1

A Bilevel Programming Model for the Reactive Power Optimization

Zhong-Ping Wan, Heng Fan, Shu-Fen Wang and Guang-Min Wang

Abstract— A new bilevel programming model combined problem is that Minimize transmission losses or other ap-
the reactive power optimization and the voltage stability propriate objective functions(e.g. multi-objective [1], [3], [16]),
is presented in this paper. A hybrid algorithm, based on while satisfying a given set of physical and operating con-
the genetic algorithm and simulated annealing algorithm to
solve this bilevel programming model of the reactive power straints. They are all single mathematical programming prob-
optimization, is constructed. Testing results of applying the lem. To attain reactive balance in the electric power system,
algorithm to an IEEE-30 bus system and an IEEE-6 bus system the reactive balance in whole electric networks should be
shows feasibility and ef ciency of algorithm. Consequently, rstly satis ed, then reactive power balance in local lines and
blind adjustment of local voltage in nowadays electric power transformer substation be satis ed. In practical electric power
system is solved effectively, at the same time, the dispatch center
adjusts and controls the voltage level of bus and active power system, in order to satisfy their own pro t, some local lines and
losses successfully. transformer station adjust themselves blindly without consider-
ing the pro t of the other lines and transformer station, so the
Keywords— reactive power optimization; voltage stability; voltage improvement in local area is at the cost of the voltage
bilevel programming model; genetic algorithm; simulated an- decrease of others, which makes the problem in whole system
nealing algorithm.
more instability. The characteristic of the bilevel programming
begins from the unitary scale, considering not only the whole
systems’ pro t but also local area’s pro t, which is suitable to
I. I NTRODUCTION
solve the present problem. Furthermore, control of each level
One of the important operating tasks of power utilities is accord with the physical characteristic and operation situation
to keep voltage within an allowable range for high quality in the electric power system, which is combined with the
customer services. In recent years, with the local electric center of dispatchingpower planttransformer substation.
network’s improvement, the scale of electric power system In this paper,on the basis of the bilevel programming theory
becomes larger and larger, the complexity increases continu- and method [2] , combining with the electric power system’
ously, the request to voltage stability and reactive optimization s characteristic, a bilevel programming model for the re-
reaches a new level. From the voltage distortion in Florida active power optimization is presented. The reactive power
in 1970’s to America and California’s serious power cut optimization problem in electric power system is a com-
in August of last year, more and more electric network plex mixed-integer nonlinear programming problem. Many
accidents with the character of voltage distortion result in algorithms have been used to solve this problem, such as
enormous economic losses and confusion of social life, which the reduced gradient method, successive linear program-
attracts more attention of electric power engineers. The volt- ming, successive quadratic programming([6], [9] and refer-
age quality is one of important indexes of electrical energy, ences), the Newton method,etc. Recently, many effective algo-
one of the basic task in electric power system is to ensure rithms have been presented, such as the arti cial intelligence
the consumers’ voltage stability. The voltage adjustment and algorithms[4],[8],[9],[11],[14],[15],[17] , the particle swarm opti-
reactive distribution in the electric power system have a close mization methods[5],[18] , fuzzy mathematics method [10],[12]
relationship, the level of the system’s voltage is the idiographic more effective algorithm with Time-Varying load demand [4] .
embodiment of reactive balance between supply and demand. Here,considering the character of the reactive power optimiza-
The transmission of reactive power not only causes power tion, the genetic algorithm and simulated annealing algorithm
losses but also produces a large voltage descent along the is adopted to solve our bilevel programming model in this
transmission networks. Considering the relationship between paper.
the reactive balance and voltage stability, reactive optimization
becomes the focus again.Therefore,the reactive power and II. M ATHEMATICAL MODEL OF BILEVEL PROGRAMMING
voltage control is obviously very important problem in a
A. The introduction of bilevel programming
distribution system[5],[7],[13],[16] .
Generally, The model for solving reactive power optimization The bilevel programming is a special case of multiple-level
programming[2] , which has the hierarchy characteristic, the
This work was partially supported by the National Natural Science Foun- upper level and the lower level has close relations, restrict one
dation of China(under Grant:70371032) another, and has its own characters. In a bilevel programming,
Zhong-Ping Wan and Shu-Fen Wang are with School of Mathemat-
ics and Statistics, Wuhan University, Wuhan, 430072, China (e-mail: zp- upper decision maker(leader) optimizes his objective function
[email protected]). independently and is affected by the reaction of the lower level
Heng Fan is with Central Southern China Electric Power Design Institute, maker(follower) who makes his decision after the former.Their
Wuhan, 430071, China (e-mail: Fan [email protected]).
Guang-Min Wang is with Institute of System Engineering, Wuhan Univer- objective functions are generally con ict each other. However,
sity, Wuhan, 430072, China (e-mail: [email protected]). most problems encountered in practice fall into the situation in
 2

which they depend on the degree of interactive or cooperation The objective function of the lower level is the minimization
between the two levels,but the information between them is power losses, the objective function of the upper level is the
incomplete and vague. Simultaneously,we notice that although minimization of each bus voltage deviating from the normal
the lower level maker cannot control the decision of the upper level, that is the voltage stability. In a word, we have the
level maker, nal decision of the lower level does eventually following mathematical model based bilevel programming for
in uence the upper level and overall results. the reactive power optimization(BP):
The model presented in this paper applies this character of
the bilevel programming in the reactive power optimization of min F = max{F1 (v1 , x, y), F2 (v2 , x, y), · · · , FN (vN , x, y)}
x∈X
the electric power systemwhich not only satisfy the dispatch-
ing center controls to the voltage level of the whole system,  min
but also considers each individual reactive power optimization.  vi ≤ vi ≤ vimax , i ∈ N
The objective of upper level is the voltage stability(control), s.t. q min max
≤ qGi ≤ qG , i ∈ Ng
 Gmin
i i
max
the objective of the lower level is the active power losses Ti ≤ Ti ≤ Ti , i ∈ N t
minimization, the upper communicates and adjusts with the
lower by the OPF formulation. where Fi (vi , x, y) = max |vi − vˆi | and y solves
X
min ploss = gij [vi2 + vj2 − 2vi vj cos(θi − θj )
B. The reactive power optimization of the bilevel program- y∈Y
i∈N
ming
The following is bilevel Structure of the reactive Power 
 4pP i = p G i − pD i −
Optimization. 


 v i vj (Gij cos θij + Bij sin θij ) = 0

 j∈N


The controls of the 
dispatch center =⇒ The adjusts of the generator
and the transformer tap s.t.
 4qP

vi
i = q G i − q Di −
vj (Gij sin θij − Bij cos θij ) = 0

 j∈N

⇓ 





qimin ≤ qGi ≤ qimax , i ∈ Nr
pmin ≤ pGi ≤ pmax Gi , i ∈ N g
 Gi max
|θij | ≤ θij
⇓ ⇓ ⇓
The rst The second The kth where vˆi is the expected voltage level of the dispatch center,
··· N denotes the numbers of the system bus, Ng denotes the
compensative bus compensative bus compensative bus
numbers of generator bus, Nr denotes the bus sum with com-
Fig. 1 Bilevel Structure of the reactive Power Optimization. pensation device, Nt denotes the bus sum with transformer, gij
The dispatch center is a managing department which con- denotes conductance in each lines, Gij and Bij is the mutual
trols and adjusts the bus voltage level without transcending the conductance and susceptance between bus i and j respectively,
permissible areas, by the analyzing data about each bus from θij denotes phase angle difference between bus i and j, θij max

the upper level to make macroscopical control and synthetical denotes the maximum limit of phase angle difference between
adjustment. Because the transformer tap mainly adjusts the bus i and j, pDi and qDi is the real power load and reactive
voltage level, the ploughing into or removal of the capacitor power load of the bus i respectively, pGi is real power supply
mainly affects the reactive power dispatching of electric sys- bus i, qGi is reactive power supply bus i, qi is reactive power
tem. The relationship between the reactive power dispatching supply with reactive power facility facility, vi is the voltage
and the voltage stability results in the relationship between the level, Ti is the adjustable sums of transformer tap.
transformer and capacitor sets. Furthermore, if the capacitor
sets consider the voltage constraints, which gives little and
bad effects on the voltage regulating. So the transformer tap III. T HE H YBRID A LGORITHM
is the decision variable of the upper level, the capacitor sets
is the decision variable of the lower level to control reactive A. A Overview
power balance, the adjusts of generator can be the decision Genetic algorithms(GA) are a powerful technique based on
variable of the upper level, also be the decision variable of biological concept and are very suitable foe solving optimiza-
the lower level with its simple and adjusts, in this model it is tion problems. Genetic algorithms have become increasingly
the decision variable of the upper level. popular in recent years in science and engineering disciplines.
The limited adjusts of generator: Simulated annealing(SA) is a powerful method for solving
min max
qG i
≤ q Gi ≤ q G i
, i ∈ Ng combinatorial optimization problems. They have been success-
The limited adjusts of transformer: fully applied to many problems in the area of power system
Timin ≤ Ti ≤ Timax , i ∈ Nt analysis and control,i.e. unit commitment,reactive power op-
The limited adjusts of each compensative bus: timization etc. Here, a hybrid algorithm combining genetic
qimin ≤ qGi ≤ qimax , i ∈ Nr algorithm with simulate annealing algorithm for solving the
(BP) problem is presented.
 3

B. The solution procedure Table 1. Voltage data of IEEE-30 bus system

bus No. upper limit lower limit initial state optimal result
Step 1 Gives the initial parameter of the network. 1 1.10 0.90 1.0500 1.1000
Step 2 Forms the bus impedance matrix Y , the initial Jacobi 2 1.10 0.90 1.1000 1.0800
matrix B1 and B2 . 3 1.10 0.90 0.8638 0.9905
Step 3 Inputs the known bus data, the decision variable of 4 1.10 0.90 0.9628 1.0053
5 1.10 0.90 0.9089 0.9810
the upper level, by solving the bus bar equationthe
6 1.10 0.90 0.9448 0.9962
upper initial groups about the generic algorithm. Total transmission loss: 7.34Mvar.
Step 4 Consider each upper initial individual, by the sim-
ulated annealing algorithm and the genetic algo-
rithm,we solve the lower level problem. crossing rate mutating rate and single point crossover
4.11 By the genetic algorithm, the decision vari- 3) The decision variable change in hybrid algorithm, but
able of the lower level, also the complement randomly change in the genetic algorithm, select a group
variable changes, solves the bus bar equa- of random variable of the independence in the simulated
tion, we have the lower initial groups. annealing algorithmon the base of those random variable
4.12 Compute the value function, makes of the independence,change by some regular.
selection-crossover-mutation, then, so we
have the new generation groups. IV. T HE ACCOUNT OF EXAMPLE AND RESULT ANALYSIS
4.13 If it is the generation which is over, go to
A. IEEE-6 bus system and IEEE-30 bus system
the Step 5, otherwise go to the Step 4.12.
4.21 By the simulated annealing algorithm, the The IEEE-6 bus system concludes two generators and two
decision variable of the lower level, also tunable transformers. Choice bus 4 and bus 6 as reactive power
the complement variable changes, solves the compensated bus. IEEE-30 concludes six generators, 4 tunable
bus bar equation, we have the lower initial transformers and 9 reactive power compensated buses. To the
groups. IEEE-6 bus and IEEE-30 bus systems, the voltage range is
4.22 Random variable, by its and the new point 0.90-1.10 (p.u.) and 0.95-1.05 (p.u.),respectively,the tunable
makes selection with some probability, we transformers’ transformation ratio range is 0.95-1.05(p.u.), and
have the new point. the standard power is 100Mvar. Numerical results are exhibited
4.23 If it is constrains condition which is over, in the Table 1 and Table 2, respectively.
go to the step e, otherwise go to the step
4.22. B. Comparison the genetic algorithm with the simulated an-
Step 5 By the solution to the lower problem, we can have nealing algorithm
the suiting value each upper group about the genetic The above two tables give us the voltage of the bus systems.
algorithm, also evaluates groups, then makes selec- The solving of the lower programming adopts the genetic
tion crossovermutation, so we have the new upper algorithm at IEEE-6 bus system while the simulated annealing
groups. algorithm at IEEEE-30 bus system. To the little system, we can
Step 6 If it is constrains condition which is over, we can converge to at larger probability by adopting randomly selected
output the result of the upper and lower objective genetic algorithm, but the calculation speed is not too slow.
function each bus voltage level, otherwise go to d, For the large system, on the choice of the calculation speed
continues to iterative. and the convergence to , we squint towards the calculation
speed. So the lower adopts simulated annealing algorithm.
C. Some remarks on the algorithm In the literature, the transmission loss of IEEE-6 bus system
1) Because the hierarchy characteristic of the bilevel pro- decreases from 11.6Mvar to 8.88Mvar. The decreasing am-
gramming in itself, when we solve the problem, if the plitude is 23%. The transmission loss of IEEE-30 bus system
upper and lower level also use the genetic algorithmto decreases from 8.865Mvar to 7.34Mvar.The decreasing ampli-
large system, the complication of solving process is tude is 17.1%. In this paper, the transmission loss of IEEE-
high. For example,the upper level have fty groupsthe 6 bus system decreases from 11.15Mvar to 8.97Mvar. The
sum of evolution generation is twentythe lower level decreasing amplitude is 19.5%. The transmission loss of IEEE-
have thirty groups, the sum of evolution generation is 30 bus system decreases from 9.04Mvar to 7.33Mvar.The
twenty,the sum of solving bus bar equation is over ten decreasing amplitude is 18.9%. Through comparison, it shows
thousands. Considers this problemthe solving process that the model and numerical results of this paper are feasible.
use the simulated annealing algorithm in large lower More detailed analysis that relates to this reactive optimization
level, the complication obviously decrease, but the small bilevel programming model will be given in other paper.
system can also use the genetic algorithm in both level,
it is easy to converge the optimization solution in the V. C ONCLUSION
overall situation. A bilevel programming model of the reactive power opti-
2) When using the genetic algorithm, we make some gener- mization is presented in this paper. And a hybrid algorithm for
ations be the over condition in both level, use the stable solving the model is proposed by using the genetic algorithm
 4

Table 2. Voltage data of IEEE-30 bus system [10] R.-H. Liang, and Y.-S. Wang, Fuzzy-Based Reactive Power and Voltage
bus No. upper limit lower limit initial state optimal result Control in a Distribution System, IEEE Trans. Power Syst., 2003, 18(2)
1 1.05 0.95 1.000 1.0246 610-618.
2 1.05 0.95 1.000 1.0197 [11] Y.Liu, L.Ma, and J.Zhang, GA/SA/TS hybrid algorithms for reactive
3 1.05 0.95 1.0100 1.0163 power optimization, Power Engineering Society Summer Meeting, 2000.
4 1.05 0.95 1.0000 1.0043 IEEE, Volume:1, 16-20 July 2000 245-249.
5 1.05 0.95 1.0000 1.0168 [12] C.-T. Su, and C.-T. Lin, Fuzzy-Based Voltage/Reactive Power Schedul-
ing for Voltage Security Improvement and Loss Reduction, IEEE Trans.
6 1.05 0.95 0.9900 1.0025 Power Syst., 2001, 319-323.
7 1.05 0.95 1.0000 1.0276 [13] B.D.Thukaram, et al., Optimal Reactive Power Dispatch Algorithm for
8 1.05 0.95 1.0000 1.0122 Voltage Stability Improvement, International Journal of Electrical Power
9 1.05 0.95 0.9999 1.0076 & Energy Systems, 1996,18(7)461-468.
10 1.05 0.95 1.0000 1.0326 [14] A.J.Urdaneda, J.F.Gmez, E.Sorrentino, L.Flores, and R.Diaz, A hybrid
11 1.05 0.95 1.0000 1.0067 genetic algorithm for optimal reactive power planning based upon suc-
cessive linear programming, IEEE Trans. Power Syst., 1999, 19(4) 1292-
12 1.05 0.95 0.9700 1.0018 1298.
13 1.05 0.95 1.0050 0.9964 [15] J.G.Vlachogiannis, and N.D.Hatziargyriou, Reinforcement Learning for
14 1.05 0.95 0.9600 0.9971 Reactive Power Control, IEEE Trans. Power Syst.,2004, 19(3) 1317-1325.
15 1.05 0.95 1.0000 0.9905 [16] B.Venkatesh, G.Sadasivam, and M.Abdullah Khan, A New Optimal
16 1.05 0.95 0.9600 0.9913 Reactive Power Scheduling Method for Loss Minimization and Voltage
17 1.5 0.95 0.9800 0.9931 Stability Margin Maximization Using Successive Multi-Objective Fuzzy
18 1.05 0.95 1.0000 0.9824 LP Technique, IEEE Trans. Power Syst., 2000, 15(2), 844-851.
[17] W.Yan, S. Lu, and David C.Yu, A Novel Optimal Reactive Power Dis-
19 1.05 0.95 0.9999 0.9849 patch Method Based on an Improved Hybrid Evolutionary Programming
20 1.05 0.95 1.0000 1.0067 Technique, IEEE Trans.Power Syst., 2004,19(2) 913-918.
21 1.05 0.95 0.9900 1.0018 [18] H. Yoshida, K. Kawata, Y. Fukuyama, S.Takayama, and Y.Nakanishi,
22 1.05 0.95 1.0000 0.9964 A Particle Swarm Optimization for Reactive Power and Voltage Con-
23 1.05 0.95 1.0000 0.9971 trol Considering Voltage Security Assessment, IEEE Trans.Power Syst.,
24 1.05 0.95 0.9999 1.0099 2000,15(4) 1232-1239.
25 1.05 0.95 1.0000 0.9817
26 1.05 0.95 1.0000 0.9864 VI. B IOGRAPHIES
27 1.05 0.95 1.0200 1.0252 Zhong-Ping Wan is a Professor in School of Mathematics and Statistics
28 1.05 0.95 1.0365 1.0344 at the Wuhan University, China. He received the B.S. degree and the
29 1.05 0.95 1.0058 1.0363 M.S.degree in Mathematics from Wuhan University of Technology and
30 1.05 0.95 1.0420 1.0425 Xidian University, China, respectively. His present research interests include
Total transmission loss: 7.33Mvar. stochastic programming,bilevel programming and semi-in nite programming
etc.,as well as their applications of in the electric power systems optimization
operation and the electric power markets.

Heng Fan is a Senior engineer in Central Southern China Electric Power

and simulated annealing algorithm. The results of the system Design Institute, China. She received the B.S. degree in power system
studies demonstrate the The feasibility of the model and from Huazhong University of Science and Technology, China. Her present
algorithm. interests include power system optimization, operational planning etc.

Guang-Min Wang is a Ph.d.student at Wuhan University, China. He

R EFERENCES received the B.S. degree and the M.S.degree in Mathematics from Wuhan
University, China, respectively. He is currently pursing the Ph.D. degree in
[1] B.Baran,J.Vallejos,R.Ramos, and U.Fernandez,Multi-objective reactive Institute of System Engineering at Wuhan University. His present research
power compensation, Transmission and Distribution Conference and interests include Optimization algorithms,genetic algorithm application in the
Exposition, 2001 IEEE/PES, Volume:1, 28 Oct.-2 Nov. 2001 Pages:97- electric power system.
101.
[2] J.F.Bard, Practical Practical Bilevel Optimization Algorithms and Appli-
cations, Kluwer Academic Publishers,1998.
[3] Y.-L.Chen, An Interactive Fuzzy-Norm Satisfying Method for Multi-
Objective Reactive Power Sources Planning, EEE Trans.Power Syst.,
2000, 15(3) 1154-1160.
[4] Y.Deng,X.Ren, C.Zhao, and D.Zhao, A Heuristic and Algorithmic Com-
bined Approach for Reactive Power Optimization With Time-Varying
Load Demand in Distribution Systems,IEEE Trans.Power Syst., 2002,
17(4) 1068-1072.
[5] Y.Fukuyama and H.Yoshida, A particle swarm optimization for reactive
power and voltage control in electric power systems, Evolutionary Com-
putation, 2001. Proceedings of the 2001 Congress on , Volume: 1, 27-30
May 2001 Pages:87-93.
[6] N.Goudinin, Reactive power optimization successive quadratic program-
ming method, IEEE Trans. Power Syst., 1998, 14(4) 1219-1225.
[7] S.Hao, A Reactive Power Management Proposal for Transmission Oper-
ators, IEEE Trans. Power Syst., 2003, 18(4) 1373-1381.
[8] W.Huang, C.Xu, J.Zhang, and S.Hu, Study of reactive power optimiza-
tion based on immune genetic algorithm, Transmission and Distribution
Conference and Exposition, 2003 IEEE PES, Volume: 1, 7-12 Sept. 2003
Pages:186-190.
[9] K.Y.Lee, and F.F.Yang, Optimal reactive power planning using evolu-
tionary algorithms: A comparative study for evolutionary programming,
evolutionary strategy, genetic algorithm and linear programming, IEEE
Trans. Power Syst., 1998, 13, 101-108.