You are on page 1of 10

This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON POWER SYSTEMS 1

Hourly Electricity Demand Response in the


Stochastic Day-Ahead Scheduling of Coordinated
Electricity and Natural Gas Networks
Xiaping Zhang, Mohammad Shahidehpour, Fellow, IEEE, Ahmed Alabdulwahab, and Abdullah Abusorrah

Abstract—This paper studies the role of hourly economic de- Index of natural gas supply contracts.
mand response in the optimization of the stochastic day-ahead Index of nodes in natural gas network.
scheduling of electric power systems with natural gas transmission
constraints. The proposed coordinated stochastic model (referred Index of buses in electric network.
to as EGTran) considers random outages of generating units Index of triangle.
and transmission lines, and random errors in forecasting the
day-ahead hourly loads. The Monte Carlo simulation is applied to Variables:
create multiple scenarios for representing the coordinated system
uncertainties. The nonlinear natural gas network constraints are Status indicator of generating unit at time .
linearized and incorporated into the stochastic model. Numerical Generation dispatch of a unit at time .
results demonstrate the benefits of applying the hourly economic
demand response for enhancing the scheduling coordination of Start up and shut down cost of a unit at time .
natural gas and electricity networks. It is demonstrated that Cost function of generating unit .
electricity demand response would offer a less volatile hourly load
profile and locational marginal prices, and less dependence on Cost of natural gas contract at time .
natural gas constraints for the optimal operation of electric power Natural gas consumption of contract at time
systems. The proposed model for EGTran could be applied by
grid operators for the hourly commitment and dispatch of power
.
system units. Gas delivery amount of supplier at time .
Index Terms—Hourly demand response, mixed-integer pro- Gas load of load at time .
gramming, natural gas transmission system, stochastic secu- Load demand in the th block of the stepwise
rity-constrained unit commitment. demand bid at bus at time , in MW.
Adjustable load at bus at time , in MW.
NOMENCLATURE State of curtailment at bus at time , 1 when
curtailed and 0 otherwise.
Index: ON time of load at bus at time , in hour.
Index of scenarios. OFF time of load at bus at time , in hour.
Index of hours. Binary indicator variable of the th triangle.
Index of generating units. Power flow through branch .
Index of natural gas suppliers. Bus voltage angle.
Index of natural gas loads. Natural gas flow from node to .
Index of electricity transmission branches. Pressure of node.
Index of buses. Constants:
Probability of scenario .
Manuscript received September 07, 2014; revised November 09, 2014; ac- Number of scenarios.
cepted December 30, 2014. This work was supported in part by the Grant No.
Gr/34/6 from the Deanship of Scientific Research (DSR) at King Abdulaziz Number of load buses.
University in Saudi Arabia . Paper no. TPWRS-01217-2014. Number of time periods.
X. Zhang is with the Galvin Center for Electricity Innovation at Illinois In-
stitute of Technology, Chicago, IL 60616 USA. Reactance between bus and .
M. Shahidehpour is with the Galvin Center for Electricity Innovation at Illi-
nois Institute of Technology, Chicago, IL 60616 USA, and also with the Renew- Power flow limits of branch .
able Energy Research Group, King Abdulaziz University, Jeddah, Saudi Arabia. Fuel function coefficient of gas-fired unit.
A. Alabdulwahab and A. Abusorrah are with the Electrical and Computer
Engineering Department and affiliated with the Renewable Energy Research Number of natural gas suppliers.
Group, King Abdulaziz University, Jeddah, Saudi Arabia.
Color versions of one or more of the figures in this paper are available online Number of natural gas loads.
at http://ieeexplore.ieee.org. Price of natural gas for contract .
Digital Object Identifier 10.1109/TPWRS.2015.2390632

0885-8950 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON POWER SYSTEMS

Contracted amount of take or pay gas readily available. This supply situation often happens during
contract . cold seasons when natural gas companies supply the natural gas
Set of gas-fired generating units. for heating purposes to residential customers with fixed natural
Set of nodes connected with . gas transmission services. Such conditions necessitate appro-
Set of node pair for pipeline from priate strategies for mitigating the increasingly intense natural
node to node . gas transmission constraints [4]. The hourly economic demand
Minimum and maximum of natural gas response would play a major role in such cases as grid operators
load. optimize the stochastic security- constrained hourly schedule
Minimum and maximum pressure. with natural gas constraints.
Demand-side participation could offer an effective option in
Minimum and maximum of gas injection.
critical circumstances when the natural gas supply to distributed
Number of blocks of energy demand. generating plants becomes scarce. Demand-side participation
Marginal benefit of the th block of the bid, includes on-site storage and demand response [5]. We specifi-
in $/MW. cally consider the application of hourly economic demand re-
Maximum load at bus at time . sponse in this study. The participating consumers would re-
Expected price-responsive load . duce their hourly demands voluntarily in response to electricity
Minimum and maximum curtailed load. market prices [6]. In restructured power systems, nodal prices
may vary with time and location, so customers could adjust
Pick-up or drop-off rate of load, MW/min. their load profiles corresponding to price volatilities. Customers
Minimum ON/OFF time of load, hour. could shift less critical loads to hours with more moderate prices
Max energy change at bus in the or curtail loads, though the second option is usually undesirable.
scheduling horizon. Hourly demand response could change the hourly load profile,
Load participation factor. reduce peak load periods, and provide an economic operation
Number of triangles . schedule [7]. The impact of price-based demand response on
market clearing and locational marginal prices of power systems
Constants in the th triangle. using network-constrained unit commitment model is quantita-
Node-gas supplier incidence matrix. tively analyzed in [8]. The economic demand response offers an
Node-gas load incidence matrix. important strategy for mitigating intense natural gas constraints
Bus-generator incidence matrix. faced by power system operators.
The natural gas transmission capacity may not always be
Bus-electrical load incidence matrix. fully available for serving gas-fired power plants because it
Bus-branch incidence matrix. would have certain obligations for serving residential and com-
mercial loads. When natural gas and electricity transmission
flows reach their peaks at the same time (e.g., Northeastern
I. INTRODUCTION U.S. cold winter peaks in 2013), the natural gas supply to
power plants might be curtailed and fuel price spikes could

N ATURAL gas is becoming rapidly the optimal choice occur. The fuel curtailment could lead to natural gas-fired
for fueling new generating units in electricity grid [1]. plant shutdown, higher power system operation costs, and
The additional fleet of natural gas combined-cycle plants has even jeopardize power system security at zones with a large
emerged over the past few years with the retirement of less number of natural gas-fired units. In such circumstances,
environmentally friendly coal generating plants. This transi- hourly economic demand response would offer several security
tion is driven by the combination of low natural gas prices, options to grid operators for managing outages, and incentives
integration of additional renewable units which depend on to customers to flatten their hourly load profiles for economic
the flexibility offered by natural gas units, and increasingly purposes. Accordingly, some of the installed natural gas-fired
inevitable tougher environmental standards in many parts units which utilize expensive fuel would not be committed or
of the world. In electricity markets, the independent system fully dispatched in the scheduling period, which would further
operator (ISO) executes security-constrained unit commitment lead to a more economical solution schedule.
(SCUC) for minimizing the operation cost on the premise of The interdependency issues of electricity and natural gas
satisfying network security constraints. For grid operators, the transmission systems have received significant attention in
interdependence of natural gas and electricity grid operations various papers. A coordinated model for assessing the im-
is a critical and growing concern, which is exacerbated by the pact of natural gas transmission systems on power system
addition of a large number of natural gas-fired generators [2]. security was proposed in [9]; the natural gas network was
The natural gas supply availability and fuel price volatility modeled by daily and hourly limits on pipelines, sub-areas,
would directly affect the hourly generation dispatch, cost of plants, and generating units. The coordinated scheduling of
supplying the hourly load, and electric power system reliability interdependent hydrothermal power system and natural gas
[3]. Natural gas-fired power plants usually depend on interrupt- transmission system was considered in [10] by applying Aug-
ible natural gas transmission services, which could be problem- mented Lagrangian relaxation method. A two-phase nonlinear
atic if a sufficient supply of natural gas to power plants is not
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

ZHANG et al.: HOURLY ELECTRICITY DEMAND RESPONSE 3

optimization model was proposed in [11] to model the coordi- We use a two-state continuous-time Markov chain model in
nated operation of natural gas and power systems for reliability EGTran to represent available and unavailable states of power
studies. The long-term integrated planning of interdependent system components. In a two-state model, the underlying dis-
natural gas and electric transmission systems was analyzed tribution functions for representing the duration of operation or
in [12]. The piecewise linear approximation of nonlinear repair state is assumed to be exponential. The repair and failure
flow-pressure relations were modeled in the unit commitment rates for the th component in period are represented by and
problem [13]. The impact of natural gas contracts and con- , respectively. The steady state availability of the th gener-
straints on a generating company's midterm risk-constrained ating unit is and its unavailability is .
hydrothermal scheduling was studied in [14]. An integrated The associated conditional probabilities for the th component
model for day-ahead unit commitment model with nonlinear are defined as
natural gas transmission constraints was proposed in [15]. A
two-stage midterm stochastic unit commitment model to opti-
mize coordinated water and natural gas supplies was proposed
in [16], which considered system components random outages,
load forecast errors, and water inflow uncertainty.
This paper considers the hourly demand response in the
short-term stochastic model (referred to as EGTran) as an
economic option for optimizing the day-ahead scheduling of
interdependent electricity and natural gas transmission systems.
The natural gas transmission constraints representing the fuel We generate a successive series of states by drawing samples
availability to natural gas-fired generation are introduced in of time to failure or time to repair with exponential
the hourly dispatch of the stochastic SCUC. The coordinated distribution functions for representing generation units and
system uncertainties include the hourly load forecast errors, transmission line outages. A power system state represents
which follow a truncated normal distribution, and forced a combination of individual states of components. We apply
outages of generating units and transmission lines, which are in each scenario for repre-
represented by independent Markov processes. The nonlinear senting component availabilities in which indicates
natural gas network constraints are converted into a set of linear the th component is available at time and indicates
constraints by using a piecewise linear approximation in the otherwise.
3-D Euclidean space and the resulting problem is formulated A low-discrepancy Monte Carlo simulation method Latin hy-
as a mixed integer linear programming (MILP) problem. The percube sampling (LHS) technique is employed to decrease the
proposed coordinated EGTran model for electricity and natural variance of simple Monte Carlo simulation [18]. LHS stratifies
gas transmission systems would maximize the total expected the input probability distributions and the effect is that each
social welfare of stochastic scenarios while satisfying the sample (the data of each simulation) is constrained to match
prevailing security constraints. the input distribution very closely. So we can use a relatively
The rest of this paper is organized as follows: The stochastic smaller number of samples to reach the same convergence.
simulation is presented in Section II. The coordinated model The day-ahead load forecast error can be represented by hy-
is formulated in Section III. Section IV presents and in detail perbolic distribution function, normal distribution function, or
discusses a 6-bus system and the IEEE 118-bus system. The truncated normal distribution function. The truncated normal
conclusion drawn from the study is provided in Section V. distribution is more practical since it would eliminate extrem-
ities of random loads for large forecast errors. In this paper, a
II. STOCHASTIC SIMULATION OF COORDINATED NATURAL truncated normal distribution for load forecast errors is repre-
GAS AND ELECTRIC POWER TRANSMISSION SYSTEMS sented by zero mean and a standard deviation that is 5% of the
hourly load forecast. The probability distribution function of the
In this paper, a set of possible scenarios are generated based truncated normal distribution is given as
on the Monte Carlo simulation method for the modeling of un-
certainties in EGTran. Load forecast errors and random out-
ages of generating units and transmission lines are considered
in all scenarios. Each scenario represents a possible state for the
availability of power system components and an hourly power
system load along with natural gas and power system constraints
for supplying the given hourly load quantities in each scenario. For the load uncertainty, the scheduling horizon is divided
A sequential Monte Carlo simulation is applied to model into several time intervals, and a few scenarios (say ) are gen-
random power system outages in the scheduling period. The erated in each interval (say ) based on the historical data. The
EGTran simulation is based on the sampling of the probability scenario tree will have scenarios each with a possibility of
distribution of the component state duration. We assume the .
probability distributions associated with mean time to failure The computation burden ascends rapidly with the number
(MTTF) and mean time to repair (MTTR) are available and the of scenarios in large-scale scheduling problems. Therefore, a
power system is at the normal state initially [17]. scenario reduction method is adopted as a tradeoff between
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON POWER SYSTEMS

the computation efficiency and the modeling accuracy [19].


The scenario reduction algorithms would determine a scenario
subset and assign new probabilities to the preserved scenarios
such that the corresponding reduced probability measure is
the closest to the original measure in terms of a certain prob-
ability distance. Simultaneous backward and fast forward
reduction algorithms are introduced in [20]. In our work, we
used SCENRED for the scenarios reduction, which is a tool
provided by General Algebraic Modeling System (GAMS) and Fig. 1. Stepwise demand response bidding curve.
contains several reduction algorithms. The GAMS outcome
would present a smaller number of scenarios with a reason-
able approximation of the original system. After reduction, B. System and Units Constraints
scenarios are retained and a weight is assigned to each 1) Power Balance Constraint:
scenario that reflects the possibility of its occurrence. The
sum of probabilities for all scenarios is equal to one, that is
(2)
.
The natural gas price for generators is fixed within a day if we
choose a spot market for natural gas. Therefore, the gas price 2) Generating Unit Constraints: Physical constraints on ca-
uncertainty is not considered in our day-ahead optimal sched- pacity limits, ramp rate limits, startup/ shutdown characteristics,
uling. However, in a midterm optimal scheduling or long-term min ON/OFF time, etc. [21].
planning, the natural gas price would play a very critical role 3) Power Transmission Constraints: Constraint (3) shows
because the gas price usually exhibits a high correlation with the power balance at each bus. Constraint (4) represents the
the weather and the season. power flow from bus to bus , which is limited by the trans-
mission line capacity (5). Constraint (6) sets the voltage angle
of the reference bus to zero:
III. FORMULATION OF EGTRAN
Hourly economic demand response is introduced into the pro- (3)
posed stochastic model of EGTran for managing the interde- (4)
pendency of power systems with natural gas transmission con-
straints. EGTran would determine the hourly unit commitment (5)
and economic demand response schedules for maximizing the (6)
expected social welfare while meeting the prevailing electricity
and natural gas constraints. The frame of the model is described
as follows: C. Demand Response Constraints
Max: Expected Social Welfare We consider price responsive loads for the demand response
s.t. program via price-sensitive load consumption curves. The price
1) Power system and generator units constraints responsive load participants can control the energy consumption
2) Hourly demand response constraints in response to the market prices by curtailing or shifting loads
3) Natural gas fuel and network constraints to other operating hours [22]. A price responsive load bid con-
sists of several blocks of stepwise hourly energy consumption
A. Objective Function quantity and corresponding prices as shown in Fig. 1.
Here, OA, OB, and OC represent the scheduled load by
The EGTran objective (1) is to maximize the expected so-
market operator, expected price responsive load ( ), and
cial welfare over the entire scheduling horizon. The first term of
maximum load ( ), respectively. The scheduled load by
the objective function represents the customer gross surplus, the
market operator is the sum of block demands at bus at time .
second term is the contract cost of natural gas supply to gas-fired
is the adjustable load, which is defined as the difference
units, and the third term is the generation cost of other thermal
between the expected price responsive load and the scheduled
units which includes fuel cost, start-up, and shut-down cost.
load.
is assigned to each scenario which reflects the possibility of oc-
The correlation between block demand and adjustable load at
currence. The joint objective function for scenarios represents
each period is given as
the scenario-based approximation of uncertainties applied to the
original power system:
(7)

The constraints on adjustable load are expressed as

(1)
(8)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

ZHANG et al.: HOURLY ELECTRICITY DEMAND RESPONSE 5

where indicates the status of demand response at bus at (15)


time in scenario , 1 means the load is shifted while 0 means
the scheduled load is equal to the expected load. is pos-
2) Natural Gas Network Constraints: There are certain
itive when the load is shifted out from bus at hour while
similarities between electricity and natural gas transmission
it is negative when load is shifted to bus at hour . Both ad-
systems. Both are designed to supply end users through their
justable load and its status are decision variables in
respective transmission system. The natural gas transmission
the model.
system is comprised of natural gas wells, storage facilities,
The total curtailment for each price responsive load at bus
transmission pipelines (high pressure) and distribution pipelines
in the scheduling horizon is limited to a predefined amount.
(low pressure), and natural gas customers. As one of the largest
The setting value of the right term would indicate whether the
and most complex nonlinear systems, the natural gas transmis-
load is curtailed or shifted in the scheduling horizon. By setting
sion system can be represented by its steady state and dynamic
to 0, the curtailed load at certain time periods will be fully
characteristics [23]. The steady state mathematical model
shifted to other periods:
comprised of a group of nonlinear equations is presented in our
stochastic day-ahead scheduling model. From the mathematical
(9) viewpoint, the steady-state natural gas problem will determine
the state variables including nodal pressures and flow rates in
The adjusted load between two successive periods is individual pipelines based on the known injection values of
restricted by load pick-up/drop-off rates (10). The load natural gas supply and load.
pick-up/drop-off rates represent the ramping capability for Supply and Load: : Natural gas suppliers are gas well and
restored/curtailed loads, which is correlated with customer load storage facilities which provide the natural gas through its trans-
characteristics: mission network. Supplies are modeled as positive gas injec-
tions at related nodes. The lower and upper limits of gas sup-
pliers at each period are given as
(10)
(16)
The minimum on/off times in (11) indicate that a specific load
is supplied or curtailed in the scheduling horizon. Minimum Natural gas consumers are classified into industrial, commer-
on time defines the number of consecutive hours that the load cial, and residential loads with different priorities. Natural gas-
would be supplied after it is restored. Minimum off time rep- fired power plants are industrial customers which use natural
resents the minimum number of consecutive hours that a load gas to generate electricity. The priority of residential load is
would be off after it is curtailed: higher than that of industrial loads in the natural gas scheduling
horizon. Natural gas loads are represented as negative injections
at related nodes with lower and upper limits:
(11)
(17)

D. Natural Gas System Constraints Flow Conservation: The steady-state natural gas injection
at each node is equal to the flow extracted from the node. The
1) Natural Gas Fuel Constraints for Gas-Fired Units:
flow conservation (18) ensures the nodal balance at the natural
Gas-fired unit is the largest industrial consumer of natural gas
gas transmission system:
system which links the natural gas system with the electric
power system. A power plant that owns the gas-fired units may
sign firm or flexible contracts with natural gas suppliers. For (18)
each gas-fired unit, we model the natural gas supply contract
as flexible contract in this model. Constraints (12)–(15) would
Pipeline: Natural gas is delivered to customers via
provide the coupling between the electric power and the natural
pipelines. The pipelines include passive pipeline (regular) and
gas system operations. The fuel cost of each natural gas-fired
active pipeline. The compressors in active pipelines would
unit depends on its natural gas consumption and the price
increase the pressure difference between the corresponding
of natural gas (12). Here, the natural gas consumption is de-
nodes for increasing the transmission capacity. Natural gas
termined by the hourly generation dispatch of natural gas-fired
flows in pipelines are dependent on factors such as the length
units (13). Each natural gas supply contract for supplying
and diameter of pipelines, operating temperatures, pressures,
natural gas-fired units is considered as a natural gas load in the
type of natural gas, altitude change over the transmission path,
natural gas network model (14). Equation (15) indicates that
and the roughness of pipelines.
the natural gas consumption by a gas-fired unit cannot exceed
In the proposed EGTran model, we consider the regular
the contracted amount in a natural gas day:
pipeline for simulating the natural gas pipeline system. The
(12) natural gas flow between gas nodes and is a quadratic
function of the pressure at the two end nodes:
(13)
(14) (19)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON POWER SYSTEMS

(20)
(21)

where is the pipeline constant that depends on the temper-


ature, length, diameter, friction, and natural gas compositions.
Similar to electric transmission lines with bus voltage limits,
the natural gas network would maintain the nodal pressure (21)
within the prescribed limits for customers.
In (19), the natural gas flow is formulated as a nonlinear func-
tion of natural gas flow and nodal pressure. A fast forward sub-
stitution method which results from Newton-Raphson method
could be applied which requires a large number of iterations for
an approximation solution that is sensitive to the initial natural
gas operating point [15]. If the initial point is not close to the
global optimal point, the final solution may be a local optima.
The nonlinear pipeline flow rate constraints would have a signif-
icant impact on the computation burden in the stochastic model.
In EGTran, we convert the natural gas flow (19)–(20) into a
set of linear constraints by using a piecewise linear approxima-
tion in a 3-D Euclidean space, which is briefly discussed below.
Given a node pair , the feasible region of node pressure
is expressed as

(22)

The piecewise linear approximation would divide the feasible


region (22) into convex polytopes (triangles). A plane in a 3-D
Euclidean space is then generated in each polytope to represent
the relationship between the natural gas flow and the nodal pres- Fig. 2. Stochastic SCUC with natural gas transmission constraints.
sure. The natural gas flow when is expressed as

A few scenarios are retained by scenario reduction and each sce-


nario represents a possible state for the electric power system.
(23) The stochastic SCUC solution would provide the gas consump-
where , , and are constants in the th triangle; and tion by natural gas-fired units. Accordingly, the natural gas net-
are node pressures in the th triangle; and is binary indicator work subproblem would check the natural gas transportation
variable. The following linear constraints are introduced based feasibility in each scenario. If any violation is encountered, nat-
on the above notations: ural gas network constraints are fed back to the master problem
for a revised SCUC solution. This process will be repeated until
(24) a feasible solution of the coordinated system is obtained for all
scenarios.
(25) IV. NUMERICAL SIMULATIONS OF THE EGTRAN MODEL
(26) We use two case studies consisting of a 6-bus power system
with 6-node natural gas system and the IEEE 118-bus system
(27) with 10-node natural gas system to test the proposed model. The
proposed coordinated model stated in (1)–(28) is solved using
the ILOG CPLEX 11.0 in Microsoft VisualC#.NET on an Intel
(28)
Xeon Server with 64 GB of RAM.

Equations (1)–(28) represent the stochastic EGTran problem for A. Six-Bus System
integrating the hourly demand response into the natural gas net- The 6-bus system depicted in Fig. 3 has three natural gas-
work model. The EGTran problem is converted into an MILP fired units, seven transmission lines, and three loads. The unit
unit commitment problem and natural gas network subproblem startup and shutdown costs are assumed negligible and equal
as shown in Fig. 2. Here, the input data including hourly load to zero in this case. We consider the same fuel price for the
forecast and the forced outages of generating units and trans- three gas-fired units. The 6-node natural gas system is given
mission lines are used to generate a large number of scenarios. in Fig. 4, which has 5 pipelines, 2 natural gas suppliers, and
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

ZHANG et al.: HOURLY ELECTRICITY DEMAND RESPONSE 7

TABLE I
HOURLY SCHEDULE OF CASE 1 OF 6-BUS SYSTEM

Fig. 3. Six-bus power system.

TABLE II
HOURLY SCHEDULE OF CASE 2 OF 6-BUS SYSTEM

Fig. 4. Six-node natural gas system.

5 natural gas loads. Natural gas Loads 1, 5, 3 are determined


by the hourly generation dispatch of natural gas-fired units 1,
2, and 3. Natural gas loads 2 and 4 represent other kinds of gas
loads. The test data for electricity and natural gas systems are
provided in motor.ece.iit.edu/data/Gastranssmion6_1.xlsx.
Three cases are presented to illustrate the hourly economic
demand response applied to the stochastic day-ahead scheduling
of electric power systems with natural gas transmission con-
straints.
Case 1: Deterministic base case without considering the Fig. 5. Hourly dispatch of unit 1 in Cases 1–2.
hourly demand response.
Case 2: Hourly demand response is introduced into the
model. shifted to other periods within the same scheduling horizon. The
Case 3: Stochastic conditions are considered in Case 2. hourly commitment schedule is shown in Table II. Since the de-
These cases are discussed as follows. mand response has shifted the peak load to other periods, the
Case 1: The hourly demand response is not considered in expensive generating unit 3 is not committed in the scheduling
this case. Natural gas transmission constraints are based on the horizon. Also the total generation dispatch of unit 2 is decreased
linear model discussed in Section III. The hourly commitment from 405.22 MWh in Case 1 to 244.93 MWh in Case 2 within
schedule is shown in Table I in which the hour 0 represents the schedule horizon.
the initial state. When we ignore natural gas constraints, the The hourly generation dispatch comparison of the cheapest
cheapest unit 1 is committed during the entire scheduling unit 1 in Cases 1–2 is shown in Fig. 5. Here, unit 1 has a more
horizon, while the expensive units 2 and 3 are only committed volatile schedule when natural gas constraints are considered.
between hours 11 and 20. However, the expensive unit 2 is At hour 10, the pressure at nodes 4 and 5 is at its minimum when
committed between hours 1 and 23 when the natural gas trans- natural gas transmission constraints are considered in Case 1,
mission constraints are enforced. The total generation dispatch which resulted in a lower dispatch. In comparison, the unit 1
of cheapest unit 1 is decreased from 4529.8 MWh to 4388.9 dispatch is less volatile in Case 2 when the hourly demand re-
MWh. The daily operation cost is increased to $85 687 as com- sponse is introduced. The daily production cost is reduced to
pared $81 317 without the natural gas transmission constraints. $80 070 in Case 2, which is lower than that in Case 1. In this
Case 2: In this case, hourly demand response is introduced case, the peak load periods are shifted to other periods, which
at all load buses to analyze its impact on the hourly genera- also reduce locational marginal price (LMP) and transmission
tion unit scheduling with natural gas transmission constraints. congestion.
The load participation factor, which is defined as the ratio of Fig. 6 shows that the hourly load profile is more flat when de-
available price responsive load to the expected price responsive mand response is incorporated. A flat load profile corresponds
load, is fixed at 0.2. The total load curtailments in the sched- to lower LMPs. For comparison, the average LMP profile of
uling horizon is set to zero, which means the load can only be the six buses is shown in Fig. 7 in which the LMPs are higher
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON POWER SYSTEMS

TABLE III
COMPARISON OF RESULTS FOR DIFFERENT SCENARIO

Fig. 6. Hourly load profile without and with DR.

random load in the entire scheduling horizon and operation cost


for different scenarios with corresponding probabilities.
The expected operation cost in this case is $79 778, which is
lower than that of the deterministic solution (e.g., $80 070). As
shown in Table III, the scenario cost is also calculated based on
the dispatch of generating unit in various scenarios. The produc-
tion cost varies in scenarios because each scenario represents a
combination of random loads over the entire scheduling period,
which is different from the forecasted value.
Fig. 7. Hourly average LMP in Cases 1–2.
The daily production costs in Cases 1–3 are compared in
Fig. 9. In Case 1, the total cost is increased from $81 317 to
$85 687 since the inclusion of natural gas constraints has re-
sulted in a modified hourly dispatch. When the hourly demand
response is considered in Case 2, the total cost is reduced to
$80 070. This is because the hourly demand response has shifted
the peak load to other periods when the supply of natural gas is
less limited. The results show that the hourly demand response
can provide more economic unit dispatch and reduce the pro-
duction cost by shaving the consumption of natural gas at peak
hours. The computation time for 11 scenarios of 6-bus system
Fig. 8. Hourly gas consumption without and with DR. is 2078 s while the deterministic solution takes only 229 s.
A higher modeling accuracy can be obtained by including
more scenarios along with a considerably higher computation
between hours 12 and 21 in Case 1. However, the demand re- burden to solve the EGTran problem. Meanwhile, the lineariza-
sponse in Case 2 has lowered the LMPs as peak loads are shifted tion of pipeline flow rate constraints also brings in ancillary vari-
to off-peak hours. ables to the problem. So the selection of an appropriate number
The total natural gas consumption profile of the three units of scenarios would represent a tradeoff between the computa-
in case 1 and case 2 is shown in Fig. 8. It presents a similar tion efficiency and the modeling accuracy.
profile as electricity load. A peak load appears at hour 17 in
case 2. When the hourly demand response is incorporated, the B. Modified IEEE 118-Bus System
natural gas consumption profile could be shaved more flat. The The modified IEEE-118 bus system has 54 thermal genera-
natural gas-fired power plants are usually supplied as industrial tors including 8 natural gas-fired units, 186 branches, and 91
customers with lower priority in natural gas schedule. So a flat load buses. The total capacity of natural gas-fired units is 725
natural gas consumption profile of gas-fired units would avoid MW, which is 10% of total generation capacity. The peak load
to peak simultaneously with residential load and decrease the of 6000 MW occurs at hour 21. Hourly demand programs at all
chance of inadequate supplies of natural gas. load buses, and random outages of generating unit and transmis-
Case 3: In this case, uncertainties are introduced to the sion lines, load forecast errors are considered. The natural gas
EGTran model. The load forecasting error follows a truncated transmission system is composed of 10 nodes and 10 pipelines
normal distribution with a mean value equal to the load forecast as shown in Fig. 10. L1-L8 are gas loads consumed by gas-fired
and standard deviation of 5% of the mean value. Power system units 1-8, respectively. L9-L12 are fixed residential gas loads.
component outages are not considered in this small example. The test data for electricity and natural gas systems are given in
The hourly demand response is included in the stochastic motor.ece.iit.edu/data/Gastranssmion_118_10.xls.
model. We created 6561 scenarios initially, which are reduced Table IV presents the deterministic solution and stochastic
to 11 by the scenario reduction technique for computation solution of three stochastic scenarios with its probabilities. In
efficiency. Table III presents the results including the sum of this 10-node natural gas system, the cheapest natural gas unit
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

ZHANG et al.: HOURLY ELECTRICITY DEMAND RESPONSE 9

cost as well as the scenario cost. The scenario cost difference


between three scenarios is mainly caused by the random load
consideration in each scenario.
The computation time for the deterministic problem of the
118-bus system is about 350 s. However, the computation time
increases exponentially in the stochastic model because the sto-
chastic model would introduce additional variables associated
with the number of scenarios. Meanwhile, the linearization of
natural gas pipeline flow would also require a large number of
Fig. 9. Daily production cost comparison for Case 1–3.
auxiliary variables to solve the gas distribution problem. Hence,
we reduce the number of stochastic scenarios from 6561 to 3
which would be computed in 10 330 s. In practical cases, we
would offer more scenarios and reduce the computation time
further by applying heuristic rules for representing specific
cases and considering parallel processing that would speed up
the solution process further. We may also resort to enforcing
the natural gas network constraints simply on selected gas-fired
units which are located in gas supply-limited regions.
In addition, several methods for considering the uncertainty
in the unit commitment problem were proposed in recent years
which included interval and robust optimization methods [24],
[25]. One attractive aspect of these methods is that they can
Fig. 10. Ten-node natural gas system. solve the unit commitment problem with less added variables
and more moderate information on potential uncertainties. For
TABLE IV a large-scale power system problem considering its interdepen-
COMPARISON OF RESULTS FOR DIFFERENT CASES dency with a natural gas system, these recent methods may offer
better alternatives for addressing the computation challenges.

V. OBSERVATIONS
Based on the numerical results, we list our observations:
1) Natural gas transmission constraints could impact the unit
commitment and cause the daily operation cost to increase.
2) Hourly demand response offers a flat load profile by
shifting peak loads to other periods. It could shave the
natural gas consumption, lower LMPs, and reduce the
system operation cost.
3) The stochasticity factor and the nonlinear model of natural
G6 cannot always operate at full capacity due to the natural gas gas flow have increased the computation burden. However,
node pressure constraints. The pressure at natural gas node 8 selecting an appropriate number of scenarios and a sim-
has reached its minimum 310Psig during most of the periods, plified gas flow model could speed up the solution as a
which would further limit the natural gas supply to unit G6. tradeoff between the computation efficiency and the mod-
Meanwhile, the other cheap unit G4 is located at the same lo- eling accuracy.
cation (node 5) where a high priority residential load L11 is lo-
cated. Furthermore, node 5 is distant from the natural gas sup- VI. CONCLUSIONS
plier node. The natural gas consumption of G4 is tightly lim- This paper proposed a short-term stochastic model of EGTran
ited by the natural gas constraints. However, the hourly de- which incorporates the hourly demand response for managing
mand response program introduces a flat natural gas consump- the interdependency of electricity and natural gas systems. The
tion profile and lowers the LMPs by shifting the peak loads. paper applies the MIP approach for solving the coordinated
So the operation cost does not change much as compared to electricity and natural gas model and considers the random out-
$1 699 510 without considering natural gas constraints. Without ages of generating units and transmission lines, as well as load
considering the random outages and load forecast errors, the de- forecast errors. The natural gas-fired units would link the two
terministic daily production cost is $1 701 014. coordinated systems. The natural gas network is modeled as a
In the stochastic model when component outages and load un- set of hourly linear constraints on fuel contracts, natural gas
certainties are both considered, the expected operation cost for suppliers, load pickup/drop-offs, and pipeline flows. The nat-
the three retained scenarios is $1 703 304. The random outage ural gas transmission constraints could impact the hourly unit
occurs in scenario 2 for unit 25 at hours 9 and 10. However, commitment and cause the daily operation cost to increase. The
the expensive unit 25 is not committed during the scheduling case studies demonstrate that the hourly demand response is an
horizon, so this outage would not affect the expected operation effective strategy to lower the hourly LMPs and further reduce
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON POWER SYSTEMS

the operation cost by offering a flat load profile. The coordinated [16] L. Wu and M. Shahidehpour, “Optimal coordination of stochastic
stochastic model provides an appropriate way to consider the hydro and natural gas supplies in midterm operation of power sys-
tems,” IET Gener., Transm., Distrib., vol. 5, pp. 577–587, 2011.
impact of the system uncertainties. The computation burden in [17] NERC Generator Availability Data System (GADS) [Online]. Avail-
the stochastic case is increased significantly with the inclusion able: http://www.nerc.com/pa
of natural gas system constraints. In practical cases, we would [18] L. Wu, M. Shahidehpour, and T. Li, “Cost of reliability analysis based
on stochastic unit commitment,” IEEE Trans. Power Syst., vol. 22, no.
offer more scenarios by simplifying gas constraints on specific 2, pp. 800–811, May 2007.
units and periods based on gas supply positions, introducing [19] L. Wu, M. Shahidehpour, and T. Li, “Stochastic security-constrained
heuristic rules for representing specific cases, and applying par- unit commitment,” IEEE Trans. Power Syst., vol. 22, no. 2, pp.
800–811, May 2007.
allel processing that would speed up the solution process. [20] N. Gröwe-Kuska, H. Heitsch, and W. Römisch, “Scenario reduction
and scenario tree construction for power management problems,” in
ACKNOWLEDGMENT Proc. IEEE Power Tech Conf., Bologna, Italy, Jun. 2003.
[21] M. Shahidehpour, H. Yamin, and Z. Li, Market Operations in Electric
The authors acknowledge and appreciate the technical and fi- Power Systems. New York, NY, USA: Wiley, 2002.
nancial support for the grant number Gr/34/6 provided by DSR. [22] H. Mohsenian-Rad and A. Leon-Garcia, “Optimal residential load con-
trol with price prediction in real-time electricity pricing environments,”
IEEE Trans. Smart Grid, vol. 1, no. 2, pp. 120–133, Sep. 2010.
REFERENCES [23] S. An, “Natural gas and electricity optimal power flow,” Ph.D. disser-
[1] EIA, Annual Energy Review, 2011 [Online]. Available: http://www. tation, Oklahoma State Univ., Dept. Elect. Eng., Stillwater, OK, USA,
eia.doe.gov/emeu/aer/ 2004.
[2] R. D. Tabors, S. Englander, and R. Stoddard, “Who's on first? The co- [24] C. Zhao and Y. Guan, “Unified stochastic and robust unit commit-
ordination of gas and power scheduling,” Electricity J., vol. 25, no. 5, ment,” IEEE Trans. Power Syst., vol. 28, no. 3, pp. 3353–3361, Aug.
pp. 8–15, Jun. 2012. 2013.
[3] M. Shahidehpour, Y. Fu, and T. Wiedman, “Impact of natural gas in- [25] L. Wu, M. Shahidehpour, and Z. Li, “Comparison of scenario-based
frastructure on electric power systems,” Proc. EEE, vol. 93, no. 5, pp. and interval optimization approaches to stochastic SCUC,” IEEE
1042–1056, May 2005. Trans. Power Syst., vol. 27, no. 2, pp. 913–921, May 2012.
[4] ISO New England, Natural Gas and Power Generation in New England
[Online]. Available: http://www.iso-ne.com/ Xiaping Zhang received the B.S. and M.S. degrees in electrical engineering
[5] R. H. Patrick and F. A. Wolak, “Real-time pricing and demand side from Shanghai Jiaotong University, China, in 2006 and 2009, respectively. Cur-
participation in restructured electricity markets,” in Proc. Conf. Retail rently she is pursuing the Ph.D. degree at the Illinois Institute of Technology,
Participation in Competitive Power Markets, Stanford Univ., 2001. Chicago, IL, USA.
[6] A. C. Tellidou and A. G. Bakirtzis, “Demand response in electricity
markets,” in Proc. 15th Int. Conf. Intell. Syst. Appl. Power Syst., 2009,
pp. 1–6.
[7] H. Wu, M. Shahidehpour, and A. Al-Abdulwahab, “Hourly demand re- Mohammad Shahidehpour (F'01) is the Bodine Chair Professor and Director
sponse in day-ahead scheduling for managing the variability of renew- of Robert W. Galvin Center for Electricity Innovation at the Illinois Institute
able energy,” IET Gener., Transm., Distrib., vol. 7, no. 3, pp. 226–234, of Technology, Chicago, IL, USA. He is also a Research Professor with the
2013. Renewable Energy Research Group, King Abdulaziz University, Jeddah, Saudi
[8] L. Wu, “Impact of price-based demand response on market clearing Arabia.
and locational marginal prices,” IET Gener., Transm., Distrib., vol. 7, Dr. Shahidehpour is the recipient of the Honorary Doctorate for the Poly-
no. 10, pp. 1087–1095, 2012. technic University of Bucharest in Romania.
[9] T. Li, M. Eremia, and M. Shahidehpour, “Interdependency of natural
gas network and power system security,” IEEE Trans. Power Syst., vol.
23, no. 4, pp. 1817–1824, Nov. 2008.
[10] C. Liu, M. Shahidehpour, and J. Wang, “Application of augmented La-
grangian relaxation to coordinated scheduling of interdependent hy- Ahmed Alabdulwahab received the Ph.D. degree in electrical engineering
drothermal power and natural gas system,” IET Gener., Transm., Dis- from the University of Saskatchewan, Saskatoon, SK, Canada.
trib., vol. 4, pp. 1314–1325, Aug. 2010. He is an Associate Professor in the Department of Electrical Engineering
[11] J. Munoz, N. Jimenez-Redondo, J. Perez-Ruiz, and J. Barquin, “Natural and Computer Engineering and affiliated with the Renewable Energy Research
gas network modeling for power systems reliability studies,” in Proc. Group, King Abdulaziz University, Jeddah, Saudi Arabia. He has been a consul-
IEEE/PES General Meeting, Jun. 2003, vol. 4, pp. 23–26. tant to the Electricity and Co-Generation Regulatory Authority in Saudi Arabia
[12] S. Hecq, Y. Bouffioulx, P. Doulliez, and P. Saintes, “The integrated and a visiting scientist at Kinectrics and the University of Manchester. His field
planning of the natural gas and electricity systems under market condi- of interest includes power system planning and reliability evaluations.
tions,” in Proc. IEEE Porto Power Tech. Conf., Porto, Portugal, 2001.
[13] M. Urbina and Z. Li, “A combined model for analyzing the interdepen-
dency of electrical and gas system,” in Proc. 39th North Amer. Power
Symp. (NAPS '07), 2007. Abdullah Abusorrah received the Ph.D. degree in electrical engineering from
[14] C. Sahin, Z. Li, and M. Shahidehpour, “Impact of natural gas on risk- the University of Nottingham, U.K.
constrained midterm hydrothermal scheduling,” IEEE Trans. Power He is an Associate Professor in the Department of Electrical Engineering and
Syst., vol. 26, no. 2, pp. 520–528, May 2011. Computer Engineering at King Abdulaziz University, Jeddah, Saudi Arabia. He
[15] C. Liu, M. Shahidehpour, Y. Fu, and Z. Li, “Security-Constrained unit is the coordinator of the Renewable Energy Research Group and member of the
commitment with natural gas transmission constraints,” IEEE Trans. energy center foundation committee at King Abdulaziz University. His field of
Power Syst., vol. 24, no. 3, pp. 1523–1535, Aug. 2009. interest includes power quality and system analyses.

You might also like