544 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 30, NO.

1, JANUARY 2015

Look Ahead Robust Scheduling of Wind-Thermal

System With Considering Natural Gas Congestion
Cong Liu, Member, IEEE, Changhyeok Lee, and Mohammad Shahidehpour, Fellow, IEEE

Abstract—Natural gas pipeline congestion will impact on the fuel

adequacy of several natural gas fired generating units at the same
time. This letter focuses on the development of a robust optimiza-
tion methodology for the scheduling of quick start units when con-
sidering natural gas resource availability constraints. Natural gas
transmission will be approximated by linear constraints, and the Fig. 1. Constant traveling time of natural gas flow.
linepack capacity of pipelines will be considered in the proposed
model. Case studies show the effectiveness of the proposed model
and algorithms.
U.S. that has a large number of gas-fired quick-start generating units.
Index Terms—Natural gas electric coordination, renewable en-
We integrate the linearly approximated natural gas flow constraints into
ergy, robust optimization, unit commitment.
the proposed robust optimization tool. This model can lead to more
secured commitments of quick-start gas-fired units under load variation
I. INTRODUCTION and potential renewable ramping events.

N ATURAL gas fired generating units play an important role in

an electric power system. The quick start capabilities and fast
ramping attribute of gas fired generating units like hydro units are cru-
II. MODELING OF NATURAL GAS TRANSMISSION SYSTEM
The transient natural gas flow can be represented as partial differen-
cial to compensate renewable generation forecast errors, contingen- tial equations (PDEs) for time and position which are dependent natural
cies, and load variations. Natural gas fuel availability will affect power gas density, mass flow, flow velocity, and pressure [2]. Although the
system operation and security. If natural gas pipeline congestion oc- PDEs can be transformed into nonlinear algebraic difference equations
curs, the gas operator is likely to limit the amount of the natural gas [3], [4], the optimization with those constraints is still nonconvex. It has
delivered to gas-fired generating plants because most of them hold in- been found to be reasonable to use a linear DC power-flow model in
terruptible transportation contracts [1], [2]. In addition, renewable en- power markets, and this allowed us to clear the markets on given time-
ergy uncertainty will result in uncertain natural gas usage of gas-fired frame or interpret the results. Similarly, reasonable assumptions can be
generating units [3]. Therefore, in operating day closed to the real time, made to establish a linear approximation for the natural gas flow.
it is necessary to include natural gas constraints and renewable uncer- Developing a linear model depends on appropriate tradeoffs and
tainty into the look-ahead scheduling problem. approximation. For example, one simplifying assumption would be
Robust optimization models the uncertainty using a deterministic set isothermic conditions. In addition, because gas pipeline flow is turbu-
(e.g., set of possible scenarios or range of possible values for the un- lent, we may assume that natural gas horizontal axial velocity is con-
certain parameters) without any probabilistic description. It provides stant [6], [7]. With these two assumptions, the natural gas flow model
a robust solution that is immune to any possible scenario of the un- can be approximated using a linear model.
certainty set, which is an important aspect in the security constrained A natural gas pipeline is shown in Fig. 1. Equation (1) represents
scheduling of electric power systems. The robust optimization often mass balance at gas node . is linepack capability representing
solves the so-called mini-max bilevel problem, which finds the solu- natural gas amount stored in pipeline at time period . repre-
tion minimizing worst-case cost or infeasibility that is maximized over sents the gas inflow of gas node through pipeline . If the practical
the uncertainty set. gas mass flow of a pipeline is from to , is positive, other-
In [2]–[4], different models are proposed to deal with the combined wise, it is negative. can represents natural gas load of power
optimization of electricity and natural gas scheduling problem. In this generation, industry, company and residence. The sum of inflow minus
letter, the proposed model is for the look-ahead robust scheduling of the usage of natural gas equals to the total change of linepack capacity.
gas-fired units in a utility or independent system operator (ISO) in the Equation (2) denotes gas mass flow in given traveling speeds through a
pipeline as shown in Fig. 1. Formulation (3) represents the upper bound
and the lower bound of the line packing of a pipeline:
Manuscript received September 27, 2013; revised March 24, 2014; accepted
May 05, 2014. Date of publication June 06, 2014; date of current version De-
cember 18, 2014. The submitted manuscript has been created by UChicago
Argonne, LLC, Operator of Argonne National Laboratory (“Argonne”). Ar-
gonne, a U.S. Department of Energy Office of Science laboratory, is operated (1)
under Contract No. DE-AC02-06CH11357. This work was supported by the
U.S. Department of Energy, Office of Electricity Delivery and Energy. Paper (2)
no. PESL-00135-2013. (3)
C. Liu is with Argonne National Laboratory, Decision and Information Sci-
ence, Argonne, IL 60439 USA.
C. Lee is with Northwestern University, Evanston, IL 60208 USA.
M. Shahidehpour is with the Illinois Institute of Technology, Chicago, IL Natural gas used for power generation is represented as loads in (1). We
60616 USA. can integrate the linear constraints (1)–(3) into the robust optimization
Digital Object Identifier 10.1109/TPWRS.2014.2326981 framework presented in the next section.

0885-8950 © 2014 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.
LIU et al.: LOOK AHEAD ROBUST SCHEDULING OF WIND-THERMAL SYSTEM WITH CONSIDERING NATURAL GAS CONGESTION 545

III. ROBUST SCHEDULING MODEL TABLE I

RESULTS FOR THE 118-BUS SYSTEM

(4)

With the presence of uncertainty, the unit commitment decision

needs to be made before the uncertainty revealed; however, the power
generation can be determined after the uncertain parameters are TABLE II
COMPUTATIONAL PERFORMANCE OF ROBUST MODEL
observed as a recourse. Under robust feasibility criteria, this leads
to the two-stage robust scheduling problem as shown in (4), where
and are the uncertainty
sets of load and wind, represents commitment decisions in the first
stage. corresponds to and , the forecasted load and wind.
The second stage problem is to check the feasibility of the constraint
including natural gas transmission constraints for
any given and . In the second part, we consider multi-period robust scheduling
represents the second stage variables. problems and study their computational performance. In look-ahead
The solution of the proposed model uses the column-and-constraint scheduling problem, the time is closed to real-time operation. The
generation algorithm [5]. Master problem and subproblem are solved time horizon in optimization problem is usually less than or equal to 4
iteratively. The worst cases are selected through solving subproblem periods. The wall clock time and the number of iteration for multi-pe-
and are added into master problem as new constraints. The subproblem riod look-ahead scheduling problems are exhibited in Table II. When
is max-min bilevel programming problem. After taking the dual of the number of time periods increases, the computational time also
inner problem, the subproblem becomes a bilinear programming increases rapidly. However, the computational time is acceptable in
problem. We transform bilinear program into mixed integer linear the framework of a look-ahead scheduling problem.
program after introducing binary variables.
V. CONCLUSION
IV. CASE STUDIES
A robust optimization methodology-based look-ahead scheduling of
We apply the model into the IEEE 118-bus system that include quick-start generating units with natural gas transmission constraints
3 areas, 54 generators, and 186 transmission lines. The penetration is proposed. The method successfully guarantees that there are enough
level of wind power generation is 20%. We assume the interval natural gas resources supplied to gas-fired generating units to balance
uncertainty set around the forecast values as follows: 20% of the variation and intermittence of renewable energy, especially in the
the wind power generation for the each one of three wind farms each case when there is natural gas pipeline congestion. Linear approxi-
time period, and 3% of the total load for each time period. A mation of natural gas flow model is proposed and used in the sim-
pipeline supply natural gas to 20 gas-fired generating units in area 2. ulation. By using the proposed robust method, the operator does not
We solve the problem to the optimality. need to sample scenarios during the simulation. Since the worst cases
In the first part, we consider one time period and compare the dif- will be selected and checked during the process, the unit commitment
ference between results of deterministic unit commitment and robust of quick-start units are immunized to any scenarios in uncertainty set.
unit commitment. The forecasted load for the next operation period is Case studies based on the IEEE 118-bus system are given and illus-
3733 MW. Line pack capacity is crucial for the real-time operation of trate the effectiveness and computational performance of the proposed
gas-fired generating units. Different initial line pack will be assumed method.
in the case studies.
Case 1) Deterministic scheduling with higher initial linepack REFERENCES
Case 2) Deterministic scheduling with lower initial linepack [1] ISO New England, CIGRE 2008 Case Study: Electric & Natural
Case 3) Robust scheduling with higher initial linepack Market interdependencies within New England, Sep. 2008.
Case 4) Robust scheduling with lower initial linepack [2] C. Liu, M. Shahidehpour, and J. Wang, “Coordinated scheduling of
With higher initial line pack in Case 1, 12 gas-fired generating units electricity and natural gas infrastructures with a transient model for
natural gas flow,” Chaos, vol. 21, no. 2, Jun. 2012.
and 8 non-gas-fired generating units are committed to produce elec- [3] M. Qadrdan, M. Chaudry, J. Wu, N. Jenkins, and J. Ekanayake, “Im-
tricity. If the initial line pack is low in Case 2, only 10 gas-fired gener- pact of a large penetration of wind generation on the GB gas network,”
ating units are committed due to less amount of natural gas contained Energy Policy, vol. 38, no. 10, pp. 5684–5695, 2010.
in the pipeline. However, the total number of committed units is 25 [4] S. An, Q. Li, and T. W. Gedra, “Natural gas and electricity optimal
power flow,” in Proc. IEEE/PES Transmission and Distribution Conf.
as shown in Table I. More gas fired and non-gas-fired generating units
Expo., 2003, vol. 1, pp. 7–12.
in Case 2 is committed compared to Case 1. In Case 3 and Case 4, [5] B. Zeng and L. Zhao, “Solving two-stage robust optimization problems
two-stage robust scheduling methodology is employed to determine using a column-and-constraint generation method,” Oper. Res. Lett.,
unit commitment. Robust methodology guarantees that any scenarios vol. 41, no. 5, pp. 457–461, 2013.
in the uncertainty set are feasible. Case 3 commits 13 gas generating [6] R. L. Street, G. Z. Watters, and J. K. Vennard, Elementary Fluid Me-
chanics, 7th ed. New York, NY, USA: Wiley, 1996.
units, which is one more than Case 1. Compared to Case 2, Case 4 [7] M. Y. Damavandi, I. Kiaei, M. K. Sheikh-El-Eslami, and H. Seifi,
commits two more non-gas generating units to cover the wind and load “New approach to gas network modeling in unit commitment,” En-
uncertainty as shown in Table I. ergy, vol. 36, no. 10, pp. 6243–6250, 2011.