Expert Systems with Applications 36 (2009) 6118–6126

Contents lists available at ScienceDirect

Expert Systems with Applications

journal homepage: www.elsevier.com/locate/eswa

Reactive power and voltage control based on general quantum genetic algorithms
John G. Vlachogiannis a,*, Jacob Østergaard b
a
Center for Electric Technology (CET), Department Electrical Engineering, Technical University of Denmark (DTU), Elektrovej Building 326, Room 122, DK-2800, Kgs. Lyngby, Denmark
b
Center for Electric Technology, (CET) Department Electrical Engineering, Technical University of Denmark (DTU), Elektrovej Building 325, Room 054, DK-2800 Kgs. Lyngby, Denmark

a r t i c l e i n f o a b s t r a c t

Keywords: This paper presents an improved evolutionary algorithm based on quantum computing for optimal
Reactive power control steady-state performance of power systems. However, the proposed general quantum genetic algorithm
Steady-state performance (GQ-GA) can be applied in various combinatorial optimization problems. In this study the GQ-GA deter-
Meta-heuristic techniques mines the optimal settings of control variables, such as generator voltages, transformer taps and shunt
Genetic algorithm
VAR compensation devices for optimal reactive power and voltage control of IEEE 30-bus and 118-bus
Quantum mechanics computation
systems. The results of GQ-GA are compared with those given by the state-of-the-art evolutionary com-
putational techniques such as enhanced GA, multi-objective evolutionary algorithm and particle swarm
optimization algorithms, as well as the classical primal-dual interior-point optimal power flow algorithm.
The comparison demonstrates the ability of the GQ-GA in reaching more optimal solutions.
Ó 2008 Elsevier Ltd. All rights reserved.

1. Introduction world nonlinear constrained optimization problems. Specifically

During the history of science of computational intelligence
many evolutionary algorithms (EA) were proposed having more
or less success in solving various nonlinear engineering optimiza-
tion problems. Among them the best are considered to be the pop-
ular particle swarm optimization (PSO) (Kennedy & Eberhart,
1995), the ant-colony systems (ACS) (Dorigo, 1992) and the cul-
tural algorithms (Reynolds, 1994). In the last years the effort is
continued by the same and other researchers generating more
effective EA. The reason for the growing development of EA is that
mathematical optimization methods, such as nonlinear program-
ming, quadratic programming, Newton–Raphson based tech-
niques, sequential unconstrained minimization and interior point
algorithms, have failed in handling non-convexities and non-
smoothness in engineering optimization problems. The main
advantage of EA is that they do not require the objective functions
and the constraints to be differentiable and continuous (Esmin,
Lambert-Torres, & De Souza, 2005; Lee, 2005; Lee & El-Sharkawi,
2003; Lee & El-Sharkawi, 2002; Vlachogiannis & Lee, 2006a; Vla-
chogiannis, 2006). However, their main problem remains the same,
the achievement of the global best solution in a short computing
time.
The two above-mentioned aspects sparked off the introduction
of a more robust EA based on quantum mechanics to solve real-
* Corresponding author. Tel.: +45 4525 3579; fax: +45 4588 1295.
E-mail addresses: iv@elektro.dtu.dk (J.G. Vlachogiannis), joe@elektro.dtu.dk (J.
Østergaard).
in this paper, the reactive power and voltage control problems
are solved by means of a quantum computing inspired genetic
algorithm. In general, quantum computing was introduced in the
early 1980s by Feynmann (1986, 1982) and Beinoff (1980). Quan-
tum computers will operate on superposition of all classical search
states, allowing them to evaluate properties of all states in about
the same time a classical machine requires for a single evaluation.
Superposition is described by a state vector S (represented by sym-
bol-ket jSi), consisting of complex numbers, called amplitude
amplifications (Hogg & Portnov, 2000). Under these circumstances,
quantum computing in the future could play a significant role in
computer science. Recent researches (latest 1990s) face quantum
computing as a new evolutionary technique reducing the complex-
ity of global optimization problems. They can be classified in two
fields: One focuses on generating new quantum algorithms using
evolutionary techniques such as genetic programming (Malossini,
Blanzieri, & Calarco, 2004; Rylander, Soule, Foster, & Alves-Foss,
2001; Spector, Barnum, Bernstein, & Swamy, 1999) and the other
concentrates on quantum-inspired evolutionary computing for
classical computers (Han & Kim, 2000, 2002, 2004; Narayanan &
Moore, 1996; Wang, Tang, & Wu, 2005; Zhang, Li, Jin, & Hu,
2004). In the last field, some quantum genetic algorithms (QGA)
have been recently proposed for some combinatorial optimization
problems, such as travelling salesman problem (Narayanan &
Moore, 1996), knapsack problem (Han & Kim, 2000, 2002, 2004),
filter design (Zhang et al., 2004) and numerical optimization prob-
lem (Wang et al., 2005).
In this paper, a QGA named general quantum genetic algorithm
(GQ-GA) for combinatorial optimization problems in power engi-

0957-4174/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2008.07.070
 J.G. Vlachogiannis, J. Østergaard / Expert Systems with Applications 36 (2009) 6118–6126 6119

neering is introduced. The proposed GQ-GA is characterized by X

Nd

theoretical background and search capability compared with J 2 ¼ VDðx; uÞ ¼ jV i  V sp

i j; ð2Þ
i¼1
state-of-the-art QGA (Han & Kim, 2000, 2002, 2004; Narayanan &
Moore, 1996; Wang et al., 2005; Zhang et al., 2004) and other clas- where Vi is the voltage at load bus-i, V sp
i is the pre-specified refer-
sical GA and meta-heuristic evolutionary techniques such as PSO. ence value at load bus-i, which is usually set at the value of
These achievements are based on the concept of quantum theory 1.0 pu, and Nd is the number of load buses.
that one quantum state (q-gene) can represent at least the super- As search space in both problems, the following two vectors are
position of two single states. So, one individual (q-chromosome) considered:
in GQ-GA can represent many states at the same time and there
xT ¼ ½V L1 ; V L2 ; . . . ; V LNd ; Q G1 ; Q G2 ; . . . ; Q GNg ; SL1 ; SL2 ; . . . ; SLNL ; ð3Þ
are weak relationships between individuals (q-chromosomes)
T
since each one of them is determined by current best solution u ¼ ½V G1 ; V G2 ; . . . ; V GNg ; t 1 ; t2 ; . . . ; t NT ; Q C 1 ; Q C 2 ; . . . ; Q C NC ; ð4Þ
and its probability, that is, the history of individual (q-chromo-
where x is the vector of depended variables consisting of load bus
some) up to date (Hogg & Portnov, 2000).
voltages VL, generator reactive power outputs QG, and transmission
Specifically, the proposed GQ-GA algorithm aims to determine
line loadings SL, and u is the vector of the control variables consist-
the optimal settings of control variables, such as voltage magni-
ing of generator voltages VG, transformer tap settings t, and shunt
tudes, transformer taps and shunt VAR compensation devices con-
VAR compensations QC.
sidered as q-chromosomes of GQ-GA for two optimization
The equality constraints of both optimization problems are typ-
problems, namely minimization of (a) real power losses in trans-
ical load flow equations as follows:
mission lines and (b) sum of voltage deviations on load busses. Re-
sults of GQ-GA on the networks of IEEE 30-bus and 118-bus system PGi  P Di  fPi ðx; uÞ ¼ 0; ð5Þ
are compared to those given by other evolutionary computational Q Gi  Q Di  fQ i ðx; uÞ ¼ 0; ð6Þ
techniques such as (a) the enhanced GA (Bakirtzis, Biskas, Zoumas,
& Petridis, 2002) (next called classical GA), (b) multi-objective EA where fPi and fQ i are the real and reactive power flow equations at
(Abido & Bakhashwain, 2005), (c) hybrid H-PSO (Esmin et al., bus-i, respectively; P Gi and Q Gi are the generator real and reactive
2005), (d) global variant (PSO-PC) based on passive congregation power at bus-i, respectively; PDi and Q Di are the load real and reac-
(Vlachogiannis, 2006), (e) local variant (CLONEPAC) PSO based on tive power at bus-i, respectively.
passive congregation (Vlachogiannis, 2006) and (f) PSO based on The inequality constraints in both problems represent the sys-
coordinated aggregation (CA) (Vlachogiannis & Lee, 2006a, tem operating constraints:
2006b), as well as classical primal-dual interior-point OPF algo-
rithm (De Souza, Honorio, Torres, & Lambert-Torres, 2004). The  Generation constraints: Generator voltages VG and reactive power
comparison demonstrates the superior performance of GQ-GA in outputs QG are restricted by their limits as follows:
finding more optimal solutions.
The paper is organized as follows: the problems of reactive V min max
Gi 6 V Gi 6 V Gi ; i ¼ 1; 2 . . . ; NG; ð7Þ
power and voltage control are formulated in Section 2. Section 3 Q min 6 Q Gi 6 Q max i ¼ 1; 2; . . . ; NG; ð8Þ
Gi Gi ;
presents the basic concept of quantum computing. Section 4 intro-
duces the GQ-GA algorithm. Performance evaluation of GQ-GA in where NG is the number of generators.
comparison with the other evolutionary computational and classi-
cal algorithms is presented in Section 5. Final conclusions and fur-  Switchable VAR constraints: Switchable VAR compensations QC
ther research are outlined in Section 6. are restricted by their limits as follows:

Q min max
Ci 6 Q Ci 6 Q Ci ; i ¼ 1; 2; . . . ; NC; ð9Þ
2. Reactive power and voltage control
where NC is the number of switchable VAR sources.
The proposed GQ-GA is tested and compared with other EA and
conventional OPF algorithms on optimal steady state performance  Transformer constraints: Transformer tap settings t are bounded
of power systems in terms of minimization of (a) power losses in as follows:
transmission lines and (b) sum of voltage deviations on load busses
while satisfying several equality and inequality constraints. Since t min
i 6 ti 6 t max
i ; i ¼ 1; 2; . . . ; NT; ð10Þ
the main focus of this paper is the performance evaluation of the where NT is the number of transformers.
first introduced GQ-GA, two nonlinear optimization problems are
separately studied. It is noticeable that in the case of minimization  Functional operating constraints: This term refers to the con-
of sum of voltage deviations, the objective function is very sensi- straints of load voltages at load buses VL and transmission line
tive to the control variables. Thus, a clearer picture of the effective- loadings SL as follows:
ness of the proposed algorithm is given.
The first objective is to minimize the real power losses in trans- V min
Li 6 V Li 6 V max
Li ; i ¼ 1; 2; . . . ; Nd; ð11Þ
mission lines that can be expressed as
SLi 6 Smax
Li ; i ¼ 1; 2; . . . ; Nl: ð12Þ
X
Nl
The inequality constraints (8), (11) and (12) are included in the
J 1 ¼ PLoss ðx; uÞ ¼ Pl ; ð1Þ
l¼1
objective functions (1) and (2) as penalty factors.

where x is the vector of depended variables, u is the vector of con-
trol variables, Pl is the real power losses at line-l and Nl is the num-
ber of transmission lines.
The second objective is to optimize the voltage profile of the
power system. This is realized by minimization of the sum of volt-
age deviations at load buses that can be expressed by
3. Quantum computing concept
The basic concept of quantum computing is addressed in this
section (Han & Kim, 2000, 2002, 2004):
The smallest unit of information stored in a two-state quantum
computer is called a quantum bit or qubit. A qubit may be in the
6120 J.G. Vlachogiannis, J. Østergaard / Expert Systems with Applications 36 (2009) 6118–6126

‘‘1” state, in the ‘‘0” state, or in any superposition of the two (Hey, where A00. . .0, A00. . .1, . . . , A01. . .1, . . . , A11. . .1 are defined as
1999). The state of a qubit can be represented as (Fig. 1):
A00...0 ¼ a1 a2 . . . am ; A00...1 ¼ a1 a2 . . . bm ; . . . ; A01...0
jSi ¼ aj0i þ bj1i; ð13Þ
¼ a1 b2 . . . am ; . . . ; A11...1 ¼ b1 b2 . . . bm : ð18Þ
where a and b are complex numbers that specify the probability
amplitudes of the corresponding states and therefore are called
amplitude amplifications (Hey, 1999). The ja(t)j2 gives the probabil- 4. General quantum genetic algorithm
ity that the qubit will be found in the ‘‘0” state and jb(t)j2 gives the
probability that the qubit will be found in the ‘‘1” state. The state of In this section, the GQ-GA for reactive power and voltage con-
a qubit can be changed by the operation of a quantum gate. A quan- trol problems is performed. However, the presented GQ-GA can
tum gate is a reversible gate and can be represented as a unitary be applied in various other combinatorial optimization problems.
operator U acting on the qubit basis states satisfying U+U = UU+, The steps of the proposed GQ-GA are
where U+ is the Hermitian adjoint of U. There are several quantum
gates, such as the NOT gate, controlled NOT gate, rotation gate, Had- Step 1: Set t = 0.
amard gate, etc. (Hey, 1999). If there is a system of qubits, the sys- Step 2: Translate each control variable to a q-chromosome. It is
tem can represent many states at the same time. However, in the formed as shown in Fig. 1. If m binary bits are needed to
act of observing a quantum state, it collapses to a single state. A encode a q-chromosome, we will use a register of m-
number of different representations can be used to encode the solu- genes to represent all the possible states 2m of the quan-
tions onto individuals in evolutionary computation. The representa- tum machine, namely the possible states in the feasible
tions can be classified broadly as: binary, numeric, and symbolic. region of the optimization problem. In this study, there
The proposed GQ-GA uses the binary representation, adopted by are three regions in each q-chromosome (Fig. 2), one
Han and Kim (2004, 2002, 2000), called q-gene in this study, for for each set of control variables: (a) Gn-vector of gener-
the probabilistic representation of the smallest unit of information ation PV bus voltage magnitudes, VG, (b) tn-vector of
and a q-chromosome as a string of q-genes, which are defined transformer tap settings, t and (c) Q C n -vector of shunt
below: VAR compensations QC. Encoding is performed using dif-
A q-gene is defined as the smallest unit of information, and is ferent q-gene lengths for each set of control variables
defined with a pair of complex numbers (a, b) as depending on the desired accuracy (Fig. 2).
Step 3: Initialize a population of n q-chromosomes:
½a; bT ; ð14Þ
2 2
where jaj or jbj gives the probability that the q-gene will be found
in ‘‘0” or ‘‘1” state, respectively. Normalization of the state to unity Q ð0Þ ¼ fq1 ð0Þ; q2 ð0Þ; . . . ; qn ð0Þg: ð19Þ
guarantees that: jaj2 + jbj2 = 1.
Each q-chromosome with length m (m q-genes) is represented as
A q-gene may be in the ‘‘1” state, in the ‘‘0” state, or in a linear
    
superposition of the two (Fig. 1). The q-chromosome as a string of  a1 ð0Þ  a2 ð0Þ   am ð0Þ 
qi ð0Þ ¼   ...  ði ¼ 1; 2; . . . ; nÞ; ð20Þ
q-genes is defined as b ð0Þ  b ð0Þ   b ð0Þ 
1 2 m
    
a1  a2   am where aj(0), bj(0) are numbers representing amplitude amplifica-
   ; ð15Þ
b1  b2   bm tions of fundamental q-genes ‘‘0” or ‘‘1”, respectively, and jaj(0)j2
jaj j2 þ jbj j2 ¼ 1 ðj ¼ 1; 2 . . . ; mÞ: ð16Þ or jbj(0)j2 gives the probability the q-gene to be found in ‘‘0” or
‘‘1” state (16).
The representation of q-chromosome has the advantage of repre-
senting a linear superposition of all possible states. In general, the Step 4: For all q-chromosomes qi(0) (i = 1, 2, . . . , n) initialize the
state of a q-chromosome can be represented as amplitude amplifications at
jSi ¼ A00...0 j00 . . . 0i þ A00...1 j00 . . . 1i þ    þ A01...0 j01 . . . 0i    
 aj ð0Þ   r j 
þ    þ A11...1 j11 . . . 1i; ð17Þ    qffiffiffiffiffiffiffiffiffiffiffiffiffi 
 ¼ ðj ¼ 1; 2; . . . ; mÞ: ð21Þ
 bj ð0Þ   1  r 2j 

The random numbers (rj) could follow the normal distribution

N(0, 1) as Han and Kim proposed in (Han & Kim, 2000, 2002,
2004). These numbers next will be compared with the possibilities
ja jj2 or jbjj2 causing the collapse of quantum states to a single state
in classical computers. One main contribution of this paper is that
the normal distribution that random numbers (rj) follow (Han &
Kim, 2000, 2002, 2004; Narayanan & Moore, 1996; Wang et al.,
2005; Zhang et al., 2004) is modified incorporating knowledge we
possibly have gain for the specific problem. This is discussed in
the following step.

Step 5: Make randomly a set of binary solutions P(0) by observing

Q(0) states:

Pð0Þ ¼ fp1 ð0Þ; p2 ð0Þ; . . . ; pn ð0Þg; ð22Þ

where pi(0) is formed by selecting each fundamental q-gene:

Fig. 1. Basic quantum-bit (q-gene). pl ð0Þ ¼ ðp1 p2 . . . pm Þ ðl ¼ 1; 2; . . . 2m Þ: ð23Þ

 J.G. Vlachogiannis, J. Østergaard / Expert Systems with Applications 36 (2009) 6118–6126 6121

Fig. 2. Q-chromosome for reactive power and voltage control.

In a quantum computer, in the act of observing a quantum state, it Step 8.2.1: For each pj(t) generate a random number (rj) using
collapses to a single state. However, in a classical computer the (24).
observed states are produced using the probability of ja jj2 or jbjj2 Step 8.2.2: If rj < [max(0, bj(0))]2 set pj(t) at value of ‘‘1” else ‘‘0”.
as follows: Step 8.2.3: Evaluate each pi(t) (i = 1, 2, . . . , n) running load flow
(5),(6) and calculating the objective function either
For each pj generate a random number (rj) if rj < [max(0, bj(0))]2 (1), (2).
set pj at value of ‘‘1” else ‘‘0”. Obviously, the negative amplitude Step 8.3: Store the global best solution f(b) and the best q-chro-
amplifications (bj(0) < 0) are further ignored for the exploration mosome (b) among pi(t) (i = 1, 2, . . . , n).
of search space. This occurs since negative amplitude amplifica- Step 8.4: Update each q-chromosome qi(t) using the quantum
tions are produced mainly by observations pi(0) with objective rotation gate Ri as follows:
function values far from the global best achieved.
Let us assume that we have a little bit of information about the
search space of problem to be explored. Then, we can see that this " # " #
prior knowledge can be easily put into the initial values of q-genes.
aj ðt þ 1Þ aj ðtÞ
¼ Ri ðtÞ  ðj ¼ 1; 2; . . . ; mÞ; ð25Þ
Following the sensitivity analysis introduced in (Vlachogiannis & bj ðt þ 1Þ bj ðtÞ
Lee, 2005) in the examined problems, it is concluded that the  
cosDhi sinDhi
knowledge of search space is better explored/exploited if random Ri ðtÞ ¼ ði ¼ 1; 2; . . . ; nÞ: ð26Þ
sinDhi cosDhi
numbers (rj) follow the empirical formula of normal distribution,
which is powered by ln(n). Where n is the number of q-chromo- The rotation gate Ri is used as a Q-gate in GQ-GA. Dhi is a rotation
somes. The empirical possibility distribution formula increases angle (Fig. 1) of each q-chromosome toward either j0i or j1i state
the possibility q-genes to be initialized/collapsed at value of ‘‘1”. depending on its sign. The rotation angle Dhi is related to the nor-
In this way, the area of search space with high values of control malized difference Dfi between achievements of each one of q-chro-
(decision) variables can be better explored/exploited. In other mosomes and the global best optimum:
words, in this study, the global optimum solution is ‘‘hidden” near
Dhi ¼ Dfi  signðab  aj Þ  sign½bj  sin Dfi  aj  ð1  cos Dfi Þ;
the area of search space, where there are high values of control
variables than the rest space. Indeed, for instance large amounts ð27Þ
of voltage magnitudes and dispersed shunt VAR compensation in- where
crease the reactive power supply in the system, resulting in the re-  
lief of generators and so the total real power losses and sum of f ðbÞ
Dfi ¼ p  1  ði ¼ 1; 2; . . . ; nÞ; ð28Þ
voltage deviations are reduced. Consequently, in this study, the f ðpi ðtÞÞ

random numbers which drive the collapse of quantum states of þ1 if ab P aj ;
signðab  aj Þ ¼ ð29Þ
q-chromosomes in a classical computer are calculated by the formula 1 if ab < aj ;
r j ¼ Nð0; 1ÞlnðnÞ : ð24Þ sign½bj  sin Dfi  aj  ð1  cos Dfi Þ
(
þ1 if bj  sin Dfi P aj  ð1  cos Dfi Þ ð30Þ
It is remarkable that the proposed possibility distribution (24) fol- ¼
lowed by the collapsed q-genes in this study has similar shape as 1 if bj  sin Dfi < aj  ð1  cos Dfi Þ:
the quantum Plank radiation formulae followed by the density of
The product of two sign functions in (27) indicates the direction of
neutrons in the earth’s gravitational field (Barenco et al., 1995; Cyb-
rotation Dhi so as aj(t + 1) to be more closed to ab(t). Specifically, the
enco, 2001; Nesvizhevsky et al., 2005).
first sign takes into account the difference between aj(t) and ab (t)
and the second one the location of q-chromosome j (Fig. 1). If it is
Step 6: Evaluate each pi(0) (i = 1, 2, . . . , n) running load flow (5),(6)
located in the first/third quadrant the sign is set to positive (+) else
and calculating the objective function either (1), (2).
to negative (). These features gives to the proposed GQ-GA the
Step 7: Store the global best solution f(b) and the best q-chromo-
possibility of a general EA applied in any combinatorial optimiza-
some (b) among pi(0) (i = 1, 2, . . . , n). .
tion problem. Then, it can be specialized on each of them choosing
Step 8: While not termination criterion (Maximum number of
only the appropriate possibility distribution (24) guiding the q-
iterations = 200) do.
genes to collapse in the classical computers.
Begin
5. Performance evaluation
Step 8.1: t = t + 1.
Step 8.2: Make P(t) solutions by observing Q(t  1) states as The results of GQ-GA in optimal steady-state performance on
follows: IEEE 30-bus and 118-bus systems is compared with those given
6122 J.G. Vlachogiannis, J. Østergaard / Expert Systems with Applications 36 (2009) 6118–6126

by classical GA (Bakirtzis et al., 2002), multi-objective EA (Abido &
Bakhashwain, 2005), hybrid H-PSO (Esmin et al., 2005), global
(PSO-PC) and local variant (CLONEPAC) based on passive congrega-
tion PSO algorithms (Vlachogiannis, 2006), the PSO algorithm
based on coordinated aggregation (CA) (Esmin et al., 2005; Vlacho-
giannis & Lee, 2006b) and the classical primal-dual interior-point
OPF algorithm (IP-OPF) (De Souza et al., 2004). Specifically, all
above algorithms, in shortness called competitors, need to handle
two optimization problems, namely minimization of real power
losses in transmission lines (1) and sum of voltage deviations on
load busses (2). The objective functions (1) and (2) are high penal-
ized (penalty factor: 1000) when the operating constraints (8), (11)
and (12) are violated. The inequality constraints (7), (9) and (10)
are satisfied by encoding the control variables in q-chromosomes.
The criterion for the comparison is the achievement of global opti-
mum solution in the shortest computing time.
(Esmin et al., 2005; Vlachogiannis & Lee, 2006a, 2006b; Vlacho-
giannis, 2006) are those, which lead them faster in the best
achievement (see Tables 1 and 2 for IEEE 30-bus and 118-bus sys-
tem, respectively). In both case studies, the number of particles is
set at value of 30 for all PSO competitors.
All above stochastic parameters were selected by means of sen-
sitivity analysis tables. Sensitivity analysis was performed with the
parameters in Tables 1 and 2 following a previous work (Vlacho-
giannis & Lee, 2005). The average and minimum of objective func-
tions was estimated in up to 1500 iterations in 100 trials for each
competitor. The maximum number of iterations of PSO competi-
tors is set at value of 200. It is emphasized that in both case studies
PSO competitors consider the control variables as continuous. Fi-
nally, for comparison purposes a classical IP-OPF algorithm was
developed based on study (De Souza et al., 2004) for IEEE 30-bus
and 118-bus systems.

5.1. Case study 5.3. Results

The topology and the complete data of the IEEE 30-bus and 118-
bus systems can be found in http://www.ee.washington.edu/re-
search/pstca/pf30/pg_tca30bus.htm and http://www.ee.washing-
ton.edu/research/pstca/pf118/pg_tca118bus.htm respectively. The
network of IEEE 30-bus consists of 6 generators, 41 lines, 4 trans-
formers and 2 capacitor banks. In the transformer tests, tap set-
tings are considered within the interval [0.9, 1.1]. The available
reactive powers of capacitor banks are within the interval [0, 30]
MVAr and they are connected to busses 10 and 24. Voltages are
considered within the range of [0.95, 1.1]. In this case the search
space has 12 dimensions, namely the 6 generator voltages, 4 trans-
former taps and 2 capacitor banks. In the case of IEEE 118-bus sys-
tem the decision space has 75 dimensions. The network consists of
54 generators, 9 transformers, 12 capacitor banks and 186 lines. In
the transformer tests, tap settings are considered within the inter-
val [0.9, 1.1]. The available reactive powers of capacitor banks are
within the range of [0, 30] MVAr. Voltages are considered within
the range of [0.95, 1.1].
In the case of IEEE 30-bus system, the GQ-GA converges in 48
iterations achieving the least real power loss of 5.0402 MW
(Fig. 3). The total CPU time is 1.922 s (Table 3). The best case of clas-
sical GA achieves 5.0943 MW. It converges in 106 iterations and the
total CPU time is 11.36 s. The PSO competitors converge between 17
and 56 iterations and their final real power losses range between
5.0921 MW (CA) and 5.0938 MW (H-PSO). The total CPU time of
PSO competitors ranges between 0.858 s (PSO-PC) and 2.24 s (H-
PSO). The multi-objective EA (Abido & Bakhashwain, 2005) conver-
gences in about 70 iterations and its optimum objective value is
5.1065 MW. There is no information about the execution time of
multi-objective EA on the reactive power control problem (Abido
& Bakhashwain, 2005). The classical IP-OPF achieves the highest
real power loss of 5.1129 MW in a total CPU time of 0.636 s. The fi-
nal settings of decision variables proposed by competitors for this
case study are given in Table 3. For comparison purposes this table
also presents the initial settings of decision variables.

Table 1
5.2. Parameters of competitors
Optimum values of stochastic parameters used by PSO competitors for IEEE 30-bus
system
In the case of IEEE 30-bus system, for both GQ-GA and classical
Stochastic parameter H- PSO- CLONEPAC CA
GA (Bakirtzis et al., 2002) the binary length of each control variable
PSO PC
is set at: 11 for each one of six generator voltages; 10 for each one
of four transformer taps and 11 for each one of two capacitor Social parameter (c1) 0.55 0.60 0.60 –
Cognitive parameter (c2) 0.55 0.60 0.60 –
banks. Since the binary length of voltage magnitudes at PV busses Passive congregation parameter (c3) – 0.50 0.40 –
is Gn = 66, transformers tap settings is tn = 40 and shunt VAR com- Lower limit of inertia weighting factor 0.1 0.7 0.4 0.1
pensations is Q C n ¼ 22, the total length of each q-chromosome is (wmin)
128 (Fig. 2). In other words, one q-chromosome consists of 128 Upper limit of inertia weighting factor 1.0 0.9 0.9 1.0
(wmax)
q-genes and can represent simultaneously 2128 binary states. The
Number of search intervals (Nr) 15 15 15 15
population of classical chromosomes and q-chromosomes is set Number of particles (N) 30 30 30 30
at value of 80 and 30, respectively, and the maximum number of
generations is 200.
In the case of IEEE 118-bus system, for both GQ-GA and classical
GA (Bakirtzis et al., 2002) the binary length of each control variable Table 2
Optimum values of stochastic parameters used by PSO competitors for IEEE 118-bus
is set at: 14 for each one of 54 generator voltages; 12 for each one
system
of 9 transformer taps and 12 for each one of 12 capacitor banks.
Since the binary length of voltage magnitudes at PV busses is Stochastic parameter H- PSO- CLONEPAC CA
PSO PC
Gn = 756, transformers tap settings is tn = 108 and shunt VAR com-
pensations is Q C n ¼ 144, the total length of each q-chromosome is Social parameter (c1) 0.60 0.55 0.55 –
Cognitive parameter (c2) 0.60 0.55 0.55 –
1008 (Fig. 2). In other words, one q-chromosome consists of 1008
Passive congregation parameter (c3) – 0.30 0.50 –
q-genes and can represent simultaneously 21008 binary states. The Lower limit of inertia weighting factor 0.1 0.1 0.1 0.1
population of classical chromosomes and q-chromosomes is set at (wmin)
value of 100 and 80, respectively, and the maximum number of Upper limit of inertia weighting factor 1.0 0.9 0.9 1.0
generations is 200. (wmax)
Number of search intervals (Nr) 20 20 20 20
The results of multi-objective EA are presented as those given in
Number of particles (N) 30 30 30 30
(Abido & Bakhashwain, 2005). The parameters of PSO competitors
 J.G. Vlachogiannis, J. Østergaard / Expert Systems with Applications 36 (2009) 6118–6126 6123

Furthermore, the feasibility of GQ-GA in voltage control of IEEE

30-bus system is tested. In this case, the GQ-GA converges in 67
iterations achieving the lowest sum of voltage deviation (VD) of
0.0792 pu (Fig. 4). The total CPU time is calculated at 2.682 s (Table
4). The best case of classical GA achieves 0.1352 pu. It converges in
124 iterations and the total CPU time is 13.33 sec. The PSO compet-
itors converge between 15 and 74 iterations and their final sum of
voltage deviations range between 0.1245 pu (CA) and 0.1393 pu
(H-PSO). The total CPU time of PSO competitors ranges between
0.76 s (CA) and 3.736 s (CLONEPAC). The multi-objective EA (Abido
& Bakhashwain, 2005) converges in about 110 iterations and its
optimum objective value is 0.1477 pu. There is no information
Fig. 4. Convergence of GQ-GA in voltage control of IEEE 30-bus system.
about the execution time of the last one in voltage control problem
(Abido & Bakhashwain, 2005). The IP-OPF converges in total CPU
time of 0.89 s achieving the highest sum of voltage deviation of this case study, it is remarkable the difference between the excel-
0.1733 pu. The initial and the final settings of decision variables lent achievement of GQ-GA and those of other competitors.
proposed by all competitors are also given in Table 4. Finally, the feasibility of GQ-GA in voltage control of IEEE 118-
In the case of IEEE 118-bus system, the GQ-GA converges in 118 bus system is tested. In this case, the GQ-GA converges in 134 iter-
iterations achieving the least real power loss of 122.2227 MW (Fig. ations achieving the lowest sum of voltage deviation (VD) of
5). The total CPU time is 12.592 s (Table 5). The classical GA 1.2198 pu (Fig. 6). The total CPU time is calculated at 14.303 s (Ta-
achieves 131.9657 MW. It converges in 143 iterations and the total ble 6). The classical GA achieves 1.2917 pu. It converges in 131 iter-
CPU time is 15.375 s. The H-PSO and PSO-PC converge in 93 and 89 ations and the total CPU time is 14.407 s. The PSO competitors
iterations, achieving real power losses of 131.9146 and converge between 85 and 95 iterations and their final sum of volt-
131.9083 MW, respectively. The total CPU time of the H-PSO and age deviations range between 1.2755 pu (CA) and 1.2976 pu (H-
the PSO-PC are 26.04 and 28.09 s, respectively. The CLONEPAC PSO). The total CPU time of PSO competitors ranges between
and CA converge in 43 and 71 iterations, achieving real power 25.480 s (H-PSO) and 29.984 s (PSO-PC). The mathematical pro-
losses of 131.9010 and 131.8638 MW, respectively. The total CPU gramming based IP-OPF is the fastest algorithm since it converges
time of CLONEPAC and CA are 13.572 and 22.453 s, respectively. within 5.53 s (11 iterations). However, it gives the worst solution
The IP-OPF is the fastest algorithm since it converges within of 1.3023 pu. Due to the space limitation Table 6 gives 15 out of
4.068 s (eight iterations). However, it gives the worst solution of 75 values of decision variables as proposed by competitors. In all
132.1097 MW. Due to the space limitation Table 5 gives 15 out cases the total CPU time is calculated in a 1.4-GHz Pentium-IV PC.
of 75 values of decision variables as proposed by competitors. In
5.4. Comparison

Comparing the results of competitors in reactive power and

Fig. 3. Convergence of GQ-GA in reactive power control of IEEE 30-bus system.
voltage control of IEEE 30-bus and 118 bus systems, it is concluded
that the proposed GQ-GA achieves better solutions in a satisfactory
execution time. Although multi-objective EA and PSO competitors
are continuous optimization techniques, they cannot achieve the
global optimum solution and prematurely converge in a short com-
puting time. The multi-objective EA can not be successfully used to
find solutions in problems having a semi-smooth or non-convex
solution space. The main advantage of the GQ-GA over PSO com-
petitors is that it optimally progresses its population regulating
only one stochastic parameter (24) instead of those in Tables 1
and 2 for PSOs. The GQ-GA compared to the classical GA has much

Table 3
Results of competitors in reactive power control of IEEE 30-bus system

Decision variable, objective, convergence Initial settings EA IP-OPF H-PSO PSO-PC CLONEPAC CA Classical GA GQ-GA
VG1 1.0500 1.050 1.0500 1.0408 1.0027 1.0102 0.9500 1.0499 1.0488
VG2 1.0450 1.044 1.0500 1.0500 1.0339 1.0145 0.9500 1.0405 1.0389
VG5 1.0100 1.024 1.0500 0.9500 1.0484 1.0248 0.9500 1.0174 1.0221
VG8 1.0100 1.026 1.0334 0.9500 1.0107 1.0300 0.9622 1.0245 1.0256
VG11 1.0500 1.093 1.0500 1.0500 1.0610 1.0198 0.9753 1.0499 1.0333
VG13 1.0500 1.085 1.0149 1.0000 0.9762 1.0060 1.0500 1.0492 1.0359
t69 0.9780 1.078 0.9933 1.0329 1.0185 1.0498 0.9966 1.0656 1.0099
t610 0.9690 0.906 1.0593 1.0132 1.0348 0.9832 1.0500 0.9747 1.0194
t412 0.9320 1.007 1.0087 1.0007 0.9972 1.0014 1.0006 0.9993 0.9755
t2728 0.9680 0.959 0.9971 1.0069 1.0056 1.0061 1.0073 1.0650 1.0171
QC10 0.19 0.19 0.15253 0.18938 0.18728 0.17168 0.25000 0.28593 0.25324
QC24 0.043 0.043 0.08926 0.06281 0.06487 0.06221 0.06253 0.04411 0.06902
Real power losses (MW) 5.3786 5.1065 5.1129 5.0938 5.0928 5.0921 5.0933 5.0943 5.0402
Achieved iteration 70 5 56 17 19 18 106 48
Total CPU time (s) 0.636 2.240 0.858 0.959 0.910 11.360 1.922
6124 J.G. Vlachogiannis, J. Østergaard / Expert Systems with Applications 36 (2009) 6118–6126

Table 4
Results of competitors in voltage control of IEEE 30-bus system

Decision variable, objective, convergence Initial settings EA IP-OPF H-PSO PSO-PC CLONEPAC CA Classical GA GQ-GA
VG1 1.0500 1.037 1.0500 1.0500 0.9875 1.0075 1.0050 0.9826 1.0303
VG2 1.0450 1.027 0.9910 1.0500 1.0155 1.0034 0.9500 0.9914 0.9892
VG5 1.0100 1.013 0.9614 0.9500 1.0132 1.0363 1.0500 0.9800 1.0123
VG8 1.0100 1.008 0.9598 0.9500 1.0393 1.0500 1.0500 1.0343 1.0039
VG11 1.0500 1.030 1.0500 1.0500 1.0181 1.0278 1.0021 0.9808 1.0376
VG13 1.0500 1.007 0.9500 1.0156 1.0058 1.0279 1.0279 1.0079 0.9966
t69 0.9780 1.054 0.9973 1.0335 1.0375 1.0254 1.0287 0.9745 1.0409
t610 0.9690 0.907 1.0859 0.9532 0.9527 0.9655 0.9500 1.0212 0.9954
t412 0.9320 0.928 1.0008 0.9941 0.9935 0.9933 0.9929 0.9807 0.9528
t2728 0.9680 0.945 1.0048 1.0222 1.0221 1.0182 1.0248 0.9960 1.0265
QC10 0.19 0.19 0.11072 0.11131 0.11116 0.08120 0.00467 0.13600 0.13908
QC24 0.043 0.043 0.15928 0.00734 0.00000 0.04187 0.00636 0.14479 0.04660
Sum of voltage deviations (pu) 0.4993 0.1477 0.1733 0.1393 0.1356 0.1340 0.1245 0.1352 0.0792
Achieved iteration 110 7 48 40 74 15 124 67
Total CPU time (s) 0.890 1.920 2.019 3.736 0.760 13.330 2.682

Fig. 5. Convergence of GQ-GA in reactive power control of IEEE 118-bus system.

Fig. 6. Convergence of GQ-GA in voltage control of IEEE 118-bus system.

better performance and search capability. The main disadvantage

of classical GA is that it can not represent/evaluate a linear super-
position of binary states, therefore the quality of the provided solu-
tion and the execution time is drastically deteriorated with the
increase of the chromosome length, i.e., the system size and com-
plexity of search space. In addition, the linear superposition of bin-
ary states and the evolutionary mechanism based on quantum
rotation gate in GQ-GA surmounts the main problem of the corre-
lated control variables from which all classical GA suffer. Finally, a
fair comparison of GQ-GA with classical IP-OPF was performed in
these case studies, where their solution spaces are semi-smooth.
The final results indicate that, even in easy spaces for IP-OPF, the
proposed GQ-GA has better search capability. In general, the IP-
OPF algorithm suffers from bad initial condition, termination con-
dition, and optimality criteria, and in most cases is unable to solve
non-linear quadratic objective functions (De Souza et al., 2004).
It should be underlined that meta-heuristic techniques such as
GQ-GA and PSO are more attractive on problems with more com-
plexities, non-smoothness and non-convexities, where classical
GA, multi-objective EA and conventional IP-OPF fail. However,
the main problem of meta-heuristics remains their convergence
time especially in large scale problems, where there is a need of
more agents in order to explore successfully the decision space.

Table 5
Results of competitors in reactive power control of IEEE 118-bus system

Decision variable, objective, convergence IP-OPF H-PSO PSO-PC CLONEPAC CA Classical GA GQ-GA
VG1 0.95500 1.00997 1.07183 1.01687 1.02433 1.02437 1.05457
VG4 0.99800 1.03899 0.99734 1.05270 1.01117 1.02576 1.06346
VG34 0.98400 1.03325 1.03031 1.01955 1.01595 1.02978 1.05595
VG65 1.00500 1.02210 1.02626 1.04786 1.00494 0.99658 1.05437
VG92 0.99000 1.05364 1.02712 1.04645 1.05149 1.01678 1.06934
VG113 0.99300 0.99132 1.07186 0.99906 1.03860 1.03669 1.07394
t85 0.99726 0.99786 0.99561 0.99133 0.99435 0.99090 0.99665
t2625 1.02806 1.09041 1.09048 1.10000 1.10000 1.10000 1.10000
t3017 1.01440 1.01243 1.01258 1.01476 1.01238 1.01576 0.99456
t6869 0.92988 0.92064 0.93099 0.91905 0.93057 0.94667 0.90596
t8180 1.00317 0.95415 0.95940 0.95797 0.95945 0.92437 0.91457
QC34 0.18699 0.18757 0.23130 0.20304 0.26978 0.20564 0.04358
QC48 0.15299 0.14560 0.22609 0.13747 0.10773 0.17568 0.12569
QC74 0.20441 0.17504 0.20603 0.26237 0.30000 0.19700 0.30000
QC110 0.16283 0.14891 0.23381 0.13298 0.15823 0.14234 0.10000
Real power losses (MW) 132.1097 131.9146 131.9083 131.9010 131.8638 131.9657 122.2227
Achieved iteration 8 93 89 43 71 143 118
Total CPU time (s) 4.068 26.040 28.090 13.572 22.453 15.375 12.592
 J.G. Vlachogiannis, J. Østergaard / Expert Systems with Applications 36 (2009) 6118–6126 6125

Table 6
Results of competitors in voltage control of IEEE 118-bus system

Decision variable, objective, convergence IP-OPF H-PSO PSO-PC CLONEPAC CA Classical GA GQ-GA
VG1 1.04292 1.00607 1.04261 1.06480 1.08477 1.08523 1.10000
VG4 0.99948 0.99380 0.95000 1.04374 1.01488 1.01012 1.07569
VG34 1.01500 1.04010 1.05522 1.00416 1.01101 1.03436 1.07235
VG65 0.98968 1.07434 1.07772 1.01818 0.95000 1.06550 0.95000
VG92 0.98540 1.03054 1.05265 1.05754 1.10000 1.06000 1.10000
VG113 0.98000 0.98768 1.02342 0.96913 0.97463 0.98498 0.99236
t8–5 1.01205 0.98273 0.99738 0.97877 0.98634 0.99045 0.99368
t26–25 0.96345 0.94625 0.95184 0.98998 0.92555 0.94479 0.97386
t30–17 0.99290 0.97344 0.96860 0.96746 0.97947 0.98235 0.99723
t68–69 0.94491 0.94614 0.93147 0.91404 0.90000 0.90000 0.90000
t81–80 0.96677 0.94051 0.95095 0.94855 0.94696 0.95978 0.95859
QC34 0.19144 0.21290 0.25216 0.18419 0.23172 0.22448 0.18576
QC48 0.19588 0.30000 0.21317 0.17862 0.21554 0.30568 0.16487
QC74 0.14417 0.10367 0.10721 0.07908 0.00000 0.16026 0.00000
QC110 0.15323 0.12998 0.14607 0.06590 0.25232 0.12868 0.20577
Sum of voltage deviations (pu) 1.3023 1.2976 1.2849 1.2889 1.2755 1.2917 1.2198
Achieved iteration 11 91 95 85 85 131 134
Total CPU time (s) 5.530 25.480 29.984 26.828 26.881 14.407 14.303

Fortunately, the rapid progress of computer systems and especially
the quantum computers will overcome this problem.
6. Conclusions
This paper presented a new quantum computing inspired ge-
netic algorithm for optimal steady-state performance of power
systems. Although the proposed GQ-GA has similar encoding
method with the classical GA, it can actually represent probabilis-
tically much more states. Also its evolutionary mechanism is com-
pletely different and much more effective than classical GA. The
quantum population is progressed by the operation of simple
mathematical arrays (quantum gates). Specifically, the GQ-GA
introduced a general quantum rotation gate applied to any combi-
natorial optimization problem. In addition a prior knowledge for
the specific problem was easily incorporated to the initial condi-
tion of GQ-GA. This resulted in a new possibility distribution that
q-genes collapse. The new distribution has similar shape with
quantum Plank radiation formulae followed by collapsed quantum
states of elementary particles in physics. The GQ-GA was applied in
reactive power and voltage control problems of IEEE 30-bus and
118-bus systems. The results were compared with those obtained
by other state-of-the-art evolutionary computational algorithms
such as enhanced GA, multi-objective EA and PSO as well as the
classical IP-OPF demonstrating the excellent performance of GQ-
GA in finding the global best solution. Further, the proposed GQ-
GA can be easily applied in the forthcoming quantum computers,
where the execution time will be annihilated, since the linear
superposition of quantum states will be simultaneously composed
and collapsed. Also GQ-GA can be modified for application in a con-
ventional computer network and tested in other practical optimi-
zation problems of power systems.
References
Abido, M. A., & Bakhashwain, J. M. (2005). Optimal VAR dispatch using a
multiobjective evolutionary algorithm. International Journal of Electrical Power
and Energy Systems, 27(1), 13–20.
Bakirtzis, A. G., Biskas, P. N., Zoumas, C. E., & Petridis, V. (2002). Optimal power flow
by enhanced genetic algorithm. IEEE Transactions on Power Systems, 17(2),
229–236.
Barenco, A., Bennett, C. H., Cleve, R., Di Vincenzo, D. P., Margolus, N., Shor, P., et al.
(1995). Elementary gates for quantum computation. Physical Review A, 52,
3457–3467.
Beinoff, P. (1980). The computer is a physical system: A microscopic of Hamiltonian
model of computers as presented by Turing machines. Journal of Statistical
Physics, 22, 563–591.
Cybenco, G. (2001). Reducing quantum computations to elementary unitary
operations. IEEE Computers, 27–32.
De Souza, A. C. Z., Honorio, L. M., Torres, G. L., & Lambert-Torres, G. (2004).
Increasing the loadability of power systems through optimal-local-control
actions. IEEE Transactions on Power Systems, 19(1), 188–194.
Dorigo, M. (1992). Optimization, learning and natural algorithms. Ph.D. thesis,
Politecnico de Milano, Italy.
Esmin, A. A. A., Lambert-Torres, G., & De Souza, A. C. Z. (2005). A hybrid particle
swarm optimization applied to loss power minimization. IEEE Transactions on
Power Systems, 20(2), 859–866.
Feynmann, R. P. (1982). Simulating physics with computers. International Journal of
Theoretical Physics, 21, 467–482.
Feynmann, R. P. (1986). Quantum mechanical computers. Foundations of Physics,
16(6), 507–531.
Han, K.-H., & Kim, J.-H. (2000). Genetic quantum algorithm and its application to
combinatorial optimization problem. Proceedings of Congress on Evolutionary
Computation, 1354–1360.
Han, K.-H., & Kim, J.-H. (2002). Quantum-inspired evolutionary algorithm for a class
of combinatorial optimization. IEEE Transactions on Evolutionary Computation,
6(6), 580–593.
Han, K.-H., & Kim, J.-H. (2004). Quantum-inspired evolutionary algorithms with a
new termination criterion, He Gate, and two-phase scheme. IEEE Transactions on
Evolutionary Computation, 8(2), 156–169.
Hey, T. (1999). Quantum computing: An introduction. Computing and Control
Engineering Journal, 10(3), 105–112.
Hogg, T., & Portnov, D. (2000). Quantum optimization. Information Science, 128,
181–197.
Kennedy, J., & Eberhart, R. (1995). Particle swarm optimization. Proceedings IEEE
International Conference on Neural Networks, 1942–1948.
Lee, K. Y. (2005). Tutorial on intelligent optimization and control of power systems.
In Proceedings the 13th international conference on intelligent systems application
to power systems (ISAP), Arlington, VA, USA.
Lee, K. Y., & El-Sharkawi, M. A. (2002). Modern heuristics optimization techniques
with applications to power systems. In Proceedings IEEE PES general meeting,
Tutorial Paper 02TP160.
Lee, K. Y., & El-Sharkawi, M. A. (2003). A tutorial course on evolutionary
computation techniques for power system optimization. In Proceedings IFAC
symposium on power plants and power system control, Seoul, Korea.
Malossini, A., Blanzieri, E., & Calarco, T. (2004). QGA: A quantum genetic algorithm.
Technical Report DIT-04-105, Trento, Italy <http://www.dit.unitn.it> (accessed
January 2008).
Narayanan, A., & Moore, M. (1996). Quantum-inspired genetic algorithms.
Proceedings of IEEE International Conference on Evolutionary Computation, 61–66.
Nesvizhevsky, V. V., Petukhov, A. K., Borner, H. G., Baranova, T. A., Gagarski, A. M.,
Petrov, G. A., et al. (2005). Investigation of the neutron quantum states in the
earth’s gravitational field. Journal of Research of the National Institute of
Standards and Technology, 110(3), 263–267.
Reynolds, R. G. (1994). An introduction to cultural algorithms. Proceedings of the third
annual conference on evolutionary programming. River Edge, NJ: World Scientific.
pp. 131–139.
Rylander, B., Soule, T., Foster, J. A., & Alves-Foss, J. (2001). Quantum evolutionary
programming. In Proceedings of international conference on genetic and
evolutionary computation (GECCO) (pp. 1005–1011). Morgan Kaufman.
Spector, L., Barnum, H., Bernstein, H. J., & Swamy, N. (1999). Finding a better-than-
classical quantum AND/OR algorithm using genetic programming. Proceedings
of Congress on Evolutionary Computation, 2239–2246.
Vlachogiannis, J. G. (2006). Constricted local neighborhood particle swarm
optimization with passive congregation for reactive power and voltage
control. Electric Power Components and Systems, 34(6), 509–520.
6126 J.G. Vlachogiannis, J. Østergaard / Expert Systems with Applications 36 (2009) 6118–6126
Vlachogiannis, J. G., & Lee, K. Y. (2006b). Coordinated aggregation particle swarm
optimization applied in reactive power and voltage control. In Proceedings IEEE
PES general meeting, Montreal Canada, paper 06GM0780.
Vlachogiannis, J. G., & Lee, K. Y. (2005). Contribution of generation to transmission
system using parallel vector evaluated particle swarm optimization. IEEE
Transactions on Power Systems, 20(4), 1765–1774.
Vlachogiannis, J. G., & Lee, K. Y. (2006a). A comparative study on particle swarm
optimization for optimal steady-state performance of power systems. IEEE
Transactions on Power Systems, 21(4), 1718–1728.
Wang, L., Tang, F., & Wu, H. (2005). Hybrid genetic algorithm based on quantum
computing for numerical optimization and parameter estimation. Applied
Mathematics and Computation, 171, 1141–1156.
<http://www.ee.washington.edu/research/pstca/pf30/pg_tca30bus.htm> (accessed
January 2008).
<http://www.ee.washington.edu/research/pstca/pf118/pg_tca118bus.htm>
(accessed January 2008).
Zhang, G. X., Li, N., Jin, W. D., & Hu, L. Z. (2004). A novel quantum genetic algorithm
and its application. ACTA Electronica Sinica, 32, 476–479.