6

A Simple Hybrid Particle Swarm Optimization

Wei-Chang Yeh
Department of Industrial Engineering and Engineering Management
National Tsing Hua University
Taiwan, R.O.C.

Open Access Database www.i-techonline.com

1. Introduction
As a novel stochastic optimization technique, the Particle Swarm Optimization technique
(PSO) has gained much attention towards several applications during the past decade for
solving the global optimization problem or to set up a good approximate solution to the
given problem with a high probability. PSO was first introduced by Eberhart and Kennedy
[Kennedy and Eberhart, 1997]. It belongs to the category of Swarm Intelligence methods
inspired by the metaphor of social interaction and communication such as bird ocking and
sh schooling. It is also associated with wide categories of evolutionary algorithms through
individual improvement along with population cooperation and competition. Since PSO
was rst introduced to optimize various continuous nonlinear functions, it has been
successfully applied to a wide range of applications owing to the inherent simplicity of the
concept, easy implementation and quick convergence [Trelea 2003].
PSO is initialized with a population of random solutions. Each individual is assigned with a
randomized velocity based to its own and the companions ying experiences, and the
individuals, called particles, are then own through hyperspace. PSO leads to an effective
combination of partial solutions in other particles and speedens the search procedure at an
early stage in the generation. To apply PSO, several parameters including the population
(N), cognition learning factor (cp), social learning factor (cg), inertia weight (w), and the
number of iterations (T) or CPU time should be properly determined. Updating the velocity
and positions are the most important parts of PSO as they play a vital role in exchanging
information among particles. The details will be given in the following sections.
The simple PSO often suffers from the problem of being trapped in local optima. So, in this
this paper, the PSO is revised with a simple adaptive inertia weight factor, proposed to
efficiently control the global search and convergence to the global best solution. Moreover, a
local search method is incorporated into PSO to construct a hybrid PSO (HPSO), where the
parallel population-based evolutionary searching ability of PSO and local searching
behavior are reasonably combined. Simulation results and comparisons demonstrate the
effectiveness and efficiency of the proposed HPSO.
The paper is organized as follows. Section 2 describes the acronyms and notations. Section 3
outlines the proposed method in detail. In Section 4, the methodology of the proposed
HPSO is discussed. Numerical simulations and comparisons are provided in Section 5.
Finally, Concluding remarks and directions for future work are given in in Section 6.
Source: Advances in Evolutionary Algorithms, Book edited by: Witold Kosiski, ISBN 978-953-7619-11-4, pp. 468, November 2008,
I-Tech Education and Publishing, Vienna, Austria

www.intechopen.com

118

Advances in Evolutionary Algorithms

2. Acronym and notations

Acronym:
PSO : Particle Swarm Optimization Algorithm
SPSO : Traditional PSO
IPSO : An improved PSO proposed in [Jiang et. al. 2007]
HPSO : The proposed Hybrid PSO
Notations:
D : The number of dimensions.
N : The number of particles in each replication.
T : The number of generations in each replication.
R : The total number of independent replications.
r : The random number uniformly distributed in [0, 1].
cp, cg : The cognition learning factor and the social learning factor, respectively.
w : The inertia weight.
xt,i,j : The dimension of the position of particle i at iteration t, where t=1,2,,T,
i=1,2,,N, and j=1,2,,D.
Xt,i : Xt,i=(xt,i,1,,xt,i,D) is the position of particle i at iteration t, where t=1,2,,T,
i=1,2,,N, and j=1,2,,D.
vt,i,j : the dimension of the velocity of particle i at iteration t, where t=1,2,,T,
i=1,2,,N, and j=1,2,,D.
Vt,i : Vt,i=(vt,i,1,,vt,i,D) is the velocity of particle i at iteration t, where t=1,2,,T,
i=1,2,,N, and j=1,2,,D.
Pt,i : Pt,i=(pt,i,1,,pt,i,D) is the best solution of particle i so far until iteration t, i.e., the
pBest, where t=1,2,,T, i=1,2,,N, and j=1,2,,D.
Gt : Gt=(gt,1,,gt,D) the best solution among Pt,1,Pt,2,,Pt,N at iteration t, i.e., the gBest,
where t=1,2,,T.
F() : The fitness function value of .
U(),L() : The upper and lower bounds for , respectively.

3. The PSO
In PSO, a solution is encoded as a finite-length string called a particle. All of the particles
have tness values which are evaluated by the tness function to be optimized, and have
velocities which direct the ying of the particles [Parsopoulos et. al. 2001]. PSO is initialized
with a population of random particles with random positions and velocities inside the
problem space, and then searches for optima by updating generations. It combines the local
and global search resulting in high search efficiency. Each particle moves towards its best
previous position and towards the best particle in the whole swarm in every iteration. The
former is a local best and its value is called pBest, and the latter is a global best and its value
is called gBest in the literature. After nding the two best values, the particle updates its
velocity and position with the following equation in continuous PSO:

www.intechopen.com

vt,i,j = wvt-1,i,j+cprp,i,j(pt-1,i,j-xt-1,i,j)+cprp,i,j(gt-1,j-xt-1,i,j),

(1)

xt,i,j=xt-1,i,j+vt,i,j.

(2)

119

A Simple Hybrid Particle Swarm Optimization

The values cprp,i,j and cgrg,i,j determine the weights of the two parts, and cp+cg is usually
limited to 4 [Trelea 2003]. Generally, the value of each component in Xt,i and Vt,i can be
clamped to the range [L(X), U(X)] and [L(V), U(V)], respectively, to limit each particle to
ensure its feasibility.
For example, let
X3,4=(1.5, 3.6, 3.7, -3.4, -1.9),

(3)

V2,4=(0.4, 0.1, 0.7, -2.7, -3.5),

(4)

P3,4=(1.6, 3.7, 3.5, -2.1, -1.9),

(5)

G3=(1.7, 3.7, 2.2, -3.5, -2.5),

(6)

Rp=(0.21, 0.58, 0.73, 0.9, 0.16),

(7)

Rg=(0.47, 0.45, 0.28, 0.05, 0.77),

(8)

L(X)= (0, 0, 0, -3.6, -3),

(9)

U(X)=(2, 4, 4, 0, 0),

(10)

L(V)=( -4, -4, -4, -4, -4),

(11)

U(V)=( 4, 4, 4, 4, 4),

(12)

w=.9,

(13)

cp=cg=2.

(14)

V3,4=(0.59, 0.296, -0.502, -0.1, -4.074).

(15)

Then, from Eq.(1), we have

Since -4.074<-4, V3,4 needs to be adjustmented in the following:

V3,4=(0.59, 0.296, -0.502, -0.1, -4).

(16)

X4,4=(2.09, 3.896, 3.198, -3.5, -5.974),

(17)

X4,4=(2.0, 3.896, 3.198, -3.5, -3.0)

(18)

Under the guidance of Eq.(2),

and

after the adjustment according to the upper/lower-bounds of X.

We conducted the preliminary experiments, and the complete computational procedure of
the PSO algorithm can be summarized as follows.
STEP 1: Initialize: Initialize parameters and population with random positions and
velocities.

www.intechopen.com

120

Advances in Evolutionary Algorithms

STEP 2: Evaluation: Evaluate the fitness value (the desired objective function) for each
particle.
STEP 3: Find the pBest: If the fitness value of particle i is better than its best fitness value
(pBest) in history, then set current fitness value as the new pBest to particle i.
STEP 4: Find the gBest: If any pBest is updated and is better than the current gBest, then set
gBest to the current value.
STEP 5: Update and adjustment velocity: Update velocity according to Eq.(1). Adjust the
velocity to meet its range if necessary.
STEP 6: Update and adjustment position: Update velocity and move to the next position
according to Eq.(2). Adjust the position to meet their range if necessary.
STEP 7: Stopping criterion: If the number of iterations or CPU time are met, then stop;
otherwise go back to STEP 2.

4. The proposed HPSO

To overcome the weakness of PSO for local searches, this paper aims at creating HPSO by
combining PSO, local search (LS), and vector based (VB) with a linearly varying inertia
weight. The PSO part in the proposed HPSO is similar to the SPSO proposed in section 3.
Hence, only the differences, i.e. the linearly varying inertia weight, LS and VB, are
elaborated in this section.
4.1 Initial population
The initial population is generated randomly in the feasible space such that its lower/upper-bounds are satisfied. To construct a direct relationship between the problem domain
and the PSO particles in this study, the ith dimension in the pariticle stands for the value of
the ith variable in the solution.
4.2 The linearly varying inertia weight
One of the most important issues to nd the optimum solution effectively and efficiently
while designing the PSO algorithm is its parameters. The inertia weight represents the
inuence of previous velocity which provides the necessary momentum for particles to
move across the search space. Hence, the inertia weight dictates the balance between
exploration and exploitation in PSO [Jiang et. al. 2007]. Shi and Eberhart (2001) made a
signicant improvement in the performance of the PSO with a linearly varying inertia
weight over the generations, which linearly varies from 0.9 at the beginning of the search to
0.4 at the end. Thus the linearly varying inertia weight is adapted in the proposed HPSO to
achieve trade-off between exploration and exploitation, i.e. the inertia weight of the ith
generation is
wi=U(w)-(i-1)[U(w)-L(w)]/N.

(19)

4.3 Vector based PSO

The underlying principle of the traditional PSO is that the next position of each particle is a
compromise of its current position, the best position in its history so far, and the best
position among all existing particles. The vector synthesis is the original mathematical
foundation of PSO, as shown in the following figure.

www.intechopen.com

121

A Simple Hybrid Particle Swarm Optimization

Vi t 1

Pi t

xit +1

Vi t

C p Rp
Gt

Cg Rg

xit
Fig. 1. The vector synthesis of PSO.
Eqs.(1) and (2) are more complicated than the concept of the vector synthesis in increasing
the diversity of the dimensions of each particle. Hence, the following equations are
implemented in the proposed HPSO instead of Eqs.(1) and (2):
Vt,i=wiVt-1,i+cprp(Pt-1,i-Xt-1,i)+cgrg(Gt-1-Xt-1,i),

(20)

Xt,i=Xt-1,i+Vt,i.

(21)

In the traditional PSO, each particle needs to use two random number vectors (e.g., Rp and
Rq) to move to its next position. However, only two random number (e.g., rp and rq) are
needed in the proposed HPSO. Besides, Eqs(20) and (21) are very easy and efficient in
deciding the next positions for the problems with continuous variables. For example, let P3,4,
X3,4, P3,4, G3, L(X), U(X) are the same as defined in the example in Section 1. Assume rp=0.34
and rg=0.79. From Eq.(19),
wi=0.9-(4-1)[0.9-0.4]/1000=0.8985.

(22)

Plug wi, rp, rg and the other required value into Eq.(20), we have
V3,4=(0.6974, 0.28985, -1.56505, -1.75795, -3.97275),

(23)

V3,4=(0.6974, 0.28985, -1.56505, -1.75795, -3.97275),

(24)

where X4,4 is adjustmented from

(2.1974, 3.88985, 2.13495, -5.15795, -5.87275).

(25)

4.4 Local search method

One of the major drawbacks of PSO is is its very slow convergence. To surmount this
drawback, to guide the search towards unexplored regions in the solution space and to
avoid being trapped into local optimum, LS is implemented for constructing the proposed
HPSO.
In PSO, proper control of global exploration and local exploitation is crucial in nding the
optimum solution efficiently [Liu 2005]. A local optimizer is applied to the best particle (i.e.
gBest) for each run in order to push it to climb the local optimum [Liu 2005]. With the hybrid
method, PSOs are used to perform global exploration around particles except the gBest to
maintain population diversity, while the local optimizer is used to perform local
exploitation to the best particle. Since the properties of PSOs and conventional local
optimizers are complementary, HPSOs are often better than either method operating alone
from the computation exprements shown in Section 5.

www.intechopen.com

122

Advances in Evolutionary Algorithms

The proposed LS is very simple and similar to the famous local improvement method the
pairwise exchange procedure. In LS, the ith dimension of both the current best particle of all
population (i.e., gBest) are replaced by the current best particle of the jth particle (i.e., pBest).
If the fitness function value is improved, the the current gBest is updated accordingly.
Otherwise, there is no need to change the current gBest. The above procedure in the
proposed HPSO is repeated until all dimensions in the gBest are performed.
To minimize the number of duplicated computations of the same fitness function in LS, only
one non-gBest is randomly selected to each dimension of gBest in the local search. The
complete procedure of the local search part of the proposed HPSO can be summarized as in
the following:
STEP 0. Let d=1.
STEP 1. Let n=1.
STEP 2. If Gd=Pt,n or gt,d=pt,n,d, go to STEP 4. Otherwise, let F*=F(Gd), Gd=Pt,n, and gt,d=pt,n,d.
STEP 3. If F(Gd) is better than F*, then let F*=F(Gd). Otherwise, let gt,d=g.
STEP 4. If n<N, let n=n+1 and go to STEP 2.
STEP 5. If d<D, let d=d+1 and go to STEP 1.

5. Numerical examples
To evaluate the performance of the proposed algorithms, four famous benchmark
optimization problems [Jiang et. al. 2007] are used, which are described as follows.
2
2 2
Rosenbrock f1 ( x ) = i =1 100 ( xi +1 xi ) + ( xi 1) [-30,30]n

n
f 2 ( x ) = i =1 xi2 10 cos ( 2 xi ) + 10 [-5.12,5.12]n
Rastrigrin

Function

Griewark

n 1

f3 ( x) =

Formula

x
n
1
n
x 2 i =1 cos( i ) + 1

i =1 i
4000
i

Range

[-600,600]n

Optima Solution
0

(1,,1)

(0,,0)

(0,,0)

Table 1. Benchmark functions.

Features of the above three functions are the following: Rosenbrock is an unimodal function
and its variables are strongly dependent and gradient information often misleads
algorithms; Rastrigin is ultimodal function with many local optima; Griewank is strongly
multi-modal with significant interactions between its variables (caused by the product term)
and the number of local minima increases with dimensionality [Jiang et. al. 2007].
These problems are implemented using the proposed HPSO, SPSO, and the best-known
PSO (IPSO) proposed in by Jiang et. al. (2007) with regard to these three benchmark
problems and the results of the experiments were compared. The proposed HPSO, SPSO
and IPSO were implemented in C programming language on an Intel Pentium 2.6 GHz PC
with 2G memory. To facilitate fair comparison, the number of iterations, the number of runs,
and the populations for the proposed HPSO, SPSO and IPSO are taken directly from Jiang
et. al. (2007). All these methods use a linearly varying inertia weight over the generations,
varying from 0.9 at the beginning of the search to 0.4 at the end. In addition, cg=cp=2,
Xmax=Vmax=UB and Xmin=Vmin=LB are used.
As in Jiang et. al. (2007), the proposed HPSO, SPSO and IPSO were tested on four group
problems (population sizes of 20, 40, 80, and 160). The population sizes of each group are
equal and each group problem contains 3 data sets. Each data set was first run with 3 sets of
dimensions: 10, 20, and 30 and the corresponding maximum number of generations are set

www.intechopen.com

A Simple Hybrid Particle Swarm Optimization

123

as 1000, 1500 and 2000, respectively. Hence, there are 12 different test sets to each
benchmark problem as follows:
Set A1: N=20, D=10, T=1000;
Set A2: N=20, D=20, T=1500;
Set A3: N=20, D=30, T=2000;
Set B1: N=40, D=10, T=1000;
Set B2: N=40, D=20, T=1500;
Set B3: N=40, D=30, T=2000;
Set C1: N=80, D=10, T=1000;
Set C2: N=80, D=20, T=1500;
Set C3: N=80, D=30, T=2000;
Set D1: N=160, D=10, T=1000;
Set D2: N=160, D=20, T=1500;
Set D3: N=160, D=30, T=2000;
Each algorithm with each set of parameter is executed in 50 independent runs. The average
tness values of the best particle found for the 50 runs for the three functions are listed in
Table 2. The shaded number shows the best result with respect to the corresponding
function and the set. From the Table 2, it can be seen that IPSO outperforms the HPSO in
Rosenbrock functions for POP=20 and 80. However, the remaining cases of Rosenbrock
functions and for the Griewark function, the proposed HPSO has almost achieved better
results than SPSO and IPSO. Furthermore HPSO is superior to SPSO and IPSO at all
instances of the Rastrigrin function.
Rosenbrock
PSO
IPSO
HPSO
A1
42.6162 10.5172 3.3025
A2
87.2870 75.7246 124.3305
A3
132.5973 99.8039 122.7829
B1
24.3512 1.2446
0
B2
47.7243 8.7328 0.0797
B3
66.6341 14.7301 120.7434
C1
15.3883 0.1922 0.0797
C2
40.6403 1.5824 60.3717
C3
63.4453 1.5364 4.7461
D1
11.6283 0.0598
0
D2
28.9142 0.4771
0
D3
56.6689 0.4491
0
Average 39.48832 3.22272 20.66896
SET

Rastrigrin
Griewark
PSO
IPSO HPSO PSO
IPSO HPSO
5.2062 3.2928
0
0.0920 0.0784 0.0071
22.7724 16.4137 0.4975 0.0317 0.0236 0.0168
49.2942 35.0189 1.0760 0.0482 0.0165 0.0190
3.5697 2.6162
0
0.0762 0.0648 0.0002
17.2975 14.8894
0
0.0227 0.0182 0.0026
38.9142 27.7637
0
0.0153 0.0151 0.0012
2.3835 1.7054
0
0.0658 0.0594
0
12.9020 7.6689
0
0.0222 0.0091
0
30.0375 13.8827
0
0.0121 0.0004
0
1.4418 0.8001
0
0.0577 0.0507
0
10.0438 4.2799
0
0.0215 0.0048
0
24.5105 11.9521
0
0.0121 0.0010
0
15.67783 9.50649
0
0.03396 0.02483 0.00044

Table 2. Mean Fitness function values 50 independent runs.

The final statistical result including the Success Rate, the fitness function values, CPU times
and convergence iterations of all 50 runs related to Rosenbrock function, Rastrigrin function,
and Griewark function are listed in Tables 3-5. The Success Rate is defined to be the
percentage of the number of nal searching solution that is equal to the global optimal value
in 50 independent runs. Convergence iterations denote the number of iterations required for
convergence. These data are divided into three categories: maximum, minimum, average,
and standard deviations (denoted by max, min, mean, and std., respectively).

www.intechopen.com

124

SET
A1
A2
A3
B1
B2
B3
C1
C2
C3
D1
D2
D3

Advances in Evolutionary Algorithms

Success
Rate
92%
90%
82%
100%
98%
96%
98%
98%
98%
100%
100%
100%

Fitness Function Value

max
min
mean
std
153.17
0
3.3025
21.65
3018.59 0
124.3305 597.18
3018.59 0
122.7829 597.14
0
0
0
0
3.99
0
0.0797
0.56
3018.59 0
120.7434 597.52
3.99
0
0.0797
0.56
3018.59 0
60.3717 426.89
237.31
0
4.7461
33.56
0
0
0
0
0
0
0
0
0
0
0
0

Running Time (sec.)

max min Mean std
0.03 0.02 0.02 0.00
0.12 0.05 0.06 0.01
0.29 0.09 0.11 0.04
0.05 0.04 0.05 0.00
0.25 0.09 0.14 0.03
0.47 0.19 0.23 0.07
0.13 0.07 0.11 0.01
0.46 0.24 0.37 0.04
0.91 0.44 0.66 0.10
0.27 0.21 0.25 0.02
1.00 0.78 0.89 0.05
2.41 1.52 1.78 0.17

Convergence Iterations
max min mean
std
962 520 720.3 73.77
1264 520 1130.3 141.50
1987 914 1603.5 268.45
787 607 660.1 39.34
1129 813 1024.7 64.66
1635 1176 1477.9 114.45
782 556 602.7 37.37
1089 649 946.9 60.43
1457 1217 1346.7 56.36
681 533 567.4 28.35
1048 818 887.3 47.31
1452 1073 1232.0 73.77

Table 3. Experimental results on Rosenbrock function of 50 independent runs.

SET
A1
A2
A3
B1
B2
B3
C1
C2
C3
D1
D2
D3

Success
Rate
100%
98%
96%
100%
100%
100%
100%
100%
100%
100%
100%
100%

Fitness Function Value

max min mean std
0
0
0
0
24.87 0 0.4975 3.52
28.92 0 1.0760 5.34
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0

Running Time (sec.)

max min mean std
0.05 0.04 0.05 0.00
0.16 0.15 0.16 0.00
0.37 0.33 0.34 0.01
0.09 0.08 0.08 0.00
0.27 0.25 0.25 0.00
0.58 0.53 0.55 0.01
0.17 0.15 0.16 0.00
0.49 0.45 0.47 0.01
1.01 0.92 0.96 0.02
0.32 0.30 0.31 0.00
0.92 0.86 0.89 0.01
1.91 1.74 1.79 0.03

Convergence Iterations
max min mean std
112 5
30.3 20.52
329 11 56.1 49.95
357 19 86.4 74.91
58
7
25.4 12.34
104 11 36.4 20.96
196 20 47.7 28.55
56
5
19.3 11.02
106 9
32.9 17.43
127 18 45.2 25.74
38
5
15.6 7.43
63
9
26.4 10.87
97
9
34.7 15.96

Table 4. Experimental results on Rastrigrin function of 50 independent runs.

SET
A1
A2
A3
B1
B2
B3
C1
C2
C3
D1
D2
D3

Success
Rate
86%
74%
76%
98%
96%
96%
100%
100%
100%
100%
100%
100%

Fitness Function Value

max min mean std
0.10 0 0.0071 0.02
0.13 0 0.0186 0.04
0.21 0 0.0190 0.04
0.01 0 0.0002 0.00
0.07 0 0.0026 0.01
0.03 0 0.0012 0.01
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0

Running Time (sec.)

max min mean std
0.08 0.08 0.08 0.00
0.30 0.27 0.28 0.01
0.72 0.64 0.67 0.03
0.14 0.13 0.14 0.00
0.49 0.44 0.45 0.01
1.06 0.99 1.00 0.01
0.26 0.25 0.26 0.00
0.81 0.78 0.80 0.01
1.74 1.67 1.70 0.02
0.51 0.49 0.50 0.00
1.53 1.46 1.49 0.01
3.18 3.04 3.10 0.03

Convergence Iterations
max min mean std
711
6 110.7 169.22
1401 18 262.1 303.67
812 15 275.0 272.40
518
4
53.1 74.79
823 10 90.7 151.76
738 10 95.7 140.38
124
2
20.5 23.07
120 10 40.8 26.87
131
8
48.1 28.07
30
6
14.8 6.09
104
9
29.0 16.99
77
9
32.9 14.44

Table 5. Experimental results on Griewark function of 50 independent runs.

www.intechopen.com

A Simple Hybrid Particle Swarm Optimization

125

As the dimension increases, the solution space get more complex, and PSO algorithm is
more likely to be trapped into local optima. Experimental data shown in Table 2 does not
clearly indicate that the HPSO outperforms the other PSOs in the measures of average
fitness function values. However, the Success Rates are all over 74%. Therefore, the
proposed HPSO can nd global optima with very high probability, and it is concluded that
HPSO has the strongest exploration ability and it is not easy to be trapped into local optima.
Table 3 shows that the proposed HPSO uses only 3.18 seconds in worst case and 0.6 seconds
in average. Thus, HPSO is very effective, efficient, robust, and reliable for complex
numerical optimization.

6. Conclusions
A successful evolutionary algorithm is one with a proper balance between exploration
(searching for good solutions), and exploitation (refining the solutions by combining
information gathered during the exploration phase). In this study, a new hybrid version of
PSO called HPSO is proposed. The HPSO constitutes a vector based PSO method with the
linearly varying inertia weight, along with a local search.
A novel, simpler, and efficient mechanism is employed to move the gBest to its next position
in the proposed HPSO. The HPSO combines the population-based evolutionary searching
ability of PSO and local searching behavior to effciently balance the exploration and
exploitation abilities. The result obtained by HPSO has been compared with those obtained
from traditional simple PSO (SPSO) and improved PSO (IPSO) proposed recently.
Computational results show that the proposed HPSO shows an enhancement in searching
efficiency and improve the searching quality. In summary, the results presented in this work
are encouraging and promising for the application of the proposed HPSO to other complex
problems.
Further analysis is necessary to see how other soft computing method (e.g., the genetic
algorithm, the taboo search, etc.) react to local searches for future researchers who may want
to develop their own heuristics and to make further improvements. Our research is still very
active and under progress, and it opens the avenues for future efforts in this directions such
as: how to adjust parameters, increase success rates, reduce running times, using other local
search, and the aggregation of different and new concepts to PSO.

7. References
B. Liu, L. Wang, Y.-H. Jin, F. Tang, D.-X. Huang (2005), Improved particle swarm
optimization combined with chaos, Chaos Solitons & Fractals, Vol. 25, 2005, pp.
12611271.
I.C. Trelea (2003), The particle swarm optimization algorithm: convergence analysis and
parameter selection, Information Processing Letters, Vol. 85, 2003, pp. 317325.
J. Kennedy and R.C. Eberhard (1995), Particle swarm optimization, Proceedings of IEEE
International Conference on Neural Networks, Piscataway, NJ, USA, 1995, pp. 1942-1948.
J. Kennedy and R.C. Eberhard and Y. Shi, Swarm intelligence, San Francisco, CA: Morgan
Kaufmann; 2001.
J. Kennedy and R.C. Eberhart (1997), A discrete binary version of the particle swarm
algorithm, Systems, Man, and Cybernetics, Computational Cybernetics and
Simulation, IEEE International Conference, Vol. 5, No. 12-15, 1997/10, pp. 4104-4108.

www.intechopen.com

126

Advances in Evolutionary Algorithms

J. Moore and R. Chapman (1999), Application of particle swarm to multiobjective

optimization, Department of Computer Science and Software Engineering, Auburn
University.
K.E. Parsopoulos, V.P. Plagianakos, G.D. Magoulas, M.N. Vrahatis (2001), Improving
particle swarm optimizer by function stretching, Advances in Convex Analysis and
Global Optimization, 2001, 445457.
R.C. Eberhart and Y. Shi (2001), Particle Swarm Optimization: Developments, Application
and Resources, Proceedings of the 2001 Congress on Evolutionary Computation,
Seoul, South Korea, Vol. 1, pp. 81-86.
Y. Jiang, T. Hu, C. Huang, and X. Wu (2007), An improved particle swarm optimization
algorithm, Applied Mathematics and Computation, Vol. 193, pp. 231239.

www.intechopen.com

Advances in Evolutionary Algorithms

Edited by Xiong Zhihui

ISBN 978-953-7619-11-4
Hard cover, 284 pages
Publisher InTech

Published online 01, November, 2008

Published in print edition November, 2008

With the recent trends towards massive data sets and significant computational power, combined with
evolutionary algorithmic advances evolutionary computation is becoming much more relevant to practice. Aim
of the book is to present recent improvements, innovative ideas and concepts in a part of a huge EA field.

How to reference

In order to correctly reference this scholarly work, feel free to copy and paste the following:
Wei-Chang Yeh (2008). A Simple Hybrid Particle Swarm Optimization, Advances in Evolutionary Algorithms,

Xiong Zhihui (Ed.), ISBN: 978-953-7619-11-4, InTech, Available from:

http://www.intechopen.com/books/advances_in_evolutionary_algorithms/a_simple_hybrid_particle_swarm_opti
mization

InTech Europe

University Campus STeP Ri

Slavka Krautzeka 83/A
51000 Rijeka, Croatia
Phone: +385 (51) 770 447
Fax: +385 (51) 686 166
www.intechopen.com

InTech China

Unit 405, Office Block, Hotel Equatorial Shanghai

No.65, Yan An Road (West), Shanghai, 200040, China
Phone: +86-21-62489820
Fax: +86-21-62489821