中国科技论文在线 http://www.paper.edu.

cn

Mathematics and Computers in Simulation 68 (2005) 57–65

A hybrid method of chaotic particle swarm optimization and

linear interior for reactive power optimisation
Jiang Chuanwena,∗ , Etorre Bompardb
a
Department of Electrical Engineering, Shanghai Jiaotong University,
Shanghai 200030, PR China
b
Department of Electrical and Computer Engineering, Polectenico Di Torino, Italy

Received 18 June 2004; received in revised form 28 September 2004; accepted 4 October 2004
Available online 18 November 2004

Abstract

The reactive power optimization is an effective method to improve voltage level, decrease network losses and
maintain the power system running under normal conditions. This paper provides a method combining particle
swarm optimization (PSO) with linear interior point to handle the problems remaining in the traditional arithmetic
of time-consuming convergence and demanding initial values. Furthermore, since chaotic mapping enjoys certainty,
ergodicity and stochastic property, the paper introduces chaos mapping into the particle swarm optimization, the
paper presents a new arithmetic based on a hybrid method of chaotic particle swarm optimization and linear interior
point. Thanks to the superior overall exploration ability of particle swarm optimization and the local exploration
ability of linear interior point within the neighborhood of the optimal point, the new method can improve the
performance of both convergence and results’ precision. Tested by IEEE-30, the new method provided in this
paper is proved effective and practical in the optimization of shunt capacitors and tap position of load-ratio voltage
transformer.
© 2004 IMACS. Published by Elsevier B.V. All rights reserved.

Keywords: Reactive power optimization; Chaos; PSO; Linear interior point

∗
Corresponding author. Tel.: +86 21 54747501; fax: +86 21 54747501.
E-mail addresses: jiangcw sjtu@yahoo.com.cn (J. Chuanwen), ettore.bompard@polito.it (E. Bompard).

0378-4754/$30.00 © 2004 IMACS. Published by Elsevier B.V. All rights reserved.

doi:10.1016/j.matcom.2004.10.003

转载
中国科技论文在线 http://www.paper.edu.cn

58 J. Chuanwen, E. Bompard / Mathematics and Computers in Simulation 68 (2005) 57–65

1. Introduction

As a key measure of power system operation, reactive power optimal compensation can affect
the security and economy of the grid. A reasonable distribution of reactive power will decrease net-
work losses, improve the voltage level as well as maintain the grid operating under normal con-
ditions. For the reactive power optimization is a non-linear planning problem with multi-object,
multi-constraint and multi-variable including continuous variables, such as node’s voltage, and dis-
crete ones, like the tap position of a transformer, the entire procedure of optimization is very com-
plex and has been a focus of many researchers for a long time. Lots of optimal technologies have
been used to solve this problem, including the algorithm based on gradient [1] or based on math-
ematic optimization like linear planning method [2], non-linear planning method [3], the Bender’s
discrete method [4], etc. Generally, there are some defects in those methods to some extent, such
as the sensitivity to initial value, the un-stability of convergence and the complication of algorithm.
Nowadays, some random search algorithms have been proposed to solve the non-linear optimiza-
tion [5–8], for example genetic algorithms (GAs), evolution strategies (ESs), simulated annealing
(SA) and evolutionary programming (EP). Particularly, with its sound exploration ability both global
and local, a new evolution technology, named particle swarm optimisation (PSO), has become the
new focus of research [9–11]. The achievements encourage people to make further research in this
field.
Considering the characteristics of reactive power optimization, the paper provides a novel method,
combining particle swarm optimization with linear interior point to handle the problems remaining in the
traditional arithmetic of time-consuming convergence and demanding initial values. Furthermore, since
chaos mapping enjoys certainty, ergodicity and stochastic property, the paper introduces chaos mapping
into the particle swarm optimization, and thus presents a new arithmetic based on a hybrid method of
chaotic particle swarm optimization and linear interior point. Tested by IEEE-30, the new method provides
in this paper is proved effective and practical in the optimization of shunt capacitors and tap position of
load-ratio voltage transformer.
The paper was organized as follows: Section 2 is the reactive power optimization model in a distribution
grid; The hybrid method of chaotic particle swarm optimization and linear interior for reactive power
optimization is introduced in Section 3; numerical examples were shown in Section 4; Section 5 is the
conclusion.

2. Reactive power optimization problems

The goal of reactive power optimization is to minimize real power losses and improve voltage profile
by setting generator bus voltages, VAR compensators and transformer taps.
The model of reactive power optimization of power systems can be described as follows.
Objective function:


 
f = min PGi − PLi (1)
i ∈ NG i ∈ NL
中国科技论文在线 http://www.paper.edu.cn

J. Chuanwen, E. Bompard / Mathematics and Computers in Simulation 68 (2005) 57–65 59

where f represents the minimization of transmission losses; NG and NL the numbers of generators and
load buses, respectively, PG the total active power generated in MW and PL is the total load of the system
in MW.
The constrains are described as:
(1) Power flow equation:

Pi − Vi Vj (Gij cos θij + Bij sin θij ) = 0
j∈i

Qi − Vi Vj (Gij sin θij − Bij cos θij ) = 0 (2)
j∈i
(i, j = 1, 2, · · · , n)
(2) The upper and lower limits of state variable:
Vi,min ≤ Vi ≤ Vi,max , i = 1 : Nbus
(3)
QGi,min ≤ QGi ≤ QGi,max , i = 1 : NG
(3) The upper and lower limits of control variable:
VGi,min ≤ VGi ≤ VGi,max , i = 1 : NG
QCi,min ≤ QCi ≤ QCi,max , i = 1 : Ncap (4)
Kl,min ≤ Kl ≤ Kl,max , i = 1 : Nl
where Nbus is the number of busbars, Ncap the number of shunt capacitors, Nl the number of transformers,
Pi and Qi represents separately the active and reactive power of the ith node. Vi and QGi represent vectors
of load bus voltage and reactive power generation, respectively; VGi , QCi and Kl are vectors of generator
voltage, VAR compensation and transformer ratio, respectively.

3. A hybrid arithmetic based on chaotic particle swarm and linear interior point

For the reactive power problem described by the formulae (1)–(4), it can be presented as following:

 min f (x, u)

s.t. g(x, u) = 0 (5)

 h(x, u) ≤ 0
where u is the control variable and x is the state variable. f(x, u) is the objective function, g(x, u) the load
flow equation constraints and h(x, u) is the inequation constraints.
The new arithmetic based on chaotic particle swarm and linear interior point can be classified into
two phases. First, with the large-scale exploration ability of chaotic particle swarm, the optimal flow
solution is driven into the neighborhood of the global optimal point. linearize the Eq. (5) successively
in the neighborhood of optimal point and solve the linear sub-problem by the linear primal-dual interior
point method with the predictor-corrector and thus, reach or close to the optimal solution so as to increase
the precision.
中国科技论文在线 http://www.paper.edu.cn

60 J. Chuanwen, E. Bompard / Mathematics and Computers in Simulation 68 (2005) 57–65

3.1. Chaotic particle swarm optimization process

Particle Swarm Optimization was first proposed by Kennedy and Eberhart in 1995 [12]. It aimed at
the description of birds’ behaviors by graphic method. Later on, it was developed as a general heuristic
exploration technique, which performs effective exploration through memory and feedback. With the
imitation of the behavior of bio-community, it enjoys a rapid calculation and a sound global exploration
when applied in a large-scale optimization, so it is suitable for parallel calculation.
Like evolutionary algorithms, PSO technique conducts search using a population of particles, corre-
sponding to individuals. Each particle represents a candidate solution to the problem at hand.
During the calculation, the particle is affected by three factors when it is moving in space. One of the
factors is the particle’s current velocity V(t). Another is the optimal point X* (t) where the particle has
reached before. The third factor is the optimal point X** (t) of the community or the sub-community. The
particle’s velocity is changed towards X* (t) and X** (t) in every iteration step. Meanwhile, Vi , X* (t) and
X** (t) are assigned separately a weight at random. The velocity and position is updated according to the
formula (6) and (7).
∗
vi,j (t) = w × vi,j (t − 1) + c1 × r1 × (xi,j (t − 1) − xi,j (t − 1))
∗∗
+c2 × r2 × (xi,j (t − 1) − xi,j (t − 1)) (6)

xi,j (t) = xi,j (t − 1) + vi,j (t) (7)

(i = 1, 2, . . . , n j = 1, 2, . . . , m)
where c1 and c2 are the learning factors, generally, c1 = c2 = 2. w is the given weight, r1 , r2 are the randoms
within the interval of [0,1]. t is the number of iteration, n the number of particles and m is the number of
dimensions.
Studied the influence on the convergence of PSO exerting by the parameters of w, the reference [13]
held the view that w is the key factors to affect the convergence of PSO. In fact, however, it can’t ensure the
optimization’s ergodicity entirely in phase space, because it is random in traditional PSO. Therefore, the
paper provides a new method, which introduces chaotic mapping with certainty, ergodicity and stochastic
property into particle swarm optimization so as to improve the global convergence. The parameter w is
modified by the formula (8).

w(t + 1) = 4.0 × w(t) × (1 − w(t)), w(t) ∈ (0, 1) (8)

The procedure of the improved chaotic PSO can be described as follows.

Step 1. Initialization: VGi , QCi and Kl are regarded as control variables. Set t = 0. Let Xj =
j j j
{VGi , QCi , Kl } be a particle, generate randomly n particles {Xj (0), j = 1,. . ., n} (set n to 50 in this paper).
All particles are set between the lower and upper limits. Similarly, generate randomly initial velocities of
all particles, {Vj (0), j = 1, . . . , n}, where Vj (0) = {vj,1 (0), . . . , vj,m (0)}. vj,k (0) is generated by randomly
selecting a value with uniform probability over the kth dimension [−vmax k ,v max
k ]. State variables QGi and
Vi are calculated in power flow Eq. (2) for the above particles. The constrained optimization problem
is converted into unconstrained optimization problem using penalty factor as given in the Eq. (9). Each
中国科技论文在线 http://www.paper.edu.cn

J. Chuanwen, E. Bompard / Mathematics and Computers in Simulation 68 (2005) 57–65 61

particle in the initial population is evaluated using the Eq. (9).

 2  2
J = f + φ1 (QGi − QGi,min ) + φ2 (Vi − Vi,min ) (9)
i ∈ NG i ∈ NL

where NL is load node number and φ1 , φ2 are penalty factors.

For each particle, set Xj∗ (0) = Xj (0) and Jj∗ = Jj , j = 1, . . . , n. Let J ∗∗ = min{J1∗ , . . . , Jn∗ }. Set the
particle associated with J** as the global best, X** (0).
Step 2. Velocity and position updating: Let t = t + 1. Using the global best and individual best of each
particle, the ith particle velocity and position in the jth dimension is updated using the Eqs. (10)–(12).
∗
vi,j (t) = w(t) × vi,j (t − 1) + c1 × r1 × (xi,j (t − 1) − xi,j (t − 1))
∗∗
+ c2 × r2 × (xi,j (t − 1) − xi,j (t − 1)) (10)

xi,j (t) = xi,j (t − 1) + vi,j (t) (11)

(i = 1, 2, . . . , n j = 1, 2, . . . , m)
where
w(t + 1) = 4.0 × w(t) × (1 − w(t)) (12)
Step 3. Individual and global best updating: Each particle is evaluated according to its updated position.
If Jj < Jj∗ , j = 1, . . . , n then
Xj∗ (t) = Xj (t), Jj∗ = Jj
else go to Step 3.
Search for the minimum value Jmin among Jj∗ .
If Jmin < J** then
X∗∗ (t) = Xmin (t), J ∗∗ = Jmin
else go to Step 3.
Step 4. Stopping criteria: If one of the stopping criteria is satisfied then stop; else go to Step 2.

3.2. Successive linear optimization process

Set the solution xk∗ of the Eq. (5) obtained by the chaotic particle swarm optimization as the initial
value to be successively linearized. Execute linear process at xk∗ to form linear planning subroutine.

∗ T

 min ∇g(xk ) ∇x
s.t. J(xk∗ )x = 0 (13)

 ∗ ∗
h(xk ) + H(xk )x ≤ 0
where Jij = ∂g(x)i /∂xj and Hij = ∂h(x)i /∂xj .
中国科技论文在线 http://www.paper.edu.cn

62 J. Chuanwen, E. Bompard / Mathematics and Computers in Simulation 68 (2005) 57–65

Table 1
The parameters of the shunt-wound capacitors
No. Node Rating Minimum Maximum Step
1 10 0.1 0.000 0.50 0.5
2 24 0.06 0.000 0.50 0.5
3 3 0.05 0.000 0.50 0.5
4 5 0.07 0.000 0.50 0.5
5 6 0.07 0.000 0.50 0.5
6 7 0.07 0.000 0.50 0.5

For the linear planning subroutine above, ref. [14] provided a useful method called primal-dual interior
point. If the solution of the subroutine is xk∗ , then

xk+1 = xk + xk∗ (14)

Then, execute the power flow calculation at the point xk+1 , and get the iteration point that strictly satisfies
the equation constraints, and then linearize the power flow problem (5). Repeat the procedure described
above until satisfy the requirements of ||xk+1 − xk || ≤ ε.

4. Numerical examples

During the practical calculation, the result got by the synthetic optimization is not very good, when
combine transformer’s tap position optimization with parallel compensation. It is because the parallel
compensation of reactive power plays a key part in the optimization, relieving the lack of reactive power,
while transformer’s taps can regulate the transmission of reactive power, solving the unreasonable dis-
tribution of reactive power. Furthermore, the regulation of the taps often causes faults. Therefore, the
parallel compensation plays a main part in the reactive power optimization. This paper optimizes the
reactive power by the parallel compensation and taps optimization separately.
This paper makes an experiment of IEEE-30 and executes the optimization by the method proposed by
this paper and traditional PSO separately. The parameters of shunt capacitors are shown in the Table 1.
The parameters of regulative transformers are in the Table 2. The other parameters used can be found in
reference [15]. All the parameters are in the per unit system.
For the chaotic particle swarm optimization, the initial values are chosen as: the weights c1 , c2 are 2
and w(0) = 0.48.

Table 2
The parameters of the adjustable transformers
No. Start End Ratio Minimum Maximum Step
1 6 9 0.98474 0.8500 1.1500 0.05
2 6 10 0.9629 0.8500 1.1500 0.05
3 4 12 0.98726 0.8500 1.1500 0.05
4 27 28 1.04373 0.8500 1.1500 0.05
中国科技论文在线 http://www.paper.edu.cn

J. Chuanwen, E. Bompard / Mathematics and Computers in Simulation 68 (2005) 57–65 63

Table 3
The results of capacitor optimization
Method Loss (10−2 ) Reactive change of shunt-wound capacitors
1 2 3 4 5 6
PSO 7.0516 0.0713 0.0625 0.0575 0.07 0.07 0.0717
New PSO 7.0512 0.0139 0.0647 0.0575 0.07 0.07 0.074

Fig. 1. The iteration performances of both algorithms during shunt capacitors optimisation.

The optimal results of shunt capacitors are shown in the Table 3. The Fig. 1 shows the iteration
performances of the two algorithms during the shunt capacitors optimization. And the Fig. 2 shows the
performances of the algorithms during the regulative transformers optimization. The results of the two
algorithms are compared in the Tables 3 and 4. It explicitly tells us that the new algorithm achieves a
better performance regarding efficiency and convergence rate.

Table 4
The results of the adjustable transformers optimization
Method Network loss Ratio change of adjustable transformers

1 2 3 4
PSO 0.0704386 0.983072 0.966367 0.999529 1.04273
New PSO 0.0704363 0.981692 0.969356 1.01091 1.04192
中国科技论文在线 http://www.paper.edu.cn

64 J. Chuanwen, E. Bompard / Mathematics and Computers in Simulation 68 (2005) 57–65

Fig. 2. The iteration performances of both algorithms during regulative transformers optimisation.

5. Conclusions

Based on the reactive power optimization problems, this paper provides a method combining particle
swarm optimization with linear interior point, which is a new hybrid arithmetic based on chaotic particle
swarm optimization and linear interior. It introduces chaotic mapping that enjoys certainty, ergodicity
and stochastic property, into particle swarm optimization, enhancing the global exploration ability of the
particle. Meanwhile, it fully takes the advantages of both chaotic particle swarm optimization and linear
interior point, which the former is able to jump out of the local optimal point and the latter boasts the
local exploration ability within the neighborhood of the optimal. It increases the precision and quickens
the convergence. Tested by IEEE-30, it is proved to be efficient and practical during the optimization of
shunt capacitors and tap position of load-ratio voltage transformers.

References

[1] N. Deeb, S.M. Shaidepour, Linear reactive power optimization in a large power network using the decomposition approach,
IEEE Trans. Power Syst. 5 (2) (1990) 428–435.
[2] J.S. Horton, L.L. Grigsby, Voltage optimization using combined linear programming and gradient techniques, IEEE Trans.
PAS 103 (7) (1984) 1637–1643.
[3] Mamandur KRC, R.D. Chenoweth, Optimal control of reactive power flow for improvements in voltage profiles and for
real power loss minimization, IEEE Trans. Power Syst. 100 (7) (1981) 3185–3193.
[4] C. Ying, Mingbo L: Nonlinear Primal-Dual Interior Point Algorithm for Discrete Reactive Power Optimization, Power
Syst. Autom. 25 (10) (2001) 23–27.
[5] D. Bhagwan Das, C. Patvardhan, Reactive power dispatch with a hybrid stochastic search technique, Int. J. Electr. Power
Energy Syst. 24 (9) (2002) 731–736.
中国科技论文在线 http://www.paper.edu.cn

J. Chuanwen, E. Bompard / Mathematics and Computers in Simulation 68 (2005) 57–65 65

[6] J.R. Gomes, O.R. Saavedra, A Cauchy-based evolution strategy for solving the reactive power dispatch problem, Int. J.
Electr. Power Energy Syst. 24 (4) (2002) 277–283.
[7] J.R. Gomes, O.R. Saavedra, Optimal reactive power dispatch using evolutionary computation: extended algorithm., IEEE
Proc. Gen. Transm. Distribution 146 (6) (1999) 586–592.
[8] C.A. Roa-Sepulveda, B.J. Pavez-Lazo, A solution to the optimal power flow using simulated annealing, Int. J. Electr. Power
Energy Syst. 25 (1) (2003) 47–57.
[9] M.A. Abido, Optimal power flow using particle swarm optimization, Int. J. Electr. Power Energy Syst. 24 (7) (2002)
563–571.
[10] S. Naka, T. Genji, Y. Fukuyama, A hybrid particle swarm optimization for distribution state estimation, IEEE Trans. Power
Syst. 18 (1) (2003) 60–68.
[11] J. Kenndey, The particle swarm: social adaptation of knowledge, in: Proceedings of the 1997 IEEE International Conference
on Evolutionary Computation, IEEE Service Center, Piscataway, 1997, pp. 303–308.
[12] J. Kennedy, R. Eberhart, Particle Swarm Optimization., in: Proceedings of IEEE International Conference on Neural
Networks, vol. 4, 1995, pp. 1942–1948.
[13] I.C. Trelea, The particle swarm optimization algorithm: convergence analysis and parameter selection, Inf. Process. Lett.
85 (6) (2003) 317–325.
[14] S. Mehrotra, On the implementation of a primal-dual interior point method, SIAM J. Optim. 2 (4) (1992) 575–601.
[15] Z. Boming, C. Shousun, High Electric Power Network Analysis, Publisher of Tsinghua University, Beijing, 1996, pp.
311–313.