Bacterial Foraging Optimization Algorithm with Particle

Swarm Optimization Strategy for Global Numerical

Optimization
Hai Shen

Key Laboratory of Industrial

Informatics, Shenyang
Institute of Automation,
Chinese Academy of
Sciences, China
Graduate School of the
Chinese Academy of
Sciences, China
College of Physics Science
and Technology, Shenyang
Normal University, China
shenhai@sia.cn
Yunlong Zhu
Key Laboratory of Industrial
Informatics, Shenyang
Institute of Automation,
Chinese Academy of
Sciences, China
ylzhu@sia.cn
Xiaoming Zhou
Key Laboratory of Industrial
Informatics, Shenyang
Institute of Automation,
Chinese Academy of
Sciences, China
Graduate School of the
Chinese Academy of
Sciences, China
zhouxiaoming@sia.cn
Haifeng Guo
Key Laboratory of Industrial
Informatics, Shenyang
Institute of Automation,
Chinese Academy of
Sciences, China
guohf@sia.cn
Chunguang Chang
Key Laboratory of Industrial
Informatics, Shenyang
Institute of Automation,
Chinese Academy of
Sciences, China
changchunguang@sia.cn
ABSTRACT
In 2002, K. M. Passino proposed Bacterial Foraging Optimization
Algorithm (BFOA) for distributed optimization and control. One of
the major driving forces of BFOA is the chemotactic movement of
a virtual bacterium that models a trial solution of the optimization
problem. However, during the process of chemotaxis, the BFOA
depends on random search directions which may lead to delay in
reaching the global solution. Recently, a new algorithm BFOA
oriented by PSO termed BF-PSO has shown superior in propor-
tional integral derivative controller tuning application. In order to
examine the global search capability of BF-PSO, we evaluate the
performance of BFOA and BF-PSO on 23 numerical benchmark
functions. In BF-PSO, the search directions of tumble behavior
for each bacterium oriented by the individuals best location and
the global best location. The experimental results show that BF-
PSO performs much better than BFOA for almost all test functions.
Thats approved that the BFOA oriented by PSO strategy improve
its global optimization capability.

Corresponding author.
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for prot or commercial advantage and that copies
bear this notice and the full citation on the rst page. To copy otherwise, to
republish, to post on servers or to redistribute to lists, requires prior specic
permission and/or a fee.
GEC09, June 1214, 2009, Shanghai, China.
Copyright 2009 ACM 978-1-60558-326-6/09/06 ...$5.00.
Categories and Subject Descriptors
I.2 [Computing Methodologies]: Articial Intelligence; I.2.8 [
Problem Solving, Control Methods, and Search]: [control
theory]
General Terms
Algorithms, Performance.
Keywords
Bacterial Foraging, Particle Swarm Optimization, Numerical Opti-
mization
1. INTRODUCTION
Optimization originated in the applied mathematics arena is a
process of seeking the best possible solutions for given problems.
As a powerful solution approach, it has been used in almost all
elds of engineering, nance, management as well as social sci-
ence.The general unconstrained numerical optimization problem
which was relied on by those applied elds can be dened as:
min f (x) : x R
n
,
where f : R
n
R. At present, a study of global optimization
problems has become a highly concerned topic.
Optimization techniques may follow different approaches. Re-
cently, optimization techniques inspired by biology behaviors, known
as bio-mimetic optimization algorithms, have obtained more and
more attention[1]. Some algorithms have been proposed, such as
Genetic Algorithms (GAs), Particle Swarm Optimization (PSO),
497
Ant Colony Optimization (ACO) and glowworm swarm optimiza-
tion (GSO). Bio-mimetic optimization algorithms are developed
from simulation the evolutionary process and the behaviors of biol-
ogy. They are population-based(each member stands for an biology
individual), and initialized with a population of individuals. They
utilizes the direct information "tness" instead of individuals abil-
ity to adapt to the environment. These individuals are manipulated
over many generations by ways of mimicking social behavior of
biology, in an effort to nd the optima. In comparison with other
optimization algorithms, bio-mimetic optimization algorithms have
the following characteristics:
1) The individual components are distributed and autonomous,
there is no central control, and the fault of an individual cannot
inuence solving the whole problem, these characteristics ensure
this kind of algorithms has better robustness.
2) The manner of achieves individual collaboration though non-
directly information communication make sure of the expansibility
of the algorithm.
3) They dont demand to meet the requirement of differentiabil-
ity, convexity and other conditions for mathematical description of
the problem.
4) Because of concerns merely with basic mathematical opera-
tions, therefore, they are simple and easy to be implemented on
computer.
These advantages enabled bio-mimetic optimization algorithms
to widely use in a very short period, such as power system[2][3],
vehicle routing[4][5], mechanical design[6][7] and robotics[8].
Natural selection tends to eliminate animals with poor foraging
strategies and favor the propagation of genes of those animals that
have successful foraging strategies, since they are more likely to
enjoy reproductive success. After many generations, poor forag-
ing strategies are either eliminated or shaped into good ones. This
activity of foraging led the researchers to use it as optimization pro-
cess. Based on the researches on the foraging behavior of E. coli
bacteria, Prof. K.M.Passino proposed a new evolutionary compu-
tation technique known as Bacterial Foraging Optimization Algo-
rithm (BFOA)[9]. In BFOA, the foraging (methods for locating,
handling, and ingesting food) behavior of E. coli bacteria is mim-
icked.
As an evolutionary computation technique, BFOA is also an it-
eration based optimization tool. The rst is to generate a set of ran-
dom solutions in which each solution represents bacteria, the fol-
lowing is to measure the tness of these solutions, then retain some
excellent individuals and give up other individuals according to the
tness, nally, implement certain operations on those individuals
retained. Thus, the new solutions of the next iteration were yielded
and the next iteration work begins. Until date, BFOA has success-
fully been applied to real world problems such as PID controller
design[10][11], learning of articial neural networks[12], power
system[13][14][15],numerals recognition[16], and channel equal-
izer[17]. However, during the process of chemotaxis, the BFOA
depends on random search directions which may lead to delay in
reaching the global solution.
Particle Swarm Optimization (PSO) is also a bio-mimetic opti-
mization technique developed by Eberhart and Kennedy in 1995,
which was inspired by the social behavior of bird ocking and sh
schooling. The critical concept of PSO consists of, at each time
step, changing the velocity each particle toward its best location
of individual and global best location among the individual. That
the balance between the global and local search throughout the run
make PSO become a success optimization algorithm. In past sev-
eral years, PSO has been successfully applied in many research and
application areas[18]. It is demonstrated that PSO gets better re-
sults in a faster compared with other methods. Therefore, in 2008,
W. Korani proposed an improved BFOA, namely BF-PSO[19]. The
BF-PSO algorithm borrowed the ideas of velocity updating from
PSO, the search directions specied by the tumble of bacteria were
oriented by the individuals best location and the global best loca-
tion concurrently. Then the author applied BF-PSO algorithm to
the PID parameter tuning for a set of test plants. Simulation results
demonstrate that the proposed algorithm out performance both con-
ventional PSO and BFOA.
With the purpose of further investigation the performance of the
BF-PSO, we implemented BFOA and BF-PSO on a suite of 23
functions at the same time[20]. The 23 benchmark functions used
in our experiments have been widely employed by other researchers
to bio-mimetic optimization algorithms. However they were lit-
tle tested on BFOA. When compared with the BFOA used to nd
global optimization, results show that the BF-PSO algorithm can
nd more accurate results of almost all tested problems and has a
noticeable performance.
The rest of this paper is organized as follows. Section II de-
scribes the BFOA and its implementation in detailed. Section III
provides an extensive literature survey on BFOA. In section IV, the
BF-PSO algorithm that is used to nd global optimization will be
detailed. Section V gives the 23 benchmark test functions used in
our studies. Section VI presents the experimental results and dis-
cussions on BFOA and BF-PSO. Finally, the concluding remarks
and future research directions are given in section VII.
2. THE CLASSICAL BACTERIAL FORAG-
ING OPTIMIZATION ALGORITHM
In the process of foraging, E. coli bacteria undergo four stages,
namely, chemotaxis, swarming, reproduction, and elimination and
dispersal. In search space, BFOA seek optimum value through the
chemotaxis of bacteria, and realize quorum sensing via assemble
function between bacterial, and satisfy the evolution rule of the
survival of the ttest make use of reproduction operation, and use
elimination-dispersal mechanism to avoiding falling into premature
convergence.
2.1 Chemotaxis
The motion patterns that the bacteria will generate in the pres-
ence of chemical attractants and repellents are called chemotaxes.
For E. coli, this process was simulated by two different moving
ways: run or tumble. A Bacterium alternates between these two
modes of operation its entire lifetime. The bacterium sometimes
tumbles after a tumble or tumbles after a run. This alternation be-
tween the two modes will move the bacterium, and this enables
it to "search" for nutrients. Suppose
i
( j, k, l) represent the posi-
tion of the each member in the population of S bacterial at the jth
chemotactic step, and kth reproduction step, and lth elimination
The movement of bacterium may be presented by:

i
( j + 1, k, l) =
i
( j, k, l) + C(i)( j)
Where C(i)(i = 1, 2, . . . , S ) is the size of the step taken in the ran-
dom direction specied by the tumble. ( j) was used to dene the
random direction of movement after a tumble. J(i, j, k, l) is the t-
ness, which also denote the cost at the location of the ith bacterium

i
( j, k, l) R
n
. If at
i
( j + 1, k, l) the cost J(i, j + 1, k, l) is better
(lower) than at
i
( j, k, l) , then another step of size C(i) in this same
direction will be taken. Otherwise, bacteria will tumble via taking
another step of size C(i) in random direction ( j) in order to seek
better nutrient environment.
498
2.2 Swarming
An interesting group behavior has been observed for several motile
species of bacteria including E.coli and S. typhimurium. When a
group of E. coli cells is placed in the center of a semisolid agar with
a single nutrient chemo-effector, they move out from the center in
a traveling ring of cells by moving up the nutrient gradient cre-
ated by consumption of the nutrient by the group. To achieve this,
function to model the cell-to-cell signaling via an attractant and a
repellan. The mathematical representation for E.coli swarming can
be represented by:
J
cc
(, P( j, k, l)) =
S

i=1
J
i
cc
(,
i
( j, k, l))
=
S

i=1
_

_
d
attract
exp
_

_
w
attract
p

m=1
(
m

i
m
)
2
_

_
_

_
+
S

i=1
_

_
h
repellant
exp
_

_
w
repellant
p

m=1
(
m

i
m
)
2
_

_
_

_
where is the cost function value to be added to the actual cost func-
tion. S is the total number of bacteria and p is the number of param-
eters to be optimized which are present in each bacterium. d
attract
is the depth of the attractant released by the cell and w
attract
is a
measure of the width of the attractant signal. h
repellant
=d
attract
is the
height of the repellant effect and w
repellant
is a measure of the width
of the repellant.
2.3 Reproduction
According to the rules of evolution, individual will reproduce
themselves in appropriate conditions in a certain way. For bacterial,
a reproduction step takes place after all chemotactic steps.
J
i
health
=
Nc+1

j=1
J(i, j, k, l)
Where J
i
health
is the health of bacterium i. Sort bacteria and chemo-
tactic parameters C(i) in order of ascending cost (higher cost means
lower health). For keep a constant population size, bacteria with
the highest J
health
values die. The remaining bacteria are allowed to
split into two bacteria in the same place.
2.4 Elimination-Dispersal
In the evolutionary process, elimination and dispersal events can
occur such that bacteria in a region are killed or a group is dis-
persed into a new part of the environment due to some inuence.
They have the effect of possibly destroying chemotactic progress,
but they also have the effect of assisting in chemotaxis, since dis-
persal may place bacteria near good food sources. From the evolu-
tionary point of view, elimination and dispersal was used to guaran-
tees diversity of individuals and to strengthen the ability of global
optimization. In BFOA, bacteria are eliminated with a probability
of ped.In order to keeping the number of bacteria in the popula-
tion constant, if a bacterium is eliminated, simply disperse one to a
random location on the optimization domain.
3. RELATED WORKS ON BFOA
Currently, the related works on BFOA can be divided into anal-
ysis of BFOA, application and improvement.
A. The analysis of BFOA
One major step in BFOA is chemotactic behavior. In 2008,
S.Dasgupta and S.Das et al. made an analysis of the chemotaxis
operation in BFOA [21][22]. The analysis undertaken provides im-
portant insights into the search mechanism of BFOA. The analy-
sis points out that the chemotaxis usually results in sustained os-
cillation , especially on at tness landscapes, when a bacterium
cell is close to the optima. Therefore, it is necessary to bound on
the chemotactic step-height parameter that avoids limit-cycles and
guarantees convergence of the bacterial dynamics into an optimum.
Two simple schemes for adapting the chemotactic step-height have
been subsequently proposed. In the same year, A.Abraham and
A.Biswas provided a simple mathematical analysis of the repro-
duction step used in BFOA[23]. The analysis is focus on the re-
production in a simple two-bacterial system working on a one di-
mensional tness landscape. Their analysis reveals that the repro-
duction event contributes to the quick convergence of the bacterial
population near optima.
B. The application of BFOA
R.Majhi and G.Panda et al. developed a BFOA based adap-
tive model for short term and long term forecasting of stock in-
dices[24]. The weight of the combiner is updated using the BFOA
tool. The results of the experiment indicated that the propose modal
offers computational complexity, better prediction accuracy and
lesser training time compared to those obtained fromthe MLP modal.
BFOAwas also used for solving a highly non-linear and non-convex
problem [15]. It is found that the BFOA technique succeeds in
better loss minimization compared to conventional IPSLP tech-
nique. S.Mishra and C.N.Bhende used the modied BFOA to op-
timize the coefcients of Proportional plus Integral controllers for
active power lters[25]. L.Ulagammai et al. used BFOA to train
a Wavelet-based Neural Network (WNN) and identify the inherent
non-linear characteristics of power system loads[26].
In the area of PID applications, D.H.Kim and J.H.Cho presented
an intelligent tuning method of PID controller based on BFOA[27].
Simulation results show that the object function can be minimized
by gain selection for control and the variety gain can be obtained.
B.Niu et al. designed BF-PID controller using BFOA[28]. Com-
pared with GA-PID controller, BF-PID obtained a faster settling
time, less or no overshoot and higher robustness. A multi-objective
optimization method for the parameter tuning of fuzzy PID con-
trollers is proposed using BFOA. In the proposed BFOA-tuning
method, a cost function is dened in a systematic way. The simu-
lation results show that a fuzzy PID controller designed using the
proposed BFOA has good performance[11].
C. Improved Algorithms
BFOA use function to modal the cell-to-cell signaling via an
attractant and a repellant. However, its value does not depend
on the nutrient concentration at position . In 2002, Y.Liu and
K.M.Passino used a new function to represent the environment-
dependent cell-to-cell signaling[29].
J
ar
() = exp(M J())J
cc
()
where M is a tunable parameter. Then, for swarming, the mini-
mization is J(i, j, k, l) + J
ar
(
i
( j, k, l)) . By performing social forag-
ing with chemical-attractant-induced swarming, E.coli have better
chance in locating the optimal point in a noisy environment. Con-
sidering BFOA lacks in adaptation according to the operating con-
dition, S.Mishra presented a new algorithm Fuzzy Bacterial For-
aging (FBF)[30]. FBF uses variable run length in the chemotaxis
step in place of the original constant through a Takagi-Sugeno type
fuzzy inference scheme. The resulting shows FBF has superior
performance than GA when applied to the harmonic estimation
problem. W.J.Tang et al. proposed a dynamic bacterial foraging
499
algorithm (DBFA) which aims at optimization in dynamic environ-
ments[31]. The DBFA adopts a selection scheme which enables the
bacteria to exibly adapt to the changing environment. Compared
with BFA, DBFA is able to provide satisfactory performance, and
can react to most of the environmental changes in time.
Hybridization of BFOAwith other naturally inspired meta-heuristics
has remained an interesting problem for the researchers. A.Biswas
et al. come up with an improved variant of the BFOA algorithm by
combining the PSO based mutation operator with bacterial chemo-
taxis[32]. The new algorithm, named by the authors as Bacterial
Swarm Optimization (BSO). The performance of BSO is better
than classical PSO, original BFOA and MPSO-TVAC on several
numerical benchmark functions.In 2006, D.H.Kim and J.H.Cho in-
troduces clonal selection of immune algorithm and fuzzy logic into
bacterial foraging to enhance running speed and patch of optimal
condition[33]. In 2007, they also proposed a hybrid approach in-
volving GA and BFOA. Dynamic mutation and modied simple
crossover are used in BFOA[34].
4. BF-PSO ALGORITHM
In 2008, W.Korani proposed an improved BFOA, namely BF-
PSO. The BF-PSO combines both algorithms BF and PSO. The
aims is to make use of PSO ability to exchange social information
and BF ability in nding a new solution by elimination and disper-
sal.In BFOA, a unit length direction of tumble behavior is randomly
generated. Random direction may lead to delay in reaching the
global solution. In the BF-PSO, the unit length random direction
of tumble behavior can been decided by the global best position
and the best position of each bacteria. During the chemotaxis loop,
the update of the tumble direction is determined by:
( j + 1) = w ( j) + C1 R1(Plbest-Pcurrent)
+ C2 R2 (Pgbest-Pcurrent)
Where Plbest is the best position of each bacterial and Pgbest is
the global best bacterial. The brief pseudo-code of the BF-PSO has
been provided below:
[Step 1] Initialization: Parameters Setting.
p : Dimension of the search space.
S : The number of bacteria in the population.
N
c
: Chemotactic steps.
N
s
: Swimming length.
N
re
: The number of reproduction steps.
N
ed
: The number of elimination-dispersal events.
P
ed
: Elimination-dispersal with probability.
C( i ) (i = 1, 2, , S ) : The size of the step taken in the
random direction specied by the tumble.
P( j , k , l ) : P( j , k , l ) = {
i
( j , k , l ) | i = 1, 2, ......, S }.
Generate a random vector ( j) which elements lie in
[-1,1].
C1 , C2 , R1 , R2 , w: PSO parameters.
[Step 2] Elimination Dispersal loop: l = l + 1.
[Step 3] Reproduction loop: k = k + 1.
[Step 4] Chemotaxis loop: j = j + 1.
[4.1] Take a chemotactic step for every bacterium (i).
[4.2] Compute tness function: J(i, j, k, l),
then let J
last
= J(i, j, k, l).
[4.3] Tumble: Let ( j+1) = w( j)+C1R1(Plbest-Pcurrent)
+C2 R2 (Pgbest-Pcurrent).
[4.4] Move: Let
i
( j + 1, k, l) =
i
( j, k, l) + C(i)( j).
Computer tness function: J(i, j, k, l).
Then Let J(i, j + 1, k, l) = J(i, j + 1, k, l)+
J
cc
(

( j + 1, k, l), P( j + 1, k, l)).
[4.5] Swim: Let m = 0;
while (m < N
s
)
let m = m + 1;
if J(i, j, k, l) < J
last
Let J
last
= J(i, j, k, l),
Let
i
( j + 1, k, l) =
i
( j, k, l) + C(i)( j),
Computer tness function: J(i, j, k, l),
Let J(i, j + 1, k, l) = J(i, j + 1, k, l)+
J
cc
(

( j + 1, k, l), P( j + 1, k, l)).
Else let m = N
s
.
[4.6] Go to next bacterium.
[Step 5] If ( j < N
c
), go to Step 4.
[Step 6] Reproduction: Computer the health of the bacterium i:
J
i
health
=
Nc+1

j=1
J(i, j, k, l)
Sort bacteria and chemotactic parameters C(i) in order
of ascending cost J
health
. Bacteria with the highest J
health
values die, the remaining bacteria reproduce.
[Step 7] If (k < N
re
), go to Step 3.
[Step 8] Elimination-dispersal: Eliminate and disperse bacteria
with probability P
ed
.
[Step 9] If (l < N
ed
), go to Step 2.
5. EXPERIMENTAL STUDIES
A. Test functions
To fully evaluate the performance of the BFOA and BF-PSO al-
gorithms without a biased conclusion towards some chosen prob-
lems, we employed 23 standard benchmark functions which are
given in Table I. These functions can be divided into three cate-
gories. Functions f
1
f
13
are high-dimensional problems. Func-
tions f
8
f
13
are multimodal functions where the number of local
minima increases exponentially with the problem dimension. They
appear to be the most difcult class of problems for many optimiza-
tion algorithms. Functions f
14
f
23
are low-dimensional functions
which have only a few local minima.
B. Experimental setting
The parameter setting of the BFOA and BF-PSO algorithm is
summarized as follows. The same population size S=50; the num-
ber of chemotactic steps N
c
=50; the number of reproduction steps
N
re
=2; the number of elimination-disperal events N
ed
=1; the prob-
ablity of elimination-dispersal P
ed
=0.25; the length of steps during
runs C(i)=0.1; the acceleration factors c
1
and c
2
were both 2.0, and
a decaying inertia weight w is 0.9. To make the comparison fair, the
populations for all the considered algorithms were initialized using
the same random seeds.
6. RESULT AND DISCUSSION
The average results of 30 independent runs are summarized in
Table I. Moreover, in order to be more intuitive analysis of per-
formance of BFOA and BF-PSO, the convergence results for se-
lected benchmark problems over 30 runs were attached in this pa-
per. Figures 1, 2 and 3 show the convergence results for uni-modal
functions ( f
1
f
7
), multimodal functions with many local min-
ima function ( f
8
f
13
) and multimodal functions with few local
minima function ( f
14
f
23
), respectively.
For uni-modal functions f
1
to f
7
, the BF-PSO is able to obtain
practically perfect optimization results, while BFOA has difculty
500
Table 1: The 23 Benchmark Functions, Where n is the Dimension of The Function, f
min
is the Global Minimum Value of the Function.
Test Functions n SD f
min
f
1
(x) =
_
n
i=1
x
2
i
30 [100, 100]
n
f
1
(

0 ) = 0
f
2
(x) =
_
n
i=1
|x
i
| +

n
i=1
|x
i
| 30 [10, 10]
n
f
2
(

0 ) = 0
f
3
(x) =
_
n
i=1
(
_
i
j=1
x
j
)
2
30 [100, 100]
n
f
3
(

0 ) = 0
f
4
(x) = max
i
{|x
i
|, 1 i n} 30 [100, 100]
n
f
4
(

0 ) = 0
f
5
(x) =
_
n1
i=1
(100(x
i+1
x
i
)
2
) + (x
i
1)
2
) 30 [30, 30]
n
f
5
(

1 ) = 0
f
6
(x) =
_
n
i=1
(x
i
+ 0.5)
2
30 [100, 100]
n
f
6
(

0 ) = 0
f
7
(x) =
_
n
i=1
ix
4
i
+ random[0, 1) 30 [1.28, 1.28]
n
f
7
(

0 ) = 0
f
8
(x) =
_
n
i=1
(x
i
sin(

|x
i
|)) 30 [500, 500]
n
f
8
(

420.9687) = 12569.5
f
9
(x) =
_
n
i=1
(x
2
i
10 cos(2x
i
) + 10)
2
30 [5.12, 5.12]
n
f
9
(

0 ) = 0
f
10
(x) = 20 exp
_
0.2
_
1
n
_
n
i=1
x
2
i
_
exp
_
1
n
_
n
i=1
cos 2x
i
_
+ 20 + e 30 [32, 32]
n
f
10
(

0 ) = 0
f
11
(x) =
1
4000
_
30
i=1
(x
i
100)
2

n
i=1
cos(
x
i
100

i
) + 1 30 [600, 600]
n
f
11
(

0 ) = 0
f
12
(x) =

n
{10 sin
2
(y
1
)) +
_
n1
i=1
(y
i
1)
2
[1 + 10 sin
2
(y
i+1
)]
30 [50, 50]
n
f
12
(

1 ) = 0
+(y
n
1)
2
} +
_
30
i=1
u(x
i
, 10, 100, 4)
f
13
(x) = 0.1{sin
2
(3x
1
) +
_
29
i=1
(x
i
1)
2
p[1 + sin
2
(3x
i+1
)]
30 [50, 50]
n
f
13
(

1 ) = 0
+(x
n
1)
2
[1 + sin
2
(2x
30
)]} +
_
30
i=1
u(x
i
, 5, 100, 4)
f
14
(x) =
_
1
500
+
_
25
j=1
1
j+
_
2
j=1
(x
i
a
i j
)
6
_
1
2 [65.536, 65.536]
n
f
14
(32, 32) = 0
f
15
(x) =
_
11
i=1
_
a
i

x
1
(b
2
i
+b
i
x
2
)
b
2
i
+b
i
x
3
+x
4
_
2
4 [5, 5]
n
f
15
(0.1928, 0.1908, 0.1231
,0.1358)= 0.0003075
f
16
(x) = 4x
2
1
2.1x
4
1
+
1
3
x
6
1
+ x
1
x
2
4x
2
2
+ 4x
4
2
2 [5, 5]
n
f
16
(0.08983, 0.7126) =
(0.08983, -0.7126)=-1.0316
f
17
(x) =
_
x
2

5.1
4
2
x
2
1
+
5

x
1
6
_
2
+ 10
_
1
1
8
_
cos x
1
+ 10
2 [5, 10] [0, 15]
f
17
(3.142, 2.275) =
(3.142, 2.275)=
(9.425, 2.425)=0.398
f
18
(x) = [1 + (x
1
+ x
2
+ 1)
2
(19 14x
1
+ 3x
2
1
14x
2
+ 6x
1
x
2
2 [2, 2]
n
f
18
(0, 1) = 3 +3x
2
2
)] [30 + (2x
1
+ 1 3x
2
)
2
(18 32x
1
+ 12x
2
1
+48x
2
36x
1
x
2
+ 27x
2
2
)]
f
19
(x) =
_
4
i=1
exp
_

_
3
j=1
a
i j
(x
j
p
i j
)
2
_
3 [0, 1]
n
f
19
(0.114, 0.556
,0.852)=-3.86
f
20
(x) =
_
4
i=1
exp
_

_
6
j=1
a
i j
(x
j
p
i j
)
2
_
6 [0, 1]
n
f
20
=(0.201, 0.15, 0.477,
0.275, 0.311, 0.657)=-3.32
f
21
(x) =
_
5
i=1
_
(x
i
a
i
)(x
i
a
i
)
T
+ c
i
_
1
4 [0, 10]
n
f
21
(

0 ) = 10
f
22
(x) =
_
7
i=1
_
(x
i
a
i
)(x
i
a
i
)
T
+ c
i
_
1
4 [0, 10]
n
f
22
(

0 ) = 10
f
23
(x) =
_
10
i=1
_
(x
i
a
i
)(x
i
a
i
)
T
+ c
i
_
1
4 [0, 10]
n
f
23
(

0 ) = 10
0 50 100 150 200 250 300
6
4
2
0
2
4
6
8
10
12
Generation
B
e
s
t

F
i
t
n
e
s
s
(
L
o
g

S
c
a
l
e
)


BFOA
BFPSO
(a) f1
0 100 200 300
0
5
10
Generation
B
e
s
t

F
i
t
n
e
s
s
(
L
o
g

S
c
a
l
e
)
BFOA
BFPSO
(b) f3
Figure 1: Convergence results of BFOA and BF-PSO for uni-modal functions.
501
0 50 100 150 200 250 300
2
0
2
4
6
8
10
12
Generation
B
e
s
t

F
i
t
n
e
s
s
(
L
o
g

S
c
a
l
e
)
BFOA
BFPSO
(a) f9
0 50 100 150 200 250 300
3
2
1
0
1
2
3
4
5
Generation
B
e
s
t

F
i
t
n
e
s
s
(
L
o
g

S
c
a
l
e
)
BFOA
BFPSO
(b) f10
Figure 2: Convergence results of BFOA and BF-PSO for multimodal functions with many local minima.
0 50 100 150 200 250 300
0
0.05
0.1
0.15
0.2
0.25
Generation
B
e
s
t

F
i
t
n
e
s
s
BFOA
BFPSO
(a) f15
0 50 100 150 200 250 300
10
5
0
5
10
Generation
B
e
s
t

F
i
t
n
e
s
s
BFOA
BFPSO
(b) f22
Figure 3: Convergence results of BFOA and BF-PSO or multimodal functions with few local minima.
with functions f
5
and f
6
, and the accuracy for the remaining func-
tions is also less good than the BF-PSO. According to the Figure 1
we plotted to observe the evolutionary process, in the beginning,
BFOA and BF-PSO all displays a faster convergence rate. But
when they nd solutions closer to the global optimum, BFOA ap-
peared to become trapped in a poor local optimum and unable to
escape from it, while BF-PSO is also able to improve its solution
steadily for a long time. The largest difference in performance be-
tween BFOA and BF-PSO occurs with function f
6
, the step func-
tion, which is characterized by plateaus and discontinuity. BFOA
performs poorly because search direction after tumble of bacteria
made it searching mainly in a relatively small local neighborhood,
and cannot move from one plateau to a lower one. On the other
hand, BF-PSO has a much higher probability of generating long
jumps than BFOA. Thats because that the search direction is ori-
ented concurrently by individuals best location and the global best
location enable BF-PSO to move from one plateau to a lower one
with relative ease.
For multimodal functions with many local minima f
8
to f
13
, the
BF-PSO generated signicantly better results than BFOA. Accord-
ing to the Figure 2, BFOA fell into a poor local optimum quite
early when it is just beginning to evolution. However, BF-PSO
can quickly converge toward the optima within a relatively small
number of generations due to its search direction after tumble of
bacteria.
Functions f
14
to f
23
are multimodal functions with only a few lo-
cal minima functions and the low dimensions. For these problems,
the BF-PSO procedure performs variably. On the functions f
16
to
f
20
, two algorithms yielded similar results which are all approxi-
mate the global optimal solution. Thats because these functions
represent completely similar surface characteristics, the major re-
gions of attraction are all not small local optima, such as f
18
(Figure
4). On the shekels family functions f
21
to f
23
, the performance
of BF-PSO is the worst among the all test functions. The BFOA
nds solutions closer to the global optimum, but BF-PSO doesnt
(Figure 3(b)). The problems f
21
to f
23
are characterized with large
at regions with sudden "fox holes" of varying depth (Figure 5).
The improved success rate of BFOA is due to its nature to explore
regions of interest and concentrate in the more promising ones at
each iteration. For BFOA, at the beginning of the search, there is
a number of promising "fox holes". After a short period, certain
degree of "jump" on a BFOA evolution curve occurs. Thats due to
the fact that the BFOA could sometimes experience a few genera-
tions without achieving a better solution. Such a situation can be
diagnosed. Its search direction made the "jump" very easy, and the
regions of attraction of the global minima is the smallest, so nding
the global minima become very difcult for BFOA.
502
Table 2: Comparison Between BFOA and BF-PSO on Bench-
mark Functions f
1
f
23
. All Results Have Been Averaged Over
30 Runs.
Function BFOA BF-PSO
f1 7.517710
1
5.634810
3
f2 3.9923 2.678310
1
f3 5.3897 7.108610
2
f4 1.3969 4.985310
2
f5 1.144510
2
2.744310
1
f6 1.430910
4
0
f7 2.8296 3.964410
1
f8 -4632.6 -6192
f9 9.082910
2
5.268210
1
f10 2.024910
1
6.493610
2
f11 4.514310
2
4.343710
2
f12 1.336810
2
1.8109
f13 2.6661 2.201210
1
f14 1.9920 9.980010
1
f15 1.326010
1
1.433210
3
f16 -9.818310
1
-1.0118
f17 0.4227 0.4165
f18 5.0307 3.2271
f19 -3.5766 -3.6383
f20 -2.4263 -3.2726
f21 -8.0725 -5.0429
f22 -8.1157 -5.6969
f23 -8.4042 -6.1105
2
0
2
2
1
0
1
2
0
2
4
6
8
10
x 10
5
x
1
x
2
y
Figure 4: f
18
(Goldstein-Price function).
0
5
10
0
5
10
12
10
8
6
4
2
0
x
1
x
2
y
Figure 5: f
23
(Shekel function), x
3
= x
1
, x
4
= x
2
.
7. CONCLUSIONS AND FUTURE
DIRECTIONS
In this paper, we make a detailed introduction of the underlying
ideas of BFOA, basic algorithmic structures, some major variants
proposed in the literature, and applications to optimization prob-
lems. In BFOA, because random search directions which may
lead to delay in reaching the global solution during the process of
chemotaxis of bacteria, so we also introduced an improved BFOA,
namely BF-PSO. In order to examine the performance of BF-PSO,
the BFOA and BF-PSO was investigated on 23 numerical bench-
mark functions. Fromthe simulation results, we can see this method
of search directions after tumble behavior for bacteria oriented by
PSO strategy greatly improved the optimization performance of
BFOA. The correctness and practicability of BF-PSO was proved.
Therefore, the BF-PSO has potential to be useful for other practi-
cal optimization problems. The future research effort should focus
on improving the convergence speed of BF-PSO. Also more ex-
periments are required to determine why and when the BF-PSO
methods fail on shekels family functions.
8. ACKNOWLEDGMENTS
This work was supported by the National Key of Technology RD
Program of China (Grant No.2007AA04Z189), the High-Tech Re-
search and Development Programof China (Grant No.20060104A11
18) and the Science and Technology Supporting 394 Program of
China (Grant No. 2006BAH02A09).
9. REFERENCES
[1] L. N. Decastro, F. J. Von Zuben, and Idea Group Pub et al.
Recent Developments In Biologically Inspired Computing.
IGI Publishing, Hershey, PA, USA, 2004.
[2] S. Gerbex, R. Cherkaoui, and A. J. Germond. Optimal
location of multi-type facts devices in a power system by
means of genetic algorithms. IEEE Transactions on Power
Systems, 16(3): 537544, 2001.
[3] M. A. Abido. Optimal design of power-system stabilizers
using particle swarm optimization. IEEE Transactions on
Energy Conversion, 17(3): 406413, 2002.
[4] J. E. Bell and P. R. McMullen. Ant colony optimization
techniques for the vehicle routing problem. Advanced
Engineering Informatics, 18(1): 4148, 2002.
[5] B. M. Baker and M. A. Ayechew. A genetic algorithm for the
vehicle routing problem. Computers & Operations Research,
30(5): 787800, 2003.
[6] C. A. Coello Coello. Theoretical and numerical
constraint-handling techniques used with evolutionary
algorithms a survey of the state of the art. Computer Methods
in Applied Mechanics and Engineering, 191(11-12):
12451287, 2002.
[7] R. F. Bo, R. Q. Li, and H. X. Pan. Concept optimization for
mechanical product by using ant colony system. Computer
Methods in Applied Mechanics and Engineering, 22(4):
628638, 2008.
[8] J. Wisnu, S. Kosuke, and F. Toshio. A pso-based mobile
robot for odor source localization in dynamic
advection-diffusion with obstacles environment: Theory,
simulation and measurement. IEEE Computational
Intelligence Magazine, 2(2): 3751, 2007.
[9] K. M. Passino. Biomimicry of bacterial foraging for
distributed optimization and control. IEEE Control Systems
Magazine, 22: 5267, 2002.
503
[10] A. Ali and S. Majhi. Design of optimum pid controller by
bacterial foraging strategy. In ICIT 2006: Proceedings of the
IEEE International Conference on Industrial Technology,
pages 601605, Mumbai, India, December, 2008. IEEE.
[11] H. Ch. Chen. Bacterial foraging based optimization design of
fuzzy PID controllers. In ICIC 2008: Proceedings of the 4th
International Conference on Intelligent Computing, volume
5226, pages 841849, Shanghai, China, September, 2008.
Springer-Verlag.
[12] D. H. Kim and C. H. Cho. Bacteria foraging based neural
network fuzzy learning. In IICAI 2005: Proceedings of the
2nd Indian International Conference on Articial
Intelligence, pages 20302036, Pune, India, December,
2005. IEEE.
[13] M. Tripathy and S. Mishra. Bacteria foraging based to
optimize both real power loss and voltage stability limit.
IEEE Transactions on Power Systems, 22(1): 240248, 2007.
[14] T. K. Das and G. K. Venayagamoorthy. Bio-inspired
algorithms for the design of multiple optimal power system
stabilizers: SPPSO and BFA. IEEE Transactions on Industry
Applications., 44(5): 14451457, 2008.
[15] M. Tripathy, S. Mishra, and L. L. Lai et al. Transmission loss
reduction based on FACTS and bacteria foraging algorithm.
In PPSN IX: Proceedings of the 9th International Conference
on Parallel Problem Solving from Nature, volume 4193,
pages 222231, Reykjavik, Iceland, September, 2006.
Springer-Verlag.
[16] M. Hanmandlu, A. V. Nath, and A. C. Mishra et al. Fuzzy
model based recognition of handwritten hindi numerals
using bacterial foraging. In ICIS 2007: Proceedings of the
6th Annual IEEE/ACIS International Conference on
Computer and Information Science, pages 309314,
Melbourne, Australia, July, 2007. IEEE Computer Society.
[17] B. Majhi and G. Panda. Recovery of digital information
using bacterial foraging optimization based nonlinear
channel equalizers. In ICDIM 2007: Proceedings of the First
IEEE International Conference on Digital Information
Management, pages 367372, Christ College, Bangalore,
India, December, 2006. IEEE Press.
[18] R. C. Eberhart and Y.H. Shi. Particle swarm optimization:
Developments, applications and resources. In CEC 2001:
proceedings of the IEEE congress on evolutionary
computation, pages 8186, Seoul, South Korea, May, 2001.
IEEE.
[19] W. Korani. Bacterial foraging oriented by particle swarm
optimization strategy for PID tuning. In GECCO 2008:
Proceedings of the Genetic and Evolutionary Computation
Conference, pages 18231826, Atlanta, GA, USA, July,
2008. ACM.
[20] X. Yao, Y. Liu, and G. M. Lin. Evolutionary
programmingmade faster. IEEE Transactions on
Evolutionary Computing, 3(2): 82102, July, 1999.
[21] S. Dasgupta, S. Das, and A. Abraham et al. Adaptive
computational chemotaxis in bacterial foraging algorithm. In
CISIS 2008: Proceedings of the Second International
Conference on Complex, Intelligent and Software Intensive
Systems, pages 6471, Genova, Italy, March, 2008. IEEE
Computer Society.
[22] S. Das, S. Dasgupta, and A. Biswas et al. On stability of the
chemotactic dynamics in bacterial foraging. In CSTST 2008:
IEEE/ACM International Conference on Soft Computing as
Transdisciplinary Science and Technology, pages 245251,
Paris, France, October, 2008. ACM.
[23] A. Abraham, A. Biswas, and S. Dasgupta et al. Analysis of
reproduction operator in bacterial foraging optimization
algorithm. In CEC 2008: IEEE World Congress on
Computational Intelligence, pages 14761483, Hong Kong,
June, 2008. IEEE Press.
[24] R. Majhi, G. Panda, and G. Sahoo et al. Stock market
prediction of S & P 500 and DJIA using bacterial foraging
optimization technique. In CEC 2007: IEEE Congress on
Evolutionary Computation, pages 25692575, Singapore,
September, 2007. IEEE Press.
[25] S. Mishra and C. N. Bhende. Bacterial foraging
technique-based optimized active power lter for load
compensation. IEEE Transactions on Power Delivery, 22(2):
457465, Jan, 2007.
[26] L. Ulagammai, P. Vankatesh, and P. S. Kannan et al.
Application of bacteria foraging technique trained and
articial and wavelet neural networks in load forecasting.
Neurocomputing, 70(16-18): 26592667, 2007.
[27] D. H. Kim and J. H. Cho. Adaptive tuning of PID controller
for multivariable system using bacterial foraging based
optimization. In AWIC 2005: Advances in Web Intelligence
Third International Atlantic Web Intelligence Conference,
volume 3528 of Lecture Notes in Computer Science, pages
231235, Lodz, Poland, June, 2005. Springer-Verlag.
[28] B. Niu, Y. l. Zhu, and X. X. He et al. Optimum design of PID
controllers using only a germ of intelligence. In WCICA
2006: Proceedings of the 6th World Congress on Intelligent
Control and Automation, pages 35843588, Dalian, China,
June, 2006. IEEE Press.
[29] Y. Liu and K. M. Passino. Biomimicry of social foraging
bacteria for distributed optimization: Models, principles, and
emergent behaviors. Journal of Optimization Theory and
Applications, 115(3): 603628, December, 2002.
[30] S. Mishra. A hybrid least square-fuzzy bacterial foraging
strategy for harmonic estimation. IEEE Transactions on
Evolutionary Computation, 9(1): 6173, 2005.
[31] W. J. Tang, Q. H. Wu, and J. R. Saunders. Bacterial foraging
algorithm for dynamic environments. In CEC 2006: IEEE
Congress on Evolutionary Computation, pages 13241330,
BC, Canada, July, 2006. IEEE Press.
[32] A. Biswas, S. Dasgupta, and S.Das et al. Synergy of pso and
bacterial foraging optimization: A comparative study on
numerical benchmarks. In HAIS 2007: the Second
International Symposium on Hybrid Articial Intelligent
Systems, pages 255263, Salamanca, Spain, November,
2007. Springer-Verlag.
[33] D. H. Kim and J. H. Cho. Advanced bacterial foraging and
its application using fuzzy logic based variable step size and
clonal selection of immune algorithm. In ICHIT06:
Proceedings of the 2006 International Conference on Hybrid
Information Technology, volume 1, pages 293298, Cheju
Island, Korea, November, 2006. IEEE Computer Society.
[34] D. H. Kim, A. Abraham, and J. H. Cho. A hybrid genetic
algorithm and bacterial foraging approach for global
optimization. Information Sciences, 177(18): 39183937,
2007.
504