1192 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 2, MAY 2010
Extended Benders Decomposition for Two-Stage SCUC
Cong Liu, Student Member, IEEE, Mohammad Shahidehpour, Fellow, IEEE, and Lei Wu, Member, IEEE
Abstract—This letter presents the solution of a two-stage secu-
rity-constrained unit commitment (SCUC) problem. The proposed
SCUC model could include integer variables at the second stage. A
framework of extended Benders decomposition with linear feasi-
bility and optimality cuts is proposed for the solution of mixed-in-
teger programming (MIP) problems at both stages. Test results
show the effectiveness of the proposed methodology.
Index Terms—L-shaped decomposition, MIP, SCUC.
NOMENCLATURE
Coefficients vector, matrix of MIP
problem.
Vector of ones and identity matrix.
Index of nodes in branch and bound
trees, iterations, and scenarios.
Dimension of first/second stage
problems.
Dual price function of MIP problem.
Dual price function of LP-relaxation of
MIP.
Number of scenarios and their
probabilities.
First stage, second stage, and slack
variables.
Spanning space of and variables.
Variables represent objective function
values.
Ancillary binary variable.
Dual variables.
Predefined tolerance and a large number.
I. PROPOSED MIP PROBLEM
T
HE security-constrained unit commitment (SCUC)
problem shown in Table I is solved in two stages
according to its specific structure. Generally, the first stage cor-
responds to an optimal decision, while the second stage would
examine the viability and optimality of first stage decisions.
Benders cut applications are based on a prerequisite that the
subproblem corresponding to the second stage is LP or convex
NLP [1]. In practice, variables in the second stage can be integer
or semi-continuous. Consider the following examples:
1) integer variables in power transmission control;
2) status of quick start units in post-contingency evaluation.
Manuscript received August 19, 2009. First published January 26, 2010; cur-
rent version published April 21, 2010. Paper no. PESL-00034-2009.
The authors are with the Electrical and Computer Engineering Department,
Illinois Institute of Technology, Chicago, IL 60616 USA (e-mail: cliu35@iit.
edu; ms@iit.edu; lwu10@iit.edu).
Digital Object Identifier 10.1109/TPWRS.2009.2038019
TABLE I
SCUC DECOMPOSITION STRATEGIES
We have developed an extended Benders decomposition
method which can be applied to such SCUC problems. The
SCUC problem is represented as a MIP formulation (1)–(4).
Decision variables at the first stage and at the second
stage in P could be either integer or continuous. The constraint
structure is L-shaped and no coupling among subproblems:
(1)
(2)
(3)
(4)
Nonlinear feasibility and optimality cuts were developed in
[3] via the general duality theory and showed that an L-shaped
decomposition can converge within a finite time. The formu-
lation of optimality and feasibility cuts are given in (12) and
(13). After introducing integer variables, the linear feasibility
and optimality cuts (21)–(23) would transform the two-stage
SCUC problem into a MIP set, which can be incorporated into
the master problem and lead to the global optimal solution.
As to the extended Benders decomposition, the mixed-integer
master problem (BD-MP) and subproblems (BD-SP) are given
in (5)–(8) and (9), respectively:
(5)
(6)
(7)
(8)
(9)
Here, is the current BD-MP solution. In addition, (7) and
(8) represent optimality and feasibility cuts, respectively, which
are formed by solving BD-SP. The BD-MP solution provides a
lower bound while a feasible BD-SP solution leads to an upper
0885-8950/$26.00 © 2010 IEEE