A Decomposition Based Approach for Solving a

General Bilevel Linear Programming

Xuan Liu, Member, IEEE, Zuyi Li, Senior Member, IEEE

Abstract—Bilevel optimization has been widely used in decision- 𝒔 slack variable vector
making process. However, there still lacks an efficient algorithm 𝑺 shift factor matrix of the power grid
to determine an optimal solution of a bilevel optimization 𝑼 bus-generator incidence matrix
problem, especially for a large-size problem. To bridge the gap,
this paper proposes an efficient decomposition algorithm for a 𝑽 bus-load incidence matrix
general bilevel linear programming(GBLP). The simulation 𝑾 bus-line incidence matrix
results on large-size testing system demonstrate its correctness 𝑿 branch reactance matrix
and efficiency. 𝝆, 𝝁, 𝝁′ Lagrangian vector
Index Terms—bilevel linear programming, benders 𝐱 variable vector in the lower level
decomposition, general bilevel linear programming, power 𝐲, 𝐲̂ variable vector in the upper level
systems. Note that ∆ represents incremental change and symbols in bold
represent vectors or matrices.

NOMENCLATURE
′ I. INTRODUCTION
𝑈, 𝑈 sets of extreme points
𝑉, 𝑉 ′
𝑈𝑟 , 𝑈𝑟′
sets of extreme rays
reduced sets of extreme points B ILEVEL optimization has attracted a lot of research
attentions in recent years[1]-[7]. In general, a bilevel
optimization problem can be casted into a decider-follower
𝑉𝑟 reduced sets of extreme rays
problem. In the upper level, a decision is determined. Then, the
follower in the lower makes a best response according to the
𝜀 given value for gap
determined decision. For example, the power grid
𝜏 given maximum percentage of load uncertainty
interdiction[7] is formulated as a bilevel max-min problem. In
the upper level, the attacker determines the optimal set of lines
𝑖, 𝑗 subscript: index for lagrangian multipliers to be attacked. In the lower level, the operator minimizes the
𝑘 subscript: index for lines load shedding given the set of attacked lines determined in the
upper level.
𝐾 objective value the lower level optimization problem
The most popular methods to solve a bilevel optimization
ℎ𝑘 physical flow of line 𝑘 problem are the Karush-Kuhn-Tucker (KKT) based approach
𝐿𝑚𝑎𝑥
𝑘 rating of line 𝑘 and duality based method. In the KKT based approach, the
𝑢 loading level of line 𝑘 lower level optimization problem is replaced by its KKT
𝑧𝑙𝑜𝑤𝑒𝑟 lower bound of the GBLP optimality conditions. To linearize the nonlinear
𝑧𝑢𝑝𝑝𝑒𝑟 upper bound of the GBLP complimentary constraints, additional binary variables will be
𝒄, 𝒅 coefficient vector for the objective function in the introduced to form the so called big-M constraints, which limit
upper level the computation efficiency. In the duality based approach, the
𝒆, 𝒇 coefficient vector for the objective function in the lower level is transformed into its dual problem [8]. However,
lower level the resulting nonlinear terms in the strong duality condition
make the whole problem hard to solve, especially for a large-
𝑫 load vector
size system. Thus, it is necessary for us to develop a novel
∆𝑫 false data injection vector into load measurements algorithm to get the optimal or suboptimal solution of a bilevel
𝑨, 𝒃 coefficient matrices in the upper level optimization problem.
𝑬, 𝑭, 𝑮 coefficient matrices in the lower level A little efforts [9]-[11] has been devoted to apply some
𝒉 physical line flow vector decomposition techniques to solve a bilevel optimization
𝑱 load shedding vector problem. However, we notice that these methods can be only
𝑳 line flow vector applied to a special class of the bilevel linear programming
𝑳𝒎𝒂𝒙 line rating vector with the same objectives in the lower and upper levels. So far,
𝑷 generator output vector how to solve a GBLP with different objectives in the lower
𝑷𝒎𝒂𝒙 maximum generation output vector and upper levels efficiently is still an open question. In this
paper, we propose a benders decomposition based approach to
solve a GBLP with different objective functions in the lower Note that the bilevel optimization problem has a special
and upper levels. Different from the traditional benders structure in which only variable x is transmitted to the upper
decomposition approach, the subproblem consists of two ̂, the GBLP can be solved by
level for a given y. For a given 𝒚
sequential LPs. In the first stage, the first LP is solved to get solving two sequential LPs: LP1 and LP2.
the optimal objective. Then the constraint of the objective (LP2): min 𝒄𝑻 𝒙 (10)
function will be inserted into the second LP to get the dual
variables used to generate benders cuts. subject to
The rest of this paper is organized as follows. Section II 𝒆𝑻 𝒙 = 𝑲 𝜌 (11)
proposes an decomposition algorithm to solve a GBLP. Section 𝑬𝒙 ≤ 𝑮 − 𝑭𝒚 ̂ 𝝁′ (12)
III demonstrates the proposed model with the IEEE 118-bus
where K is the objective value of the lower optimization
system. Section IV concludes the paper.
̂.
problem for a given 𝒚
II. DECOMPOSITION METHODOLOGY OF GBLP The dual problem of (10)-(12) is
In this section, we propose a decomposition algorithm to max 𝜌𝐾 + (𝝁′ )𝑇 (𝑮 − 𝑭𝒚
̂) (13)
solve a GBLP.
subject to
A. Decomposition Method for GBLP 𝜌
Consider the general bilevel linear program(GBLP) (1)-(4) [𝑒 𝐸 𝑇 ] [ 𝝁′ ] ≤ 𝑐 (14)
as follows. .
Let 𝑆 denote the feasible region of the dual problem,
(GBLP) min 𝒄𝑻 𝒙 + 𝒅𝑻 𝒚 (1) 𝜌
subject to 𝑆 = {(𝜌, 𝝁′ )|[𝑒 𝐸 𝑇 ] [𝝁′ ] ≤ 𝑐} (15)
𝑨𝒚 ≤ 𝒃 (2) The region 𝑆 has two good chatteristics:
min 𝒆𝑻 𝒙 + 𝒇𝑻 𝒚 (3)
1) It is independent with the value of the variable 𝒚 in
subject to the upper level and the value of K;
𝑬𝒙 + 𝑭𝒚 ≤ 𝑮 (4) 2) It is a fixed polygon with a finite number of extreme
points and extreme rays.
where variables 𝒙, 𝒚 are continuous variables. Note that the
objectives in the lower and upper levels could be different,
that is Let 𝑈 ′ , 𝑉 ′ denote the sets of all the extreme points and
𝒄𝑻 𝒙 extreme rays of the feasible region S, respectively.
≠𝑤
𝒆𝑻 𝒙 𝑈 ′ = {𝑢1′ , 𝑢2′ , ⋯ , 𝑢𝑛′ } 𝑉 ′ = {𝑣1′ , 𝑣2′ , ⋯ , 𝑣𝑞′ }
where 𝑤 is a scalar. This characteristic poses a huge According to [12], the optimization problem is equivalent to
challenge for developing an efficient algorithm for solving (16)-(17):
it. 𝑇
max{ 𝜌𝑖 𝐾 + (𝝁′ 𝒊 ) (𝑮 − 𝑭𝒚
̂) (𝜌𝑖 , 𝝁′ 𝒊 ) ∈ 𝑈 ′ } (16)
Note that for a fixed 𝒚 ̂ , the lower level optimization
problem becomes a linear programming. Rewrite the lower subject to
𝑇
level optimization problem (3)-(4) as
𝜌𝑗 𝐾 + (𝝁′ 𝒋 ) (𝑮 − 𝑭𝒚
̂) ≤ 0 (𝜌𝑗 , 𝝁′ 𝒋 ) ∈ 𝑉 ′ (17)
(𝐒𝐏): min 𝒆𝑻 𝒙 + 𝒇𝑻 𝒚
̂ (5)
Constraint (17) holds for ensuring the bound of the dual
subject to
problem (13)-(14).
̂
𝑬𝒙 ≤ 𝑮 − 𝑭𝒚 𝝁 (6)
Similarly, for a fixed 𝒚, the optimization problem (5)-(6) is a
Since the term 𝒇𝑻 𝒚
̂ is a constant, we have LP. Thus, its dual problem is written as
(𝐋𝐏𝟏) 𝐾 = min 𝒆𝑻 𝒙 (7) max(𝝁)𝑇 (𝑮 − 𝑭𝒚 ̂) (18)
subject to
subject to
Constraint (6)
𝑬𝑻 𝝁 ≤ 𝒄 (19)
As LP1 is a LP whose optimal value can be determined at Let T denote the feasible region of (18)-(19)
one of its extreme points.
Note that the goal of the bilevel optimization problem is to T = {(𝜇)|𝑬𝑻 𝝁 ≤ 𝒄} (20)
determine a best y that minimizes the objective value of LP3: It can be seen that the region T is also independent with the
(𝐋𝐏𝟑) min 𝒄𝑻 𝒙 + 𝒇𝑻 𝒚
̂ (8) value of the variable 𝒚 in the upper level and has a finite
min 𝒆𝑻 𝒙 (9) number of extreme points and extreme rays.
subject to 𝑈 = {𝑢11 , 𝑢12 , ⋯ , 𝑢1𝑚 } 𝑉 = {𝑣11 , 𝑣21 , ⋯ , 𝑣𝑝1 }
Constraint (6) Since the strong duality condition can be satisfied, we have
 𝐾 = max(𝝁)𝑇 (𝑮 − 𝑭𝒚
̂) (21) 3) if the subproblem is unbounded, then 𝜌∗ , (𝝁′ )∗ is the new
extreme ray that will be added into the master problem.
Introducing (21) into (16), which gives
𝑇
max 𝜌𝑖 (max(𝝁)𝑇 (𝑮 − 𝑭𝒚
̂)) + (𝝁′ 𝒊 ) (𝑮 − 𝑭𝒚
̂) (22)
Initial 𝑧𝑙𝑜𝑤𝑒𝑟 , 𝑧𝑢𝑝𝑝𝑒𝑟 , 𝜀, 𝑦
It can be verified that (22) is equivalent to
𝑇
max{𝜌𝑖 (𝝁𝒋 𝑇 (𝑮 − 𝑭𝒚
̂)) + (𝝁′ 𝒊 ) (𝑮 − 𝑭𝒚
̂) = max (𝜌𝑖 𝝁𝒋 𝑇 +
𝑇
Solve LP1 to get K
(𝝁′ 𝒊 ) ) (𝑮 − 𝑭𝒚
̂) (𝜌𝑖 , 𝝁′ 𝒊 ) ∈ 𝑈 ′ , 𝝁𝒋 ∈ 𝑈} (23)

subject to Add infeasibility cuts

Constraint (17) Feasible?
No
Note that LP2 is feasible as long as LP1 is feasible, so Yes
constraint (17) can be replaced by
Solve LP2 and update 𝑧𝑙𝑜𝑤𝑒𝑟
𝑇
(𝝁𝒋 ) (𝑮 − 𝑭𝒚
̂ ) ≤ 0 𝜇𝑗 ∈ 𝑉 (24)
Thus, the optimization problem GBLP is equivalent to (25): Solve RMP and update 𝑧𝑢𝑝𝑝𝑒𝑟
(𝐃𝐌𝐏): min 𝑧 (25)
subject to Add Optimality cuts
𝑇 ′ 𝑇 𝑻 ′ ′
z ≥ (𝜌𝑖 𝝁𝒋 + (𝝁 𝒊 ) )(𝑮 − 𝑭𝒚) + 𝒅 𝒚 (𝜌𝑖 , 𝝁 𝒊 ) ∈ 𝑈 , 𝝁𝒋 ∈ 𝑈 Converged?
(26) No
Yes
Constraints (2), (24)
Stop
B. Algorithm
As discussed, the original GBLP can be transformed into a
Fig.1 Flowchart of decomposition algorithm for GBLP
single level optimization problem DMP if we can enumerate
all its extreme points and extreme rays. In practice, however, it Proof: Since 𝑔∗ and 𝜌∗ , (𝝁′ )∗ are the optimal objective value
is very difficult to enumerate all the extreme points and of the subproblem and corresponding dual variables, we have
extreme rays for a feasible region. In particular, the size of
problem becomes large. To overcome the computation 𝑔∗ = 𝜌∗ 𝐾 + (𝑮 − 𝑭𝒚 ̂)𝑇 (𝝁′ )∗ + 𝒅𝑻 𝒚̂ (30)
difficulty, Benders in [12] developed an iterative algorithm to To ensure the bound of the subproblem, constraint (31) must
determine the optimal solution by adding a subset of extreme hold,
points and extreme rays. 𝜌∗ 𝐾 + (𝑮 − 𝑭𝒚̂)𝑇 (𝝁′ )∗ + 𝒅𝑻 𝒚 ̂ ≤ 0 (𝜌, 𝝁′ ) ∈ 𝑉 (31)
∗
Define the restricted master problem(RMP) as follows: 𝑧 is the optimal solution of the RMP, so
𝑧 ∗ = min{𝜌∗ 𝐾 + (𝑮 − 𝑭𝒚 ̂)𝑇 (𝝁′ )∗ + 𝒅𝑻 𝒚
̂} (32)
(𝐑𝐌𝐏): min 𝑧 (27)
Combining (30) and (32), we have
subject to
𝒇𝑻 𝒚̂ + 𝑧 ∗ ≥ 𝒇𝑻 𝒚
̂ + min {𝜌∗ 𝐾 + (𝑮 − 𝑭𝒚 ̂)𝑇 (𝝁′ )∗ + 𝒅𝑻 𝒚
̂}
𝑇
z ≥ (𝜌𝑖 𝝁𝒋 𝑇 + (𝝁′ 𝒊 ) )(𝑮 − 𝑭𝒚) + 𝒅𝑻 𝒚 (𝜌𝑖 , 𝝁′ 𝒊 ) ∈ 𝑈𝑟′ , 𝝁𝒋 ∈ ≥𝒇 𝒚𝑻
̂+ 𝑔 ∗
(33)
𝑈𝑟 (28) So,
𝑇 (𝒈
(𝝁𝒋 ) ̂) ≤ 0,
− 𝑬𝒙 (𝝁𝒋 ) ∈ V𝑟 (29) 𝑧 ∗ ≥ 𝑔∗ (34)
Constraints (2) If 𝑧 ∗ > 𝑔∗ ,then
where 𝑈𝑟′ , 𝑈𝑟 , V𝑟
are the reduced sets of 𝑈2 , 𝑈1 , V1 . Thus, the 𝑧 ∗ > 𝑔∗ = 𝜌∗ 𝐾 + (𝑮 − 𝑭𝒚 ̂)𝑇 (𝝁′ )∗ + 𝒅𝑻 𝒚
̂ (35)
RMP determines an upper bound for the optimal objective of Together with (32), which gives
the dual problem DMP.
min{𝜌∗ 𝐾 + (𝑮 − 𝑭𝒚 ̂)𝑇 (𝝁′ )∗ + 𝒅𝑻 𝒚 ̂} > 𝜌∗ 𝐾 + (𝑮 − 𝑭
̂)𝑇 (𝝁′ )∗ + 𝒅𝑻 𝒚
𝒚 ̂ (36)
Theorem 1: Suppose that 𝑦 ∗ , 𝑧 ∗ is the optimal solution of the This indicates that
RMP. If 𝑔∗ and 𝜌∗ , (𝝁′ )∗ are the optimal objective value of the
(𝜌, 𝜇 ′ ) ∉ 𝑈 ′ (37)
subproblem and corresponding dual variables, then we have
1) if 𝑧 ∗ > 𝑔∗ , then 𝜌∗ , (𝝁′ )∗ is the new extreme point that It is a new extreme point.
will be added into the master problem; If 𝑧 ∗ = 𝑔∗ , then we have
2) if 𝑧 ∗ = 𝑔∗ , then 𝑦 ∗ , 𝑧 ∗ is the optimal solution of the DMP; 𝑔∗ ≤ min{𝜌∗ 𝐾 + (𝑮 − 𝑭𝒚 ̂)𝑇 (𝝁′ )∗ + 𝒅𝑻 𝒚
̂} ≤ 𝜌𝐾 + (𝑮 −
𝑭𝒚̂) + 𝒅𝑻 𝒚̂(38)
 Together with (31), we can see that (𝑦 ∗ , 𝑧 ∗ ) is a feasible Step 4: Solve LP2. Suppose the optimal solution is
point of the dual problem. [𝜇𝑗 𝜌𝑖 𝝁′ 𝒊 ]𝑇 , then an upper bound of the BLP is
Consider MP is relaxed to RMP, determined
𝑧𝑅𝑀𝑃 ≤ 𝑧𝑀𝑃 (39) 𝑇
Thus, (𝑦 ∗ , 𝑧 ∗ ) is the optimal solution of the dual master z𝑙𝑜𝑤𝑒𝑟 = (𝜌𝑖 𝜇𝑗 𝑇 + (𝝁′ 𝒊 ) )(𝑮 − 𝑭𝒚
̂) + 𝒅𝑻 𝒚
̂ (46)
problem. Step 5: If the following constraint is satisfied, stop;
Finally, if the subproblem is unbounded. That is, the primal |𝑧𝑢𝑝𝑝𝑒𝑟 −𝑧𝑙𝑜𝑤𝑒𝑟 |
problem is infeasible. In this case, ≤𝜀 (47)
𝑧𝑙𝑜𝑤𝑒𝑟
𝜌∗ 𝐾 + (𝑮 − 𝑭𝒚 ̂)𝑇 (𝝁′ )∗ + 𝒅𝑻 𝒚
̂<0 (40) Otherwise, add the following cut into the master problem
As all the existing extreme rays in the RMP satisfy the 𝑇
following constraint z ≥ (𝜌𝑖 𝝁𝒋 𝑇 + (𝝁′ 𝒊 ) )(𝑮 − 𝑭𝒚) + 𝒅𝑻 𝒚 (48)
𝜌∗ 𝐾 + (𝑮 − 𝑭𝒚 ̂)𝑇 (𝝁′ )∗ + 𝒅𝑻 𝒚
̂≥0 (41) Step 6: Solve the master problem to update the variable 𝑦̂. The
Thus, upper bound of BLP is the objective value of
(𝜌, 𝝁′ ) ∉ 𝑉 ′ (42) 𝑧𝑢𝑝𝑝𝑒𝑟 = min 𝑧 (49)
It is a new generated extreme ray. subject to
Constraints (2), (28)-(29)
Theorem 2: The optimal solution of the DMP can be and then go to Step 2.
determined by solving the RMP a finite number of times. The advantage of the proposed algorithm is to reduce the
execute time by solving a small number of LPs.
Proof ： The total number of the combinations of extreme
points or extreme rays in 𝑈, 𝑈 ′ , 𝑉, 𝑉 ′ is III. CASE STUDY
𝑁 = (𝑚 + 𝑝)(𝑛 + 𝑞) In this section, we test the proposed decomposition
algorithm using the IEEE 118-bus system[13].The testing
According to Theorem 1, at each iteration, one new
model is described in section A. Simulations are carried out on
combination of extreme points or extreme rays will be added
a 3.4GHz personal computer with 8GB of RAM. The model
into the master problem. Thus, the program will stop in a finite
and algorithm are implemented in MATLAB.
number of iterations.
A. Modeling
It should be pointed that the algorithm can find the optimal
solution in 𝑁 iterations in the worst case. Luckily, the The security constrained economic dispatch(SCED) is
iterations in practice is far less than N, which garatte the performed every 5-15 minutes based on the forecasted loads
efficiency of the algorithm. This will be verified in the test and received real-time measurements. However, the high
cases. integration of renewables and malicious false data attacks
significantly increases the uncertainty of forecasted loads and
real-time measurements. It has been show that the uncertainty
According to the analysis above, the entire algorithm can be
of these data could lead to the overloading of some lines. If
described as follows:
these line overloads cannot be mitigated immediately, severe
Step 1: Initial the tolerance ε, 𝑧𝑙𝑜𝑤𝑒𝑟 = −∞, 𝑧𝑢𝑝𝑝𝑒𝑟 = +∞ and
consequences, such as cascading failures, might occur. To
̂ = 𝒚𝟎 ;
the initial𝒚 ensure the secure operation of a power system, it is very
Step 2: For the given 𝐲̂ , solve the optimization problem (43): essential for an operator to identify these lines whose power
min 𝟏𝑻 s (43) flows could exceed the flow limits after SCED. This is done by
solving the GBLP below:
subject to
max 𝑢 (50)
̂
𝑬𝒙 − 𝒔 ≤ 𝑮 − 𝑭𝒚 𝝅 (44) subject to
𝐬≥𝟎 𝑢 = |ℎ𝑘 /𝐿𝑚𝑎𝑥
𝑘 | (51)
̂. If 𝒉 = 𝑺 ∙ (𝑼 ∙ 𝑷 − 𝑽 ∙ (𝑫 − 𝑱)) (52)
The optimization problem is used to the feasibility of 𝒚
the objective of the optimal problem is zero, then 𝑦̂ is feasible. 𝟏𝑻 ∆𝑫 = 0 (53)
In this case, go to Step 3. If the objective of the optimal −𝝉𝑫𝒅 ≤ ∆𝑫𝒅 ≤ 𝝉𝑫𝒅 (0 < 𝜏 < 1) (54)
problem is greater than zero, then the subproblem (5)-(6) is
min 𝒄𝒈 𝑻 𝑷 + 𝒄𝒅 𝑻 𝑱 (55)
infeasible. In this case, we need to introduce an infeasibility
cut into the master problem. subject to
𝝅𝑻 (𝑮 − 𝑭𝒚̂) ≤ 0 (45) 𝟏𝑻 𝑷 = 𝟏𝑻 (𝑫 − 𝑱) (56)
Step 3: Solve LP1 to get the value of K. 𝑳 = 𝑺 ∙ 𝑼 ∙ 𝑷 − 𝑺 ∙ 𝑽 ∙ (𝑫 + ∆𝑫 − 𝑱) (57)
(𝐋𝐏𝟏) 𝐾 = min 𝒆𝑻 𝒙 (7) 𝑷𝒎𝒊𝒏 ≤ 𝑷 ≤ 𝑷𝒎𝒂𝒙 (58)
subject to Constraint (6) −𝑳𝒎𝒂𝒙 ≤ 𝑳 ≤ 𝑳𝒎𝒂𝒙 (59)
 𝟎 ≤ 𝑱 ≤ 𝑫 + ∆𝑫 (60) IV. CONCLUSIONS AND FUTURE WORK
Constraint (52) gives the power flows of line 𝑘 after attacks. This paper proposes an efficient decomposition algorithm to
Note that the injected false data ∆𝑫 is not included since it is solve a GBLP. This is achieved by decomposing the GBLP
not a physical load. Without loss of generality, the maximum into one master problem and a subproblem. The master
allowable attacking amount at a bus is set to 50% of its load. problem is solved to update the decision variable in the upper
The injected false data ∆𝑫 is summed to zero (53). Constraint level. And the subproblem is solved by solving two sequential
(56) represents the lower power balance equation. The line LPs to generate the necessary cut added into the master
flow under the interruption of false data ∆𝑫 is determined by problem. The simulation results on the large-size testing
constraint (57). Constraints (58)-(60) represent the lower and systems demonstrate its efficiency of the proposed
upper bound constraints of generation, line flow, and load decomposition algorithm. In the next work, we are going to
shedding, respectively. develop a decomposition approach for bilevel mixed integer
The upper level of the above bilevel problem simulates the linear programming that includes some integer variables in the
attacker’s attacking strategy, which is to determine the injected lower level. The difficult part is the strong duality condition
false data ∆𝑫 that maximizes the potential flow of line 𝑘 (51). might not be satisfied due to the existing integer variables.
The lower level simulates the system operator’s operation
strategy, which is essentially an SCED problem that minimizes REFERENCES
the total generation cost and load shedding cost (55).
[1] L. P. Garces , A. J. Conejo , R. Garcia-Bertrand and R. Romero, "A
B. Results bilevel approach to transmission expansion planning within a market
environment", IEEE Trans. Power Syst., vol. 24, no. 3, pp. 1513-1522,
The GBLP (1)-(4) is tested on the IEEE 118-bus system[13], 2009.
which consists of 118 buses and 186 lines. The uncertainty of [2] Y. Yuan, Z. Li, and K. Ren, “Modeling load redistribution attacks in
load power injection is assumed to be 50% of its actual data power systems,” IEEE Transaction on Smart Grid, vol. 2, no. 2, pp. 382–
390, Jun. 2011.
due to the interruption of cyber attacks. And the minimum [3] K. C. Almeida and F. S. Senna, "Optimal active-reactive power dispatch
outputs of generators are set to zeros. Without loss of under competition via bilevel programming", IEEE Trans. Power Syst.,
generality, we only report the simulation results of the selected vol. 26, no. 4, pp. 2345-2354, 2011.
ten lines below. [4] X.Liu, Z.Li, X.Liu, and Z.Li “Masking Transmission Line Outages via
False Data Injection Attacks,” IEEE Trans. IEEE Transactions on
TABLE 1 TESTING RESULT FOR THE IEEE 118-BUS SYSTEM Information Forensics and Security, vol. 11, no. 7, pp. 1592 - 1602, 2016.
[5] T. Li and M. Shahidehpour, "Strategic bidding of transmission-constrained
Line 𝒖 𝒈𝒂𝒑 Iterations 𝒕(𝒔) GENCOs with incomplete information", IEEE Trans. Power Syst., vol. 20,
5-11 0.59 0.31% 4 0.09 no. 1, pp. 437-447, 2005.
11-12 1.40 0.68% 6 0.06 [6] S. Fliscounakis, P. Panciatici, F. Capitanescu and L. Wehenkel,
2-12 0.60 0.36% 6 0.13 "Contingency ranking with respect to overloads in very large power
3-12 0.38 0.09% 6 0.11 systems taking into account uncertainty, preventive, corrective
7-12 0.68 0.31% 6 0.14 actions", IEEE Trans. Power Syst., vol. 28, no. 4, pp. 4909-4917, 2013.
11-13 1.03 0.55% 4 0.10 [7] J. M. Arroyo and F. D. Galiana, “On the solution of the bilevel
12-14 0.47 0.01% 7 0.14 programming formulation of the terrorist threat problem,” IEEE Trans.
13-15 0.47 0.28% 8 0.16 Power Syst., vol. 20, no. 2, pp. 789–797, May 2005.
14-15 0.48 0.01% 10 0.19 [8] A. L. Motto, J. M. Arroyo, and F. D. Galiana, “A mixed-integer LP
12-16 0.57 0.20% 10 0.19 procedure for the analysis of electric grid security under disruptive threat,”
IEEE Trans. Power Syst., vol. 20, no. 3, pp. 1357–1365, Aug. 2005.
[9] A. Delgadillo, J. Arroyo, and N. Alguacil, “Analysis of electric grid
In Table 1, the second column gives the line overloading interdiction with line switching,” IEEE Trans. Power Syst., vol. 25, no.2,
level for each line. The third column represents the existing pp. 633–641, May 2010.
gap between the primal and dual problems. And the last two [10] J. Salmeron, K. Wood, and R. Baldick, “Analysis of electric grid security
columns give the iterations and run times for the under terrorist threat,” IEEE Trans. Power Syst., vol. 19, no. 2, pp. 905–
decomposition algorithm. It can be seen that lines 11-12, 11-13 912, May 2004.
[11] C. Losada, M. Scaparra, and J.R. Hamley “Optimizing system resilience:
would be overloaded due to the uncertainty of forecasted loads A facility protection model with recovery time,” European Journal of
and received measurements after SCED. These line are Operational Research., vol. 217, no.2, pp. 519–530, Mar 2012.
identified as high-risk lines that need to be strongly monitored [12] J. F. Benders, “Partitioning procedures for solving mixed-variables
for ensuring the secure operation of a power system. And the programming problems,” Numer. Math., vol. 4, no. 1, pp. 238–252,Dec.
optimal solution of the GBLP can be determined fast. For 1962.
[13] R. D. Zimmerman, C. E. Murillo-Sánchez, and R. J. Thomas,
example, the average run time for the IEEE 118-bus system is “MATPOWER: Steady-state operations, planning and analysis tools for
0.1s. However, the KKT based approach did not converge for power systems research and education,” IEEE Transactions on Power
most lines after 5 hours. The small gap guarantees the required Systems, vol. 26, no. 1, pp. 12-19, Feb. 2011.
accuracy of the obtained solution. In addition, the algorithm
converges in a small number of iterations. For instance, the Xuan Liu received the B.S. and M.S degrees from Sichuan University, China,
maximum iterations for the IEEE 118-bus system is 10, far less in 2008 and 2011, and Ph.D. degree from Illinois Institute of Technology,
USA, in 2015, respectively. He is currently a senior research associate at the
than the iterations in the worst case. This verifies the Robert Gavin Electric Innovation Center, Illinois Institute of Technology. His
computation efficiency of the proposed decomposition research interests include smart grid security, operation and economics of
algorithm. power systems.