pricing Implications
Although Cremer et al. [6] present a high level analysis of some pricing
relationships, we have not seen any systematic analysis of the kind of price patterns
that could arise as a result of modeling gas transport dynamics on the time and
distance scales discussed here. As with any LP, a complete dual formulation could
be stated, and solution of the primal problem will automatically determine the
solution of that dual. But market participants, and market designers, will want to
understand how those prices are driven by offers and bids, and the kind of pricing
patterns that will be produced. To generate that insight, we focus on the key dual
relationships determining the way in which spatio-temporal price information
92 E.G. Read et al.
generated by solution of the LP reflects the opportunity costs of having one more
unit of natural gas available to supply at any time and place.
As always, there will be one dual pricing constraint for each primal variable, and
we can generate that constraint simply by collecting terms and summing them. Thus
if variable xi appears with coefficient aik in constraint k, then the shadow price on
constraint k will appear in the pricing constraint for commodity xi, with the same
coefficient. Also note that if a primal constraint relating to cell n and period
t contains primal variables relating to, say, cell n + 1 and/or period t + 1, then
the corresponding primal variables for period n and t must appear in primal
constraints relating to cell n1 and/or period t1. So the pricing equation for
that primal variable will involve shadow prices computed for those constraints.
Ultimately we are really only concerned to price commodities traded in the market,
in this case gas injected/extracted, and possibly end-of-day linepack.11 In other
words, ultimately, we are mainly interested in the shadow prices on constraints
(18). However, these prices depend on the prices of other (non-traded)
commodities, and all will ultimately be determined by what is effectively the
solution of a set of simultaneous equations, in the LP solution process. Thus we
need to consider some other pricing relationships as well.
First note that for a maximization objective, standard duality theory implies that
the shadow prices on < constraints will be positive, while those on > constraints
will be negative. We have expressed all upper bounds in our formulation as <
constraints, and all lower bounds as > constraints. So the shadow prices on all
upper bounds will be positive, while those on all lower bounds will be negative.
Thus adding the shadow prices on upper and lower bounds effectively creates a
composite shadow price, which will be positive if the upper limit binds, and
negative if the lower limit binds. (Similarly. the shadow price on an equality
constraint will be positive if it binds as an upper limit and negative if it binds as a
lower limit.). With that convention in mind, the pricing equations corresponding to
the variables for traded quantities, xtdi and xtsi , and for net injection ytn , are:12
ltn ¼ Bidtdi gtdi þ gtdi : xtdi (24a)
ltn ¼ Offertsi þ gtsi þ gtsi : xtsi (24b)
ltn ¼ mtn dtn þ
dtn : ytn (25)
11
By way of analogy, an electricity market formulation such as that in Alvey et al. [2] may be used
to determine prices for line capacity, and even phase angles, but we really only focus on prices for
electricity injected/extracted.
12
Primal variables are associated with each dual equation, just as dual variables were associated
with each primal equation.
An LP Based Market Design for Natural Gas 93
These conditions are easily interpreted. First, (24a) implies that the local price, l,
will equal the price for some step of the local bid/offer stack if, and only if, that bid/
offer is “marginal” at the optimum. That is, if and only if the market is free to take
one more, or one less, unit from that step at the offer/bid price because it is not up
against either the upper or lower limit of that step. The whole price system is driven
by these (marginal) bid/offer prices. Otherwise, one of the g prices will be non-zero.
For offers, the upper (or lower) limit will bind when the market price, l, is above (or
below) the offer price and (24b) merely shows how gti (or gti ) adjust to reflect that
difference. For bids to buy gas, the situation is reversed, because a positive buy
variable corresponds to a decrease in net supply, and has the opposite coefficient in
the objective function.
Second, we can take mtn , the shadow price on the mass conservation constraint
(6a) to be the system price for gas injected into the main transmission system at that
time and place i.e. ytn .13 From (25), the local price, l, will equal m if, but only if, the
market is free to take one more, or one less, unit from that location, at the local
price, because it is not up against either the upper or lower injection/extraction limit
at that location. Otherwise, if multiple participants want to inject (or extract) more
gas at n than the bottleneck constraints (19) will accommodate, d (or d) will be non-
zero, and m will be higher (lower) than l, in order to throttle injection (extraction)
back to the bottleneck capacity limit. Local participants could trade between
themselves at the l price, in order to ration limited injection/extraction capacity,
but that local price will not impact on prices anywhere else in the system.
The issue is, though, to determine the system price, m, for non-marginal
locations, where injection/extraction is constrained, or limited by upper/lower
offer/bid limits. Fundamentally, the price of a unit of gas at any point in the system,
and in time, is determined by the marginal value that gas may have in meeting
future requirements (or reducing the need for future supply) at some time and place.
In this deterministic market-clearing formulation the gas price will also be the
marginal cost of supplying gas to that point in time and space, from whatever
sources are marginal. Looking at the issue either way, the value of gas at each time
and place must be consistent with prices at adjacent times and places, which must
be consistent to prices at times and places adjacent to them, and so on, until
13
This is a slight simplification, with respect to the original formulation, because injected gas will
also have a velocity of its own, and if this was modeled it would have some small impact on the
solution of the gas transport equations. Thus, in principle, we could have differing prices for “fast
gas” and “slow gas” injected at the same time and place. But no such distinction arises here,
because the simplified formulation ignores gas velocity.
This situation is conceptually similar to that with respect to electricity injected or consumed
with differing “power factors” in electricity markets. Hogan et al. [16] discuss a regime that would
explicitly price the active and reactive components determining power factor. But real electricity
markets typically only price and trade active power, using a DC approximation to the power flow
equations, while relying on other agreements to control power factor within acceptable limits. We
assume the same to be true here, with respect to the setting of injection pressure differentials, and
hence velocities.
94 E.G. Read et al.
ultimately the entire price system is driven by a small set of marginal offer/bids. In
our simplified model these marginal prices will be for gas bought or sold in some
node in some period of the day. (If linepack trading were allowed for in the
formulation, the marginal prices be prices for end-of-day linepack, as determined
by some Packbid)
Just as for electricity markets, the equations linking all these system prices
together are the duals of the equations linking the physical quantities together in
the gas transmission system. These are the mass balance equations, the shadow
prices on which are the system gas prices, m, and the Bernoulli equations which
determine how gas flows, with shadow prices b. And the key variables, in our
simplified LP formulation, are the pressures, ptn , which (with volumes fixed)
effectively measure the mass of gas available at each point in the system, and qtn ,
measuring the flow rates between adjacent locations. Thus we must consider the
dual equations associated with these variables. We first discuss the impact which
simple limits on flows through space and time would be expected to have, as in the
case in a market for stored water, for example. We then discuss how these results
will be affected by terms arising from the Bernoulli equation and velocity limits, so
as to produce pricing effects which may not be immediately intuitive.
First, gas prices will vary over time. If gas was being stored in a static fashion,
like water in a reservoir, then we would have a simple equation linking the price of
gas in successive periods to the shadow prices on the upper/lower storage bounds.
That is, the price of gas stored in the cell would be the same in each successive
period, unless a pressure (i.e. storage) limit was binding. Conversely, an upper
(lower) mass/pressure limit would be binding if the price for gas in the next period
was higher (lower) than in the current period, giving the system incentives to
maximize (minimize) gas carried forward. Note that Eq. 1 is really only a conve-
nience, allowing simplification of some of the equations. Simplistically, each unit
of pressure in pipe cell n implies GLn units of mass there. Thus, dividing by GLn
converts prices associated with pressure in cell n, (ftn and f t ), to be compatible
n
with prices associated with mass variables for that cell (mtn from the mass balance
equation). So, in this simplified model, the price (mtn ) for injected gas (ytn ) would be
inferred from the following equation, describing the way in which those prices
evolve over time:
mtn ¼ mt1 t p~tn
n þ fn þ fn =Ln G :
t
(26)
Second, gas prices will also vary over space. Simplistically, we might expect the
price of gas moving through the pipe to be the same in each successive cell, unless a
flow limit is binding. And we might expect an upper (lower) flow limit to be binding
if (and only if) the price for gas in the next cell is higher (lower) than in the current
cell, giving the system incentives to maximize (minimize) gas flowed forward. This
would produce inter-nodal pricing impacts analogous to those arising in electricity
markets. But we do not have (mass) flow limits, per se, in this formulation, only
velocity limits in (14b), and a “friction” term in the simplified Bernoulli (Weymouth)
An LP Based Market Design for Natural Gas 95
equation (7f) which slows flow, but does not ultimately limit it. And the gas “stored”
at one time and place also influences the rate at which gas flows to other places, over
time. Thus Eq. 26 is too simplistic. To develop a more accurate representation, of the
way in which these pressure relationships affect prices, we first eliminate mass, m,
from the formulation, by substituting (1) into Eq. 4,14 to get:
Ln G p~tþ1
n ¼ Ln G p~tn þ qtn qtnþ1 þ ytn : mtn (6b)
!
X
mtn ¼ mtn1 þ btn;k þ btn1;k þ wtn1 þ wtn1 þ wtn þ wtn =2 :
k (29)
qtn
This equation tells us that the price of gas in cell n reflects the price of gas in cell
n1, upstream, plus the implied cost of moving gas from cell n1 into cell n. That
cost is determined by the shadow prices on constraints limiting flows between
adjacent cells. Since the mtn prices are for cell midpoints, the price difference is half
determined by conditions in each cell. The friction term in the Bernoulli equation
(priced at btn ) has a pervasive effect in terms of limiting and slowing inter-cell
14
And also in Eq. (21), for t ¼ T + 1, but that is not relevant here.
96 E.G. Read et al.
flows.15 The (velocity) flow limits can also play an important role, if binding. If the
upper flow limit binds, wtn will be positive, and the value of gas in the downstream
cell will be higher than in the upstream cell. But if the lower limit binds (typically at
zero) downstream gas will be worth less than upstream gas, and this change can be
quite abrupt. This may be a proper reflection of the situation, if reverse flow is
physically blocked by a check valve. Otherwise the model solution may not be
physically feasible, because there is no way to physically stop flow from reversing,
and alternative solutions should be explored, with flow restricted to be in the
opposite direction.
Second, inter-temporal price interactions are primarily determined by the dual
equations for the pressure variables.16 Because the endpoint pressure variable, ptn ,
actually appears in the Bernoulli constraints for both cells n and n1, and also in the
pressure averaging equation for both cells, the corresponding dual equation (30)
relates the prices on all four of those constraints. But we also have a pricing
equation for the midpoint pressure, (~ ptn ), which appears in the pressure averaging
equation (2a), and in the pressure and velocity bounds (13a) and (14b) for the
midpoint of cell n. So the prices for those constraints appear in the corresponding
dual equation (31), which can be re-arranged and re-scaled, to give equation (32),
describing the way in which the price of gas (mass) in cell n evolves over time.
X h i t t t
ðctn þ ctn1 Þ=2 ¼ n1;k pnk bn1;k Fnk pnk bnk
Ft t t
: ptn (30)
k
t t t t t
ctn ¼ G ½Ln mtn ½Ln mt1 þ V w þ V
w f þ t :
f p~tn (31)
n n n n n n n
n t o
mtn ¼ mt1 t G V t wt þ Vt
n þ ftn þ f n wn þ cn =GLn :
t
n n n mtn (32)
The first pair of shadow prices in { } reflects the impact of pressure (storage)
bounds in period t, as for the hypothetical simplified model discussed above. The
middle pair of terms in { } reflects the fact that having more gas in a cell increases
the pressure and hence, for a constant velocity, the rate at which gas (mass) can flow
through the cell. If either velocity limit (Vtn ,Vtn ) is binding then, it will have a non-
zero shadow price (wtn ,wtn ). If the lower velocity limit is only used to prevent flow
t
reversal, Vn will be set to zero, so that term disappears from this equation, so that
constraint plays no role in determining inter-temporal price differentials. Its shadow
price, wtn , may still contribute to inter-locational price differentials, though, via
15
As noted earlier, the Zhou and Adewumi [41] equation (7d) effectively implies a small increase
to this term, and hence a small increase to inter-spatial differentials, but the pricing effect will be
small enough to ignore if the physical impact is small enough to ignore.
16
In principle, the time derivative terms in the full Bernoulli equation, (7), would create a further
inter-temporal link between prices. But we consider that influence to be small enough to ignore if
the terms themselves are small enough to ignore.
An LP Based Market Design for Natural Gas 97
Eq. (29) above. If the upper velocity limit is binding, the wtn term will reflect a
benefit from increasing pressure so as to increase mass flow. But that strategy may
be constrained by an upper pressure limit in this cell, in which case the (f t ) term
n
associated with the upper pressure limit in Eq. 31 will rise to offset the wn term here.
t
If the binding pressure limit is in a different cell, its impact will be reflected by the
price of gas delivered to cell n rising high enough to make any further pressure
increase there unattractive.
The last term in { }, ctn , summarizes the inter-temporal pricing impacts of the
Bernoulli equation, as determined by Eq. 30. The RHS of (30) reflects the way in
which higher gas pressure at the input end of cell n, ptn , speeds the flow of gas though
cell n to cell n + 1, while inhibiting the flow of gas through to cell n from cell n1, in
accordance with the Bernoulli equations for cells n and n1. Summation over k
captures the possibility that more than one supporting hyper-plane from the piece-
wise linearization may be binding. The LHS of (30) reflects the fact that higher gas
pressure at the input end of cell n, increases midpoint pressures in both adjacent cells.
Note that (30) involves prices for two adjacent cells, and we can not simply
substitute (30) into (31) to get a complete and explicit expression for the way in
which prices in cell n evolve over time without any reference to effects in other
cells. This reflects the chain-like way in which mid- and end-point variables, and
hence prices, are linked along the pipeline. But we can substitute (31) into (30)
though, and re-arrange to get:
(
X h i t t t
ptn ¼ pt1
n þ 2 Ft p t
n1;k nk bt
n1;k Fnk pnk bnk
k
t (30a)
G Vtn ðwn þ
wtn Þ þ Vtn1 ðwtn1 þ wtn1 Þ
o
t þ ft þ f
þ ft þ f t = G½Ln þLn1
n1 n1 n n
We can think of ptn as the average price for gas in a “pseudo-cell” centered on the
boundary between cells n1 and n, and running from the midpoint of one cell to the
midpoint of the next. This price is arguably the correct price for gas injected at a cell
boundary, and we can create a cell boundary at any point where gas is to be priced.
Then Eq. 30a gives an explicit expression for the way in which the price of gas
injected at that point evolves over time, in terms of the impact gas injected there has
in both adjacent cells. Other variants can be produced by manipulating the dual
equations and/or varying the primal assumptions. For example, we may impose
pressure or velocity limits at cell boundaries, rather than at midpoints. But, since the
cells modeled can be arbitrarily short, that kind of change does not fundamentally
alter the nature of the physical outcomes, or the pricing impact of these equations.
The corresponding price terms (ftn and wtn ) just appear in basically the same form,
but in Eqs. 27 and 30 rather than in Eqs. 28 and 31.
98 E.G. Read et al.
Whatever variant of these equations is preferred, they define the key pricing
relationships, linking prices over time and space, to create a pattern of gas prices, all
driven by marginal offers and bids as discussed previously. The price of gas at any
time and place will not only reflect the value that gas will have when it is finally
delivered to the location at which it will be consumed, but also the indirect value it
may have (positive or negative) in terms of assisting or resisting the flow of gas to
other places, at other times, where it may prove to be more, or less valuable. The
Bernoulli terms mean that the price may vary from period to period, and from place
to place, even if there are no “local” or “immediate” pressure bounds limiting the
amount of gas that can be “stored” from one period to the next, and no (absolute)
flow bounds limiting the amount of gas that can be “moved” from place to place.
All other shadow prices in the dual (including shadow prices on initial and final
storage constraints, (20) and (21)), merely adjust to match that pattern.
This situation is analogous to that arising in electricity markets, where a flow
constraint on a single link will generate a distinctive “spring washer” price pattern,
first described by Ring and Read [33], implying price differentials across all links
involved in any loops in which that constraint is involved. These effects arise
because power flows according to the laws of physics, splitting across all possible
parallel paths in inverse proportion to their impedance. Prices must reflect the fact
that some part of any incremental flow will travel over the over-loaded circuit,
because it is not possible to “direct” flows to take alternative parallel paths avoiding
it. The gas system is similar in that, while valves and compressors give some degree
of control over how gas will flow, gas will flow though much of the system entirely
according to the laws of physics, not economics. Thus a single binding constraint, at
some time and place, will cause difficulty in delivering gas to various downstream
locations at, or over, various subsequent periods. Thus it will generate price
differentials across space, as in electricity markets, but also across time.
Price differentials could become quite extreme if, as sometimes happens,
extreme measures must be taken to keep the system operating. In Victoria, an
LNG stockpile is maintained near Melbourne, with stocks being gradually built up
over an extended period, so as to be available for release when required in order to
maintain pressures when demand is too high to be met by continuous supply
through the main pipeline system. The operation of that stockpile is optimized
outside the market, as is the operation of other storage facilities, such as the
Western Underground Storage Facility (WUGS). This gas is all purchased at
market prices, when they are relatively low, then re-sold at times when the marginal
value of gas, and hence the optimal gas price, must be very much higher, at the LNG
facility. Prices may be even higher at the critical time and place which actually
creates the need for such release, if that is not the LNG facility.
In electricity networks, constraint pricing effects are not the only possible
drivers of price differentials, though. If transmission system losses are modeled,
as in Alvey et al. [2], they will cause pervasive price differentials between all
locations, even when no constraints are binding. These transmission system losses
are not really analogous to the “friction loss” terms in the Bernoulli equation,
though. This is not a loss of gas, but a loss of energy, and its effect is to slow and
An LP Based Market Design for Natural Gas 99
delay gas delivery, rather than ultimately to limit it. This does represent a potential
barrier to the efficient and timely transfer of gas from producers to consumers, but
that will only imply inter-locational, or inter-temporal, price differentials in
situations where the delay, in combination with insufficient linepack in the right
part of the system, forces some other constraints, such as pressure or velocity limits,
to bind. When price differentials do occur, these terms, of themselves, typically
imply gradual change, as the cumulative effect of the Bernoulli equation on each
segment of the pipeline means that flow rates gradually decrease as distance
increases.
Compressor operation does imply gas losses, though, and these are accounted for
in Eq. 6c, in the Appendix. Thus the price of gas downstream from a compressor
must rise in proportion to the marginal gas consumption of the compressor. If
achieving the desired pressurization requires a compressor to consume 1% of the
gas passing through it, the compressor effectively converts 100 units of upstream
gas, at the upstream pressure, to 99 downstream units, at the higher downstream
pressure. So, if there were no other costs or constraints involved, they would have
the same total value, with the marginal value therefore needing to be (approxi-
mately) 1% higher on the downstream side. Abrupt price change can occur at
compressors where flow is constrained by a minimum or maximum flow limit,
though.
7 An Example
While the equations in the previous section allow us to infer how prices relate in
adjacent cells, and periods, we have not made, or seen, any systematic attempt to
determine, the variety of system wide price patterns that might emerge. But Annex
3 of Frontier Economics [11] presents some empirical analysis, based on the results
produced by the MCE of Pepper et al. [26] for a number of scenarios. That model
was developed as a pragmatic replacement for the original conceptual formulation
developed by Read and Whaley, and reported here. It ignores the time derivative
and kinetic energy terms included in the original formulation but, if these are small,
they will also have little impact on prices. While it is linearized in a different way,
the pressure/flow relationship employed in this model is essentially the same as that
in Eq. 7c above. This approximation has proved sufficient to produce a very good
approximation to physical gas flows in the system. Thus we believe the price
patterns produced by that the implemented MCE model to be indicative of the
kind of price patterns likely to emerge from any implementation of the fundamental
market design concept developed here. That is a nodal market based on an LP
representation of the underlying network realities, on a short time scale. Here we
discuss the price patterns produced for just one of the scenarios considered by
Frontier, as reported by Pepper [25].
Figure 2 gives a general locational overview of the Victorian Gas System in
terms of main pipelines and nodes. Figure 3, taken from Pepper shows the kind of
100 E.G. Read et al.
price pattern that could occur in this system for a day in which a constraint binds.
Pepper notes that prices almost always decline to a flat off-peak value at around 10
PM, with some lag in the outer portions of the system. (Hence, prices after midnight
are omitted from Fig. 3). This behavior after the evening peak reflects the fact that
An LP Based Market Design for Natural Gas 101
the intra-day linepack constraints are no long binding and the MCE only needs to
achieve a minimum system-wide linepack constraint by the end of the gas day. If
the optimization were changed to a 48 h optimization and the next gas day was also
constrained, this flattening of prices may not always occur.
The relatively high prices during this particular gas day stem from two underly-
ing causes. First, the system did not start the day from a position of unconstrained
equilibrium, but inherited a gas pressure/flow pattern which made it difficult to
meet the day’s requirements. This may be seen by the fact that the model assigns
differing values to gas in different locations, even at the start of the day. Second,
demand for gas during the peak period of the day exceeded the ability of the system
to deliver gas from the low cost supply at Longford, which was constrained by
production and gas processing plant capacity. Thus, in this solution, pressures were
expected to reach minimum allowed levels early in the evening peak period, at
Bendigo Junction and at other key points, such as the Dandenong City Gate (DCG),
towards the end of the evening peak period. Some higher cost supplies such as LNG
or stored gas are thus required to keep pressure at, or above, the minimum pressures.
Thus by 1 PM, the system is clearly struggling to get enough gas through to
Melbourne to cover requirements over the rest of the day. So prices rise over the
day, then collapse after the critical hour. Prices also rise at Longford, the main
injection point, but prices there fall earlier because gas can not reach the critical
areas in time to make any difference. Or, more exactly, the value of injecting more
gas at that place, in terms of maintaining a pressure differential to increase flow
through to the critical area by the end of the critical period falls gradually over
several hours as the end of that period approaches.
Iona is, in sense, at the opposite end of the system, at the end of the South-West
pipeline. But prices at Iona follow a similar pattern to those at Longford, presum-
ably because Iona also acts as a source from which gas can flow to Melbourne
within the critical period. Prices on this pipeline start falling earlier in the day,
though, presumably because extra gas at Iona will only have a positive impact on
Melbourne delivery if injected early in the day. Bendigo also lies at the end of a
pipeline, off the opposite side of the outer pipeline loop from Longford. But
Bendigo is not a source, and prices there follow a similar pattern to those at
Melbourne. Pepper [25] reported that there is a capacity constraint between
Melbourne (DCG) and Bendigo, restricting Bendigo gas availability during the
peak period. The Bendigo price drops sharply after 3 PM because changes in supply
or demand of gas at Bendigo no longer have much impact on the ability to meet
demand at Melbourne through to the end of the evening peak.
This fall below the end-of day value is much more marked for Springhurst, near
the neighboring state of New South Wales. There the price drops so early, and so
low, as to apparently exhibit almost the opposite of the Melbourne pattern, only
rising slowly to match the end-of-day price at the end. Pepper reports that gas is
actually flowing north, away from Melbourne, even when prices are higher in
Melbourne, at that time. As noted earlier, gas will flow in accordance with the
laws of physics and, where there is no valve to control flows, there is no reason why
that flow should necessarily enhance economic value. Exactly the same situation
102 E.G. Read et al.
arises with respect to loop flows in electricity networks, where counter-price flows
are common.
In any case, it is not the price difference at any particular time that determines
optimality. The critical issue here is maximizing flow to the major load, at
Melbourne, over the critical period. Extra gas in the Northern pipeline, early in
the day, is actually assigned a similar value to Melbourne gas because the increased
pressure can inhibit northward flows during the critical period, thus allowing the
system to meet Melbourne requirements. After 3 PM, though, the value of extra gas
there actually falls below the end-of day value. This is because the impact of one
unit of additional supply, or reduced demand at Springhurst, requires a reduction in
the flow which the model achieves by reducing pressure at the supply end of that
pipeline, thus impacting the ability to meet demand on the Northern section of the
pipeline.17
The discussion of convexity in Sect. 5 glosses over one significant point which can
prove troublesome in implementing this kind of model, whether or not it is used
for market clearing. We focused on the convexity of the LP feasible region. But
(7c) is an equality, not an inequality, and the actual physical flows are confined to
lie exactly on the boundary defined by that equation, rather than within the region
bounded by it. As with any nonlinear equation, the set of points it defines can
never be convex, even if the equation defines the boundary of a convex set. So the
physical feasible region for this problem is definitely not convex. Still, piece-wise
linearization is often used to model this kind of situation in LP models, including
those of Thomasgard, Midthun and others. This does not create a problem so long
as the objective function implies that points on the boundary are preferred
to physically infeasible interior points. It will break down, though, if that is not
the case.
A similar situation arises when a piece-wise linear representation is used to
model quadratic losses for electricity markets, as in Alvey et al. [2]. This creates a
convex LP feasible region, and solutions will lie on the appropriate boundary
provided losses are economically undesirable. This is almost always the case, but
Ring and Read [33] note that a switch must be made from piece-wise linearization
17
In this case, though, the effect is at least partly due to the fact that this section of pipeline is
represented with one pipeline segment. If the representation of the pipeline to Springhurst was
divided into multiple pipeline sections, the MCE would be better able to account for the dynamics
of the flow relationships and these prices would not fall so far.
An LP Based Market Design for Natural Gas 103
to successive linearization if the model determines that the optimal electricity price
at some point in the network would be negative. This can actually occur, in
situations where increased load (or losses) would relieve pressure on constrained
lines in a loop. And that will make it seem desirable for the optimization to propose
“solutions” which are not physically feasible, because they imply losses greater
than would actually occur, for the specified flow level. Analogous situations could
occur here, if the value of having more gas pressure, or a greater pressure differen-
tial, becomes negative at some time and place, most likely because it forces gas to
flow away from where it is needed, and there are no check valves available to stop
that occurring.18
In practice, this situation is handled by switching to successive linearization, as
discussed by Pepper et al. [26]. But a closely related situation occurs when the
model determines that flow reversal would be desirable. We have restricted
velocity, and hence pressure differentials to be positive in a defined direction,
partly because the Weymouth equation is not convex if extended into the range
where the inlet/outlet pressure difference, and hence the flow direction, reverses.
This gives us a convex optimization problem, with a unique optimum. There are
cases, though, in which a quite different alternative optimum could be considered,
with the flow on some pipe segments reversed. Gas could be compressed into a
dead-end pipeline segment, for example, and then allowed to flow back out to
meet peak demand. And the existence of a ring structure in the DTS suggests that
some locations could be supplied sometimes from one direction, and sometimes
from the other. In many cases this may not matter, in the sense that the alternative
strategies do not greatly affect economic value. But the gas system operator may
face some real “integer” choices between significantly different operating
strategies.
Ideally, an integer optimization, such as that in Martin et al. [19], could be
employed to ensure that the true optimum is found. In reality, the plausible range of
operating strategies is quite restricted, at least for this relatively simple system. If
the model is observed to force some flows to their lower limits, the operator may
make integer decisions with respect to valve and compressor settings, or just with
respect to desired flow direction on certain pipeline segments. Given those
decisions we can set the limits in (14b) so as to maintain minimum flows in the
desired direction, and re-solve using the convex linearization valid for that flow
direction but that does mean that the prices determined by the model potentially
depend on some high level strategic choices made by the operator, and that may of
significant concern, from a participant perspective.19
18
One could imagine this happening in a more extreme case of the Springhurst example above.
19
It has been suggested, though, that at least some of the observed price effects could have resulted
from sub-optimal compressor settings and from inter-temporal constraints on the bids such as
overly constrained hourly ramp rates or minimums on hourly injection quantities.
104 E.G. Read et al.
The existence of price differences, over both space and time, means that there will
be a potentially significant “settlement surplus” remaining after all accepted bids
and offers have been cleared at the prices produced by the LP optimization. This
settlement will be the sum of “rents” collected on all binding constraints. Even
when constraints do occur, though, differentials will typically still be small if those
constraints can readily be worked around, for example by adjusting compressor
settings. Running compressors creates what is effectively only a small loss of gas
from the system, and implies equally small differentials, as discussed earlier. If the
gas loss factor were constant, this price differential would be just enough to pay for
gas losses, thus making no contribution to recovering the cost of the compressor
itself. Since the price difference across a compressor reflects marginal losses, and
compressor loss functions are convex (see [26]), rents will be generated equal to the
price of gas at that point in the network, times the difference between marginal and
average losses.20 But this rent is also small.
Larger price differentials, and hence larger rents, will arise when compressors
reach their throughput limits, and/or flows are limited by the other constraints
discussed in Sect. 6. As discussed there, a single constraint, binding in a single
period, may generate price differences between various locations at various times,
and between various times at various locations. Indeed price differences can arise
even when no pipeline segment is constrained at all, in terms of absolute flow
capacity. So, rent will be collected across a great many links, and periods, where
no constraint is binding. This is analogous to the situation in electricity markets,
where a single line constraint in a loop will generate price differences, and hence
rents on all lines involved in that loop. In both cases, though, the total rent
generated by each binding constraint must equal its RHS value times its shadow
price.
In the electricity market literature, there has been much debate about the extent
to which nodal price differentials, and rents, could or should signal, incentivize, and
perhaps fund, transmission network expansion. The desire to signal and incentivize
gas network expansion was a significant consideration in developing this gas
market framework, too. But we should caution against assuming that rents derived
from inter-nodal, or inter-temporal price differences will prove sufficient, of them-
selves, to fund all optimal network enhancement. If we were to solve a joint
operation/expansion optimization, assuming compressor and pipeline, capacity to
be continuously expandable with convex costs, we would find an optimum at which
the marginal cost of expansion equaled the marginal rent assigned to compressor
capacity, by the market clearing prices, on average in NPV terms.
20
Analogously, a quadratic loss function for electricity transmission implies that marginal losses
are always twice average losses, thus generating rents equal to half the loss-induced price
differential, even for un-constrained transmission lines.
An LP Based Market Design for Natural Gas 105
8.3 Hedging
One major factor inhibiting further development of the gas market towards a nodal
pricing paradigm is that participants fear that they could be exposed to significant
price differentials, and not be able to purchase any form of insurance to cover the
implied trading risks. Following the electricity market analogy, the development of
hedging instruments similar to the “Financial Transmission Rights” (FTRs) devel-
oped by Hogan [15] gas been proposed. But a key requirement for an FTR regime to
work is that FTRs not be issued beyond what the system is (expected to be)
physically capable of delivering. Otherwise, Hogan shows that a revenue adequacy
problem arises, because the rents generated on the binding constraints will not be
sufficient to support the payments demanded by FTR holders.21 Conversely, if the
flow pattern corresponding to the set of all FTRs held by participants lies within the
convex feasible region of the market clearing formulation, the implied “FTR flows”
21
We do not expect “revenue adequacy” to cover the cost of gas actually consumed by
compressors, any more than the cost of actual losses is covered for electricity markets. In both
cases these costs must be face by traders as a residual differential, or covered in some other way.
106 E.G. Read et al.
on the lines which turn out to have binding limits in the spot market clearing
solution must be no more than their capacity. The rent required to support FTR
payments matches the FTR flows times the shadow prices on those constraints,
which must be no more than the rent collected as settlement surplus, that as
determined by the full RHS capacity of the binding limits, times their respective
shadow prices.22
The situation is essentially the same for the gas market formulation described
here, except that the “transport network” allows gas to be transferred over both time
and space, and constraints may thus be on flows from one cell to another, or on one
time period to another.23 Ignoring the possibility of line-pack, what could clearly be
supported would be FTRs with a defined delay time, hedging the difference
between the gas price in one cell and “start” period and that in another cell in that
start period, plus a specific delivery delay. If no constraints bind in such a way as to
(directly or indirectly) limit that flow we expect the two prices to be identical. But
otherwise, just as for Hogan’s electricity model, the rent collected on the flow
limiting constraints should suffice to provide hedging for the volume that can be
physically transported, with that delay.
Alternatively, we could decompose each “delayed flow” FTR into two
components: An instantaneous inter-locational FTR, as in electricity markets, and
a locationally specific inter-temporal FTR, hedging between the prices at two
different times, for the same location. Ignoring line-pack, neither of these FTR
components needs to be physically feasible, on its own. The situation is not really
very different from that arising in an electricity market for which FTRs are all
expressed with respect to some reference hub. In such a market an FTR from A to B
can be decomposed into an “A-to-Hub” component and a “Hub-to-B” component.
But the transmission system does not need to be able to support the requested
volume of flows from A to the hub, or from the hub to B, only the net flow pattern
after all requested flows have been accounted for. In the gas market case we can
think of “cell j at time t” as being analogous to a hub. Thus we can define and issue
instantaneous inter-locational FTRs, from “cell i at time t” to “cell j at time t”,
22
This holds even though a single binding constraint may generate price differentials, and hence
rents, across all lines involved in any loop in which it is involved. One way to see this is to solve
the simultaneous equation system defining power flows in terms of net nodal injections so as to
express the line flow directly in terms of net nodal injections. Since a binding constraint holds with
equality, the total rent collected on the RHS side of the constraint will be broken down into a set of
“nodal rents” on the LHS of the constraint. These nodal rents correspond to the rent collected on
that part of a notional flow from the node to a reference node which passes over the constrained
line. This representation of the constraint rents making up the settlement surplus can be used to
construct constraint based “flow gate rights”, as in Chao et al. [5], or classic FTRs, as in Hogan
[15].
23
Convexity issues will arise with respect to “integer” decisions, such as valve or compressor
settings, and possibly flow directions. But that is also true with respect to the “integer” decisions,
such as breaker or transformer settings, determining the configuration of electricity networks. In
both cases any issuer of FTRs must take care to assess the feasibility of supporting those FTRs
across the range of network configurations that might apply on the day.
An LP Based Market Design for Natural Gas 107
provided we also define and issue locational inter-temporal FTRs from “cell j at
time t” to “cell j at time t + delay”. Once issued, such instruments could not be
traded independently, but they could be traded using a market clearing optimization
that guarantees simultaneous feasibility.
The gas system can support a much wider range of FTRs than this, though,
because gas stored as line-pack can typically be released over a wide range of
intervals. Thus there is no fixed delay between the time at which gas is injected at i,
and the time it is extracted at j. So, for convenience, we could define and issue
instantaneous inter-locational FTRs, from “cell i at time t” to “cell j at time t”, and
we could also define and issue a wide variety of locational inter-temporal FTRs
from “cell j at time t” to “cell j at time r”. Here r may be greater than t, but it could
be less than t. In other words it may be possible to extract an incremental unit of gas
earlier in the day, provided we know that it will eventually be replaced by a unit
injected at time t, and arriving some time later, and that no constraints will actually
be violated in the meantime. In particular, instantaneous “trade” will often be
possible, even though instantaneous “transport” is not.
The feasible range of such transactions will be limited by binding constraints on
pressures or flows, in which case the system will need to incur the costs of re-
dispatching other injection, extraction, or compression, in order to make this
transaction possible. But the marginal cost of such re-dispatch is exactly what the
inter-nodal and inter-temporal price differences measure. And the shadow prices on
the binding constraints that determine the inter-nodal and inter-temporal price
differences will also generate the rents required to support any simultaneously
feasible pattern of inter-nodal and inter-temporal FTRs. More generally, the rents
should support any simultaneously feasible pattern of mixed spatio-temporal FTRs,
whether or not they are decomposed as in this discussion.
To date, difficulties in conceptualizing hedging arrangements have proven to be
a significant deterrent to introducing greater spatio-temporal price differentiation
into this relatively small market. But the mathematics of hedging in this kind of gas
market seem closely analogous to that in nodal electricity markets. It may seem
complex to determine simultaneous feasibility, over both space and time, but all
that is required is to notionally clear the proposed FTR trades through a version of
the spot market clearing optimization, just as in electricity markets.
and limit the potential for troublesome “loop flow” effects of the type that at least
partially motivated electricity market reform in the US, for example. Thus tradi-
tional market paradigms may be relatively more effective in the gas sector than they
were in the electricity sector. Nor is there so much need for absolute real-time
coordination.
In this particular case, the net volume traded between participants is also not
very large, since much of the gas is effectively transported on behalf of vertically
integrated participants, who inject their own (contracted) gas at one location, and
extract it at another. Still, Pepper et al. [26] describe a detailed LP optimization
model that does calculate spatio-temporal prices as above, and the dispatch
schedules associated with those prices are used. But they also describe how actual
trading prices are determined using a simplified version of the model, in which the
gas system is modeled like a simple “tank”. That is, gas injected at any location, at
any time during the day, is assumed to be able to supply demand at any other point,
and time of day. Intra-day price differentials arise because the tank model is re-run
several times during the day, but a single gas trading price is calculated each time
the model is run for (the remainder of) each trading day.24
If no transmission system constraints ever bound, the tank model would always
suffice to clear the market. Congestion certainly can occur in Victoria, and give rise
to significant pricing effects when it does, as in the example above. The tank model
under-estimates the cost of operating the real market at such times, and determines
a price which is not consistent with the costs of all suppliers or consumers.
Commercially, this is dealt with by “uplift” payments to compensate participants
mis-dispatched relative to the daily gas price. But Frontier [11] found that this did
not happen often enough to justify moving towards full inter-temporal pricing
framework developed here. And we understand that subsequent network
developments may have reduced congestion, and averted the need for further
market development along these lines.
But the reason the industry has not proceeded further with a more granular
market design is definitely not because the experimental evidence suggests that
optimal market-clearing prices would always be the same at all times and all
locations. On days when constraints bind, price differentials would appear to be
of a similar order of magnitude to those found in electricity markets, over both
space and time. But that raise a different barrier to further development, because
participants are reluctant to expose themselves to the risk implied by potentially
significant spatio-temporal price variations that may not well understood, and can
not be hedged without development of FTR instruments for which there is no
internationally established theoretical framework, or precedent. This concern is
24
Originally, the market design included end-of-day linepack trading, thus including a version of
(22) in the objective function. However, the concept was dropped, due to concerns over price
manipulation, and because the feasible end-of-day target linepack range was considered too small
and sensitive to be managed by participants.
An LP Based Market Design for Natural Gas 109
re-priced, perhaps radically, raises significant questions about the validity of deter-
ministic formulations of the type discussed here. Unlike electricity markets we can
not rely on participants to manage this in-transit stock, and the relationship between
current and future price is determined by the “hard” mathematics of physics and
duality, not by trading in futures market reliant on “softer” participant judgments.
Thus it may be that market-clearing concepts will eventually have to be developed
further, to incorporate stochastic formulations.
Pepper et al. [26] shows discusses how to deal with several complications ignored
in our simplified formulation via simple extensions of the approaches discussed in
Sect. 5. Compressors play an important role in many gas systems. By compressing
gas at one location they not only allow increased linepack storage, but increase
pressure differentials, thus increasing gas flows from one location to another.
Obviously, gas compression requires energy input, and in the Victorian system
the compressors are themselves powered by gas drawn from the gas transport
system. Although compressor fuel use is relatively low, it may be modeled as
follows. A gas powered compressor is driven by a proportion of the gas that
flows through it; increasing throughput in a pipe requires an increase in the
compressor pressure to offset the dynamic losses down the pipe. Increasing Dp
necessitates speeding up the compressor and hence increases fuel consumption. For
centrifugal gas compressors a quadratic equation relates the change in head (pres-
sure), Dp, volumetric flow at the compressor inlet qic , and impeller speed, RPM, as
follows25:
We can not use this equation directly in the formulation, though, because RPM is
not an LP variable and the equation in this form is nonlinear. Still, we can
reasonably assume that compressor operation rules will have been externally
optimized and that optimal operation will imply equations giving the minimum
gas consumption required to achieve any desired pressure/flow trade-off. Read and
Whaley [32] present a number of detailed equations and steps to determine this loss
in gas mass, during compressor operation. Basically, if we know the desired flow
rate and pressure increase across the compressor, we can calculate the required
running speed of the compressor, in RPM, from which we can determine the rate of
fuel usage, and hence the actual mass “lost” in the compressor. This loss, which is a
function of volumetric flow and pressure change, is the cost associated with
25
Constants C1C4 are normally stated by compressor manufactures as standard data.
An LP Based Market Design for Natural Gas 111
mtþ1
n ¼ mtn þ qtin qto
n þ yn Lossn ðqn ; Dpn Þ
t t ti t
(6c)
26
In this simplified representation we are assuming that compressors can be dispatched continu-
ously right down to (near) zero, with no “commitment” costs, penalties, or restrictions. This allows
us to form an LP representation with a convex feasible region. In reality there is an integer “unit
commitment” problem here, as discussed by Pepper et al. [26].
112 E.G. Read et al.
References
1. AEMO (2010) A technical guide to the Victorian gas wholesale market. http://www.aemo.
com.au/corporate/0000-0264.pdf. Accessed 7 Aug 2010
2. Alvey T, Goodwin D, Ma X, Streiffert D, Sun D (1998) A security-constrained bid-clearing
system for the New Zealand wholesale electricity market. IEEE Trans Power Syst 13
(2):340–346
3. Boyd SP, Vandenberghe L (2004) Convex optimization. Cambridge University Press,
New York
4. Breton N, Zaccour Z (2001) Equilibria in an asymmetric duopoly facing a security constraint.
Energ Econ 25:457–475
5. Chao H, Peck S, Oren S, Wilson R (2000) Flow-based transmission rights and congestion
management. Electric J 13(8):38–58
6. Cremer H, Gasmi F, Laffont JJ (2003) Access to pipelines in competitive gas markets. J Regul
Econ 24(1):5–33
7. De Wolf D, Smeers Y (1997) A stochastic version of a Stackelberg Nash-Cournot equilibrium
model. Manag Sci 43(2):190–197
8. Doane MJ, Spulber DF (1994) Open access and the evolution of the US spot market for natural
gas. JLE 37(2):477–517
9. Dorin B, Toma-Leonida D (2008) On modelling and simulating natural gas transmission
systems (Part I). J Control Eng Appl Inform 10(3):27
10. DPI: Victorian Government Department of Primary Industries (2009) The Victorian gas market.
http://new.dpi.vic.gov.au/earth-resources/industries/oil-gas/petroleum-explorers-guide-to-victoria/
the-victorian-gas-market. Accessed 2 Sep 2010
11. Frontier Economics (2003) Analysis of high level design directions for the Victorian gas
market. Report to VENCorp
12. Gabriel SA, Manik J, Vikas S (2003) Computational experience with a large-scale, multi
period, spatial equilibrium model of the North America natural gas system. Netw Spat Econ
3:97–122
13. Gabriel SA, Kiet S, Zhuang J (2005) A mixed complementarity-based equilibrium model of
natural gas markets. Oper Res 53(5):799–818
14. Gilbert RJ, Kahn EP (1996) International comparisons of electricity regulation. Cambridge
University Press, Cambridge, UK
15. Hogan WW (1992) An Efficient Concurrent Auction Model for Firm Natural Gas Transporta-
tion Capacity. Information Systems and Operational Research, Vol 30, No. 3
16. Hogan WW, Read EG, Ring BJ (1996) Using mathematical programming for electricity spot
pricing. Int Trans Oper Res 3(4):209–221
17. Johnson RB, Oren SS, Svoboda AJ (1997) Equity and efficiency of unit commitment in
competitive electricity markets. Utilities Policy 6(1):9–19
18. McCabe KA, Rassenti SJ, Smith VL (1989) Designing ‘smart’ computer-assisted markets: an
experimental auction for gas networks. Eur J Polit Econ 5(2–3):259–283
19. Martin A, Moller M, Moritz S (2006) Mixed integer models for the stationary case of gas
network optimization. Math Program Ser B 105:563–582
20. Midthun KT, Bjørndal M, Tomasgard A, Smeers Y (2007) Paper IV Capacity booking in a
Transportation Network with stochastic demand and a secondary market for Transportation Capac-
ity. www.iot.ntnu.no/winterschool11/web/material/tomasgard_paper.pdf. Accessed 27 Nov 2011
21. Midthun KT, Bjørndal M, Tomasgard A (2009) Modeling optimal economic dispatch and
system effects in natural gas networks. Energy J 30(4):155–180
22. Modisette J, Modisette J (2003) Physics of pipeline flow: energy solutions. www.energy-
solutions.com/pdf/tech_paper_Modisette_Physics_of_Pipeline_Flow.pdf. Accessed 11 July
2010
23. Murphy JJ, Dinar A, Howitt RE, Rassenti SJ, Smith VL (2000) The design of “smart” water
market institutions using laboratory experiments. Environ Resour Econ 17(4):375–394
An LP Based Market Design for Natural Gas 113
24. O’Neil RP, Williard M, Wilkins B, Pike R (1979) A mathematical programming model for
allocation of natural gas. Oper Res 27(5):857–873
25. Pepper W (2002) Stage 2-evaluation of market design packages: detailed report. Report by ICF
Consulting and Pacific Economics Group to VENCorp
26. Pepper W, Ring BJ, Read EG Starkey SR (2012) Implementation of a scheduling and pricing
model for natural gas. A. Sorokin et al. (eds.), Handbook of Networks in Power Systems II,
Energy Systems, Springer-Verlag Berlin Heidelberg
27. Prabodanie RA, Raffensperger JF, Milke MW (2009) Simulation-optimization approach for
trading point and non-point source nutrient permits. Paper presented at the 18th World IMACS
congress and MODSIM09 international congress on modelling and simulation, Cairns,
Australia, 13–17 July 2009
28. Raffensperger JF, Milke MW, Read EG (2009) A deterministic smart market model for
groundwater. Oper Res 57(6):1333–1346
29. Read EG (1989) Pricing and operation of transmission services: long run aspects. In: Turner A
(ed) Principles for pricing electricity transmission. Trans Power New Zealand, Wellington, NZ
30. Read EG (1997) Transmission pricing in New Zealand. Utilities Policy 6(3):227–236
31. Read EG (2010) Co-optimization of energy and ancillary service markets. In: Rebennack IS,
Pardalos PM, Pereira MVF, Iliadis NA (eds) Handbook of power systems. Springer-Verlag
Berlin Heidelberg, pp 307–327
32. Read EG, Whaley R (1997) A gas market model for Victoria: dispatch/pricing formulation.
Report by Putnam, Hayes & Bartlett–Asia Pacific Ltd. to VENCorp
33. Ring BJ, Read EG (1996) A dispatch based pricing model for the New Zealand electricity
market. In: Einhorn MA, Siddiqi R (eds) Electricity transmission pricing and technology.
Kluwer Academic, Boston, pp 183–206
34. Rudnick H, Palma R, Fernandez JE (1995) Marginal pricing and supplement cost allocation in
transmission open access. IEEE Tran Power Syst 10(2):1125–1132
35. Ruff LE (1997) Victorian gas market clearing logic. Report by Putnam, Hayes & Bartlett Asia
Pacific Ltd to VENCorp
36. Sioshansi F, Pfaffenberger W (2006) Electricity market reform: an international perspective.
Elsevier, Amsterdam
37. Vany AS, De Walls WD (1994) Open access and the emergence of a competitive natural gas
market. Contemp Econ Policy 12(2):77–96
38. Tomasgard A, Rømo F, Fodstad M, Midthun KT (2007) Optimization models for the natural
gas value chain, In: Hasle G, Lie KA, Quak E (eds) Geometric modelling, numerical simulation,
and optimization: applied mathematics at SINTEF, Springer-Verlag, Berlin Heidelberg,
pp 521–558
39. Wolak FA (2000) An empirical analysis of the impact of hedge contracts on bidding behavior
in a competitive electricity market. Int Econ J 14(2):1–39
40. Zheng QP, Rebennack S, Iliadis N, Pardalos PM (2010) Optimization models in the natural gas
industry. In: Rebennack IS, Pardalos PM, Pereira MVF, Iliadis NA (eds) Handbook of power
systems. Springer-Verlag Berlin Heidelberg, pp 121–148
41. Zhou J, Adewumi MA (1990) The development and testing of a new flow equation: the
Pennsylvania State University. www.psig.org/papers/1990/9504.pdf. Accessed 1 Feb 2011
42. Energy Projects Division (EPD), Dept. of Treasury & Finance, (1998) Victoria’s Gas Industry:
Implementing a competitive Structure. Information Paper No. 3, 2nd Edition, April
43. McCabe KA, Rassenti SJ, Smith VL (1990) Auction Design for Composite Goods: The
Natural Gas Industry. Journal of Economic Behavior and Organization 14:127–149. Elsevier
Science, North-Holland
Part II
Network Interactions
Energy Carrier Networks: Interactions and
Integrated Operational Planning
Abstract The integration of natural gas (NG) and electricity sectors has rapidly
increased as a consequence of the growing installation of natural gas fired power
plants (NGFPP). This has driven the need to model the interactions among the
energy carriers and to optimize energy resources management from a centralized
planning perspective. Currently, electricity and NG systems are considered in a
decoupled manner. NG prices and availabilities for the electric power generation
are used as fixed parameters for the needed coordination between both energy
sectors. This chapter presents a comprehensive literature survey of previous
research on integrated electricity and NG operational planning. The relevant
characteristics of NG and electricity systems are compared considering the physical
laws that govern the flows of these energy carriers through dedicated networks. The
interactions among the energy carriers and their networks are modeled with differ-
ent levels of detail according to the evaluated time horizon. The integrated opera-
tional planning problem of multiple energy carriers systems is comprehensively
described and formulated, covering from the long/medium-term energy resource
scheduling to the single period economic dispatch. Finally, a contribution is made
about the economic interactions between different energy carriers (electricity, NG,
and hydro energy) through opportunity costs such as water and NG values.
A. Sorokin et al. (eds.), Handbook of Networks in Power Systems II, Energy Systems, 117
DOI 10.1007/978-3-642-23406-4_5, # Springer-Verlag Berlin Heidelberg 2012
118 R. Rubio-Barros et al.
1 Introduction
Energy plays an all-encompassing and critically key role in economics and social
development. Therefore, energy planning is an essential and strategic part of
national/regional economic planning. This energy planning is driven by a set of
energy polices, which are developed to meet many interrelated and often conflicting
objectives, such as least cost of supply, security of supply and preserving the
environment [1]. The energy systems infrastructure needed to achieve these goals
is determined within the planning procedures.
This complex infrastructure, which is composed by all the technical energy
systems, allows providing the consumers with the “useful energies” or “energy
services”. Technical energy systems include production, conversion, treatment,
transport and storage facilities which comprise the supply chains from the primary
energy sources (e.g., oil, coal, natural gas, nuclear, solar, wind) to final energy
carriers required by the consumers (e.g., electricity, natural gas, water district
heating). Particularly, electric power systems perform two important functions in
the energy supply: allowing the use of primary energy sources, such as nuclear,
hydropower, and wind that otherwise they were mostly unusable; and providing for
flexibility since most of the energy sources can be converted to electricity.
While the required energy systems infrastructure is a product of the energy
planning, it does not imply some rigid action plan as in a centrally planned
economy. Conversely, because market mechanisms furnish the needed coordina-
tion in the economy, and then in the energy sectors, energy planning implementa-
tion relies mainly on market incentives, pricing methods and decentralized
competitive forces.
Before the energy crisis of the 1970s, energy was relatively cheap, and the emphasis
was more on the engineering and technological aspects. The energy system
planning was confined to various energy sectors such as electricity, oil, and coal,
with almost no coordination among them. The increasing demands were invariably
dealt with by augmenting the supply without taking into account possible
substitutions between the energy carriers [1].
Nowadays, there is consensus among policy makers that energy sector invest-
ment planning, pricing, operation and management should be carried out in an
integrated and coordinated manner in order to achieve an economic, reliable and
environmentally sustainable energy supply. A hierarchical and sequential proce-
dure is typically used to tackle this huge and complex decision-making problem.
The so-called energy models are the first stage in this hierarchical energy
planning procedure. In such models, all (or most) energy carriers are considered
in an integrated approach. Several of these energy models have been developed to
Energy Carrier Networks: Interactions and Integrated Operational Planning 119
analyze a range of energy policies and their impacts on the energy system
infrastructures and on the environment. Others are focused on the energy services
demands forecast. An overview and a classification of some of the most relevant
energy models, like TIMES (integrated MARKAL-EFOM system) [2], MESSAGE
[3], ENPEP-BALANCE [4] and LEAP [5], is included in [6]. These models are
focused on a long term planning horizon (more than 10 years) and can be tailored to
cover local, national, regional or world energy systems. The interactions between
the energy sectors and the other sectors of the economy (e.g., transport, industry,
commerce, agriculture) can be taken into account through model extensions or
represented by means of constraints.
Basically, energy models can be classified in two types according to the analyti-
cal approach used: aggregated general equilibrium (top-down) and technology
explicit (bottom-up) models. In top-down models, each energy sector is represented
by a single production function including macroeconomic variables (capital, labor)
and elasticities of substitution. Bottom-up models have a detailed description
(capacities, costs, efficiencies) of the available technologies for each energy sector,
thus the production functions are implicitly constructed [2].
Because the dimensions of the problem, energy models are not developed to
represent the characteristics of different transport modes, neither the complex
physical laws that governs electric power and natural gas (NG) systems. Only
nodal energy balances are considered. Another limitation of these models is related
to energy storage, which is oversimplified or disregarded.
More recently, new approaches to energy system planning have been presented.
They are focused on a higher technical description of some energy sectors and their
transport modes. The model presented in [7] includes the topology of several
energy systems, and the technical and economic properties of different investment
alternatives. Among other energy modes of transport, simplified electricity and NG
networks are considered. In [8] and [9] specific methodologies and tools are
proposed to address the integrated NG and electric power systems planning in
particular.
The results of energy models provide the framework for the following stages,
in which each energy carrier system is planned and operated in a decoupled
manner. Thus, specific procedures and strategies are implemented according to
specific value system, e.g., economic, technical, political, and environmental
context. Usually, single energy carrier system expansion and operation planning
are carried out considering the other energy carriers availabilities and prices as
coordinating parameters. Electric power systems are a good example of this
approach: they are planned and operated without taking into account the
integrated dynamics of the fuel infrastructures and markets, i.e., costs and
capacities of fuel production, as well as storage and transportation. The main
assumption, in which this decoupled planning and operation approach, is based
on the fact that there have not been significant energy exchanges between the
energy carriers if they are compared with the total amount of energy supplied by
each energy carrier.
120 R. Rubio-Barros et al.
Natural gas-fired power plants (NGFPPs) are the linkage between electric power
and NG systems, being producers for the former and consumers for the latter. The
growing installation and utilization of NGFPPs over the last two decades has lead to
increasing interactions between electricity and NG industries. From 1990 to 2005,
the worldwide share of NGFPPs in the power generation mix has almost doubled,
from around 10% to nearly 19%; reaching, for instance, the 54% in Argentina, the
42% in Italy and the 32% in UK [10, 11]. The installation of NGFPP has been
driven by technical, economic, and environmental reasons. The high thermal
efficiency of combined-cycle gas turbine (CCGT) power plants and combined
heat and power (CHP) units, their relatively low investment costs, short construc-
tion lead time and the prevailing low natural gas prices until 2004 have made
NGFPPs more attractive than traditional coal, oil and nuclear power plants, partic-
ularly in liberalized electricity markets. Additionally, burning NG has a smaller
environmental footprint and a lower carbon emission than any other fossil fuel.
The increasing use of NGFPPs has had a great impact on NG market. Power
generation accounted for around half of growth in gas use from 1990 to 2004; over
the most recent 5 years, this proportion rose to nearly 80% [10]. This fact is
especially notable in those countries where large capacities of NGFPPs have been
installed. Therefore, NG demand for power generation as a share of the total NG
consumption has constantly augmented during these years, reaching globally the
39%, in 2006 [12]. The other index, which also indicates the level of interrelations
between electric power and NG systems, is the share of electrical energy produced
by NGFPPs. This share depends not only on the NGFPPs installed capacity, but also
on the fuel prices and the availability of other energy resources (hydroelectricity,
wind power). For instance, in 2006, this share was globally the 20%, the 50% in
Argentina and in Italy, the 35% in UK, and the 20% in USA [11].
From an operational point of view, the NGFPPs dispatch affects the total amount
of NG consumption and its flows through the pipelines. On the other hand, the
maximum capacity of gas injection for all the system, the limited transmission
capacity of pipeline network and the priorities of use in case of NG shortages
(usually not assigned to NGFPPs), impose constraints on NGFPPs generation. The
described interactions are closely related to the regulatory frameworks of the
markets implemented in electric power and NG systems. NGFPP generation
companies participate simultaneously in both markets, therefore, they are best
suited for price arbitrage between both commodities. This means that according
to electricity and NG market prices, and its marginal production heat rate, NGFPPs
can decide to use the gas, previously contracted, and sell electricity in the power
market or resell the gas in the NG market instead of generating electricity.
Since 2005 to the first half of 2008, the raising NG prices have eroded the
competiveness of NGFPPs, decreasing the pace of growth in NG use for electricity
generation and reducing the incentives for future investments in these technologies.
Energy Carrier Networks: Interactions and Integrated Operational Planning 121
Nevertheless, as NG prices have converged to lower levels during 2009, the NGFPPs
have recovered their investment attractiveness. For the coming decade, NGFPP
capacity is estimated to continue to account for the bulk of capacity additions, thus,
the interactions between electricity and NG sectors will keep rising [13].
In this new context, in which the energy exchanges between electric power and NG
systems are significant, is essential to consider an integrated modeling of both
energy carrier systems for the planning and operation tasks. In more general
terms, the “boundary conditions” and energy exchanges between all energy carrier
systems must be carefully analyzed to determine the most suitable modeling of
these interactions. Particularly, it is important to verify if coordinating parameters
can appropriately represent these interrelations, maintaining the systems in
decoupled way; or a more detailed description is needed, leading to an integrated
modeling of the energy carrier systems.
Among all the issues, which arise from this new integrated perspective of energy
carrier systems, this chapter is focused on the modeling and operational planning
aspects. Several approaches that address the integrated modeling and analysis of
energy systems in a more comprehensive and generalized way have been presented.
These approaches consider multiple energy carriers; particularly electricity and NG
systems interactions and combined operation have been investigated.
An assessment of the impact of NG prices and NG infrastructure contingencies
on the operation of electric power systems is presented in [14] from a decoupled
modeling approach. From the market perspective, different methodologies have
been proposed to support the decision making related to: optimize NG supply
contract portfolios [15], pricing NG supply contracts [16], and price hedging
strategies between the gas and electricity markets [17]. The effects of applying
different transmission cost allocations to electricity and NG networks are analyzed
in [18, 19]. A review of the main approaches and models, which deal with the
integrated operational planning of multiple energy carrier systems, is presented in
following subsections. This review is based on the survey shown in [20]. The
different approaches are conveniently grouped according to the time horizon
considered.
subject to the following constraints: (a) Electric power balance at each node; (b)
Hydraulic balance at each water reservoir; (c) NG balance at each node and gas
storage; (d) Initial and final water and gas volumes at reservoirs; (e) Electric power
generation limits; (f) Maximum electric power flow through lines; (g) NG wells
injection limits; (h) Pipelines maximum transport capacity; (i) Bounds on storage
and turbined water volumes; (j) Bounds on storage and outflow gas volumes; (k)
Minimum up and down time of thermal units; (l) Minimum spinning reserve
requirement. To solve the integrated electricity-gas optimal short-term planning
problem an approach based on Dual Decomposition, Lagrangian Relaxation, and
Dynamic Programming is employed.
More recently, in [26] and [27], the NG network model is included in the electric
power security-constrained unit commitment problem. While in [26] the NG flows
are calculated through a nodal gas balance model, the steady-state physical laws
(pressure differences) that govern NG flows are modeled in [27]. In both
approaches, local NG storages at each NGFPP are considered. Particular and
detailed modeling of fuel switching capabilities is described in [26]. A multi-period
combined electricity and NG optimization problem is presented in [28]. The
modeling in this approach takes into account not only NG storages facilities, but
also the NG contained in the NG network, so-called line pack.
Reference [29] presents a combined NG and electricity optimal power flow. The
authors deal with the fundamental modeling of NG network, i.e., the steady-state
nonlinear flow equations and detailed gas consumption functions in compressor
stations. A complete formulation of the NG load flow problem and its similarities
with power flows are shown in detail. AC power flow modeling is applied to
determine power flows in the electricity network. The objective function is
formulated in terms of social welfare. While generation costs due to non-gas
electrical plants and gas supply costs account for the total costs, benefit to electrical
consumers and benefit to gas consumer, except NGFPPs, account for the total
benefits. Reference [30] also deals with the integrated NG and electricity optimal
power flow. Nonlinear steady-state pipelines flows and compression station are
modeled. However, the gas consumption in compressor stations is not considered.
The objective function in this approach is to minimize the sum of generation cost
due to non-gas electrical plants and cost of gas supply.
References [31] and [32] present a model to compute the maximum amount of
electric power that can be supplied by NGFPPs, subject to NG systems constraints.
Nonlinear steady-state pipelines flows and the effect of compressor stations to
enlarge the transmission capacity are included in the gas network modeling. Like
in [30], the amount of gas consumed in the compressor stations is neglected.
A comprehensive and generalized optimal power flow of multiple energy
carriers is presented in [33]. This paper presents an approach for combined optimi-
zation of coupled power flows of different energy infrastructures such as electricity,
124 R. Rubio-Barros et al.
gas, and district heating systems. A steady-state power flow model is presented that
includes conversion and transmission of an arbitrary number of energy carriers. The
couplings between the different infrastructures are explicitly taken into account
based on the new concept of energy hubs. With this model, combined economic
dispatch and optimal power flow problems are stated covering transmission and
conversion of energy. Additionally, the optimality conditions for multiple energy
carriers’ dispatch are derived, and the approach is compared against the standard
method used for electrical power systems. Reference [34] is an extension of [33],
and it also considers hydrogen as another energy carrier.
The impacts of NG transport cost in the electric power economic dispatch
problem are presented in [35]. The NG flows are modeled through the steady-
state nonlinear equations and transport cost is defined as the sum of NG consump-
tion in compression stations.
In [36], the decoupled and the combined approach for the optimal dispatch of
electric power and NG systems are compared. It is shown that a higher economic
efficiency is achieved if both energy systems are considered in an integrated
manner. Reference [37] presents an extensive analysis of the coordinating
parameters, which are the reasons for the inefficiencies in the decoupled approach.
During the last two decades, market mechanisms have been implemented in some
energy sectors to provide the essential coordination required for the efficient
planning and operation of these sectors. Under this scheme, the optimal expansion
and operational planning is mainly achieved by means of the decentralized decision
that are taken by the companies and organizations involved in these markets.
In particular, electricity and gas sectors have been liberalized to a certain extent
in many countries, introducing competition at varying degrees and at various levels
of the value chain. Essentially, this restructuring has been accomplished by
unbundling the different segments of the industries. In the electricity sector, the
production segment (generation) was separated from the service segments (trans-
mission and distribution). In the same way, the natural gas sector was split up into a
production segment (upstream) and pipeline network services (midstream and
downstream). One of the major features to ensure competition in the production
segment is that transmission and distribution companies provide open access on the
transmission network to other market participants for energy delivery that has
permitted producers to sell their product (electricity, NG) directly to end users
and marketers. Different types of markets have been established, thereby allowing
the interaction between the production sectors (supply) and the demand at different
time frames, ranging from long-term contracts to hourly spot markets.
Since a significant share of total NG consumption is used to produce electricity,
both energy carrier market prices are linked. Then, the NGFPPs play a key role in
the electricity and gas price dynamic because they are the market participants that
Energy Carrier Networks: Interactions and Integrated Operational Planning 125
allow the arbitrage between the two commodities. Liberalized markets for both
commodities promote the arbitrage and, therefore, contribute to price convergence.
The increasing links between NG and electricity also offer both a threat and an
opportunity regarding energy supply security. Flexible facilities, such as energy
storage (e.g., NG storage, water reservoirs) and fuels switching (in NGFPPs or
steam power plants) are important resources to assure the gas and electricity supply
security and to reduce prices volatility. Additionally, efficient gas and electricity
markets tend to reduce gas demand as prices increase, thus saving gas at times of
high demand or low supply.
In liberalized electricity markets, experiences show that one of the most power-
ful instruments for consumers to oppose the market power of suppliers is the
presence of a well-functioning, transparent and liquid wholesale market. It is likely
that a liquid and competitive wholesale market for NG is also a powerful tool to
counterbalance potential upstream market power in gas. There are numerous policy
challenges in establishing well-functioning gas and electricity markets to ensure
affordable and reliable energy supply.
The growth of world oil prices has produced also an increase or readjustment of
NG price. This correlation is mainly because both fuels are substitutes of each other
especially in the electricity sector. Because of that, hardly we could talk about a
disengagement of these two fuel prices. In contrast, we could consider gas prices as
regional prices, due to the lack of a unified world NG market, but a partitioned
market in which prices are subject to regional supplies and demands. However, the
trend is that NG will be a global commodity because of the development of LNG
technology and its associated markets. The economic theory shows that in a
competitive market, like a mature NG market (U.S.A., England), the price maker
is defined by short-term prices (spot price on Henry Hub or National Balancing
Point) or by standard quotations in a Stock Exchange (NYMEX, IPE). This price
reflects instantaneous match of supply and demand. However, in NG markets still a
significant part of the volumes is traded through long-term supply or demand
contracts with indexed prices over the time and penalties in any case of lack (called
deliver-or-pay or take-or-pay contracts). In NG monopolies, NG prices are calcu-
lated deducting the transmission and distribution costs from the equivalent NG
price in the final convergent energy market (usually the electricity market).
For NGFPPs, active and liquid electricity and NG markets represent a great
opportunity to increase their revenues taking advantages of all arbitrage
possibilities.
NGFPPs can now purchase NG with great flexibility, through bilateral contracts
or through the spot market. On the other hand, the wholesale electricity market
price is an important part in the decision-making process for NGFPPs. When the
market implied marginal heat rate (which is the equivalent heat-rate calculated
using the clearing price for electricity divided by the prevailing natural gas price) is
lower than the marginal production heat rate of the NGFPP, the NGFPP might want
to purchase power instead of generating it itself and sell the natural gas in the spot
market [15]. Another way of looking at the same problem is through the so-called
spark spread, which is defined as the difference, at a particular location and time
126 R. Rubio-Barros et al.
between the price of one MWh of electricity and the fuel cost of generating that
MWh. As a result, a positive spark spread indicates the power generator should
generate electricity rather than resell the previously contracted gas in the spot
market. Other service that can be provided by NGFPPs is called tolling, where a
power generator receives NG from a beneficiary and delivers electric power to the
same beneficiary in return for a service fee.
The introduction section presents the context in which the interactions among
energy carrier networks are analyzed. An overview of how energy systems are
currently planned and operated is presented. The existing interactions between
electric power and NG systems and their prospect for the coming decade are
explained. A complete review of the state of the art in the integrated operational
planning of energy carrier systems discusses about the proposed approaches to deal
with these increasing interactions. Some issues related to the introduction of market
mechanisms in electric power and NG sectors are described, along with the role of
markets in the provision of coordination for the efficient operation and planning.
Section 2 describes those energy carrier systems which are inherently associated
with dedicated networks, i.e., electric power systems, NG systems and hydrological
networks. General aspects, similarities and differences between electric power and
NG systems are discussed in detail. The modeling of electric power and NG flows in
their respective networks are extensively described and analyzed. The
characteristics of the hydrological networks are also presented.
In Sect. 3, the integrated operational planning problem of multiple energy
carriers systems is comprehensively described and formulated. This huge problem
is tackled by means of a hierarchical sequential optimization scheme, ranging from
the long/medium-term to the short-term time horizons. The single period economic
dispatch of electric power and NG systems is thoroughly discussed. Two case
studies are presented to demonstrate the advantages of the integrated operational
planning over the decoupled approach. It is important to point out the integrated
operational planning presented throughout this chapter takes into account a
centralized perspective of the issue. Consequently, the different type of markets
(spot, forward, futures) and different forms of trading (bilateral, multilateral, pool)
developed within electricity and NG markets are not modeled.
Electricity and natural gas are energy carriers, i.e., a substance or phenomenon that
can be used to produce mechanical work or heat or to operate chemical or physical
processes [38]. While natural gas is a primary energy because exists in a naturally
Energy Carrier Networks: Interactions and Integrated Operational Planning 127
occurring form and has not undergone any technical transformation, electrical
energy is a secondary energy, which is the result of the conversion of primary
energy sources.
Some energy carriers are inherently associated with dedicated networks. Then,
an energy carrier network is a set of elements that allows the transport of certain
energy between the production/conversion points to the consumption points.
Energy flows in these networks complies with physical laws, such as the relation
between voltage and current in electricity, pressures and flows in gaseous
substances. These relationships are particular for each energy system. Water
reservoirs and rivers constitute hydrological networks which are also other type
of energy carrier network. This chapter is focused in transmission networks,
which cover extensive areas, therefore, electricity, NG and hydrological energy
carrier networks are studied. Some general aspects and the mathematical formu-
lation of energy flows in these networks are presented in the following
subsections.
Even though natural gas and electricity are different energy carriers, they have a
similar organization structure (Table 1). Like in the electricity production segment,
a great diversity of technical characteristics can be found among gas suppliers. Gas
wells (commonly located at sites far from load centers) and LNG regasification
terminals (harbor locations) have capacity and operating constraints. Natural gas
transmission and distribution networks provide the same services as their electricity
counterparts. Transmission pipelines undertake the responsibility of transporting
natural gas from producers to local distribution companies or directly to large
consumers. Distribution networks generally provide the final link in the natural
gas delivery chain, taking natural gas from city gate stations and other gas supply
sources to large and small customers. Generally, there is a priority scheme for the
supply of natural gas in which residential and commercial customers take prece-
dence over large consumers (NGFPPs included).
Thermal
Pressure
NG Supply Generation
Regulator
C (coal, oil, nuclear)
Hydro
NG-fired
Generation
Unit
Distribution C Compressor C
Pipelines
NG Transmission
Demand Transmission lines
Pipelines
Electricity
C demand
NG NG
Storage Demand Wind
Generation
LNG Regas
Terminal NG-fired Electricity
Unit demand
Delivering the natural gas from gas wellhead to end customers entails pipelines,
storage facilities, compressors and valves. Figure 1 illustrates NG and electricity
systems schematically. Gas flow paths from a gas wellhead to gas-fired units and
other gas demands are shown. The electric power system includes a 3-bus network
representing the electric transmission system, hydroelectric power plants, non-gas
electrical power plants and electricity demands.
NG is transported from gas producers to customers at various locations. Three
basic types of entities are considered for the modeling of NG transmission network:
pipelines (branches), compression stations (branches) and interconnections points
(nodes). Compressors are placed at specific locations in order to increase the
pressures in these points and thus, increase the transportation capacity of the gas
network. Typically, compressors are installed not only at gas wellheads (or
regasification terminals), but also at many other transmission system locations.
However, the operating pressures are constrained by the maximum pressure
allowed in pipelines and the minimum pressure required at gate stations. Therefore,
the transmission capacity of a gas pipeline is limited.
Valves are protective and control devices whose functions are similar to
breakers, fuses, and switches in electric power systems. Isolating valves are used
to interrupt the flow and shut-off a network section. Pressure relief valves can
prevent equipment damage caused by excessive pressure. Pressure regulators can
vary the gas flow through a pipeline and maintain a preset outlet pressure.
Unlike electricity, which large scale storage is not yet technically or economi-
cally feasible, natural gas can be stored for later consumption. There are three major
types of NG storage facilities which are different in terms of capacity (working
volume) and maximum withdrawal rates:
Energy Carrier Networks: Interactions and Integrated Operational Planning 129
In electrical networks, the steady-state electric power flows are governed by Ohm’s
and Kirchhoff’s laws. These laws can be expressed by means of nodal power
balances and line power flows. The complex power balance at node j in an electrical
AC network can be stated as
X
Sj Sjn ¼ 0 (1)
n2N j
where Sj is the net complex power injected at node j (sum of the production of local
generators less the local demand), N j is the set of nodes connected to node j, and Sjn
is the outgoing complex power from node j to any other node n. The outgoing
complex power Sjk from node j to k is a function of the complex nodal voltages Uj
and Uk and the parameters of the line [39]
2
U j Uj U
Sjk ¼ k (2)
Znjk Zjk
where
1 Yjk 1
Znjk ¼ þ (3)
Zjk 2
where Zjk is the series impedances and Yjk is the shunt admittance of the P-
equivalent of the line m (with Yjk/2 at each line end). The admittance Yjk models
the capacitive effect in electric power lines and the small losses due to corona effect
[39]. Equations (1) and (2) accounts for the so-called AC power flow model which
is the most accurate representation of electric power flows. Figure 2 shows a
scheme of this model.
Active and reactive power flows can be calculated as the real and imaginary
parts of the complex power (2), respectively. In multiple energy carrier flow
130 R. Rubio-Barros et al.
Sjk Skj
m
Sj Sk
analysis, the modeling is focus on the active power flows because these are the real
amounts of the power transferred from one node to another. The active power flow
Pjk can be expressed as:
1 n 2 o
Pjk ¼ 2 Uj Rjk Uj jUk j Rjk cos yj yk Xjk sin yj yk (4)
Zjk
where Rjk and Xjk are the resistance and reactance the line m (Zjk ¼ Rjk + j Xjk), and
yj and yk are the phase angles of voltages Uj and Uk. Other factor that should be
analyzed is the efficiency of transporting energy in the form of electricity. There-
fore, losses are usually taken into account in energy systems studies, particularly
when extensive systems are considered. The series losses in the line m, which
neglect shunt losses associated to Yjk, can be stated as:
1 h 2
i
Ljk ¼ Pjk þ Pkj ¼ 2 Uj þ jUk j2 Rjk 2Uj jUk jRjk cos yj yk (5)
Zjk
Even though the active power flow Pjk through a line is a nonlinear function of
the complex voltages at the line ends, a simplified model can be reasonably used to
represent electric power flows. This is the so-called DC or MW-only power flow,
which can be obtained from (4) via a sequence of approximations [40–43]; thus:
• Losses are neglected, Ljk ¼ 0, Pjk ¼ Pkj
• Voltages at all nodes are assumed equal to 1 per unit, Ui ¼ 1 p.u. 8i
• Phase angle differences corresponding to adjacent buses are small, cos(yj –
yk) 1 and sin(yj – yk) yj – yk
• The reactance Xjk is much greater than the resistance Rjk for all the lines m,
reactance Xjk > > Rjk 8m, thus Xjk//(R2jk + X2jk) 1/Xjk
Under these approximations, the active power flow (4) from node j to k in p.u.
can be rewritten as:
1
Fm ¼ yj yk (6)
Xjk
This is the classical dc model which accuracy and applications for different
purposes are extensively discussed in [42]. Equation (6) can be extended to all
system’s branches as follows:
F ¼ BD AT u (7)
Energy Carrier Networks: Interactions and Integrated Operational Planning 131
where F is the vector of lines active power flows, BD is the diagonal matrix whose
elements are 1/Xjk, A is the node-branch incidence matrix whose Ajm element is +1
if line m enters node j, -1 if line m leaves node j, or 0 if line m is not connected to
node j and u is the vector of phase angles. Another implicit approximation assumed
in the extension of (6)–(7) is that the relation Rjk/Xjk is a constant for all the lines in
the system. On the other hand, (1) can also be expressed in its matrix form as
P ¼ A F (8)
in which P is the vector of net active power injections. The combination of (7) and
(8) leads to
0
P ¼ A BD AT u ¼ Bbus u (9)
where matrix B’bus has the same structure (sparse and symmetric) as the systems or
nodal admittance matrix, but its values being computed solely in terms of branch
reactances. Therefore, in order to calculate the active power flows through the lines,
the procedure consist in solving (9) to obtain u since P is known, and then use u in
(7) to obtain F. Alternatively, (7) and (9) can be combined, leading to a linear
relationship between active power injections and active power flows
1 h 0 i1
F ¼ BD AT A BD AT P ¼ BD AT Bbus P ¼ HP (10)
Pj Pk
½Lm ½Lm
The simplest manner to allocate theses losses is to assign all of them to the slack
node, however, this leads to cumulative error in the power flows in the vicinity of
the slack node. In order to preserve the system generation-load-loss MW balance,
the losses have to be allocated among all the nodes of the network modeling them as
equivalent injections. One way to do this is to assign half of the series losses in the
line to each line ends. Other more accurate ways to distribute losses are discussed in
[42]. Thus, the vector of nodal active power injection includes:
P ¼ PG PD P L (12)
where PG, PD and PL are the vectors of generating units’ productions, demands and
losses, respectively. Each element PLj of PL is calculated as:
X 1
PLj ¼ Lm (13)
m2M
2
j
where Mj is the set of lines connected to node j. The shunt losses can also be
considered and added in (11). Figure 3 shows a representation of DC power flow with
losses. Finally, if it is required to model different sending and receiving MW flows in
each line, it can be considered that Fij ¼ Fm þ Lm/2 and –Fji ¼ Fm Lm/2 according
to the flow references in Figs. 2 and 3.
where Wj is the net volume flow injected at node j, N j is the set of nodes connected
to node j, and Qgjn is the outgoing NG flow from node j to any other node n.
Energy Carrier Networks: Interactions and Integrated Operational Planning 133
Where
(
þ1; si s j rs k
dm ¼ (16)
1; si sj <sk
rffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
p2 Rair Tst D5m
Km ¼ (17)
64 sst fm T G Z L m
where Rair is the ideal gas constant of dry air, Tst and sst are the standard or normal
conditions of temperature and pressure, Dm the is the inside diameter of the pipe m,
fm is the dimensionless friction factor of the pipe m, T is the average gas flowing
temperature, G the is gas gravity relative to air, Z is the dimensionless gas
compressibility factor and Lm is the length of the pipe m. An additional term can
be introduced to (15) if a significant elevation difference exists between the
upstream and downstream locations
of the pipe [44] and [45]. The parameter dm
is the sign of the difference sj sk which defines the actual direction of the NG
flow. This auxiliary parameter must be included to avoid an imaginary number as
result of (15) when the conventional defined as downstream pressure sk is higher
than the upstream pressure sj. Several flow equations are in use in the NG industry
such as Weymouth and Panhandle equations, and all of them are modifications of
(15), which is called the general flow equation for steady-state gas flow.
In fact, (15) denotes that the pipeline transmission capacity is constrained to the
allowed maximum and minimum operating pressures in the different sections of the
NG network.
During transportation of NG in pipelines, the gas flow loses a part of its initial
energy due to frictional resistance which results in a loss of pressure. To compen-
sate these pressure losses and maximize the pipeline transport capacity, compressor
stations are installed in different network locations. Compressor stations consist of
compressors, which are arranged in series and parallel configurations depending on
the requirements. Several types of compressor can be used, being centrifugal
compressors the most commonly used. Many types of prime movers (e.g. electric
motors, gas turbines, gas engines, steam turbines) have found application as
compressor drivers. However, an amount of NG is withdrawn from the network,
and eventually transformed on site into electricity or heat (steam), to provide the
energy needed for the prime movers. Despite electric motors typically drive
compressors and main electric power supply is available in the compression station
134 R. Rubio-Barros et al.
area, the electric power is not extracted from electricity network because reliability
reasons. Thus, the NG consumed in a compression station Lgm can be considered as
an additional NG flow into the pipeline section, as shown in Fig. 4. This amount of
NG basically depends on the pressure added to the fluid and the volume flow rate
through the compressor. In extensive NG networks, the total NG consumed in
compressors stations can reach up to 6% of the total NG transported through the
network. The value of Lgm can be approximated as follows [33]:
Lgm ¼ KCm Qgm sj su (18)
where KCm is a constant characterizing the compression station; su and sj are the
suction and discharge pressures, respectively; Qgm ¼ Qgjk ¼ Qgsj is the flow
through the compressor. It is important to point out that (18) represents an equiva-
lent and approximate model of an entire compression station. More advanced
compression station modeling considering detailed compressor and prime movers
characteristics are presented in [29, 44] and [45]. As shown in Fig. 4, compressor
stations, like pipelines, can be modeled as branches in NG networks.
In order to make NG flows dimensionally suitable for comparison with any other
energy carrier, they should be converted to power using the NG higher heating
value. In the same way, a given volume of NG can be converted to an amount of
energy.
which is lost in transit as it loses height until reaching the next reservoir below.
Because of that, we can relate water flows and stored water volumes as power flows
and stored energy, respectively using the potential energy equation applied to the
volume of water stored.
In general, hydrological networks have certain characteristics that must be
considered, depending on the study to be carried out. In the medium and long
term planning, the main considerations to take into account are: (a) Temporal
forecasting of lateral streamflow, (b) Regulation capacity of reservoirs (annual,
seasonal, monthly, weekly, and hourly), (c) Priority use of water resources
other than energy use (ecological minimum outflows, agricultural irrigation
requirements, etc.), and (d) Evaporation and filtration losses. In the short term
dispatch, in addition, there are another features become relevant, such as:
(a) Time delay of water between coupled reservoirs, (b) Reservoirs with small
regulation capacity (weekly, daily, hourly, and several minutes), and (c) Pumped
storage facilities.
Hydrological networks are modeled by means of an series-parallel arrangement
of water reservoirs. Thus, water balances should be introduced for each reservoir at
each stage. These sets of equations (19) are called continuity equations, including
the water flow contributions of upstream hydro power plants.
X X X
Vetþ1 ¼ Vet þ Ate T Qwti T Ste T þ Qwti T þ Stu T (19)
i2 I he i2 I hu u2 E e
Vetþ1 and Vet are state variables that represent the stored volume of the water reservoir
e at the end of stage t (beginning of stage t + 1) and beginning of stage t, Ate is the
lateral streamflow arriving at the reservoir e, Ste is a slack variable that represents
the spilled outflow from reservoir e, I he is the subset of hydroelectric power plants
i associated with the water reservoir e, T is the total duration of stage t, I hu is the
subset of the upstream hydropower plants and E e is the subset of reservoirs u that
spill water into reservoir e.
Natural gas and electric power systems have a remarkable common feature which is
that extensive networks are used to transport the energy carrier from suppliers to
customers. Natural gas can also be carried in the form of Liquefied Natural Gas
(LNG), being required liquefaction trains, LNG ships and regasification terminals
to accomplish with transport and inject the gas into the network. Pipelines trans-
portation is more cost-effective over short and onshore distances, while LNG is
typically used in transcontinental carriage [10].
Although some analogies can be drawn between the electricity and NG
networks, there are several qualitative differences between both systems.
Table 2 shows the key differences between electricity and NG systems.
136 R. Rubio-Barros et al.
This section presents the mathematical formulation of the economic dispatch (ED)
of electricity an NG systems, from a centralized point of view. The formulation
considers a complete horizon, i.e. from long/medium to short term planning,
including also the single time period ED or simplified optimal power flow (OPF). At
first time these problems are considered, as state-of-the-art report, where each energy
system is optimized independently, using at most coordinating parameters like NG
Energy Carrier Networks: Interactions and Integrated Operational Planning 137
where Ci(pgi) is the production cost of a thermal generating unit i which do not use
NG as fuel (i.e., coal, nuclear, fuel oil); CGi(pgni) is the production cost of the
NGFPP i; CHi(pghi) is the opportunity cost of production of a reservoir-associated
hydropower plant i and CSE(psj) is the shortage cost related to a fictitious unit
which output power psj equals the load shedding at node j. The opportunity costs of
hydropower plants which are associated to a reservoir are given values in the ED,
and they can be calculated from the water values obtained in the medium-term
operational planning. The output powers of run-of river power plants are also
considered as given in the ED, and their operating costs usually neglected. The
cost minimization stated in (20) is subject to the following constraints:
P P P P
pgi þ pgni þ pghi þ psj þ Ajm Fm ¼ Dej þ PLj 8j (21)
i2Gj i2Gj i2Gj m
138 R. Rubio-Barros et al.
2 0 13
X X
m b
Fmax 4Hmj @ ðpgi þ pgni þ pghi Þ þ psj Dej PLj A5
j i2Gj (22)
b Fmax
m 8m
where (21) is the extended expression of (8) denoting the power balance at each
node j. The subset Gj includes all the generating units connected to node j, thus, the
first four terms of (21) accounts for the positive injection at this node; Dej is the
electricity demand at node j. The maximum transmission capacity limits for each
line m are set in (22) in which the power flows are calculated according to (10) and
(12). The generation limits are established in (23). Usually, in the ED, the minimum
output power of generating units is set to zero if it is not known which units are
committed in the analyzed time period. It is important to note that although Hmj
depend on the choice of slack node, the result of the optimization problem is
indifferent to this choice.
Some thermal generating units are capable of use a mix of fuels. If it assumed
that any fuel proportion can be burned, the fuel mixing feature can be modeled as:
X
pgi;c þ pgni bPmax
i (24)
c
where pgi,c is the partial output power produced by unit i with the fuel c different
from NG. Each generating unit that is able to fuel mixing is modeled by as many
output power variables as fuels the unit can burn. But the sum of all these variables
must be less than the unit’s maximum output power. Equation (24) presents the
simplest way of modeling fuel mixing in a generating unit. More complicated
formulations, like fuel switching, would require integer variables.
where wg is the NG injection in network of the supplier g and wsj is the fictitious NG
injection which value equals the NG load shedding at node j. This costs minimiza-
tion is constrained by the following equations
P P P P
wg þ wsj þ Qgm Qgm Lgm ¼ Dgj 8j (26)
g2W j m2I j m2Oj m2Cj
j bs j bS j
Smin 8j
max
(28)
su sj b0 8s 2 ðN Cm \ N u Þ; 8j 2 ðN Cm \ N d Þ (29)
In addition to these constraints, Eqs. (15), (16) and (18) must be taken into
account for all pipelines and for all compression stations, respectively. The NG
balance at each node j is stated in (26); where Dgj is the gas demand, W j is the
subset of gas supplier injecting in node j; I j and Oj are the subsets of pipelines
which flows are incoming and outgoing node j, respectively; and Cj is the subset of
compression stations whose suction side is connected to node j. The injection limits
for each gas supplier are set in (27). The maximum and minimum pressures allowed
at each node are established in (28). These pressure limits can vary throughout the
NG network according to pipelines characteristics and security operating rules. The
constraint (28) also applies to all the nodes in the NG network, including those
nodes which are connected to the suction side of a compression station. The gas
flow irreversibility at each compression station, which is a typical operating con-
straint, is stated in (29); where N u and N d are the subsets of nodes associated to
the suction and discharge of any compressor station, respectively; and N cm is the
subset of nodes associated to the compression station Cm. In this problem,
the decision variables are the suppliers’ injections and the pressure at each node,
which determine the NG flows.
For the simultaneous modeling and optimization of two or more energy carrier, it is
useful to combine different energy carrier nodes in a single entity. This unit, called
energy hub, provides the basic features in- and output, conversion, and storage of
the converging energy carriers [33]. Thus, energy hubs represent a generalization of
a network node, leading to an integrated view of energy carrier systems.
In an integrated electricity and NG ED, NGFPPs are the links between both
systems. On the other hand, since the ED problem considers a single time period,
the hydropower plants are modeled as thermal generating units, whose equivalent
140 R. Rubio-Barros et al.
Wind
Fossil fuels Hydro
Nuclear
Energy hub
NGFPP2
NGFPP1
NGFPPi
Sgj HR1 HR2 HRi
Natural gas network
Dgj
NG wells
From a centralized perspective, the aim of the integrated ED of electric power and
NG systems is the minimization of the total operating and shortage costs:
8 X X X 9
>
> Co ¼ Ci ðpgi Þ þ CHi ðpghi Þ þ CSE psj þ >
>
< i i j
=
min X X (30)
>
> Cg wg þ CSG wsj >>
: ;
g j
The integrated total operating cost of both systems is stated as the sum of each
energy carrier system cost (20) and (25), less the production costs of NGFPPs. The
formulation of the objective function (30) assumes that only the fuel costs account
for the production costs of thermal power plants, neglecting other variable produc-
tion costs (e.g., operation and maintenance costs). If all CGi(pgni) were also
considered, this would imply double counting these costs.
The cost minimization expressed in (30) is subject to all the constraints
described for the ED of electric power and NG systems, i.e., (15), (16), (18),
(21), (22), (25), (26), (27), (28) and (29). The interactions between the both systems
can be represented by means of two additional constraints
X
Sgj HRi ðpgni Þ ¼ Dgj (31)
i2Gj
Energy Carrier Networks: Interactions and Integrated Operational Planning 141
X
Sej þ pgni ¼ Dej (32)
i2Gj
where HRi is the heat rate function or input–output curve of NGFPP i according to
[40]. Equations (31) and (32) state the NG and electric power balances at each
energy hub j, respectively. In (31), HRi(pgni), which is the amount of NG burned in
the NGFPP i, is added as a NG demand, while in (32), the pgni produced by the
same NGFPP is an injection in the electric power system. It is important to point out
that in the integrated ED of electric power and NG systems, Dgj includes only NG
demanded for uses different to electric power generation.
The third type of parameter, which is needed to calculate the NG dispatch, is the
NG consumption at each NGFPP. Thus, the decoupled approach consists of two
stages. First, an ED of the electric power system is performed, being NG consump-
tion at each NGFPP a by-product of this procedure. Then, the ED of the NG systems
can be carried out.
However, fixed coordinating parameters cannot be suitable to represent the
existing interactions, especially if the shares of NGFPPs in the electricity supply
and in the total NG consumption are significant. The following conditions can
occur:
1. The total NG supply is not sufficient to meet the total NG demand, including the
NGFPPs’ demands. The NG supply to NGFPPs can be curtailed before than
other demands, since NGFPPs usually have lower priority of supply.
2. The limited transmission capacity in the NG network can imply that same
situation described in (1) occurs at a specific node.
3. The fixed NG prices, which determine the NGFPPs production costs, cannot
match with the NG marginal costs at nodes where NGFPPs are placed. These
142 R. Rubio-Barros et al.
In this section, a simple case study is presented to analyze the effect of coordinating
parameters and compare the performance of integrated and decoupled approaches
to electric power and NG economic dispatch.
Figure 6 depicts a two-hub electric power and NG systems. Both hubs are
connected through an electric power line and a pipeline. The hub’s limits have
Energy Carrier Networks: Interactions and Integrated Operational Planning 143
pg1 pg2
F12
ps1 ps2
pgn1 pgn2
De1 De2
HUB 2
HUB 1
w2
ws1
ws2
Qg21 sd
Dg1 Dg2
been extended for a more comprehensive representation that includes all the
production/conversion facilities located in the node. However, the modeling is
kept as described in Sect. 3.3.2.
The basic information about the production/conversion infrastructure is shown
in Table 3. The single-period ED of electric power and NG systems considered in
this example last 1 hour. NG flows (production rates, demands) are expressed as
powers (MW) instead of volumetric rates (m3/s). For this conversion, the NG higher
heating value (HHV) has been used. For instance, the Argentinean HHV is
9,300 kcal/m3 (38,937 MJ/m3), which means that 1 MWh of NG is equivalent to
92.46 m3 (1 MWh ¼ 859,845.23 kcal). The pressure limits in the hub 1 are 1.4 p.u.
and 0.6 p.u. The maximum and minimum values of the discharge pressure in the
compression station (sd) are 1.4 p.u. and 1 p.u., respectively. The reference pressure
at hub 2 is 0.85 p.u. The pipeline and compression station constants (K and KC) are
220 MW and 0.12 (dimensionless). Under these conditions, the maximum NG flow,
Qg21, is 278.28 MW, while the maximum electric power flow through F12 is
150 MW.
144 R. Rubio-Barros et al.
Summer and winter scenarios describe two different conditions for the electric
power and NG dispatch. The demands for those scenarios are shown in Table 4.
For each decoupled dispatch, the NG price for electricity generation at node 2 is
maintained at the NG marginal cost calculated in the integrated dispatch. The main
results are shown in Table 5.
In order to ensure consistency in the analysis, the total operation costs are
calculated according to (30) for the integrated and decoupled approaches. Other-
wise, the costs of the NG used by the NGFPPs would be double-counted. As it was
stated before, the origins of the economic inefficiencies in the decoupled dispatches
are due to the mismatch between the NG prices considered in ED of the electric
power system and actual NG marginal costs obtained in the ED of the NG system. As
it can be noted, lower NG prices for electricity generation at hub 1 (5.05 $/MWh to
Energy Carrier Networks: Interactions and Integrated Operational Planning 145
4.80 $/MWh) lead to more output power in pgn1 than in the integrated case
(38.11 MW to 60 MW) and to higher NG marginal at hub 1 (5.05 $/MWh to 5.29
$/MWh). The converse situation occurs for the +5% and +10% cases. It should be
noticed is that the pgn1 ranges from maximum to minimum output power with
a 5% to +10% change in the NG price at hub 1, showing major changes in the
operating schedule.
Only a combined ED of electric power and NG systems can ensure that NG
prices for electricity generation equal NG marginal costs at each node, and there-
fore, the minimum operating cost is achieved. Theoretically, the same results could
be obtained using the decoupled approach, i.e., ED of electric power and NG
systems run iteratively; however, the convergence of this procedure to a final
solution is not guaranteed. The NG marginal cost obtained in each iteration can
oscillate in a divergent manner. Thus, the convergence of the decoupled approach is
highly dependent on the starting point, i.e., the NG prices and the NGA considered
in the ED of the electric power system. The presented case study is an example that
the procedure can diverge. If the NG price at hub 1 considered in the first iteration
was 4.8 $/MWh (decoupled 5% case), the NG marginal cost at hub 1 obtained
from the NG ED would be 5.29 $/MWh. If in the next iteration, 5.29 $/MWh was
used as the NG price at hub 1 for the electric power ED, then NG marginal cost at
hub 1 obtained from the NG ED would be around 4.47 $/MWh (decoupled +5%
case). Therefore, it can be noticed that the procedure do not converge to 5.05 $/
MWh, i.e., the NG marginal cost at hub 1, obtained in the integrated approach.
The deadweight loss is the difference between the total costs obtained under the
decoupled ED and the integrated ED of electric power and NG systems. These total
operating costs differences (over-cost) are shown in the Fig. 7, in which decoupled
2.5% and +2.5% cases are also added. For the decoupled +10% case, the dead-
weight loss reaches the 1.2%. However, if the production costs of thermal units pg1
and pg2 were 20% higher than the presented in Table 3, the deadweight loss would
reach the 4.66% when the NG price at hub 1 was 27% higher than the NG marginal
cost at hub 1 calculated in the integrated case.
On the other hand, the winter scenario is suitable to evaluate the impact of total
NG availability for electricity generation NGA as coordinating parameter. Other
four cases are also presented to explain the influence of NGA:
• The integrated electricity and NG dispatch (Integrated)
• Three cases of the decoupled electricity and NG dispatch considering different
NGA:
146 R. Rubio-Barros et al.
1.30%
1.10%
0.90%
Over-cost (%)
0.70%
0.50%
0.30%
0.10%
• NGA 10% higher than the optimal NGA obtained in the integrated dispatch
(Decoupled +10%)
• NGA 20% lower than the optimal NGA obtained in the integrated dispatch
(Decoupled 20%)
• NGA 40% lower than the optimal NGA obtained in the integrated dispatch
(Decoupled 40%)
For each decoupled dispatch, the NG price for electricity generation at node 1
and 2 are maintained at the corresponding NG marginal cost calculated in the
integrated dispatch. These NG prices provide the signals for the optimal allocation
of the NGA between pgn1 and pgn2. The most important results are shown in
Table 6.
As it can be expected, a 10% higher NGA than the obtained in the integrated
approach leads to an infeasible NG dispatch, which is shown by 7.88 MW NG shortage
at node 1. In the cases with a 20% and 40% lower NGA than the obtained in the
integrated approach, the total operating cost are 0.35% and 0.47% higher than integrated
total operating cost, respectively. However, the deadweight loss increases if the
Energy Carrier Networks: Interactions and Integrated Operational Planning 147
production costs of thermal units that substitute NGFPPs production were higher. For
instance, if the production costs of thermal units pg1 and pg2 were 20% higher than the
presented in Table 3, the deadweight loss would reach the 1.45% in the cases decoupled
40%. The lower NGA impacts in a lower NGFPPs’ output power (pgn1 + pgn2) which
also means an underutilization of NG supply w2. The difference between the NGA used
in the electric power dispatch and the actual NGA is the reason of the over-costs in the
decoupled dispatches. In the integrated modeling and dispatch of electric power and NG
systems, the NGA is a by-product (not a parameter) of the optimization; thus, the
maximum economic efficiency is achieved. In the decoupled approach, the actual
NGA could be obtained within an iterative runs of electric power and NG dispatch,
however again the convergence is not guaranteed. Moreover, in the presented cases, the
estimation of the actual NGA only requires to know the NG consumption in the
compressor station (12.91 MW in the integrated case) because no NG transmission
capacities are reached, otherwise nodal NGA will be needed instead of a total NGA.
Finally, it is important to bear in mind that real cases include fixed NG prices and
NGA as coordinating parameters. Therefore the over-costs (deadweight losses) in
the decoupled approach can be higher than the presented in this case study since the
deviations have analyzed in only one coordinating parameter at each time. Also,
NG price deviations can lead to transport constraints in NG or electric power
networks, and thus, higher cost differences between the integrated and the
decoupled dispatch. However, in order to present a clear exposition, neither com-
bined deviations nor active transport constraints have been included in this chapter.
Note that in energy systems with large controllable hydropower contribution, the
scheduling of water movements between reservoirs, the availabilities and prices of
fuels, as well as the strategic network and generation maintenance plan, have
critical influences on the integrated operational planning. Similarly, NG storages,
in particular those with large working volumes, play a relevant role in seeking and
achieving the medium-term optimal decisions.
The integrated optimization of electric power and NG systems is consistent with
the optimal use of available energy resources. Thus, the combined social benefit
(overall consumer and producer surplus) obtained through this combined optimiza-
tion is greater than the sum of the benefits obtained when electric power and NG
systems are decoupled optimized.
The integrated energy systems planning with no storage units is solved by dispatching
NG suppliers/electric power generation with lower production costs to supply the energy
demands and the network losses. While there are additional factors that make the
problem complex (energy losses, transport limitations, nonlinear costs, etc.), it has two
basic characteristics: first, the dispatch is uncoupled in time, i.e. an operational decision
in the present does not affect the operating cost in the near future, thereby making the
problem separable; second, the units have a direct cost of production that depends mainly
on their own level of generation, rather than the generation level of the other units.
Conversely, the obvious feature of an integrated energy system with storage units
is to use the energy stored in the reservoirs of hydroelectric and/or NG storage units to
supply the demand, thereby avoiding the use of expensive oil fuels in thermal power
plants. However, the availability of limited amounts of energy makes the optimal
operation of integrated systems with energy storage a complex mathematical problem,
as the energy reservoirs create a coupling between operational decisions taken in the
present and the future consequences of these decisions.
Temporal couplings make the problem harder to solve, since all the stages within
the time horizon must be included in a single optimization problem. This condition
leads to the use of a large number of variables and constraints, increasing the
computing time. This complex and large-scale problem can be easier solved by
means of decomposition techniques (e.g., Benders method or Dantzig-Wolfe method),
which make possible to split the whole problem into several and smaller problems.
• The study time horizon (1–3 years) is divided into monthly (12 months) or
weekly (52 weeks) stages. Each stage is divided into subperiods (3–5 blocks) of
different time duration.
• The electricity and NG demands are represented by load duration curves
approximated with a step function. Although, there is a relationship between both
demands, usually it is assumed that they are independent and have similar profiles.
• The demands and fuel prices (fuel oil, gas oil, carbon, etc., except NG) have a
random nature, and therefore theses exogenous parameters are estimated by
means of appropriate forecast models.
• NG production costs are also uncertain, but they typically have a lower level of
dispersion.
• The watersheds have a random uncertainty on the lateral streamflows, making
necessary to apply appropriate forecasting models for each watershed.
• The production of NGFPPs is simulated by means of a linear model, using a
constant net heat rate for each NGFPP. For all other thermal units, operating
costs are also linearly modeled, taking into account the fuel prices and their net
heat rates and the corresponding fuel prices.
• The electricity transmission system is modeled through a DC flow model
without losses, which takes into account the Kirchhoff laws and the limitations
imposed by the network.
• The NG transmission system is formulated through a nodal energy balance
model, which takes into account the limits imposed by NG pipeline network,
but ignores the nodal pressures of the system.
• The preventive maintenance scheduling of generating units is considered.
• The probability of random outages of thermal generating units is taken into account
derating their maximum capacities using the forced outage rates (FOR)
• Spinning reserve requirements are also modeled derating the maximum
capacities of the generating units according to a prefixed reserve rate.
where k is the block index, bk is the number of hours at each block, t is the stage
index and at is a discount factor of stage t affected by a predetermined discount rate.
The stated objective function is subject to a set of electric systems constraints for
each block k of the stage t:
" !#
P P k;t
Fmax
m b Hmj pgi þ pgnk;t
i þ pgh k;t
i þ psk;t k;t
j Dej bFmax
m 8m
j i2Gj
(35)
X k;t
X X k;t
pgi þ pgnk;t
i þ pgh k;t
i þ psk;t
j ¼ Dej (36)
i j j
pghk;t k;t
i ¼ ri ðQw; H ÞQwi 8i (37)
0bpgk;t
i ; pgni ; pghi bPi
k;t k;t max
8i (38)
Constraints (35) and (36) model the electricity transmission system through
a DC flow model detailed above. Constraint (37) is the power equation of
the hydroelectric plant i where Qwk;ti represents the flow rate through the turbine
of the hydroelectric plant i in block k of stage t and ri the production ratio of the
hydroelectric power plant i in terms of the flow rate Qw and the hydraulic head H.
To simplify the problem, ri is modeled as a constant average production ratio
independent of the variables H and Qw. Finally, constraint (38) models the maxi-
mum capacity of power plants.
Also there are a set of constraints for each block k of stage t of the NG system:
0 X k;t X X 1
wg þ wsk;tj þ Qgm Qgm þ
B g2W j m2I j m2Oj C
B X X X C ¼ Dgk;t 8j (39)
@ A j
QOp
t
QIp t
HRi ðpgni Þ
p2P j p2P j i2Gj
Wgmin bwk;t
g bW g
max
8g (40)
m bQgm bQgm
Qgmax 8m
k;t max (41)
0bQOtp bQOmax
p 8p (43)
Equation (39) represents the nodal NG flow balance including storage inflows
and outflows, where QOtp and QIpt are the outflows (withdrawal rate) and inflows
(injection rate) of the NG storage p located on node j. P j is the subset of NG
Energy Carrier Networks: Interactions and Integrated Operational Planning 151
reservoirs located on the node j. In (39), HRi is modeled in the simplest way as
a constant and average net heat rate independent of the NGFPP production pgn.
Constraint (40) shows the limits of injection capacity of NG suppliers where
Wgmax and Wgmin are the maximum and minimum flow rate of the NG supplier g.
Constraint (41) shows the maximum carrying capacity Qgmax m of the pipeline m.
QIpmax and QOmaxp are the peak operating inflows (injection) and outflows (with-
drawal) of the NG storage p. These parameters change as the level of NG varies
within the facility, but they are modeled as constant values to simplify the problem
(constraints (42) and (43)).
In addition to the operational conditions outlined above, the complete set of
constraints also includes the time coupling constraints. These equations are
expressed through the balance equations of NG storages (44) and the continuity
equations of water reservoirs (45).
Vgtþ1
p ¼ Vgtp QOtp T þ QIpt T 8p (44)
P P P
Vetþ1 ¼ Vet þ Ate T Qwk;t
i bk Se T þ
t
Qwk;t
i bk þ Stu T 8e
k k u2E e
i 2 I he i 2 I hu
(45)
where Vgtp is a state variable that represents the stored volume of the NG reservoir
p at the beginning of stage t, Vet is a state variable that represent the stored volume
of the water reservoir e at the beginning of stage t, Ate is the lateral streamflow
arriving at the plant e, Ste is a slack variable that represents the spilled outflow of
the water reservoir e, \he is the subset of hydroelectric power plants i associated
with the water reservoir e, T is the total duration of stage t, \hu is the subset of the
upstream hydropower plants and ℰe is the subset of reservoirs u that spill into
the reservoir e. The line-pack of a pipeline, which is its intrinsic storage capacity,
can be modeled by a reservoir of small capacity (weekly, monthly capacity) in the
medium-term planning. This can be applied in very long pipelines in which
the line pack is sufficient to store a volume of NG from one stage to the
following one.
Finally, constraints (46) and (47) model the capacity of NG and water
reservoirs:
p bVgp bVgp
Vgmin 8p
t max
(46)
where Vgmin
p is the base gas (or cushion gas) of the NG storage, that is the volume of
gas intended as permanent inventory in a storage reservoir to maintain adequate
pressure and deliverability rate throughout the withdrawal season. Vgmax
p is the total
gas capacity of the storage. The difference between the total gas capacity and the
152 R. Rubio-Barros et al.
base gas is the working gas capacity of the storage, represented in constraint (46).
Vemin and Vemax are the minimum and maximum volumes of the water reservoir e.
The presented mathematical formulation address the so-called deterministic
problem, since all input parameters (demands, lateral streamflows, fuel prices,
etc.) are known values. However, as it was described before, these parameters are
uncertain. This means that the medium-term optimal operation planning is, in fact, a
stochastic optimization problem. Among other methodologies, Monte Carlo simu-
lation method can be implemented to deal with these uncertainties. Therefore, a
finite number of independent trials or simulations of the deterministic problem are
solved in order to achieve a satisfactory level of confidence in the resulting
probability distribution functions (PDF). This number depends on the tolerable
error specified for the planner which, in turn, greatly depends on the characteristics
of the problem. A detailed explanation of Monte Carlo simulation method can be
found in [55].
When a hydroelectric power plant produces power, not only incurs in a direct
monetary expenditure, but also incurs in an indirect cost associated with the
economic value of used amount of water that was previously stored. If this volume
of water was not used, it would be available for future utilization, an then, it would
save thermal generating costs or possible energy shortages. Thus, the optimal
management of the stored water must be consistent with the minimum variable
cost of thermal generation over the study period, which it is closely related to
the water opportunity cost, since any deviation from the optimal use of water at
each period produce additional thermal generation costs within the period. This
opportunity cost is known as expected water value or simply water value [50, 51]
and represents the economic value of the stored water. The water value defines the
economic suitability of accumulating water in the present for future use. One
advantage of this parameter is that it can be used, after its conversion to monetary
cost per unit of hydro-generated energy, to dispatch hydroelectric plants together
with thermal units in the short-term operation planning.
NG storage facilities introduce to the modeling of the problem similar
complexities as water reservoirs. NG storages allows the accumulation of NG in
time periods of reduced demands and/or low prices (typically during summer), for it
use in time periods of higher demands and/or prices (typically during winter).
Therefore, the optimal use of stored NG is aligned with the condition that
minimizes the sum of the present cost function (PCF) and the future cost function
(FCF). The PCF represents the production costs of NG in a given stage t and it
increases as the stored volume rise. The FCF is associated with the expected
cost of NG production, exports and shortages, starting from the end of stage
t (beginning of stage t þ 1) to the end of the study horizon. This cost, contrary
to PCF, decrease as the stored volume increase because a larger amount of stored
Energy Carrier Networks: Interactions and Integrated Operational Planning 153
PCF+FCF
Production costs
FCF PCF
Marginal cost
of production
Expected value
of NG
CProdmin
Vmin
Vmax
Optimal NG stored
decision volume
Fig. 8 PCF, FCF and expected value of natural gas for a stage
The example illustrated in Fig. 9 consists of two energy hubs (N1, N2), an electrical
system with two nodes (N1E, N2E), a transmission line (F12) connecting them, a
natural gas system with two nodes (N1G, N2G), a pipeline (Qg12) connecting them
and a hydroelectric system with two reservoirs in cascade. This example is used to
simulate the annual operational planning divided into 12 monthly stages, where
154 R. Rubio-Barros et al.
A1
V1
A2
N2E
V2 pgh1
pg2 N1E
pgh2
F12
ps pg1
De
N2 pgn2 pgn2 N1
Qg12
Dg w1
w2
ws FIOR N1G
VR
N2G
demands at each stage are divided into 4 blocks of 8, 100, 450 and 172 hours
duration. All systems are interconnected and are interdependent in their operation.
The link between electric and NG system occurs through the pgn1 and pgn2. The
link between the hydroelectric system and the electrical system is through pgh1, and
pgh2. The electric power system consists of thermal generating units with liquid
fuels (pg1 and pg2) associated with each node, whose costs are associated with the
cost of the liquid fuel used and the rate of specific fuel consumption of the unit.
Additionally, it uses an artificial unit of shortage (ps) to model the deficit of electric
power generation. The electricity load De is concentrated at the node N2E. The
system consists of NG production fields (w1, w2) associated with each node, whose
production costs are related to the cost of NG extraction. Field ws is used to
simulate NG shortages. Also, the model includes a NG reservoir VR that can deliver
NG to N2G or withdraw NG from this node steadily during each monthly
period. The hydroelectric system consists of a hydro plant V1 and run-of-water
plant V2 topologically located in cascade, with lateral streamflows A1 and A2.
Each hydroelectric generating unit (pgh1, pgh2) has an associated constant average
production ratio. Table 7 presents the data used in the problem.
Solved Cases
The example with all integrated energy systems is solved deterministically (one
scenario) using the objective function (34) and constraints (35, 36, 37, 38, 39, 40,
41, 42, 43, 44, 45, 46, 47). In order to show the optimal use of storable resources, we
present the following study cases:
• Case 1: Integrated system without NG storage and without hydro reservoirs (run-
of-river power plant).
Table 7 Example data
Electrical system
Thermal Max Capacity Specific fuel Transmission Max power flow Electricity Electric shortage cost
generation (MW) consumption system (MW) shortage ($/MWh)
pg1 80 1.1116 t/MWh F12 500 Cps 1,500
pg2 150 1.5708 t/MWh
pgn1 50 0.1935 dam3/MWh
pgn2 80 0.2150 dam3/MWh
Hydroelectric system
Hydro Max capacity Max turbined Average production Min reservoir Max reservoir Initial volume End volume
generation (MW) outflow (m3/s) ratio (MW.s/m3) volume (hm3) volume (hm3) (hm3) (hm3)
pgh1 129.6 200 0.648 200 1,200 600 600
pgh2 158.4 200 0.792 200 100 150 150
Natural gas system
NG fields Max production (dam3/day) Pipeline Max NG flow (dam3/day) Natural gas shortage Natural gas shortage cost ($/dam3)
w1 1,200 Qg12 1,000 Cws 8,000
w2 400
NG Max injection capacity Max withdrawal capacity Min volume Max volume Initial volume End volume
storage (dam3/day) (dam3/day) (dam3) (dam3) (dam3) (dam3)
VR 300 400 0 20,000 5,000 5,000
Costs and inflows
Energy Carrier Networks: Interactions and Integrated Operational Planning
Stage TG Costs ($/MWh) Fuel Costs ($/t) Lateral streamflows (m3/s) NG Costs ($/dam3)
pg1 pg2 pg1 pg2 A1 A2 w1 w2
1 77.81 117.81 70.00 75.00 84.00 11.00 108.45 228.45
2 83.56 123.97 75.17 78.92 42.00 6.50 112.37 232.86
3 85.69 126.52 77.09 80.54 36.00 5.00 113.73 234.70
4 77.29 118.53 69.53 75.45 24.00 6.00 107.74 229.22
155
1
Average price calculated as the price weighted by the duration of the block.
158 R. Rubio-Barros et al.
• The average prices of N1G and N2G in Case 2 show a relatively constant value
over the study horizon (Fig. 10a) due to the use of the NG reservoir; whereas in
Case 1, having no capacity to store NG for high demands periods, NG supply with
higher cost (w2) is needed, and thus total generation/production cost are higher
than in Case 2 (Table 8). Figure 11 clearly shows that the use of the NG storage
results in a decrease of injection of w2, reducing total NG production costs.
• The average prices of N1E and N2E in Case 3 are relatively constant during the
year due to the use of water stored in V1 (Fig. 10b), decoupling the time of
occurrence of inflows regarding the time when water is required for power genera-
tion; conversely, in Case 1, having no possibility to dam water for times of high
demands and/or less inflows, the water inflows should be used as run-of-river power
plant, generating all the natural river water flow at each given time (Fig. 12),
increasing the total operating cost. It should be noted that the amount of hydro
Fig.12 Electric demand and pgh1 supply with and without reservoir V1
Energy Carrier Networks: Interactions and Integrated Operational Planning 159
energy are the same for the two cases because there was no spilling of water in
the study period, thus the both cases can be compared.
• Case 4 and Case 5 combine the effects discussed above. Table 8 shows that
the integrated operational planning of water and NG reservoirs results in
further benefits for the system as a whole since the lowest operating costs (of
the simulated cases) are obtained. Moreover, in Case 4 electricity and NG
prices are relatively more constant over time than the prices obtained for
first three cases, even though these prices are higher in the final stages
(Fig. 10).
• Case 5 shows a small decrease in total costs compared with Case 4 (Table 8),
thereby indicating that in Case 4 more hydro storage capacity is still needed.
This analysis, despite its simplicity, shows the importance of the operational
planning of energy carriers and the advantages coming from their combined
optimization and the regional integration of these resources. The optimal sched-
uling of energy storages (water reservoirs, NG reservoirs) allows to achieve an
overall benefit for all the involved energy systems considering the limited
availability of the energy resources and the technical operating constraints of
the energy systems.
Cg (wg), and HRi(pgni) are usually modeled as quadratic functions. All generating
units have a minimum output power, in particular for thermal units, it can reach up
to 50% of the maximum output. Thermal generating units also have some technical
features that must be considered: (a) significant startup cost; (b) minimum up- and
down-time constraints; (c) maximum ramp-up and ramp-down limits. Under these
conditions, the start-ups and shutdowns are explicitly modeled by means of binary
variables, thus the problem is so-called unit commitment. The startup costs are
added to the objective function (30). Technical characteristics (b) and (c) constitute
time coupling constraints in this short-term optimization problem.
There are two alternative approaches to employ the results of the medium-term
operational planning as coordinating parameters for the optimal use of the energy
storages in the short-term scheduling [52]. One of them, called primal approach,
consists in using the optimal amount of energy (stored water and stored NG),
determined in the medium-term model, for the time horizon to be optimized in
the short-term model. This means, that (37), (45) and (47) must be included to
consider the hydro-system. Equation (45) is re-defined on an hourly basis, and the
Vet for the last simulated hour is fixed to same value obtained in the medium-term
model. In this primal approach, the CHi(pghi) are equal to zero. To consider NG
storages, QOtp and QIpt are added to (26) if NG storage facilities are located in the
corresponding nodes. Equations (42), (43), and (46) must be also introduced. The NG
storage balance constraint (44) is re-defined on hourly time steps, and the Vgpt for the
last simulated hour is fixed to same value obtained in the medium-term model.
The other approach for coordinating medium-term and short-term optimizations
is called the dual approach [53]. It consists in using the water and NG economic
values described in Sect. 3.6.5 Water values can be utilized to calculate the
opportunity cost functions CHi(pghi) according to the layout of hydrological net-
work. In a similar way, NG storages can be modelled as equivalent NG suppliers
considering supplying cost functions Cg based on the NG values.
Another complex unit commitment problem that exists is with compression
stations. Since they include several compressor units, there is an optimal operating
point that minimizes the amount of NG consumed. However, this specific issue is
out of the scope of the presented integrated short-term operational planning.
Electric power flows can be represented with the highest possible detail, i.e., the AC
power flow model described in Sect. 2.2. This implies that reactive power resources
(e.g., generating units; shunt, synchronized and static VAR compensators) should
be introduced in the modeling, as well as reactive power loads. However, if electric
power system has sufficient and well-distributed reactive power resources and
adequate reactive power flow controlling devices to keep the voltage within the
operating limits, AC power flows model can be disregarded. This assumption is in
particular suitable to deal with the integrated energy systems operational planning,
which is focused on the real quantities of energy transferred within the networks. In
Energy Carrier Networks: Interactions and Integrated Operational Planning 161
this context, the DC model considering losses, also described in Sect. 2.2, is
appropriate to model the electric power flows.
As it was introduced in Sect. 2.5, hourly NG flows with changing demands cannot
be accurately modeled by means of steady-state gas flow equations. A transient NG
flow model must be considered to take into account the typical imbalanced operation
of the NG systems, i.e., for a given hour the amount of NG injected in the NG
transmission systems does not match the total NG withdrawals. This transient model
allows to represent the line-pack, i.e., the amount of NG stored in pipeline network,
and to optimize its use for the minimization of the total operating and shortage costs.
Essentially, simplified isothermal transient NG flows can be described by means of
the followings partial differential equations [44].
@s 4 sst T Z @Qg
¼ (48)
@t p Tst D2 @x
@s2 64 s2 f G T Z
¼ 2 st2 5 Qg2 (49)
@x p Tst D Rair
@ 2 s2 fGQg @s2
¼ (50)
@x 2 pD3 Rair T Z @t
where (48) and (49) are the continuity and the momentum equations. These
equations can be written as (50) if it assumed that the NG flow at standard
conditions Qg is averaged over length in every time interval Dt. To solve these
partial differential equations several finite differences schemes can be used [44]. In
[48], Eqs. (48) and (49) are applied to the optimization of a single pipeline
operation. The classical implicit method to solve (50) can be expressed as [54, 44]
Stj jþ1 2 Sj
Stþ1 þ Stþ1
tþ1
Stþ1
j j1
¼a (51)
Dt ðDxÞ2
where S ¼ s2, a is the constant indicated in (50). The index j-1, j and j + 1 are the
contiguous nodes resulting of the segmentation of each pipeline (principle of
finite difference methods). Equation (51) must be introduced as constraints to
model transient NG flows within the integrated short-term energy systems
optimization.
Regarding water flows, if the water continuity constraint (45) is considered in
the short-term multi-energy carrier optimization (primal approach), then it is
important to take into account the time delays existing since the water is released
in the upstream water reservoir up to this water arrives to downstream water
reservoir.
Finally, the short-term operational planning must also include systems’ security
constraints. While minimum spinning (operating) reserve is considered for electric
power systems, minimum total line-pack is taken into account for NG systems.
162 R. Rubio-Barros et al.
4 Conclusions
The integration of NG and electricity energy carriers is a reality, and their strongest
interdependence must be mandatory considered in the centralized planning
modeling of energy resources management. The NGFPPs are the main protagonists
in this integration. They link both energy carriers because they can arbitrate their
participation in both markets to the best suited price from social welfare viewpoint;
otherwise, the link is through fixed parameters, losing the economic synergy. In this
context, the complexity of the centralized planning optimization problem increases.
Therefore, this challenge requires new approaches to model the energy carrier
interactions and new solution procedures to divide the entire problems into coordi-
nated subproblems. This chapter presents a comprehensive literature survey of state
of the art on integrated NG and electricity operational planning. The relevant
characteristics of NG and electricity systems are compared considering the physical
laws that govern the flows of these energy carriers through dedicated networks.
Also, the interactions among the energy carriers and their networks are modeled
with different levels of detail according to the evaluated time horizon. Finally, some
contributions about economic interactions between different energy carriers (elec-
tricity, NG, and hydro energy carriers) are presented with emphasis on the concept
of water values that can be extended to NG values when storage facilities are
available. An example of a medium-term planning is presented to show the tempo-
ral interdependencies introduced by hydro and NG reservoirs.
The formulation of integrated operational planning presented in this chapter
addresses the problem from a centralized perspective, and thus it is based on the
economic theory of perfect competitive markets. Although the economic concept of
perfect markets is never satisfied in reality, the proposed approach provides a
benchmark against which actual electricity and NG markets can be compared,
since the results obtained from a centralized optimization reflect the efficient use
of scarce energy resources.
Acronyms
AC Alternating current
CCGT Combined-cycle gas turbine
DC Direct current
ED Economic dispatch
FACTS Flexible alternating current transmission system
FCF Future cost function
LHV Lower heating value
LNG Liquefied natural gas
NGFPP Natural gas fired power plant
(continued)
Energy Carrier Networks: Interactions and Integrated Operational Planning 163
NG Natural gas
OPF Optimal power flow
PCF Present cost function
PDF Probability distribution functions
PTDF Power transfer distribution factors
SDDP Stochastic dual dynamic programming
UK United Kingdom
USA United States of America
Variables
F Active power flow of electric power lines
F Vector of lines active power flows
L Shunt losses of electric power lines
Lg NG consumed in a compression station
P Active power flow injections
P Vector of active power injections
pg Production of a thermal generating unit which do not use NG as fuel (i.e., coal,
nuclear, fuel oil)
pgh Hydropower plant production
pgn NGFPP production
PG Vector of generating unit’s production
ps Load shedding
PD Vector of demands
PL Vector of losses
Qg NG flow at a pipeline
QI Inflow (injection rate) of NG storage
QO Outflow (withdrawal rate) of NG storage
Qw Flow rate through the turbine of the hydroelectric plant
s Nodal pressure
S Spilled outflow of the water reservoir/complex power injection
U Complex voltage
V Stored volume of the water reservoir
Vg Stored volume of the NG reservoir
w NG injection of the supplier
ws NG load shedding at node
y Phase angle of complex voltage
u Vector of voltage phase angles
Functions
C(pg) Production cost of a thermal generating unit which do not use NG as fuel (i.e., coal,
nuclear, fuel oil)
CG(pgn) Production cost of the NGFPP
CH(pgh) Opportunity cost of production of a reservoir-associated hydropower plant
CSE(ps) Electric power shortage cost
C(w) NG supply costs
CSG(ws) NG shortage costs
HR(pgn) Heat rate function or input-output curve of NGFPP
r(Qw,H) Production ratio of the hydroelectric power plant
(continued)
164 R. Rubio-Barros et al.
Parameters
A Node-branch incidence matrix
A Lateral streamflow arriving at the water reservoir
b Number of hours at each block
BD Diagonal matrix whose elements are 1/X
D Inside diameter of the pipe
De Electricity demand
Dg NG demand
f Dimensionless friction factor of the pipe
FOR Force outage rate of generating unit
G Gas gravity relative to air
H Sensitivity matrix or power transfer distribution matrix
K Factor representing pipeline characteristics and NG properties
KC Constant of the compression station
L Length of the pipe
NGA NG availability for electricity generation
Qgmax Maximum transport capacity of the pipeline
QImax Peak operating inflow (injection) of the NG storage
QOmax Peak operating outflows (withdrawal) of the NG storage
R Resistance of the line
Rair Ideal gas constant of dry air
T Total duration of stage
Tst, sst Standard or normal conditions of temperature and pressure
Vmax, Vmin Maximum and minimum volumes of the water reservoir
Vgmin Base gas (or cushion gas) of the NG storage
Vgmax Total gas capacity of the NG storage
Wmax, Wmin Maximum and minimum flow rate of the NG supplier
X Reactance of the line
Y Shunt admittance of the P-equivalent line
Z Series impedance of the P-equivalent line
a Discount factor of the stage
Sets
Cj Compression stations whose suction side is connected to node j
Ee Reservoirs that spill into the reservoir e
Gj Generating units connected to node j
Ij Pipelines which flows are incoming node j
I hu Upstream hydropower plants
I he Hydroelectric power plants associated with the water reservoir e
(continued)
Energy Carrier Networks: Interactions and Integrated Operational Planning 165
References
1. Munasinghe M, Meier P (1993) Energy policy analysis and modeling. Cambridge University
Press, Cambridge, MA
2. Loulou R, Remne U, Kanudia A, Lehtila A, Goldstein G (2005) Documentation for the TIMES
Model, Part I. Energy Technology Systems Analysis Programme [Online]. Available: http://
www.etsap.org. [Accessed: 28 Agu 2009]
3. Messner S, Schrattenholzer L (2005) MESSAGE-MACRO: linking an energy supply model
with a macroeconomic model and solving it inter-actively. Energy 25:267–282
4. CEEESA (2008) Overview of the Energy and Power Evaluation Program (ENPEP-BAL-
ANCE). Center for Energy, Environmental, and Economic Systems Analysis (CEEESA),
Argonne National Laboratory, [Online]. http://www.dis.anl.gov/projects/Enpepwin.html
[Accessed: 15 Nov 2008]
5. SEI (2006) LEAP: User Guide. Stockholm Environmental Institute, [Online]. Available: www.
energycommunity.org/documents/Leap2006UserGuideEnglish.pdf [Accessed: 08 Mar 2008]
6. van Beeck N (1999) Classification of energy models. Tilburg University and Eindhoven
University of Technology, The Netherlands
7. Bakken H, Skjelbred HI, Wolfgang O (2007) eTransport: investment planning in energy
supply systems with multiple energy carriers. Energy 32:1676–1689
8. Hecq S, Bouffioulx Y, Doulliez P, Saintes P (2001) The integrated planning of the natural gas
and electricity systems under market conditions. In: Proceedings of the IEEE power engineer-
ing society PowerTech, Porto, 2001
9. Unsihuay C, Marangon-Lima JW, Zambroni de Souza AC (2007) Integrated power generation
and natural gas expansion planning. In Proceedings of the IEEE power engineering society
PowerTech, Lausanne, 2007
10. International Energy Agency (2007) Natural gas market review 2007. IEA/OECD
Publications, Paris
11. International Energy Agency (2008) Electricity information 2008. IEA/OECD Publications,
Paris
12. International Energy Agency (2008) World energy outlook 2008. IEA/OECD Publications,
Paris
13. International Energy Agency (2009) Natural gas market review 2009. IEA/OECD
Publications, Paris
14. Shahidehpour M, Fu Y, Wiedman T (2005) Impact of natural gas infrastructure on electric
power systems. Proc IEEE 93(5):1042–1056
15. Chen H, Baldick R (2007) Optimizing short-term natural gas supply portfolio for electric
utility companies. IEEE Trans Power Syst 22:232–239
16. Street A, Barroso LA, Chabar R, Mendes ATS, Pereira MV (2008) Pricing flexible natural gas
supply contracts under uncertainty in hydrothermal market. IEEE Trans Power Syst 23:1009–1017
166 R. Rubio-Barros et al.
17. Takriti S, Supatgiat C, Wu LS-Y (2001) Coordination fuel inventory and electric power
generation under uncertainty. IEEE Trans Power Syst 16:603–608
18. Morais MS, Marangon Lima JW (2003) Natural gas network pricing and its influence on
electricity and gas markets. In: Proceedings of the IEEE power engineering society
PowerTech, Bologna, 2003
19. Morais MS, Marangon Lima JW (2007) Combined natural gas and electricity network pricing.
Elec Power Syst Res 77:712–719
20. Rubio R, Ojeda-Esteybar D, Añó O, Vargas A (2008) Integrated natural gas and electricity
market: a survey of the state of the art in operation planning and market issues. In: Proceedings
of 2008 IEEE/PES transmission and distribution conference and exposition: Latin America,
Bogotá, 2008, pp 1–8
21. Quelhas A, Gil E, McCalley JD, Ryan SM (2007) A multiperiod generalized network flow
model of U.S. Integrated energy system: part I – model description. IEEE Trans Power Syst
22:829–836
22. Gil EM, Quelhas AM, McCalley JD, Voorhis TV (2003) Modeling integrated energy trans-
portation networks for analysis of economic efficiency and network interdependencies. In:
Proceedings of North American power symposium (NAPS), Rolla, 2003
23. Correia P, Lyra C (1992) Optimal scheduling of a multi-branched interconnected energy
system. IEEE Trans Power Syst 7:1225–1231
24. Bezerra B, Kelman R, Barroso LA, Flash B, Latore ML, Campodónico N, Pereira MVF (2006)
Integrated electricity-gas operations planning in hydrothermal systems. In: Proceedings of X
SEPOPE, Florianópolis 2006
25. Unsihuay C, Marangon-Lima JW, Zambroni de Souza AC (2007) Short-term operation
planning of integrated hydrothermal and natural gas systems. In: Proceedings of the IEEE
power engineering society PowerTech, Lausanne, 2007
26. Li T, Erima M, Shahidehpour M (2008) Interdependency of natural gas network and power
system security. IEEE Trans Power Syst 23:1817–1824
27. Liu C, Shahidehpour M, Fu Y, Li Z (2009) Security-constrained unit commitment with natural
gas transmission constraints. IEEE Trans Power Syst 24:1523–1536
28. Chaudry M, Jenkins N, Strbac G (2008) Multi-time period combined gas and electricity
network optimisation. Elec Power Syst Res 78:1265–1279
29. An S, Li Q, Gedra TW (2003) Natural gas and electricity optimal power flow. In: Proceedings
of the IEEE power engineering society transmission and distribution conference, Dallas, 2003
30. Unsihuay C, Marangon Lima JW, Zambroni de Souza AC (2007) Modeling the integrated
natural gas and electricity optimal power flow. In: Proceedings of the IEEE power engineering
society general meeting, Tampa, 2007
31. Mello OD, Ohishi T (2006) An integrated dispatch model of gas supply and thermoelectric
generation with constraints on the gas supply. In: Proceedings of X SEPOPE, Florianópolis,
2006
32. Munoz J, Jimenez-Redondo N, Perez-Ruiz J, Barquin J (2003) Natural gas network modeling
for power systems reliability studies. In: Proceedings of the IEEE power engineering society
PowerTech, Bologna, 2003
33. Geidl M, Andersson G (2007) Optimal power flow of multiple energy carriers. IEEE Trans
Power Syst 22:145–155
34. Hajimiragha A, Canizares C, Fowler M, Geidl M, Andersson G (2007) Optimal energy flow of
integrated energy systems with hydrogen economy considerations. In: Proceedings of bulk
power system dynamics and control – VII, Charlestone, 2007
35. Rajabi H, Mohtashasmi S (2009) Economic dispatch problem considering natural gas trans-
portation cost. Proc World Acad Sci Eng Technol 38:1482–1487
36. Ojeda-Esteybar D, Rubio-Barros R, Añó O, Vargas A (2009) Despacho óptimo integrado de
sistemas de gas natural y electricidad: comparación con un despacho desacoplado y aplicación
al sistema argentino. In: Proceedings of the XIII ERIAC, Puerto Iguazú, 2009
Energy Carrier Networks: Interactions and Integrated Operational Planning 167
Abstract This article describes the fuel transportation and storage components of
the supply chain for electricity. We focus on dispatchable generation based on
transportable fuels. Coal has very flexible transportation and storage requirements.
Natural gas requires pressurized pipelines and storage facilities; or it can be
liquefied, then stored and transported at very low temperatures, and then
revaporized. Biomass presents logistical challenges related to its relatively low
energy intensity and seasonality of supply. We review ways to model the physical
constraints and cost characteristics that govern the transportation and storage of
these fuels and examine their implications for decision models in restructured
electricity markets.
1 Introduction
A. Sorokin et al. (eds.), Handbook of Networks in Power Systems II, Energy Systems, 169
DOI 10.1007/978-3-642-23406-4_6, # Springer-Verlag Berlin Heidelberg 2012
170 S.M. Ryan and Y. Wang
could help determine how to best exploit these resources to meet residual demand
after nuclear, hydro and intermittent generation is taken into account. Considering
fuel transportation along with electricity transmission allows comprehensive
comparisons of the economic and environmental impacts of different methods for
moving energy from the location of fuel extraction to the point of its end-use as
electric energy. For example, to satisfy incremental demand for electricity using
coal, the options include “coal by rail”, in which coal is transported to the load
center where it is used in a coal generation plant; “coal by wire”, in which
electricity is delivered by high-voltage transmission line from a mine-mouth
plant; “coal to gas by pipeline”, in which coal is gasified and the synthetic gas is
delivered by pipeline to a combined cycle plant near the load; and “coal to gas by
wire”, in which electricity is transmitted to the load from a gasifier and combined
cycle plant at mine-mouth [8].
We examine the physical, economic and environmental characteristics of
transporting each fuel and generating electricity from it. Where applicable, we
also survey optimization models that have included these features. Economies of
scale influence the tradeoffs between investment and operating costs as well as the
prevalence of fuel purchases in long term contracts rather than spot markets. Fuel
transportation costs must be weighed against electricity transmission losses when
determining generating plant size and location. Uncertainties concerning future
voluntary efforts and regulations to limit carbon and other emissions affect capacity
investment decisions. Deregulated fuel and electricity markets increase the volatil-
ity of prices both upstream and downstream of each generator. Understanding these
characteristics and interactions helps the modeler decide where and to what extent
to include discrete, nonlinear and stochastic components in electricity supply chain
optimization models.
Models have been constructed in government and academia to simulate the energy
system in the U.S. and used to study the interactions between fuel and electricity
subsystems.
Beginning in the mid-1970s, the U.S. Department of Energy’s Energy Informa-
tion Administration (EIA) and its predecessor, the Federal Energy Administration
(FEA), developed a sequence of computer-based, medium-term energy modeling
systems to analyze domestic energy markets and the relationships among electric
energy and fuels. The first of these, the Project Independence Evaluation System
(PIES) [9] was employed by the FEA prior to 1982. It was designed to provide a
framework for the development of a national energy policy through quantitative
analysis and projections of the energy system. PIES incorporated effects of fuel
price sensitivity, fuel competition, technology restriction or improvement, resource
limitations, economic factors, and regional variations on the energy system. In
1982, PIES was updated to the Intermediate Future Forecasting System (IFFS) [10].
172 S.M. Ryan and Y. Wang
In contrast to the PIES, IFFS was partitioned by fuel to avoid complex integration
issues and balance the workload among the staff in charge of submodels. In 1993,
the IFFS was replaced by the National Energy Modeling System (NEMS) [11].
Coupling advanced modeling and optimization techniques with the latest comput-
ing technology, NEMS combines and processes more energy information than its
predecessors and therefore is more capable with projections. The system is used to
test different assumptions about energy markets and to evaluate the potential
impacts of new and advanced energy production, conversion and consumption
technologies.
Some recent efforts have been made to formulate and analyze optimization
or equilibrium models of the supply chain composed of both fuel and electricity.
Quelhas et al. [12] formulated a generalized minimum cost network flow model
of the bulk energy transportation system in the U.S. When applied in a case
study of a recent year, the results suggested that the cost to supply a fixed
demand for electricity could be lowered by substituting more coal for natural
gas and increasing interregional electricity trade [13]. However, the authors
acknowledged that gas turbines may have been underutilized as peaking
generators in the case study due to load aggregation. The deterministic nature
of the model also may have misrepresented decision-making in the context of
fuel price volatility. Wang and Ryan [14] implemented a two-stage stochastic
programming version of the model in a rolling horizon simulation with gas price
forecast updates and found that the simulated results matched the actual fuel
mix much more closely.
Liu and Nagurney [15] formulated a network equilibrium model of fuel and
electricity markets and applied it in case studies of the New England region of
the U.S. They assumed fuel prices would be constant over the medium-term
study period while electricity prices could fluctuate, and that the electric power
market was perfectly competitive. The results showed the effects of fuel price
variations on electricity price variations; revealed interactions among the
markets for power, natural gas and fuel oil; and illustrated the impacts of
electricity demand changes on the markets for both electric power and natural
gas. Ryan et al. [16] combined a simple fuel transportation model with an
equilibrium model for the electricity market in which generators could
strategically set their output levels based on both their marginal fuel costs and
the geographic variation of electricity locational marginal prices. The equilib-
rium model was validated by comparison with a detailed computational agent
simulation [17] and the effect of demand price sensitivity on generator market
power was investigated [18].
While NEMS is very comprehensive, it is not available for researchers outside of
the U.S. government to use or modify for their own purposes. Many sophisticated
models of electricity markets have been developed by academic researchers, but
few incorporate the supply, transportation, and storage of fuels in any detail. In the
next section, for several different fuels we highlight characteristics of the supply
chain and review examples of optimization or equilibrium models for both the fuel
itself and its use in generating electricity.
Costs and Constraints of Transporting and Storing Primary Energy 173
3.1 Coal
In the U.S. in 2009, 94% of coal consumption was for electric power. Over half of
the production was of Western, low-sulphur coal, while 32% came from the
Appalachia region in the Eastern U.S. and 14% from the country’s interior [19].
Coal is easy to transport and store. It is transported mostly by rail but also barge,
truck, and other means including slurry pipeline. Multimodal transportation is used
for the longest distances of over 1,000 miles, while rail is used for shipments
averaging 800 miles, barge for distances of approximately 300 miles, and truck or
other means for the shortest distances [20]. Most rail shipments are in dedicated
“unit trains” of over 100 cars that transport only coal to a single location [21]. At the
end of 2009, coal stocks in the electric power industry equaled 19% of the amount
consumed for the year, but this was a record high level attributable to low demand
during the year caused by economic downturn, weather patterns that reduced
demand, and lower prices for natural gas [19].
Most coal consumed by electric utilities is purchased in long-term contracts,
which by definition span at least 1 year but can last up to 50 years. In 1997 the
average length of a utility coal contract was nearly 16 years [20]. Joskow conducted
a series of studies on such contracts negotiated in the 1960s and 1970s to determine
the reasons for their prevalence and the strength of their provisions for price
adjustments when spot market prices dropped. He found that contract duration
increased with the level of relationship-specific investment, including (1) site
specificity as in mine-mouth generation plants “typically developed simultaneously
with the mines themselves”; (2) physical asset specificity such as generators
designed to use a specific type of coal; and (3) dedicated assets such as capacity
built by the supplier to meet demand from a specific customer [22]. The most
common type of contract was “base price plus escalation (BPE),” which specified a
price based on market conditions at contract signing and an adjustment formula that
incorporated exogenous input-price indexes, such as for labor, materials and
supplies, explosives, electricity, and general inflation [23]. The contracts also had
“take-or-pay” provisions; i.e., minimum and maximum quantities that the buyers
were obligated to take [24].
MacDonald studied how long term coal contracts were modified in the 1980’s in
response to falling prices for both coal itself and rail transportation. Economies of
174 S.M. Ryan and Y. Wang
scale are significant in rail transportation as rail costs per ton moved in a dedicated
unit train decrease with the train size [25]. Dennis studied the relationship between
coal delivered prices and transportation rates before and after railroad deregulation
occurred in 1980. Based on a spatial equilibrium model, under which the variation
in prices across regions depends only on the variation in transportation rate, he
showed that the standard deviation (across regions) of delivered coal prices had
dropped mostly steadily from the mid-1970s to 1996, and concluded that rail
deregulation had resulted in lower transportation rates for coal [26].
The prevalence of long-term coal contracts results from economies of scale in
both transportation and generation and the resulting need for both supplier and
consumer to make significant investments in their capability to handle large
volumes of material. It follows that generation capacity investment models should
include fuel purchase and transportation contracts along with physical plant
decisions. An example of such a model is that of Bienstock and Shapiro [27],
who included the acquisition of coal contracts in their stochastic programming
study of electric utility capacity expansion. Possible investments in the first stage of
a 10-year study horizon included coal plants with and without flue gas
desulphurization, a gas plant, an oil plant, a peaking plant, two different long-
term coal contracts, and a slurry pipeline. Coal purchases on the spot market were
included in the second-stage, operational variables. Uncertainties included demand,
air pollution (SO2) control limits, the date the slurry pipeline would be completed,
and fuel prices. Integer decision variables were used to model fixed and nonlinear
costs for adding plants and evaluating fuel purchase contracts.
Natural gas is a major source of electricity generation through the use of gas
turbines and steam turbines. Natural gas burns more cleanly than other fossil
fuels, such as oil and coal, and produces no sulfur dioxide and less carbon dioxide
than coal. In addition, gas-fired generating units have distinct advantages of high
efficiency, fast response and shorter installation time. It is projected that, by 2035,
46% of the new generation capacity in the U.S. will use natural gas as the preferred
fuel [28]. It is transported primarily by pipeline and has many uses other than
electricity generation – only 29% is currently used for electricity generation in the
U.S. Restructuring in the past decades in Europe and by the Federal Energy
Regulatory Commission (FERC) in the U.S. has reduced the vertical integration
in the gas industry.
Gas distribution utilities formerly purchased most of their gas from pipeline
transmission companies through long-term contracts which typically were com-
posed of commodity charge, demand charge and winter requirement charge. The
latter two charges, together with a “take-or-pay” clause, induce distribution utilities
to use storage facilities and add interruptible industrial customers to increase the
ratio of the average daily purchase to the peak-day purchase [39]. The restructuring
of the natural gas industry that began with FERC Order 436 (1985) and was
substantially completed with Order 636 (1992) has changed gas transportation
patterns and rates [40]. The role of pipeline companies was transformed from
merchants to nondiscriminatory carriers so that local distribution companies
(LDCs) can now purchase gas directly from the producers. Duann [41] discussed
the economic effect of drastic increases in the amount of gas directly purchased by
LDCs, which induced more responsive adjustments of gas production and transpor-
tation capacity. New market flexibility has contributed to greater efficiency in the
use of the gas industry infrastructure. As the unbundling of pipeline sales, transpor-
tation, and storage increased competition among gas suppliers, LDCs face more
complicated options in gas procurement, transportation and storage. However, they
see opportunities to achieve cost savings by building a least-cost gas supply
portfolio [42].
Structural changes in the natural gas industry have increased the importance of
systematic analysis of supply combinations by LDCs. Optimization models
designed for this problem are usually multi-period cost-minimizing linear programs
with supply, demand and storage constraints. Choices are made between firm and
interruptible pipeline reservations, firm and interruptible sales, spot purchase and
long term contract, injections and withdrawals from storage [43].
The natural gas demand market is divided into submarkets of residential,
commercial, and industrial end users. Gas demand mostly depends on weather
conditions and is therefore fundamentally highly variable. Guldmann and Wang
[42] used a two-stage stochastic mixed-integer linear program to incorporate
uncertain demand in the gas procurement and delivery decision process, where
contracts are determined in the first stage and the operational variables are selected
as recourse to the demand scenarios. The authors proposed a simulation method
which approximates the total cost as a response function of demand by sampling
and regression. This approach provides a convenient way to conduct sensitivity
analyses over parameters other than demands. Unlike most prevailing models
where deliverability, the maximum amount of gas that can be withdrawn from or
injected into storage each day, is given as exogenous, another linear stochastic
program proposed by Bopp et al. [44] explicitly considers it as a decision variable
that closely ties and impacts other variables. A case study of the LDC at Huntsville,
Alabama, showed that in order to achieve minimum cost, deliverability must be
carefully chosen according to the minimum level of firm transportation
requirements and the availability of spot purchases. The total costs decrease if
178 S.M. Ryan and Y. Wang
more storage facilities (thus more rates of deliverability) are available. The model
suggests that LDCs negotiate for a rate of deliverability close to the optimal rate. In
a 1 year horizon model that aids a utility in selecting gas contracts, Butler and Dyer
[45] adopted a daily-weekly-monthly method to aggregate future periods and yet
still obtain insights regarding current daily operational decisions.
While most studies in the literature focus on the U.S. natural gas industry,
Contesse et al. [46] developed a mixed-integer model for Metrogas, a Chilean
LDC, in support of its daily operations. This model incorporates special features
that do not appear in the U.S. gas system, including make-up generation and
recovery, the aging of generated make-up, the adoption of linepack to handle
demand fluctuation and the inclusion of fines for different types of deficits and
overruns. Evidently, the use of linear programs in selecting an optimal gas supply
portfolio is easily adapted to various industrial specifications. Restrictions on the
desired level of detail provided are usually due to constraints on computational
power available.
Restructuring of the gas markets in the U.S. and Europe has reduced the amount of
vertical integration and provided opportunities for different entities in the supply
chain to pursue their own objectives. Because decision-making is distributed among
multiple agents whose objectives may not be perfectly aligned, equilibrium models
are more appropriate than system optimization from a single perspective. More-
over, while a major goal of restructuring was to increase competition, the large
market shares held by a few entities have allowed those players to exert oligopolis-
tic power. Gabriel et al. [47] reviewed models of gas markets in the U.S. and
formulated a nonlinear complementarity model in which gas marketers compete as
Nash-Cournot players and other participants including pipeline, production, storage
reservoir and peak gas operators as well as consumers are price-takers. Gabriel and
Smeers [48] noted that the mixture of regulated and non-regulated behavior that is
present in both U.S. and European markets is naturally captured in complementarity
models. In the EU, market power is held by gas producers rather than marketers, as
captured in equilibrium models reviewed and extended by Lise and Hobbs [49].
The market price of natural gas directly affects the commitment, dispatch and cost
of power supplies. Interruptions or outages in gas network could greatly reduce the
available supply of electricity and jeopardize the power system security. The power
system, on the other hand, also has a significant impact on the natural gas market
because it uses 29% of total gas consumption. When the demands for electricity and
natural gas peak at the same time because of, for example, severe weather
Costs and Constraints of Transporting and Storing Primary Energy 179
condition, the price could rise sharply in the gas market, which results in higher
generation costs.
Shahidehpour et al. [50] summarized the structure of the U.S. natural gas
transportation system and analyzed its interdependency with the electricity system.
The authors introduced constraints of gas supply capacities and emission limits into
the security-constrained unit commitment (SCUC) model. They studied the fast
response of combined-cycle units compared to thermal units, the impact of gas
price on gas-fired generation and the impact of gas infrastructure outages on the
locational marginal price (LMP) of electricity. Li et al. [51] further investigated the
inter-relationship of the two systems by incorporating linear natural gas network
constraints into the optimization of SCUC for a vertically integrated utility’s daily
operation planning. Liu et al. [52] improved the model in [51] by using nonlinear
equations to represent constraints of the natural gas transmission system. The
complex nonlinear mixed-integer problem was solved by decomposition
techniques.
Takriti et al. [53] presented a dynamic program for unit commitment which also
accounts for fuel constraints and prices that vary with uncertain future power
demand and price. In [54], the authors focused on hedging between the natural
gas market and electric power market. A common strategy for a gas utility to take
this advantage is to purchase power generating assets so that it can turn some of its
supply into electricity to profit from soaring electricity prices in the summer when
gas prices are low. Scenarios of gas price and demand and electricity price are
produced as input for a stochastic program. The optimal solution advises a utility on
when to buy or sell natural gas and when to convert gas into electric power. Instead
of a cost-minimization-based framework, a risk-cost trade-off was considered by
Chen and Baldick [55]. An integrated simulation-optimization algorithm was
proposed to solve for optimal gas supply portfolio that maximized the utility
function for a power generation entity. Geidl and Andersson [56] introduced a
new concept of “energy hub”, which includes basic elements of connections,
converters and storage. An energy hub can represent real facilities such as power
plants and large buildings. It enables the integration and optimal dispatch of
multiple energy carriers and products.
natural gas distribution system. The chain is very capital intensive. Huge cryogenic
tanks, around 70 m in diameter and 45 m high, holding over 100,000 m3, are
required to store LNG, and specialized tanker ships are required to transport it [57].
These ships are powered by steam turbines that use boil-off gas in a gas boiler [59],
supplemented if necessary by fuel oil. Tanker capacity was 130,000 m3 in 2004 but
projected to increase to 200,000 m3 by 2020 [60]. In 2007, there were five LNG
terminals in the U.S. and 45 proposed ones, 18 of which had been approved by
FERC.
LNG is attractive to suppliers as a way to monetize natural gas for export. For
both thermodynamic efficiency and low cost, LNG facilities must be large scale.
Long contracts, on the order of decades, are required to justify the investments.
Transport and liquefaction account for approximately 85% of the cost to deliver
LNG to the vaporization terminal. The cost is less dependent on distance than is the
cost of pipeline gas [57]. Although natural gas-based generation is promoted for
environmental reasons, Jaramillo et al. [59] found that the life-cycle air emissions
of gas are not much lower than those of coal when LNG is included, due to
emissions caused by the liquefaction and transportation stages. With advanced
generation technology and carbon capture and storage, coal would have similar
emissions to gas in various forms.
Cayrade [60] studied pipeline and LNG gas infrastructure necessary to meet
Europe’s growing demand, motivated by GHG reduction. He found that LNG is
competitive with pipelines at long distances; however, clean coal could also be
competitive. Because of the large scales involved, infrastructure investment must
overcome uncertainty about future gas prices as well as other factors. However,
LNG at competitive prices was available in Europe in 2009 [61]. Zheng and
Pardalos [62] formulated a two-stage stochastic mixed integer program for simul-
taneously expanding a transmission pipeline network and locating LNG terminals.
Discrete expansion sizes and choices among candidate facility locations were
modeled with binary variables while uncertain demands and supply limits were
formulated as discrete probabilistic scenarios. For either a risk-neutral or a risk-
constrained objective, the presence of integer variables in the second stage required
the development of a specialized solution algorithm.
Despite the significance of economies of scale in its associated facilities, LNG
can be a viable alternative at smaller scales as well. Kuwuhara et al. [58] developed
a model to optimize the supply of LNG to remote Amazonian regions of Brazil as a
possibility to replace diesel generators by gas turbines. Transport would be by ship
on local rivers to towns located between 1,000 and 1,500 km from the liquefaction
terminal, with travel times of 72 h or more. The model was a linear program with a
single nonlinear constraint that also involved integer variables representing the
number of trips per month and the number of ships required. It found the optimal
capacities of the liquefaction and vaporization plants, including their storage tanks,
and the optimal capacities of the ships. The cost was most sensitive to variation in
cost of the liquefaction plant and least sensitive to changes in cost of the storage
tanks. The conclusion was that LNG could be economically competitive to diesel
for supplying electricity to small, stand-alone networks.
Costs and Constraints of Transporting and Storing Primary Energy 181
3.3 Biomass
Intergovernmental Panel on Climate Change, Sims et al. [66] examined the carbon
emission and mitigation costs of fossil fuel, nuclear power, and renewable energy
resources. They projected that renewable energy would continue to grow but, in the
absence of significant government intervention, still would provide less than 2% of
the electricity market share in the year 2020. Gasification improves thermal effi-
ciency and, because biomass fuels contain oxygen and little sulphur, they are easier
and cheaper to gasify than coal. Biomass fuels could provide rural areas in particu-
lar with renewable energy. The report concluded that in the US, biomass could
contribute to CO2 mitigation, especially where forest or agricultural residues are
available at very low or negative costs. Gan and Smith [67] used a computable
general equilibrium model to simulate coal prices under CO2 emission reduction
and tax schemes and compared the costs of electricity generation by woody biomass
with coal – specifically, between BIGCC, conventional pulverized coal, and
integrated coal gasification combined cycle units. For sources of biomass they
considered logging residues and poplar plantations. The generic BIGCC unit
studied was smaller than either of the coal plants because of the biomass transpor-
tation costs. Without any mandated reductions in CO2 emissions, the biomass
production costs would have to be reduced substantially to compete with coal,
but imposing a CO2 emission tax would lower that requirement. Woody biomass
energy is nearly CO2-neutral but improving the productivity of production,
harvesting and transport systems is necessary for adoption on a large or wide
scale. However, in some locations, electricity generation by biomass alone or
cofiring biomass is already competitive on a small scale.
A recent study by the EIA examined a combined 25% RPS with a 25%
renewable fuel standard for transportation by 2025 [68]. In addition to requiring
that at least 25% of electricity sales be produced from renewable sources by 2025,
the RPS would establish a market for renewable energy credits. The NEMS-based
results showed a dramatic shift from coal and natural gas to biomass and wind
power. Advanced biomass generation (i.e., BIGCC) would be required, but when
full-scale commercial plants will be available was still uncertain. The major
impediment is the front end handling and processing of the feedstocks, which limits
utilization to below 60% because of frequent jams. Also, the supply and cost of
biomass energy crops will be critical. Because renewable capacity would offset new
capacity of more efficient coal and natural gas generation, the RPS “actually
increases the amounts of coal and natural gas consumed per kilowatt hour of
electricity generated” [68, p.26]. The impact of the RPS was found to be very
sensitive to assumptions about prices for fossil fuels and availability of advanced
biomass technology.
Despite the challenges posed by economies of scale, logistics of fuel transporta-
tion, and uncertainties about fuel price and technology development, optimization
models for expanding biomass generation or integrating it with other forms of
electricity generation seem to be absent in the literature. We expect that, as concern
for reduction of carbon emission increases, such studies will be very valuable for
determining how to most efficiently increase the utilization of this renewable and
relatively clean energy source.
Costs and Constraints of Transporting and Storing Primary Energy 183
References
1. Energy Information Administration (2010) International energy outlook 2010. U.S. Depart-
ment of Energy, Washington, DC
2. Energy Information Administration (2010) Electric power annual 2008. Department of
Energy, Washington, DC
3. Casten S (2010) Fuel swap: natural gas as a near-term CO2 mitigation strategy. Public Utilities
Fortn 148(4):40–44
4. Federal Energy Regulatory Commission (2005) The Western energy crisis, the Enron bank-
ruptcy, and FERC’s response. http://www.ferc.gov/industries/electric/indus-act/wec/chron/
chronology.pdf. Accessed Sept 2009
184 S.M. Ryan and Y. Wang
5. Commonwealth of Massachusetts (2006) Electricity price, reliability, and markets report 2005.
6. Federal Energy Regulatory Commission (2008) 2007 state of the markets report
7. Clayton M (2006) Enough coal on hand to keep US cool? Christian Science Monitor, 25 May
2006
8. Bergerson JA, Lave LB (2005) Should we transport coal, gas or electricity: cost, efficiency and
environmental implications. Environ Sci Technol 39(16):5905–5910
9. Hogan WW (1975) Energy policy models for project independence. Comput Oper Res
2:251–271
10. Murphy FH, Conti JJ, Shaw SH, Sanders R (1988) Modeling and forecasting energy markets
with the intermediate future forecasting system. Oper Res 36(3):406–420
11. Energy Information Administration (2003) The national energy modeling system: an over-
view. U.S. Department of Energy, Washington, DC
12. Quelhas A, Gil E, McCalley JD, Ryan SM (2007) A multiperiod generalized network flow
model of the U.S. Integrated energy system: part I – model description. IEEE Trans Power Syst
22(2):829–836
13. Quelhas A, Gil E, McCalley JD (2007) A multiperiod generalized network flow model of the
U.S. Integrated energy system: part II – simulation results. IEEE Trans Power Syst 22(2):
837–844
14. Wang Y, Ryan SM (2010) Effects of uncertain fuel costs on optimal energy flows in the U.S.
Energy Syst 1:209–243
15. Liu Z, Nagurney A (2009) An integrated electric power supply chain and fuel market network
framework: theoretical modeling with empirical analysis for New England. Nav Res Log
56(7):600–624
16. Ryan SM, Downward A, Philpott AB, Zakeri G (2010) Welfare effects of expansions in
equilibrium models of an electricity market with fuel network. IEEE Trans Power Syst 25
(3):1337–1349
17. Ryan SM (2009) Market outcomes in a congested electricity system with fuel supply network.
In: IEEE Power Engineering Society General Meeting, Calgary
18. Ryan SM (2009) Demand price sensitivity and market power on a congested fuel and
electricity network. In: IEEE Power & Energy Society General Meeting, Minneapolis,
25–29 July 2010
19. Freme F (2010) U.S. coal supply and demand: 2009 review (trans: Administration EI). U.S.
Department of Energy, Washington, DC
20. U. S. Energy Information Administration Coal transportation information. http://www.eia.doe.
gov/cneaf/coal/ctrdb/ctrdb.html. Accessed Jul 2010
21. U.S. Energy Information Administration (2006) Coal production in the United States – an
historical overview. U.S. Department of Energy, Washington, DC
22. Joskow PL (1987) Contract duration and relationship-specific investments: empirical evidence
from coal markets. Am Econ Rev 77(1):168–185
23. Joskow PL (1988) Price adjustments in long-term contracts: the case of coal. J Law Econ
31:47–83
24. Joskow PL (1990) The performance of long-term contracts: further evidence from coal
markets. Rand J Econ 21(2):251–274
25. MacDonald JM (1994) Transactions costs and the governance of coal supply and transporta-
tion agreements. J Transp Res Forum 34(1):63–74
26. Dennis SM (1999) Using spatial equilibrium models to analyze transportation rates: an
application to steam coal in the United States. Transp Res Part E 35:145–154
27. Bienstock D, Shapiro JF (1988) Optimizing resource acquisition decisions by stochastic
programming. Manage Sci 34(2):215–229
28. Energy Information Administration (2010) Annual energy outlook 2010. Energy Information
Administration, Washington, DC
29. O’Neill RP, Williard M, Wilkins B, Pike R (1979) A mathematical programming model for
allocation of natural gas. Oper Res 27(5):857–873
Costs and Constraints of Transporting and Storing Primary Energy 185
30. De Wolf D, Smeers Y (2000) The gas transmission problem solved by an extension of the
simplex algorithm. Manage Sci 46(11):1454–1465
31. De Wolf D, Smeers Y (1996) Optimal dimensioning of pipe networks with application to gas
transmission networks. Oper Res 44(4):596–608
32. Rı́os-Mercado RZ, Wu S, Scott LR, Body EA (2002) A reduction technique for natural gas
transmission network optimization problems. Annals Oper Res 117:217–234
33. Martin A, M€oller M, Moritz S (2006) Mixed integer models for the stationary case of gas
network optimization. Math Program 105:563–582
34. Chebouba A, Yalaoui F, Smati A, Amodeo L, Younsi K, Tairi A (2009) Optimization of natural
gas pipeline transportation using ant colony optimization. Comput Oper Res 36:1916–1923
35. Andre J, Bonnans F, Cornibert L (2009) Optimization of capacity expansion planning for gas
transportation networks. Eur J Oper Res 197:1019–1027
36. Djebedjian B, Shahin I, El-Naggar M (2008) Gas distribution network optimization by genetic
algorithm. In: Ninth International Congress of Fluid Dynamics & Propulsion, Alexandria,
2008
37. Steinbach MC (2007) On PDE solution in transient optimization of gas networks. J Comput
Appl Math 203:345–361
38. Midthun KT, Bjorndal M, Tomasgard A (2009) Modeling optimal economic dispatch and
system effects in natural gas networks. Energy J 30(4):155–180
39. Guldmann J-M (1983) Supply, storage, and service reliability decisions by gas distribution
utilities: a chance-constrained approach. Manage Sci 29(8):884–906
40. Energy Information Administration (2009) Major legislative and regulatory actions
(1935–2008). http://www.eia.doe.gov/oil_gas/natural_gas/analysis_publications/ngmajorleg/
ngmajorleg.html. Accessed Sep 2010
41. Duann DJ (1991) Direct gas purchases by local distribution companies: supply reliability and
cost implications. J Energy Dev 15(1):61–91
42. Guldmann J-M, Wang F (1999) Optimizing the natural gas supply mix of local distribution
utilities. Eur J Oper Res 112:598–612
43. Avery W, Brown GG, Rosenkranz JA, Wood RK (1992) Optimization of purchase, storage and
transmission contracts for natural gas utilities. Oper Res 40(3):446–462
44. Bopp AE, Kannan VR, Palocsay SW, Stevens SP (1996) An optimization model for planning
natural gas purchases, transportation, storage and deliverability. Omega 24(5):511–522
45. Butler JC, Dyer JS (1999) Optimizing natural gas flows with linear programming and
scenarios. Decis Sci 30(2):563–580
46. Contesse L, Ferrer JC, Maturana S (2005) A mixed-integer programming model for gas
purchase and transportation. Annals Oper Res 139:39–63
47. Gabriel SA, Kiet S, Zhuang J (2005) A mixed complementarity-based equilibrium model of
natural gas markets. Oper Res 53(5):799–818
48. Gabriel S, Smeers Y (2006) Complementarity problems in restructured natural gas markets. In:
Seeger A (ed) Recent advances in optimization, Lecture notes in economics and mathematical
systems. Springer, Berlin, pp 343–373
49. Lise W, Hobbs BF (2009) A dynamic simulation of market power in the liberalised European
natural gas market. Energy J 46(Special Issue):119–135
50. Shahidehpour M, Fu Y, Wiedman T (2005) Impact of natural gas infrastructure on electric
power systems. Proc IEEE 93(5):1042–1056
51. Li T, Eremia M, Shahidehpour M (2008) Interdependency of natural gas network and power
system security. IEEE Trans Power Syst 23(4):1817–1824
52. Liu C, Shahidehpour M, Fu Y, Li Z (2009) Security-constrained unit commitment with natural
gas transmission constraints. IEEE Trans Power Syst 24(3):1523–1536
53. Takriti S, Krasenbrink B, Wu LS-Y (2000) Incorporating fuel constraints and electricity spot
prices into the stochastic unit commitment problem. Oper Res 48(2):268–280
54. Takriti S, Supatgiat C, Wu LS-Y (2002) Coordinating fuel inventory and electric power
generation under uncertainty. IEEE Trans Power Syst 17(1):13–18
186 S.M. Ryan and Y. Wang
55. Chen H, Baldick R (2007) Optimizing short-term natural gas supply portfolio for electric
utility companies. IEEE Trans Power Syst 22(1):232–239
56. Geidl M, Andersson G (2007) Optimal power flow of multiple energy carriers. IEEE Trans
Power Syst 22(1):145–155
57. Thomas S, Dawe RA (2003) Review of ways to transport natural gas energy from countries
which do not need the gas for domestic use. Energy Convers Manage 28:1461–1477
58. Kuwahara N, Bajay SV, Castro LN (2000) Liquefied natural gas supply optimisation. Energy
Convers Manage 41:153–161
59. Jaramillo P, Griffin WM, Matthews HS (2007) Comparative life-cycle air emissions of coal,
domestic natural gas, LNG, and SNG for electricity generation. Environ Sci Technol 41(17):
6290–6296
60. Cayrade P (2004) Investments in gas pipelines and liquefied natural gas infrastructure. What is
the impact on the security of supply. The Fondazione Eni Enrico Mattei Note di Lavoro Series.
Fondazione Eni Enrico Mattei
61. BP (2010) BP statistical review of world energy
62. Zheng QP, Pardalos PM (2010) Stochastic and risk management models and solution algo-
rithm for natural gas transmission network expansion and LNG terminal location planning. J
Optim Theory Appl 147:337–357
63. Haq Z (2002) Biomass for electricity generation. Energy Information Administration,
Washington, DC
64. Dornburg V, Faaij APC (2001) Efficiency and economy of wood-fired biomass energy systems
in relation to scale regarding heat and power generation using combustion and gasification
technologies. Biomass and Bioenergy 21:91–108
65. Caputo AC, Palumbo M, Pelagagge PM, Scacchia F (2005) Economics of biomass energy
utilization in combustion and gasification plants: effects of logistic variables. Biomass and
Bioenergy 28:35–51
66. Sims REH, Rogner H-H, Gregory K (2003) Carbon emission and mitigation cost comparisons
between fossil fuel, nuclear and renewable energy cost resources for electricity generation.
Energy Policy 31:1315–1326
67. Gan J, Smith CT (2006) A comparative analysis of woody biomass and coal for electricity
generation under various CO2 emission reductions and taxes. Biomass and Bioenergy
30:296–303
68. Energy Information Administration (2007) Energy and economic impacts of implementing
both a 25-percent renewable portfolio standard and a 25-percent renewable fuel standard by
2025. Energy Information Administration, Washington, DC
Integrated Optimization of Grid-Bound Energy
Supply Systems
A. Sorokin et al. (eds.), Handbook of Networks in Power Systems II, Energy Systems, 187
DOI 10.1007/978-3-642-23406-4_7, # Springer-Verlag Berlin Heidelberg 2012
188 S. Prousch et al.
Energy demand in municipal areas represents the highest share of the total energy
demand. Therefore, municipal areas have a key role in fulfilling political aims.
Additionally, the same company usually operates the different energy supply grids
in municipal areas. Thus, an integrated optimization of these grids offers a high
potential for cost reductions.
To optimize and evaluate municipal energy supply concepts, the scope of
observation must be chosen in a way that all technical, economical and environ-
mental consequences of planning decisions can be detected. Therefore, the techni-
cal system boundaries include all equipment needed to build a municipal energy
supply system. This article focuses on low voltage electricity grids, medium-
pressure natural gas grids and district heating grids.
For the separate planning of energy supply grids, electrical and thermal demand
of customers are exogenous data. When planning whole energy supply systems, the
energy demand of customers can be crucially influenced by the choice of their
heating systems. Therefore, in this article, different types of heating systems are
also within the technical system boundaries.
Economical effects resulting from planning decisions include investment costs,
operational costs as well as energy costs discussed in subsequent section of this
paper.
Integrated Optimization of Grid-Bound Energy Supply Systems 191
Fulfilling the supply task is the basis and aim of energy supply system planning.
Geographical positions of customers, their electrical and thermal energy demand as
well as the topology of the supply area characterize the supply tasks.
Customers of municipal areas can be subdivided into distributed generation and
loads, which deliver respectively demand electrical and/or thermal energy. While
the optimal positioning of local network stations and pressure regulating stations as
well as distributed generation units is a task of network planning, geographical
positions of loads are fixed. Differences between various customers are given by
their electrical and thermal characteristics. Especially, the individual load profiles
are of relevance.
Due to political incentives, penetration of distributed generation is increasing. It
generally concerns systems with co-generation or the use of renewable energy
sources like global radiation, wind or hydro energy. For municipal energy supply,
especially CHP-units are well suited, since they are usually run by fossil energy
carriers and therefore not depended on fluctuating availability of renewable energy
sources. Furthermore, it is possible to erect these units within municipal areas in
contrast to wind energy generation and hydro power plants.
Due to their different characteristics, loads in municipal areas can be divided
into the categories business and household. Load profiles of business customers are
usually measured und therefore well known for network planning. However, annual
energy demand and rated power are the only known information for household
customers. Furthermore, household customers have a highly stochastic load behav-
ior, which differs from customer to customer. Therefore, in [2] stochastic models
have been developed to determine the relevant load for dimensioning grids.
In the category natural gas supply, conventionally common boilers cover the
heat demand. Alternative concepts with gas-run heat pumps or solar units allow
using renewable energies. Solar heating systems can only contribute a fraction of
the total heat demand. Conventional boilers usually cover the significantly higher
part. Gas-run heat pumps are commonly operated in combination with geothermal
units and are capable of covering the total energy demand without the necessity of
additional boilers. Using renewable energy sources for heat supply instead of
natural gas leads to a reduced utilization of existing natural gas grids and thereby
to a reduced economical efficiency there off.
Covering heat demand and part of electricity demand via small-scale CHP-units
describes another alternative concept in the category natural gas supply. Generated
electricity is either fed into the grid or used locally.
The district heating supply concept consists of co-generators providing thermal
energy, which is distributed via district heating grids. Generated electricity is fed
into the medium voltage grid.
Besides an electrical grid, all electric energy supply concepts require no further
grids. In these concepts electric heating systems e.g. electrical heat pumps supply
the heat demand. Table 1 gives an overview of the considered concepts in this
article.
Energy supply grids serve to transport and distribute different energy carriers such
as electricity, natural gas and district heat. On one hand, the supply task is
influencing the design of energy supply grids, on the other hand, degrees of freedom
in network planning as network structures or choice of equipments as well as
technical constrains have major influence.
There are several analogies between electricity, natural gas and district heating
grids. Underlying physical laws are generally different, though transferring models
and methods is often possible. This section discusses analogies and differences of
electricity, natural gas and district heating networks.
The fundamental problem structure of network calculation is comparable for the
different grids. The load flow occurs from source to drain via conductors. The analogy
to the voltage u of an electricity network is the pressure p in natural gas and district
heating networks. There is also an analogy between current I and flow rate w.
A significant difference exists in the physical laws. In contrast to electricity, in
natural gas and heat grids active and reactive power is inexistent. Additionally, gas
is compressible. Therefore, above-mentioned analogies between voltage, current,
pressure and flow rate are only valid under the assumption of incompressible gas
(constant density). An additional state variable in district heating grids is the
temperature T.
For a given load situation and network structure (e.g. as depicted in Fig. 1), a
network calculation determines the unknown state variables (Table 3).
Voltages and pressures must fulfill the load flow equations introduced in Table 2.
With the knowledge of all mass flows in the investigated grid, calculating
temperatures of the network nodes is possible.
Technical grid constrains can be verified with the results of the network
calculation.
i m
. .
m .
Δ u, Δ S Δp Δ p, Δ Φ
n
194 S. Prousch et al.
The pressure within a district heating grid has to be above a minimum pressure
guaranteed to customers and has to be above vapor pressure of the heat carrier with
a safety margin. This is also valid for the static pressure, which has to be regulated
by the pressure maintenance. Furthermore, to prevent damaging of equipments, the
upper pressure limit must not be exceeded.
Depending on the type of pipeline, considering different inlet temperatures is
mandatory. Besides limitations due to temperature resistance of equipments, in
view of economical advantages (heat dissipation, heat production costs) an inlet
temperature of 90 C should not be exceeded. Besides inlet temperature, return
temperature is an important factor for the efficiency of district heating grids.
Transmission capacity, energy demand for pumping, heat dissipation and efficiency
of co-generators depend on this parameter.
Furthermore, a minimum flow rate of about 0.8 m/s is necessary to guarantee a
minimum recirculation preventing a cooling of the heat carrier. Observing a
maximum flow rate of about 2 m/s prevents objectionable noise emissions.
Degrees of freedom of the optimization method are number and position of local
network stations, pressure regulation stations and co-generators as well as the
choice of routes and equipments. In addition, the heating system of customers is a
degree of freedom.
Practical purposes (e.g. stock keeping, acquisition) as well as specifications for
equipments are limiting the number of available types of equipments in network
planning. Furthermore, heating systems and distributed generation units are offered
in standardized sizes, thus investigations in this article are limited to typical systems
for different types of customers (e.g. household in one family house or apartment
house, business customers).
The aim of the optimization method introduced in this article is finding an
economically optimal, technically valid and energy efficient development strategy
for municipal energy supply systems. This requires considering parameters of the
supply task and their future development as well as all degrees of freedom and
boundary conditions when planning energy supply systems.
Existing methods to optimize energy supply systems either only allow planning
electricity, natural gas or district heating grids separately or are only able to solve
small-scale problems due to high computing times. A fundamental advantage of the
developed method is an appropriate modeling of the supply task as well as of
technical units while allowing an integrated planning of realistic energy supply
systems. Figure 2 illustrates the requirements on the optimization method including
all relevant constrains.
Computation time of the optimization method should not increase exponentially
with the number of degrees of freedom. Assumptions and simplifications to reduce
computation time are only valid as long as the quality of the solution is not affected.
The target function of the optimization is minimizing the capital and operational
expenditures (Table 4).
with
X
NLNS
C_ CAPEX ¼ a_ LNS; i þ bLNS; i cLNS; i SLNS; i
i¼1
X
NPRS
þ a_ LNS; i þ bLNS;i cPRS;i V_ PRS; i
i¼1
X
NCoGen:
þ a_ CoGen:; i þ bCoGen:; i cCoGen:; i Pth; CoGen:; i
i¼1
X
N cabel
þ a_ cabel; i þ bcabel; i li ccabel; i (9)
i¼1
NX
PL heat
þ a_ PL heat; i þ bPL heat; i li cPL heat; i
i¼1
NX
PL NG
þ a_ PL NG; i þ bPL NG; i li cPL NG; i
i¼1
X
NHS
þ a_ HS; i þ bHS; i cHS; i Pth;HS; i
i¼1
198 S. Prousch et al.
and
X
NCustomer
C_ OPEX ¼ T_ m; el:; Customer; i Pel:; Customer; i kel:; Customer; i
i¼1
þT_ m; th:; Customer; i Pth:; Customer; i k_th:; Customer; i (10)
X
NCoGen:;i
T_ m; el:; CoGen:; i; i Pel:; CoGen:; i; i kel:; feedin; i
i¼1
The newly developed system model bases on the system model to design electric
grids introduced in [1]. This model has been successfully applied in long term
planning of electric grids and shows that it can easily be adapted for designing other
energy supply grids [3]. This allows a consistent modeling of all relevant grids in
energy supply systems and makes an integrated planning of grid-bound energy
supply systems possible that uses synergies and considers cost-shifting between the
different sectors.
A hierarchical model with different levels of abstraction is developed to repre-
sent a real energy supply system. The highest level in the developed model consists
of streets of houses and junction points. A street of houses consists of customers that
are connected via routes. For all customers within a street of houses a uniform
demand of electricity and heat is assumed. Junction points contain network nodes of
a single or several energy supply grids. They represent the connection between
streets of houses and connection points for transformers, pressure regulators and co-
generators.
As mentioned above all energy supply systems have an electricity supply via the
electrical distribution grid. Therefore, in this model every customer is connected to
an electrical network node. To supply the heat demand of customers several
different heating systems can be allocated. Each of them requires a connection to
a corresponding network node. The model assumes that all customers within the
same street of houses are equipped with a uniform heating system.
Routes are differentiated in cable and pipeline routes as well as overhead
line routes. Cable and pipeline routes consist of a ditch with lines of single or
several energy supply grids. Overhead line routes contain electric lines exclusively.
Figure 3 illustrates the developed system model.
Input data of the optimization is a complete system model containing all degrees
of freedom and constrains. However, the result of the optimization describes a
specific condition of the model representing an energy supply system. The algo-
rithm delivers information describing if specific components (lines, transformers,
pressure regulators, heating systems etc.) are realized and which type should be
Integrated Optimization of Grid-Bound Energy Supply Systems 199
street of houses
customer
PR LNS
heating system
customer
JP JP heat
electricity heating system
heating system
heating system
route
customer
customer
electricity line
JP JP
natural gas line
heating system
LNS co-gen. district heat line
customer
used. Thereby, every component can have a variety of different conditions of which
one is optimal in terms of the planning task. Attached to every condition are the
technical data as well as the capital and operational expenditures of the
corresponding system.
The optimization method defines conditions for every component in a way that
the resulting energy supply system complies with all constrains and has minimal
total cost simultaneously. The presented task corresponds with a combinatorial
optimization since all optimization variables can only have integer values and the
solution space contains a limited number of possible combinations.
4 Optimization Method
The system model described in the prior section allows transferring the planning of
municipal energy supply systems into a combinatory optimization problem. Solu-
tion of such problems require high computing times, which is why even solving
small-scale problems within a feasible computing time is already challenging.
Generally, methods to solve combinatory optimization problems are divided in
exact and heuristic methods. The category of exact methods includes methods,
which determine solutions that can be proven to be mathematically optimal. The
category of heuristics contains methods, which have a methodological search
adapted to the problem structure.
The fundamental advantage of an exact optimization method is the guaranteed
optimality when a solution is found. Although the branch and bound method as well
as various derived variants of it offer run time advantages compared with
200 S. Prousch et al.
alternative exact methods, only small-scale problems can be solved. The system
model results in such a complex combinatory optimization problem, that the use of
exact methods is infeasible. Therefore, a heuristic optimization method is chosen to
allow solving large-scale problems.
Since in long-term planning of energy supply grids it turned out to be advanta-
geous to have the possibility to choose out of a range of similar cost efficient
network structures, population based approaches proved to be particularly suitable.
Furthermore, they offer the possibility to realize parallel processing for a additional
reduction of computing time.
For integrated planning of energy supply grids, genetic algorithms seem to be
advantageous since their fundamental structure is mainly problem independent,
which allows their application in various different problems. This allows an
integrated planning of different energy supply.
The nomenclature of genetic algorithms is very similar to that of evolution
theory and genetics. Table 5 provides an overview of the nomenclature.
Genetic algorithms solve an optimization problem by iteratively evaluating all
individuals of a population and selecting the ones with the highest fitness as parents
for the next generation. By combining attributes of parent individuals with a high
fitness, the fitness of individuals is advancing from generation to generation.
Therefore, the gen string of a new individual is a composition of the gen strings
of parent individuals. [14] gives a detailed description of genetic algorithms.
The fundamental concept of genetic algorithms does not allow an implicit
consideration of constrains in the problem structure. However, as aforementioned,
when designing energy supply systems a variety of constrains and planning
guidelines have to be considered. Consequently, methodologies are needed to
deal with invalid solutions. An obvious approach is excluding individuals that
violate boundary conditions from the solution space with the result that all consid-
ered individuals comply with the given boundary conditions. This is not feasible for
problems with a large number of constrains like the problem at hand. A more
suitable approach to consider boundary conditions in genetic algorithms is the
application of penalty functions. Invalid solutions are panelized in the fitness
determination, e.g. with a penalty function proportional to extend and type of
violation. This reduces the fitness of invalid individuals and consequently their
probability of selection.
For optimization problems with a large number of invalid solutions with regard
to the solution space, application of repair functions to convert invalid to valid
Integrated Optimization of Grid-Bound Energy Supply Systems 201
Technical evaluation
Repair
Local
Solution valid?
search no
yes
Genetic
Local search done? operators
no
yes
Economical evaluation
Abort?
no
yes
Economical-ecological evaluation
Abort?
no
yes
Solution output
In the next step, costs for each energy supply system are determined and
attributed to the corresponding systems as quality rating. All systems are arranged
according to their quality in order to compose a new genetic pool for the stochastic
generation of new system drafts. After completion of a generation the best
individuals are transferred unchanged into the new population, new individuals
replace the others. Therefore, genetic operators – e.g. selection, crossover, mutation –
are applied.
In case no improvement of the solution is achieved for several generations, the
most cost efficient energy supply systems, which comply with all technical
constrains are handed over to the overlaying genetic algorithm.
In the overlaying genetic algorithm, the economical and environmental evalua-
tion determines the overall fitness of every energy supply system. After completion
of a generation, an abortion criterion is checked. In case it is not fulfilled, again
genetic operators are applied otherwise the most efficient energy supply systems are
exported.
area. In the identified load centers, the minimum necessary number of stations is
allocated.
The selection of optimal energy supply concepts mainly depends on the supply
task, especially on the load density. Thus, to determine initial solutions for the first
optimization loop these parameters are determined. Based on empirical values or
estimations regarding the economy of different energy supply concepts depending
on the mentioned parameters, potentially optimal concepts for each street of houses
can be determined.
The optimal network structure for low voltage and district heating grids is a
radial network structure. A minimum spanning tree connecting all customers with
the shortest path represents the optimal design of radial networks. Therefore, it is
obvious to design low voltage and district heating grids based on the theoretically
achievable length of a minimum spanning tree. However, a minimum spanning
tree can conflict with technical constrains. Thus, an adequate number of initial
solutions, with structures similar to the minimum spanning tree, is necessary. To
determine a minimum spanning tree different algorithms are available since it is a
common problem of graph theory. In the developed method, the prim algorithm
is applied. To determine an adequate large number of initial solutions spanning
trees with restriction of single routes are generated. Thus, according to a
given probability in every step of the prim algorithm the shortest possible path
is restricted. Thereby, numerous degenerated minimum spanning trees can be
generated.
To guarantee that the selection of initial solutions does not limit the solution
space improperly, further initial solutions are stochastically generated.
In this section the functionality and performance of the newly developed method is
demonstrated by an exemplary application to a close to reality planning problem.
Therefore, for a modeled supply task, reflecting all relevant features of a real supply
task, the optimal network structures, dimensions of equipments and heating systems
are determined.
After introducing the modeled supply task, the conventional energy supply
system is designed at optimal costs. It serves as a reference to evaluate the potential
206 S. Prousch et al.
The considered supply area (see Fig. 5) exhibits a predominant urban character. The
total area is approximately 0.6 km2 in size. It includes 5,118 inhabitants, whose
electrical and thermal energy demand is distributed on 224 nodes.
A very heterogeneous load density characterizes the supply task. While a very
dense building structure predominate in the center of the supply area, single-family
houses are dominating in the peripheral area resulting in a comparable low load
density.
Assuming a conventional energy supply the total electrical load amounts
4.4 MW and the total thermal load 15.2 MW. The electrical energy demand
amounts 4.4 GWh/a and the thermal 12.4 GWh/a. The different customer types
are described in Table 9.
Due to the geographically limited expansion of the supply area, only a single
network level for each grid is considered. Only low voltage part of the electricity
grid is considered. The design of natural gas supply is limited to medium pressure
level.
Customer type 1
Customer type 2
Customer type 3
Junction point
Optional routes
100 m
Usable routes are limited to the pathway of roads. On every route, realizing lines
of every energy supply grid is allowed. Optional positions of stations for the
connection to overlaid network levels or feed-in of district heating are the 40
junction points depicted in Fig. 5.
In accordance with the praxis of network planning, the number of possible types
of equipments is limited. Table 10 shows technical data of the optional equipments.
Furthermore, technical constrains listed in Table 11 are considered.
For natural gas, h-gas with a calorific value of 11.3 kWh/mN3 and a medium
temperature of 5 C in the relevant peak load situation is assumed.
The district heating grid is operated with a temperature of 90 C and a pressure of
eight bars.
Local network
station
Pressure regulating
station
Customer type 1
Customer type 2
Customer type 3
Electricity line
Natural gas line
In this section the conventional energy supply system, called reference system in
the following, is determined. Thus, the choice of the heating system is no degree of
freedom but fixed to a condensing boiler. Consequently, in an integrated optimiza-
tion the low voltage and medium pressure natural gas grids are designed. The
determined energy supply system is depicted in Fig. 6.
While the low voltage grid shows a pure radial network structure, the natural gas
grid is characterized by a weekly meshed network structure. Due to the integrated
planning the path of lines of the electricity and natural gas grid are coordinated
optimally. To achieve minimal costs almost exclusively the same routes are used.
The minimum number of stations to fulfill the supply task is realized. Further-
more, stations are exclusively located in load centers. Thereby, the energy supply
system is designed to the limits of technical constrains, guaranteeing a cost minimal
solution without any overcapacities.
This section presents the energy supply system at optimal costs determined with the
newly developed method. In this calculation, heating systems of customers are not
fixed but result of optimization. Figure 7 shows the results of the optimization.
The determined energy supply system shows broad differences when compared
to the reference system. While the structure of electricity supply remains almost
unchanged, different energy supply concepts to cover the heat demand are chosen
depending on the load density. In regions of high load densities, grid-bound heat
supply turns out to be most cost-efficient.
Two district heating grids are installed in regions of particularly high load
densities. In these regions, district heating supply benefits of the high local energy
demand. Thus, in spite of high capital expenditures for pipelines and co-generators,
Integrated Optimization of Grid-Bound Energy Supply Systems 209
Local network
station
Pressure regulating
station
Co-generator
Customer type 1
Customer type 2
Customer type 3
Electricity line
Natural gas line
District heat line
1.5 6
electricity
106 /a natural gas 103 T/a
grids
0 0
Conv. ESS Opt. ESS Conv. ESS Opt. ESS
compared to other concepts, the total costs can be decreased due to low operational
expenditures.
At the border of the supply area, electrical heat supply turns out to be advanta-
geous. Due to low load density, an additional grid besides the electricity grid is not
economical.
Figure 11 gives an overview of the annual total costs divided into capital and
operational expenditures. They add up to 1.4 million €/a. Compared with the
conventional energy supply a cost reduction of 11% can be achieved. Besides the
economical results of the optimization, the results of the environmental evaluation
are depicted in Fig. 8. The annual carbon dioxide emissions can be reduced by 23%
compared with the reference system.
It must be pointed out that in the reference system, the conventional energy
supply is already designed at optimal costs. Therefore, the achieved cost reduction
only results of an optimal choice of energy supply concepts. Existing, historically
developed energy supply systems usually show particularly more inefficient
210 S. Prousch et al.
5 Conclusions
Due to current developments in the energy sector, new challenges for the planning
of grid-bound energy supply systems arise.
In the past, energy supply systems have been planned based on subjective
experiences of planning engineers without the specification of long-term objectives
and therefore often exhibit oversized dimensioning and inefficient supply concepts.
This leads high costs- and energy-inefficiencies.
Computer-aided optimization procedures provide important assistance in the
planning process. While the development and application of such optimization
procedures for the separate planning of single energy supply grids is widely adopted
and state-of-the-art, the integrated planning of grid-bound energy supply systems is
currently subject to research.
The presented approach enables the long-term planning of energy supply
systems including electricity, natural gas and district heating grids as well as
heating systems of customers. Moreover, it allows consideration of all relevant
technical constrains and planning guidelines.
Exemplary applications could demonstrate the method’s capability and the
advantages through applying this method for long-term planning of energy supply
systems. The comparison of a conventional with an optimized energy supply
systems shows possible gains of cost- and energy-efficiency.
References
1. Maurer C, Haubrich H-J (2006) Planning of high voltage networks under special consideration
of uncertainties of load and generation. In: Proceedings of CIGRE, RWTH Aachen University,
Germany
2. Egger H (2008) Strukturmerkmale f€ ur die vergleichende Bewertung von Niederspan-
nungsnetzen. Klinkenberg Verlag, Aachen
Integrated Optimization of Grid-Bound Energy Supply Systems 211
3. H€ubner M, Haubrich H-J (2008) Long-term planning of natural gas networks. In: IEEE 5th
international conference on the European electricity market (EEM), Lissabon
4. Sandou G, Olaru S (2009) Particle swarm optimization based NMPC: An application to district
heating networks, vol 384, Nonlinear model predictive control. Springer, Berlin/Heidelberg,
pp 551–559
5. Bakken B et al (2002) Energy distribution systems with multiple energy carriers. In: Sympo-
sium gas and electricity networks, Brazil
6. Geidl M, Andersson G (2007) Integrated modeling and optimization of multi-carrier energy
systems. ETH, Z€ urich
7. E Handschin et al (2007) Gekoppelte optimale Auslegung von Strom-, Gas- und W€armenetzen 7.
VDI-Fachtagung “Optimierung in der Energiewirtschaft”
8. Gil EM et al (2003) Modeling integrated energy transportation networks for analysis of
economic efficiency and network interdependencies. In: North American power symposium,
Rolla
9. de Mello OD, Ohishi T (2004) Natural gas transmission for thermoelectric generation problem.
In: X Symposium of specialists in electric operational and expansion planning, Rio de Janeiro,
Brazil
10. S€oderman J, Pettersson F (2006) Structural and operational optimisation of distributed energy
systems. Appl Therm Eng 26(13):1400–1408
11. Hecq S et al (2001) The integrated planning of the natural gas and electricity systems under
market conditions. In: IEEE PowerTech proceedings, Porto
12. Shahidehpour M, Fu Y, Wiedman T (2005) Impact of natural gas infrastructure on electric
power systems. Proc IEEE 93(5):1042–1056
13. An S, Li Q, Gedra TW (2003) Natural gas and electricity optimal power flow. In: IEEE PES
transmission and distribution conference, Dallas
14. Michalewicz Z (1994) Genetic algorithms + data structures ¼ evolution programs, 2nd edn.
Springer, Berlin/Heidelberg/New York
Index
Page references in Roman denote Vol. I and Italic page references denote Vol. II.
A B
ACF. See Auto correlation function (ACF) Balancing, 286, 287, 296
Active power reserve, 556, 558–561, Bellman equation, 309
563–572, 576 Benefit function, 67, 83
Advanced metering infrastructure (AMI), Bernoulli equation, 83, 84, 88, 89, 94–99, 111
477, 478, 491, 493 Best response functions, 25–26, 28
Agents negotiation, 209 Bidding, 41–57
Aggregation, 450–454 behavior, 263–267, 276
All-electric supply, 190 strategies, 61–86
Allowance auction, 71, 74, 82, 83, 86 Bilateral, 5–7, 11–13
Allowance penalty, 86 contracts, 177–179, 181–183, 186,
American option, 326, 335, 339 241–261, 560, 563, 564
Ancillary service, 555–576 obligation, 242
Andes community, 345–365 Bi-level programming, 49, 56
ANPDI. See Average nodal price index Biomass, 168, 179–181
(ANPDI) Bottom-up models, 290, 292–293
ANSI, 495, 497 Box-Cox transformation, 159
Ant colony optimization, 396, 402–403 Branch and bound algorithm, 411–413, 426
Approximate dynamic programming, 435–464
ARIMA. See Autoregressive integrated
moving average (ARIMA) C
ARIMA model, 152–158, 161, 162, 168 Capability curve, 559, 561, 562, 572, 575, 576
ARMA. See Auto-regressive moving average Capacity constraint, 265–268, 270, 272, 273,
(ARMA) 275–277
ARMAX model, 157 Capacity expansion, 63, 77, 172
Artificial intelligence, 175, 176, 209 Capacity investment, 264, 274–277
Assessment metrics, 6, 17–19 Capacity-proportional differentiation scheme,
Auctions, 41–57 233, 237
Auto correlation function (ACF), 91, 93, 106, Capacity tariff, 226– 233, 236, 237
162, 168, 169 Cap-and-trade, 63, 70, 71, 74–76
Autoregressive integrated moving average Capital investment, 266, 274, 277
(ARIMA), 104–106, 108, 113–114, CDF. See Customer damage function (CDF)
116–119 CHP-units, 189, 190
Auto-regressive moving average (ARMA), Closed-loop, 274
104, 105, 107 Coal, 167–172, 178–180
Average nodal price index (ANPDI), 371, 375, CO2 allowance(s), 63, 64, 82, 86
376, 378, 382, 384 Cointegration, 160, 170
A. Sorokin et al. (eds.), Handbook of Networks in Power Systems II, Energy Systems, 213
DOI 10.1007/978-3-642-23406-4, # Springer-Verlag Berlin Heidelberg 2012
214 Index
Economies of scale, 167, 172, 178–181 Expected unserved energy (EUE), 519, 520
EISA. See Energy Independence and Security Exponential smoothing model, 154–155
Act (EISA)
Elasticity of demand, 287, 293
F
Elasticity of substitution, 290
Failure risk, 436, 462
Electricity auctions, 263–266, 269–274
Fanning equation, 31
Electricity forward contracts, 62, 63, 65,
Feature selection, 100–102, 108–113
70, 77
Financial transmission rights (FTRs), 63, 64,
Electricity industry, 319
67–70, 72–74, 76–81, 83, 86, 396,
Electricity market, 3–36, 173–210, 241–261,
422–423
263–278, 413–423
Flexibility limits., 27
Electricity price, 89–120, 304–306, 311, 312,
Flow
316–318, 320, 324–331
limits, 94, 96
Electricity price forecasting, 165–169
mass, 83, 85, 89, 94, 97, 111
Electricity supply, 196, 206
velocity, 88, 96
Electricity supply chain, 169
Forecasting error, 168, 169
Electric power flow, 122, 123, 129–132, 145,
Frequency response, 557, 558, 563
160–161
FTR auction, 69, 70, 77, 86
Electric power system, 118, 119, 121, 126, 128,
FTRs. See Financial transmission rights (FTRs)
131, 135–138, 141, 145
Future evolution, 176, 177
Electric-system integration, 346
Electric transportation, 490, 497, 503
Energy, 174–177, 179, 181, 183–186, 189, G
199–203, 205, 207–209 GA. See Genetic algorithm (GA)
Energy carrier networks, 117–162 Game theoretic model, 63, 64, 66, 71–75, 77,
Energy carriers, 117–162 84–86
Energy efficiency, 492, 503 Game theory models, 5, 6, 16, 19, 22–25, 33
Energy exchange, 354, 355, 357, 358, 360 Gas
Energy flows in hydrological networks, market, 78–81, 105, 106, 109
134–135 optimization, 81, 108, 111
Energy hubs, 124, 139–140, 153 pipeline, 80, 82, 83, 105
Energy Independence and Security Act Gas flow
(EISA), 492, 493 problem, 15
Energy management, 490–492, 495 steady state, 32
Energy market, 346, 556, 564, 565, 570, 572 Gas Monthly data forecasts, 162–164
Energy planning, 505–521 Gasification, 179–180
Energy production, 326 Gas supply units (GSUs)
Energy supply grids, 188–191, 196, 198, 201, capacitated, 66, 67, 71–74
203, 208 gas network, 66, 68–70, 73–75
Energy systems planning, 118–119 location problem, 66–75
Energy tariff, 225, 228–232, 234, 236, 237 uncapacitated, 71–74
England and Wales, 270 GBM. See Geometric Brownian motion (GBM)
Environmental benefits, 357, 362–364 Generation
Equilibrium, 414 capacity, 349
models, 170, 176, 181 distributed/Generation, decentralized,
solution, 67, 81 470–472, 476, 483
Equity, 337, 339, 340 expansion problem, 506, 514, 518–520
EUE. See Expected unserved energy (EUE) fluctuating, 467, 477, 484
Evolutionary heuristics, 55, 56 planning problem, 508–509, 521
Expansion, 395–430 unit commitment, 543
Expansion planning, 395–430, 508–509, 521 Generator, 557–559, 561, 562, 564, 566–576
Expected energy not supplied, 506 Genetic algorithm (GA), 44–46, 49, 55–57,
Expected profit, 255, 261 198–203, 396–403, 408, 413, 426
216 Index
Geometric Brownian motion (GBM), 305, 306 Intermittent generation, 169, 286, 298
Greedy randomized adaptive search procedure Interoperability, 477–479, 492–503
(GRASP), 396, 403–405, 408 Investment, 323–342
Greenhouse gases, 362 cost, 305, 310, 312, 357, 359, 360
Grid-bound energy supply, 185–208 opportunity, 306, 309, 311, 317, 319
Grid cost, 213, 219, 220, 224, 237 threshold, 306–313, 318
Grid, smart grid, 467–472, 475, 477–482, 484 timing, 303–320
GSUs. See Gas supply units (GSUs) ISO. See Independent system operator (ISO)
Iterative search algorithm, 25–28, 30, 33
H
HAPP. See Hourly Alberta pool price (HAPP) J
Heat pump, 190 Java, 178, 189, 191
Heat supply, 185–190, 207
Hedging, 79, 105–107
Heteroskedasticity, 160, 162 L
Heuristic optimization, 43, 44 Latin America, 346
Heuristics, 197, 396–408 Liberalization, 39
High-voltage transformer, 435–464 Linearization
HOEP. See Hourly Ontario energy price (HOEP) piecewise, 6, 16–19, 25
Homoskedasticity, 160 successive, 19–21
Hourly Alberta pool price (HAPP), 95 Linear programming, 77–111
Hourly Ontario energy price (HOEP), 91–94, formulation, 4, 5, 19
108–119 Linear value function approximation,
Hydro, 274, 275, 277 441, 443
Hydro-production, 53, 56, 57 Linepack, 5, 10, 13–18, 20, 28, 30–34
Hydro-thermal production, 55 Liquefied natural gas, 177–178
LMP. See Locational marginal price (LMP)
Load flow, 191, 202
I Load forecasting, 152
IEC, 495, 497, 503 Load profile, 78–80
IEEE, 495, 497, 501, 503 Load shifting, 471, 477
IEEE 24-bus RTS, 371, 377–391 Local search, 200, 203
IEEE reliability test system, 507 Local transformer, 213, 224, 225, 229–231,
IETF, 497, 503 233, 237, 238
Incentive, 468, 472–475, 477, 478, 480, 481, 484 Locational marginal price (LMP), 420–422
Independent supply point, 213, 215, 219, 220, LOLC. See Loss of load cost (LOLC)
224, 237, 238 Long-run, 274, 277
Independent system operator (ISO), 42–44, Long-term contracts, 171, 175, 181
46–49, 56, 57, 488, 497, 503 Long-term planning, 186, 187, 198, 208
Inflexibility restrictions, 27 Loop flow effects, 108
Integrated electricity and natural gas economic Losses, 398, 399, 406, 420, 423, 426
dispatch, 139–141 Loss of load cost (LOLC), 508–510, 516
Integrated medium-term operational planning, Loss of load cost coefficient, 509–510
147–159 Loss of load probability, 506
Integrated operational planning of multiple Low voltage grids, 191–192, 206
energy carrier systems, 121 Low voltage line, 213
Integrated short-term operational planning, LP relaxation, 396, 397
159–161
Integration of natural gas and electricity
sectors, 117 M
Interactions between electric power and natural MAPE. See Mean absolute percentage error
gas systems, 120–121 (MAPE)
Interconnection, 345–365 Marginal cost, 65, 77–79
Index 217
Operations-and-failure costs, 360, 361 Power generation dispatch, 524, 526, 527, 530,
OPF. See Optimal power flow (OPF) 532, 534, 542, 548, 550
Opportunity cost, 307, 311, 560–562, 565, 568, Power interruption cost, 509
570, 572, 576 Power plant, 303–320
Optimal location Power system, 175–177, 209, 210, 555–576
analytical model, 67–69 economics, 525, 526, 529, 532, 533,
application, 64, 69–70, 75 538–544, 546, 548–550
concepts and notation, 66–67 modeling, 525–527, 531, 534, 536–538,
Optimal power flow (OPF), 65, 66, 69, 74, 543–545, 547, 548
81, 524–526, 529, 531, 534–536, network constraints, 508
540, 545, 549 operations, 524–526, 528, 529, 533,
of electricity and natural gas systems, 123 543–545, 547–549
Optimal stopping, 306 reliability, 505–521, 524–526, 528–534,
Optimization, 396, 402–403, 406, 408–429 539, 542, 549, 550
methods, 188, 194, 208 Power transmission
Option FTR, 67–69 control, 524–526, 529, 539, 549
Option value, 306, 309, 311–314, 316, economics, 525, 526, 529, 532,
317, 319 538–544, 549
Outages, 286, 288, 289 operations, 525, 526, 528, 533, 543,
Over the counter (OTC), 473 547–549
Own-price elasticity, 290 planning, 525, 529–534, 547
scheduling, 528–529, 543
switching, 523–551
P Pre-dispatch demand (PDD), 110–111
Parallel optimization, 57–58, 203 Pre-dispatch prices, 110–111
Parameter estimation, 327–334 Pre-processing, 100
Partial auto correlation function (PACF), Pressure
135, 160, 162, 168, 169 difference, 10, 14, 25
Payoff matrix, 72–74, 81, 85 stages, 38, 41, 42, 47, 50–55
Peak load, 293 Pressure/flow relationships, 13
Peak reduction, 489–490 Price-demand scatter, 144
Phasor measurement unit (PMU), 489, 493 Price(s)
Piecewise estimation, 334 consumer bids, 15
Piece-wise linearization, 80, 88, 90, 91, 102 elasticity, 267
Piecewise linear value function approximation, forecast, 241, 258
437, 450–454 heteroskedasticity, 131, 133, 135
Pipeline, 173–175, 178 marginal value, 19
friction losses, 89 nodal, 5, 29, 30
loop, 101 signals, 282, 285, 287
network, 80, 105 spikes, 91, 95–96, 99, 100, 102, 103, 109,
segment, 11, 14, 18, 19, 21, 22, 25 117, 119
segments, 82, 103 supplier offers, 15
system, 15, 16, 19, 20, 25 trading, 4
Pivotal supplier, 264, 266–269, 273, 276 volatility, 96–98
PJM Interconnection, 436 Price-taking, 46, 50, 52, 56
Planned outages, 110 Pricing
Planning principles, 58 nodal, 94, 109
Plant sizing, 307 spatio-temporal, 91, 107–109
p-Median problem, 437, 454–456 Priority list, 371, 372, 379, 382
PMU. See Phasor measurement unit (PMU) Privacy, 468, 477, 479–484
Pool, 4, 5, 10–12, 19–33, 177–179, 181–183, Procurement, 473
186, 210 Producer models without strategy, 43, 49–55
Index 219
Storage, 468, 471, 477, 478, 483, 490–492, Treasury bill auction, 276
497, 500 Trend, 130, 134
Strategic bidding, 3–36, 176 Trinomial lattice, 342
Strategies, 174, 176–178, 181–184, 186, 187,
192, 200, 202, 203, 205, 207, 208
Successive iteration, 20, 21, 26, 28, 29 U
Supply bid, 65, 70, 73, 76, 78–83 Uncertainty, 304, 307, 308, 311, 316, 317,
Supply function equilibrium, 263–268, 276 319, 320
Supply-side economic benefits, 357–359 regulatory, 129
Supply task, 38, 40, 43, 50, 53, 54, 56 volumetric, 128–129
System operator, 177, 179–181 Uniform-price auction, 263, 264, 269–273, 276
Systems marginal costs, 354–357 Unit-commitment, 42–46, 52, 55–57, 177
Unit root test, 160, 170
Unobserved Components model, 156–157, 163
T
Tabu search, 396, 405–408
Tariff component, 230–232, 235, 236
Taylor’s expansion, 6 V
Technology choice, 303–320 Value function, 436–438, 440–454, 464
Tendering process, 563, 564 Valves
Thermal, 270, 275, 277 check, 5, 25–26, 28
Time-of-use (TOU), 283, 291 regulator, 10, 25–26
Time-of-use pricing, 473 VARX model, 157, 158, 161, 164
Time series, 91–93, 96, 99, 100, 102, 104–106, Vehicle-to-grid, 471
113–116, 152–160, 162, 164, 170 Virtual power players, 174, 178, 183–184
Transfer function, 104–108, 113–119 Volatility, 129–132, 144, 146, 147, 305,
Transfer function model, 165–168, 170 311–314, 317, 318
Transformer failure, 435–437, 455 Voltage control, 557, 572, 576
Transformer replacement, 435, 439, 440, 464
Transmission, 62–64, 66, 67, 77, 79, 86,
167–169, 172–175, 177, 178, 181, W
395–430 Weekends, 141–143
congestion, 127–128 Weymouth equation, 90, 103
constraint, 277 Weymouth panhandle equation, 32
investments, 352, 361 Wind power generation, 323–342
owner, 436, 438, 454, 462 Wind power penetration, 141, 146, 147
Transportation logistics, 180 Winner’s curse, 270