See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/281189951

A new bio-inspired optimisation algorithm: Bird Swarm Algorithm

Article in Journal of Experimental & Theoretical Artificial Intelligence · June 2015

DOI: 10.1080/0952813X.2015.1042530

CITATIONS READS

473 5,722

5 authors, including:

Xian-Bing Meng X.Z. Gao

Guangdong University of Technology 168 PUBLICATIONS 3,150 CITATIONS
15 PUBLICATIONS 1,584 CITATIONS
SEE PROFILE
SEE PROFILE

Lihua Lu Yu Liu
Shanghai Dianji University Chinese University of Hong Kong
2 PUBLICATIONS 477 CITATIONS 230 PUBLICATIONS 10,222 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Xian-Bing Meng on 12 August 2018.

The user has requested enhancement of the downloaded file.

 Journal of Experimental & Theoretical Artificial
Intelligence

ISSN: 0952-813X (Print) 1362-3079 (Online) Journal homepage: http://www.tandfonline.com/loi/teta20

A new bio-inspired optimisation algorithm: Bird

Swarm Algorithm

Xian-Bing Meng, X.Z. Gao, Lihua Lu, Yu Liu & Hengzhen Zhang

To cite this article: Xian-Bing Meng, X.Z. Gao, Lihua Lu, Yu Liu & Hengzhen Zhang (2015): A
new bio-inspired optimisation algorithm: Bird Swarm Algorithm, Journal of Experimental &
Theoretical Artificial Intelligence, DOI: 10.1080/0952813X.2015.1042530

To link to this article: http://dx.doi.org/10.1080/0952813X.2015.1042530

Published online: 17 Jun 2015.

Submit your article to this journal

Article views: 65

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=teta20

Download by: [Central South University] Date: 07 December 2015, At: 20:30
 Journal of Experimental & Theoretical Artificial Intelligence, 2015
http://dx.doi.org/10.1080/0952813X.2015.1042530

A new bio-inspired optimisation algorithm: Bird Swarm Algorithm

Xian-Bing Menga,b*, X.Z. Gaoc, Lihua Lud,e, Yu Liub* and Hengzhen Zhanga
a
College of Information Engineering, Shanghai Maritime University, Shanghai, P.R. China; bChengdu
Green Energy and Green Manufacturing R&D Center, Chengdu, P.R. China; cDepartment of Electrical
Engineering and Automation, Aalto University School of Electrical Engineering, Aalto, Finland; dSchool
of Computer Science, Fudan University, Shanghai, P.R. China; eCollege of Mathematics and Information
Science, Zhengzhou University of Light Industry, Zhengzhou, P.R. China
Downloaded by [Central South University] at 20:30 07 December 2015

(Received 31 March 2014; accepted 12 March 2015)

A new bio-inspired algorithm, namely Bird Swarm Algorithm (BSA), is proposed for
solving optimisation applications. BSA is based on the swarm intelligence extracted
from the social behaviours and social interactions in bird swarms. Birds mainly have
three kinds of behaviours: foraging behaviour, vigilance behaviour and flight
behaviour. Birds may forage for food and escape from the predators by the social
interactions to obtain a high chance of survival. By modelling these social behaviours,
social interactions and the related swarm intelligence, four search strategies associated
with five simplified rules are formulated in BSA. Simulations and comparisons based
on eighteen benchmark problems demonstrate the effectiveness, superiority and
stability of BSA. Some proposals for future research about BSA are also discussed.
Keywords: bird swarms; swarm intelligence; social behaviours; social interactions;
Bird Swarm Algorithm; optimisation

1. Introduction
Optimisation problems abound in the real world. There are many different optimisation
problems, which are continuous, discrete, linear, nonlinear, non-smooth or non-convex. The
continuously differentiable problems can be attacked by the traditional methods, such as
gradient-based methods. If the problems are very complex, such as non-convex or non-
differentiable, they may not be efficiently solved by some traditional methods. Despite there
being still several methods that can deal with some complex problems, they may not get optimal
results in reasonable computational effort. For example, the NP problems cannot be solved by
traditional methods in an affordable time.
Meta-heuristic methods come into being as an alternative to the traditional optimisation
methods. With their merits of finding acceptable solutions in an affordable time and being
tolerant of non-convex and non-differentiable, nature-inspired meta-heuristic algorithms have
attracted great research interest during the recent years.
Nature has been providing unlimited inspiration for human to learn and design new meta-
heuristic algorithms. Inspired by the abstraction of the natural evolution and selection of
biological systems, GA (Kuo & Lin, 2013), DE (Das & Suganthan, 2011), Cultural Algorithm
(Jin, 2001), Mimetic Algorithm (Smith, 2007), etc. were proposed. PSO (Rezaee Jordehi &

*Corresponding authors. Email: [email protected]; [email protected]

q 2015 Taylor & Francis
 2 X.-B. Meng et al.

Jasni, 2013), Ant Colony Optimization (Dorigo & Stutzle, 2004), Artificial Fish Swarm
Algorithm (Gao, Wu, Zenger, & Huang, 2010), Artificial Bee Colony (Karaboga & Basturk,
2007), Glowworm Swarm Optimization (Krishnanand & Ghose, 2009), Cuckoo Search (Yang &
Deb, 2014), Bat Algorithm (Yang, 2010), Krill Herd (Gandomi & Alavi, 2012), etc. were
inspired by the swarm intelligence of social animals. Inspired by the plant, Invasive Weed
Optimization (Mehrabian & Lucas, 2006) and Flower Pollination Algorithm (Yang,
Karamanoglu, & He, 2014) were designed. Harmony Search (Manjarres et al., 2013), Charged
System Search (Kaveh & Talatahari, 2010), Brain Storm Optimization (Shi, 2011), etc. were
inspired by physical principles and nature phenomena.
Though many algorithms were developed to deal with optimisation applications, there still
exists no universal algorithm. Thus, the research of finding more efficient algorithms is still in
progress.
In this paper, a new bio-inspired algorithm, namely Bird Swarm Algorithm (BSA), is
Downloaded by [Central South University] at 20:30 07 December 2015

proposed as a new optimisation method. BSA is simplification of the social behaviours and
social interactions in bird swarms. It mimics the birds’ foraging behaviour, vigilance behaviour
and flight behaviour. Thus, the swarm intelligence can be efficiently extracted from the bird
swarms to optimise problems.
The rest of the paper is organised as follows. Section 2 illustrates the details of BSA.
Simulations and comparisons are presented in Section 3. Finally, some conclusions and
discussions are given in Section 4.

2. The Bird Swarm Algorithm

2.1 Biological fundamentals
Many bird species are gregarious, such as finches. They may roost communally, forage and fly in
flocks (Anderson, 2006). These behaviours are considered as emergent behaviours arising from
simple rules such as separation, alignment and cohesion. Through the simplest social interaction,
the swarm behaviours can develop complex motions and interactions.
Foraging in flocks, birds may get more information than their own senses can gather, and
have survival advantage and good foraging efficiency. If one bird finds some food patches,
others may feed from them (Kennedy, Eberhart, & Shi, 2001). While foraging, birds often
aggregate in response to predation threat (Krause & Ruxton, 2002). They frequently raise their
heads and scan their surroundings. These behaviours, interpreted as vigilance behaviour
(Anderson, 2006), may be conducive to detecting predators (Lima & Dill, 1990). Studies
showed that birds would randomly choose between foraging and keeping vigilance (Bednekoff
& Lima, 1998). Birds often give alarm calls when they detect a predator (Pulliam, Pyke, Caraco,
& Pulliam, 1982). Thus, the whole group would fly off together. It can reasonably conclude that
birds in a flock have a better chance of detecting a potential threat than a single one. A reduction
in individual vigilance with an increase in group size is so widespread in many birds (Ekman,
1987; Elgar & Catterall, 1981; Sullivan, 1984). In other words, a bird can spend more time
foraging as group size increases without affecting the increased risk of being attacked by the
predator (Beauchamp, 1998; Roberts, 2003).
The birds on the periphery of a group have more chance of being attacked by the predators
than those in the centre. Studies suggested that animals foraging in the centre of flock may move
to their neighbours to protect themselves from being attacked by the predators (Pulliam, 1973).
Each bird would try to move towards the centre of the flock as they perceive it. This motion,
however, may be affected by the interference induced by competing among the bird swarms
(Beauchamp, 2003). Thus, birds may not directly move towards the centre of the swarm.
 Journal of Experimental & Theoretical Artificial Intelligence 3

Birds obviously would fly to another site for foraging or just for escaping from predators.
After they arrive at a new site, they would search for food again. It is usually observed that there
exist producers and scroungers in flock-feeding birds. Producers actively search for food, while
scroungers just feed from the food found by the producers (Barnard & Sibly, 1981; Giraldeau &
Caraco, 2000). Individuals usually use alternative behavioural strategies, and choose between
producing and scrounging (Liker & Barta, 2002; Coolen, Giraldeau, & Lavoie, 2001; Johnson,
Giraldeau, & Grant, 2001; Koops & Giraldeau, 1996). Studies suggested that the birds with low
reserves will often be scroungers, while the ones with high reserves would be the producers
(Barta & Giraldeau, 2000).
For flock-feeding birds, the social behaviours give the social animals a survival advantage
(Sirot, 2006). Each bird can benefit from the social interactions to achieve the optimal survival
and good foraging efficiency. The essence behind the social behaviours and the social
interactions is the swarm intelligence, which can be associated with designing a new
Downloaded by [Central South University] at 20:30 07 December 2015

optimisation method to optimise the objective problems.

2.2 Bird Swarm Algorithm theory

The aforementioned social behaviours can be simplified by some idealised rules as follows.
(1) Each bird can switch between the vigilance behaviour and foraging behaviour. Whether
bird forages or keeps vigilance is modelled as a stochastic decision (Rule 1).
(2) While foraging, each bird can promptly record and update its previous best experience
and the swarms’ previous best experience about food patch. This experience can also be
used to search for food. Social information is shared instantaneously among the whole
swarm (Rule 2).
(3) When keeping vigilance, each bird would try to move towards the centre of the swarm.
This behaviour can be affected by the interference induced by the competition among
swarm. The birds with the higher reserves would be more likely to lie closer to the centre
of the swarm than those with the lower reserves (Rule 3).
(4) Birds would periodically fly to another site. When flying to another site, birds may often
switch between producing and scrounging. The bird with the highest reserves would be a
producer, while the one with the lowest reserves would be a scrounger. Other birds with
reserves between the highest and lowest reserves would randomly choose to be producer
and scrounger (Rule 4).
(5) Producers actively search for food. Scroungers would randomly follow a producer to
search for food (Rule 5).
Given these rules formulated verbally, the new meta-heuristic algorithm by precise
mathematical model can be developed. The basic flowchart of the BSA is shown in Figure 1.
All N virtual birds, depicted by their position xti ði [ ½1; . . . ; NÞ at time step t, forage for
food and fly in a D-dimensional space.

2.2.1 Foraging behaviour

Each bird searches for food according to its experience and the swarms’ experience. Rule 2 can
be written mathematically as follows:



xtþ1
i;j ¼ x t
i;j þ p i;j 2 x t
i;j £ C £ randð0; 1Þ þ g j 2 x i;j £ S £ randð0; 1Þ ;
t
ð1Þ
 4 X.-B. Meng et al.

Initialize N solutions and

the related parameters

Generate new No No
Keep
solutions Flying? Foraging?
vigilance

Yes
Yes
Forage for food
Downloaded by [Central South University] at 20:30 07 December 2015

Yes
Producer? Search for new
food patch

No

Scrounger: Forage by
following the
producer

Fitness evaluation of
the new solutions

Update solutions

Stop criterion No
reached?

Yes
Post process
results

Figure 1. Basic flowchart of BSA.

where j [ ½1; . . . ; D, rand (0, 1) denotes independent uniformly distributed numbers in (0, 1).
C and S are two positive numbers, which can be respectively called as cognitive and social
accelerated coefficients. pi;j is the best previous position of the ith bird and gj the best previous
position shared by the swarm.
For simplicity, the Rule 1 can be formulated as a stochastic decision. If a uniform random
number in (0, 1) is smaller than P ðP [ ð0; 1ÞÞ, a constant value, the bird would forage for food.
Otherwise, the bird would continue vigilance.
 Journal of Experimental & Theoretical Artificial Intelligence 5

2.2.2 Vigilance behaviour

Given the Rule 3, birds would try to move to the centre of the swarm, and they would inevitably
compete with each other. Thus, each bird would not directly move towards the centre of the
swarm. These motions can be formulated as follows:



xtþ1
i;j ¼ x t
i;j þ A1 mean j 2 xt
i;j £ randð0; 1Þ þ A2 p k;j 2 x i;j £ randð21; 1Þ;
t
ð2Þ
 
pFiti
A1 ¼ a1 £ exp 2 £N ; ð3Þ
sumFit þ 1
  
pFiti 2 pFitk N £ pFitk
A2 ¼ a2 £ exp ð4Þ
j pFitk 2 pFiti j þ 1 sumFit þ 1
Downloaded by [Central South University] at 20:30 07 December 2015

where k ðk – iÞ is a positive integer, which is randomly chosen between 1 and N. a1 and a2 are
two positive constants in [0, 2], pFiti denotes the ith bird’s best fitness value and sumFit
represents the sum of the swarms’ best fitness value. 1, which is used to avoid zero-division
error, is the smallest constant in the computer. meanj denotes the jth element of the average
position of the whole swarm.
When a bird moves towards the centre of the swarm, it will inevitably compete with each
other. For simplicity, the average fitness value of the swarm is considered as the indirect effect
induced by the surroundings when a bird moves to the centre of the swarm. Given the fact that
each bird wants to stand at the centre of swarm, the product of A1 and rand(0,1) should not be
more than 1. Here, A2 is used to simulate the direct effect induced by a specific interference
when a bird moves to the centre of the swarm. If the best fitness value of a random kth bird
(k – i) is better than that of the ith bird, then A2 . a2, which means that the ith bird may suffer a
greater interference than the kth bird. Though there are some randomness and unpredictability,
the kth bird would be more likely to move towards the centre of the swarm than the ith bird.
In this paper, it solves the minimal optimisation problems, unless otherwise indicated. Thus, the
smaller the fitness value is, the better the fitness value will be.

2.2.3 Flight behaviour

Birds may fly to another site in response to predation threat, foraging or any other reasons. When
arrived at a new site, they would forage for food again. Some birds acting as producers would
search for food patches, while others try to feed from the food patch found by the producers. The
producers and scroungers can be separated from the swarm according to Rule 4. The behaviours
of the producers and scroungers can be described mathematically as follows, respectively:

i;j ¼ xi;j þ randnð0; 1Þ £ xi;j ;

xtþ1 ð5Þ
t t


xtþ1
i;j ¼ xt
i;j þ x t
k;j 2 x i;j £ FL £ randð0; 1Þ;
t
ð6Þ

where randn(0,1) denotes Gaussian distributed random number with mean 0 and standard
deviation 1, k [ ½1; 2; 3; . . . N; k – i. FLðFL [ ½0; 2Þ means that the scrounger would follow
the producer to search for food.
For simplicity, we assume that each bird flies to another place every FQ unit interval. Here,
FQ is a positive integer.
 6 X.-B. Meng et al.

2.2.4 The computational procedure and computational complexity of BSA

Based on the aforementioned descriptions, BSA can be summarised as the pseudo code shown in
Table 1.
The computational complexity of BSA can be estimated. According to Figure 1 and Table 1,
the procedures of generating, evaluating and updating new solutions need the highest
computational overhead of executing BSA. The computational complexities of other steps are
rather simple, and can be neglected. Thus the ‘worst-case’ complexities can be computed by
considering the procedures of generating, evaluating and updating new solutions.
The time complexity of generating, evaluating and updating new solutions is the same,
namely OðMNÞ. To summarise, the overall computational complexity of the BSA is OðMNÞ.

3. Validation and comparison

Downloaded by [Central South University] at 20:30 07 December 2015

In this section, 18 benchmark problems (shown in Table 2) (Yang, 2014) are optimised to
investigate the effectiveness, efficiency and stability of BSA. These problems contain uni-
modal, multi-modal, high-dimensional and low-dimensional cases. The formulae of BSA reveal
a natural relationship between BSA, and PSO and DE. Moreover, PSO and DE are two famous
meta-heuristic algorithms; thus, the two algorithms will be used as the comparison algorithms.
In all case studies, the statistical results would be used to investigate the BSA’s superiority by

Table 1. Pseudo code of BSA.

Bird Swarm Algorithm (BSA)
Input: N: the number of individuals (birds) contained by the population
M: the maximum number of iteration
FQ: the frequency of birds’ flight behaviours
P: the probability of foraging for food
C, S, a1, a2, FL: five constant parameters
t ¼ 0; Initialise the population and define the related parameters
Evaluate the N individuals’ fitness value, and find the best solution
While (t , M)
If (t % FQ–0)
For i ¼ 1: N
If rand(0,1) , P
Birds forage for food (Equation 1)
Else
Birds keep vigilance (Equation 2)
End if End for
Else
Divide the swarm into two parts: producers and scroungers.
For i ¼ 1: N
If i is a producer
Producing (Equation 5)
Else
Scrounging (Equation 6)
End if End for
End if Evaluate new solutions
If the new solutions are better than their previous ones, update them
Find the current best solution
t ¼ t þ 1; End while
Output: the individual with the best objective function value in the population
 Downloaded by [Central South University] at 20:30 07 December 2015

Table 2. Benchmark problems.

Problem Bounds Optimum
P 2
F1 ¼ 20 i¼1 xi
[2 100,100] 0
2
P20
Pi [2 100,100] 0
F2 ¼ i¼1 j¼1 xj
F3 ¼ maxi f j xi j; 1 # i # 20 } [2 100,100] 0
P Q
F4 ¼ 20 jxi j þ 20 i¼1 jx j [2 10,10] 0
Pi¼1  2 i
F5 ¼ 20 bx i þ 0:5c [2 100,100] 0
Pi¼1 iþ1
F6 ¼ 20 i¼1 jxi j
[2 1,1] 0
2
P 4
P20 2
P20 20 [25,10] 0
F7 ¼ i¼1 xi þ i¼1 0:5ixi þ i¼1 0:5ixi
P 2
F8 ¼ 20 i¼1 ði xi Þ
[2 5.12, 5.12] 0
P20 2
F9 ¼ i¼1 ðxi 2 10 cosð2 pxi Þ þ 10Þ [2 5.12, 5.12] 0
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P  
1
P20 2 1 20 2 [232, 32] 0
F10 ¼ 220 expð20:2 20 i¼1 xi 2 exp 20 i¼1 cos 2pxi þ 20 þ e
P Q

1 20 2 20 p
xiffi [2600, 600] 0
F11 ¼ 4000 i¼1 xi 2 i¼1 cos i
þ1
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
F12 ¼ xsin jy þ 1 2 xjcos jy þ 1 þ xj þ ðy þ 1Þcos jy þ 1 2 xjsin jy þ 1 þ xj [2512, 512] 2 511.7

2
[2 5.12, 5.12] 2 3600
F13 ¼ 2 0:05þðx3 2 þy 2 Þ 2ðx 2 þ y 2 Þ2
F14 ¼ 0:5*ðx 4 þ y 42 16* ðx 2 þ y 2 Þ þ 5* ðx þ yÞÞ [25, 5] 2 78.3323
F15 ¼ 0:26 x21 þ x22 2 0:48x1 x2 [210, 10] 0
F16 ¼ ð1:5 2 x1 þ x1 x2 Þ2 þ ð2:25 2 x1 þ x1 x22 Þ2 þ ð2:625 2 x1 þ x1 x32 Þ2 [2 4.5, 4.5] 0
Journal of Experimental & Theoretical Artificial Intelligence

2 2
F17 ¼ 1 þ sin2 ðx1 Þ þ sin2 ðx2 Þ 2 0:1e2ðx1 þx2 Þ [210, 10] 0.9
P5 P5
F18 ¼ i¼1 ½icosðði 2 1Þx þ iÞ j¼1 jcosððj þ 1Þy þ jÞ þ ðx þ 1:42513Þ2 þ ðy þ 0:80032Þ2 [210, 10] 2176.1375
7
 8 X.-B. Meng et al.

comparing with PSO and DE. For a fair comparison, all of the parameters of these methods, such
as the population size, maximum number of iterations, etc., are set to be the same. The
population size is 6 times and 5 times the specific dimension of F1– F11 and F12 – F18,
respectively. All algorithms would run 100 independent trials with 1000 iterations. All the
experiments using Visual Studio 2010 are performed on a computer with 3.2 GHz quad-core
processor and 2.0 GB of RAM in Windows 7 operating system. The related parameter values of
each algorithm are given in Table 3.
According to the statistic results given in Table 4, it is obvious that BSA is superior to PSO
and DE. PSO and DE may trap into local optima for solving F2, F9, F12, F13, F14, F16, F17
and F18. DE may also be susceptible to premature convergence when solving F3 and F10.
As for the accuracy and stability, BSA can yield more accurate and stable results than those of
PSO and DE for solving all the problems except F5, which can be solved by these three
algorithms with the comparable precision.
Downloaded by [Central South University] at 20:30 07 December 2015

The trade-off between accuracy and convergence time of these three algorithms are
presented in Figure 2. Compared with the results of PSO and DE for solving F1, F2, F3, F5, F7–
F11 and F13, BSA can significantly improve convergence speed and reduce calculation amount
without affecting the accuracy of the results. BSA can also yield comparable convergence
performance with PSO and DE for solving the other problems except F12. The convergence
performance of BSA for solving F12 is slightly worse than those of PSO and DE.
Having analysed the details about PSO and DE, it can be concluded that time complexity of
PSO and DE is the same as that of BSA. The theoretical time complexity of these three
algorithms is OðMNÞ.
Based on the comparison results given in Table 4 and Figure 2, it can be reasonably
concluded that the performance of the BSA outperforms that of PSO and DE in terms of
optimisation accuracy, efficiency, stability and convergence performance.
The superiority of BSA over PSO and DE should be the case. In fact, PSO and the mutation
operator of DE are the special cases of the BSA under appropriate simplifications. The formula
describing the birds’ foraging behaviour in BSA is similar to that of PSO. If we set FQ as an
infinite number and P ¼ 1, BSA essentially becomes the standard PSO. Moreover, the formula
representing the scroungers’ behaviours in BSA corresponds to the mutation operator in DE.
Besides possessing the merits of PSO and DE, BSA has its own distinguishing features. BSA has
four different search strategies and can flexibly adjust the different search strategies, thus
making BSA strike better balance between exploration and exploitation than PSO and DE.

4. Conclusions and discussions

Imitating the best feature in nature to solve optimisation problems is an ongoing work. In the
present work, Bird Swarm Algorithm, inspired from the bird swarms, is proposed as a new
method for solving optimisation applications. Birds mainly have three kinds of behaviours,
including foraging behaviour, vigilance behaviour and flight behaviour, and can obtain a

Table 3. The parameter values of the three algorithms.

Algorithm Parameter values

PSO c1 ¼ c2 ¼ 1.49445, w ¼ 0.729
DE CR ¼ 0.9, F ¼ 0.6
BSA C ¼ S ¼ 1.5,a1 ¼ a2 ¼ 1, P [[0.8,1], FL[[0.5,0.9], FQ ¼ 3
 Journal of Experimental & Theoretical Artificial Intelligence 9

Table 4. Comparison of statistical results obtained by BSA, PSO and DE.

Problem Algorithm Best Mean Worst SD

F1 PSO 0 0 0 1.8988e 2 35
BSA 0 0 0 0
DE 0 0 0 1.94304e 2 12
F2 PSO 191.61632 1181.68129 4637.12542 81.9389
BSA 0 0 0 0
DE 14.57637 147.07872 555.31305 10.5381
F3 PSO 0 1.6e 2 7 1.41e 2 6 2.31673e 2 8
BSA 0 0 0 1.91005e 2 130
DE 0.00375 3.08942 30.27557 0.430579
F4 PSO 0 0 0 1.10567e 2 20
BSA 0 0 0 8.78619e 2 133
DE 4.3e 2 7 2.57e 2 6 4.075e 2 5 4.32847e 2 7
Downloaded by [Central South University] at 20:30 07 December 2015

F5 PSO 0 0 0 0
BSA 0 0 0 0
DE 0 0 0 0
F6 PSO 0 0 0 4.78569e 2 60
BSA 0 0 0 0
DE 0 0 0 0
F7 PSO 0 0 0 6.58358e 2 11
BSA 0 0 0 0
DE 0.00001 0.00013 0.00156 2.03004e 2 5
F8 PSO 0 0 0 2.13096e 2 37
BSA 0 0 0 0
DE 0 0 0 4.15726e 2 14
F9 PSO 10.94454 21.26284 41.78822 0.600484
BSA 0 0 0 0
DE 8.41884 22.70527 43.97510 0.706925
F10 PSO 0 0 0 1.31168e 2 16
BSA 0 0 0 0
DE 0 0.35702 11.64977 0.170962
F11 PSO 0 5e 2 8 5.34e 2 6 5.36538e 2 8
BSA 0 0 0 0
DE 0 0 0 1.05399e 2 12
F12 PSO 2 511.70773 2 504.67360 2 456.50873 0.917353
BSA 2 511.70773 2 511.56605 2 510.08006 0.047209
DE 2 511.70773 2 508.77971 2 440.30385 0.923235
F13 PSO 2 3600 2 2884.94048 22748.78234 31.5129
BSA 2 3600 2 3600 2 3600 0
DE 22748.782 2 2748.782 2 2748.782 4.59341e 2 13
F14 PSO 278.33233 2 77.20139 2 64.19561 0.387394
BSA 278.33230 2 78.33230 2 78.33210 4.44555e 2 6
DE 278.33230 2 77.91132 2 64.19561 0.216586
F15 PSO 0 0 0 5.10838e 2 79
BSA 0 0 0 0
DE 0 2.487e 2 5 2.15633e 2 3 2.17425e 2 5
F16 PSO 0 0.15241 0.76207 0.0307907
BSA 0 0 0 1.22573e 2 7
DE 0 0.03792 0.76218 0.0155155
F17 PSO 0.9 0.97700 1 0.00425083
BSA 0.9 0.9 0.9 7.85006e 2 17
DE 0.9 0.99884 1.054495 0.0027342
F18 PSO 2176.1376 2 172.8113 2 103.6089 1.21073
BSA 2176.1375 2 176.1366 2 176.1313 1.1779e 2 4
DE 2176.1376 2 140.4818 249.6449 3.29307
 10 X.-B. Meng et al.

x 104 Convergence curves of F1 x 105 Convergence curves of F2

4 4
BSA BSA

PSO PSO
3 3
DE DE

Best fitness
Best fitness

Optimum Optimum
2 2

1 1

0 0
Downloaded by [Central South University] at 20:30 07 December 2015

0 200 400 600 800 1000 0 200 400 600 800 1000
iterations iterations

Convergence curves of F3 x 104 Convergence curves of F4

80 8
BSA BSA

60 PSO 6 PSO

DE DE
Best fitness
Best fitness

Optimum Optimum
40 4

20 2

0 0
0 200 400 600 800 1000 0 200 400 600 800 1000
iterations iterations

x 104 Convergence curves of F5 Convergence curves of F6

4 1
BSA BSA

PSO 0.5 PSO

3
DE DE
Best fitness

Best fitness

Optimum Optimum
2 0

1 –0.5

0 –1
0 200 400 600 800 1000 0 200 400 600 800 1000
iterations iterations

Figure 2. Convergence curves of the three algorithms for solving problems F1 – F18.
 Journal of Experimental & Theoretical Artificial Intelligence 11

Convergence curves of F7 Convergence curves of F8

400 1000

BSA BSA
800
300 PSO PSO

DE DE
Best fitness

Best fitness
600
Optimum Optimum
200
400

100
200

0 0
Downloaded by [Central South University] at 20:30 07 December 2015

0 200 400 600 800 1000 0 200 400 600 800 1000
iterations iterations

Convergence curves of F9 Convergence curves of F10

250 30
BSA BSA
200
PSO 20 PSO

DE DE
Best fitness
Best fitness

150
Optimum Optimum
10
100

0
50

0 –10
0 200 400 600 800 1000 0 200 400 600 800 1000
iterations iterations

Convergence curves of F11 Convergence curves of F12

400 0

BSA BSA

300 PSO PSO

DE –200 DE
Best fitness

Best fitness

Optimum Optimum
200

–400
100

0 –600
0 200 400 600 800 1000 0 200 400 600 800 1000
iterations iterations

Figure 2. Continued.
 12 X.-B. Meng et al.

Convergence curves of F13 Convergence curves of F14

0 0

BSA BSA

–1000 PSO –20 PSO

DE DE
Best fitness

Best fitness
–2000 Optimum –40 Optimum

–3000 –60

–4000 –80
0 200 400 600 800 1000 0 200 400 600 800 1000
Downloaded by [Central South University] at 20:30 07 December 2015

iterations iterations

Convergence curves of F15 Convergence curves of F16

1.5 3
BSA BSA

PSO PSO
1 DE 2 DE
Best fitness
Best fitness

Optimum Optimum

0.5 1

0 0
0 200 400 600 800 1000 0 200 400 600 800 1000
iterations iterations

Convergence curves of F17 Convergence curves of F18

1.5 50
BSA
BSA
PSO 0
PSO

1 DE
DE
Best fitness

Best fitness

–50
Optimum
Optimum

–100
0.5

–150

0 –200
0 200 400 600 800 1000 0 200 400 600 800 1000
iterations iterations

Figure 2. Continued.
 Journal of Experimental & Theoretical Artificial Intelligence 13

survival advantage by the social interactions with others. BSA is developed to optimise
problems using the swarm intelligence extracted from the bird swarms.
The innovation in this paper lies not only in efficiently extracting the swarm intelligence
from the bird swarms to optimise problems, but also in making BSA an innate integration
algorithm. PSO and the mutation operator of DE are the special cases of BSA under appropriate
simplifications. The four search strategies in BSA make BSA easily extensible. Besides
obtaining the merits of PSO and DE, BSA has its own merits. By mimicking birds’ vigilance
behaviours and the producers’ behaviours, BSA can have good diversity, and thus efficiently
avoid prematurity. The performance of BSA is compared with that of PSO and DE. Numerous
experiments on 18 benchmark problems show that BSA is more accurate, efficient and robust
than PSO and DE.
In fact, we just proposed a uniform framework of BSA. Various representations can be used
to interpret the aforementioned rules, and develop new variants of BSA.
Downloaded by [Central South University] at 20:30 07 December 2015

How to design an efficient algorithm is still required to be studied in the future. The essence
of meta-heuristic algorithm is to look for the balance between exploration and exploitation.
To summarise, there mainly exist four categories to improve the performance of BSA.
The first category of BSA’s variants can achieve improvements by adjusting parameter
configurations. For the parameter C, S and FL, the analyses and conclusions of those
corresponding parameters in PSO and DE can be used. It must emphasise that BSA is different
from PSO and DE. The parameters should be analysed along with the other related parameters.
Meanwhile, we may try to let the frequency of birds’ flight behaviours (FQ) change
dynamically.
The second category may be realised by introducing auxiliary search strategies to enhance
BSA’s performance. For example, the LévyFlights may be incorporated into the proposed
algorithm when birds fly to other sites.
Defining neighbourhood topologies and introducing multi-swarm techniques into the
algorithm are another two feasible ways to enhance the performance of BSA. Besides improving
the algorithm mechanism itself, hybridising other metaheuristic algorithms and other
optimisation methods into BSA is also a good way to improve the proposed algorithms.
Moreover, future work can also focus on solving discrete, combinatorial and more complex
optimisation applications using BSA and its variants.

Disclosure statement
No potential conflict of interest was reported by the authors.

Funding
This work was supported by the New Academic Staff Program of Shanghai Maritime University under
Grant GK2013089.The authors also gratefully acknowledge the useful suggestions of the reviewers and
editors, which have processed and improved the presentation.

References
Anderson, T. R. (2006). Biology of the ubiquitous house sparrow: From genes to populations. Oxford:
Oxford University Press.
Barnard, C. J., & Sibly, R. M. (1981). Producers and scroungers: A general model and its application to
captive flocks of house sparrows. Animal Behavior, 29, 543– 550. doi:10.1016/S0003-3472(81)
80117-0
 14 X.-B. Meng et al.

Barta, Z., & Giraldeau, L. A. (2000). Daily patterns of optimal producer and scrounger use under predation
hazard: A state-dependent dynamic game analysis. American Naturalist, 155, 570– 582. doi:10.
1086/303342
Beauchamp, G. (1998). The effect of group size on mean food intake rate in birds. Biological Reviews, 73,
449– 472. doi:10.1017/S0006323198005246
Beauchamp, G. (2003). Group-size effects on vigilance: A search for mechanisms. Behavioral Processes,
63, 111– 121. doi:10.1016/S0376-6357(03)00002-0
Bednekoff, B. A., & Lima, S. L. (1998). Randomness, chaos and confusion in the study of antipredator
vigilance. Trends in Ecology and Evolution, 13, 284– 287. doi:10.1016/S0169-5347(98)01327-5
Coolen, I., Giraldeau, L. A., & Lavoie, M. (2001). Head position as an indicator of producer and scrounger
tactics in a ground-feeding bird. Animal Behavior, 61, 895– 903. doi:10.1006/anbe.2000.1678
Das, S., & Suganthan, P. N. (2011). Differential evolution: A survey of the state-of-the-art. IEEE
Transactions on Evolutionary Computation, 15, 4 – 31. doi:10.1109/TEVC.2010.2059031
Dorigo, M., & Stutzle, T. (2004). Ant colony optimization. Cambridge, MA: MIT Press.
Downloaded by [Central South University] at 20:30 07 December 2015

Ekman, J. (1987). Exposure and time use in willow tit flocks: The cost of subordination. Animal Behavior,
35, 445– 452. doi:10.1016/S0003-3472(87)80269-5
Elgar, M. A., & Catterall, C. P. (1981). Flocking and predator surveillance in house sparrows: Test of an
hypothesis. Animal Behavior, 29, 868–872. doi:10.1016/S0003-3472(81)80022-X
Gandomi, A. H., & Alavi, A. H. (2012). Krill herd: A new bio-inspired optimization algorithm.
Communications in Nonlinear Science Numerical Simulation, 17, 4831– 4845. doi:10.1016/j.cnsns.
2012.05.010
Gao, X. Z., Wu, Y., Zenger, K., & Huang, X. L. (2010). A knowledge-based artificial fish-swarm algorithm.
In 13th IEEE international conference on computational science and engineering (pp. 327– 332).
Hong Kong: IEEE Computer Society.
Giraldeau, L. A., & Caraco, T. (2000). Social foraging theory. Princeton, NJ: Princeton University Press.
Jin, X. D. (2001). Solving constrained optimization problems using cultural algorithm and regional
schemata (Doctorate dissertation). Detroit, MI: Wayne State University.
Johnson, C. A., Giraldeau, L. A., & Grant, J. W. A. (2001). The effect of handling time on interference
among house sparrows foraging at different seed densities. Behavior, 138, 597– 614. doi:10.1163/
156853901316924494
Karaboga, D., & Basturk, B. (2007). A powerful and efficient algorithm for numerical function
optimization: Artificial bee colony (ABC) algorithm. Journal of Global Optimization, 39, 459– 471.
doi:10.1007/s10898-007-9149-x
Kaveh, A., & Talatahari, S. (2010). A novel heuristic optimization method: Charged system search. Acta
Mechanica, 213, 267– 289. doi:10.1007/s00707-009-0270-4
Kennedy, J., Eberhart, R. C., & Shi, Y. H. (2001). Swarm intelligence. Burlington, MA: Morgan Kaufmann.
Koops, M. A., & Giraldeau, L. A. (1996). Producer –scrounger foraging games in starlings: A test of rate-
maximizing and risk-sensitive models. Animal Behavior, 51, 773– 783. doi:10.1006/anbe.1996.0082
Krause, J., & Ruxton, G. D. (2002). Living in groups. Oxford: Oxford University Press.
Krishnanand, K. N., & Ghose, D. (2009). Glowworm swarm optimization for simultaneous capture of
multiple local optima of multimodal functions. Swarm Intelligence, 3, 87 – 124. doi:10.1007/s11721-
008-0021-5
Kuo, H. C., & Lin, C. H. (2013). A directed Genetic Algorithm for global optimization. Applied
Mathematics and Computation, 219, 7348 –7364. doi:10.1016/j.amc.2012.12.046
Liker, A., & Barta, Z. (2002). The effects of dominance on social foraging tactic use in house sparrows.
Behavior, 139, 1061– 1076. doi:10.1163/15685390260337903
Lima, S. L., & Dill, L. M. (1990). Behavioral decisions made under the risk of predation: A review and
prospectus. Canadian Journal of Zoology, 68, 619– 640. doi:10.1139/z90-092
Manjarres, D., Landa-Torres, I., Gil-Lopez, S., Del Ser, J., Bilbao, M. N., Salcedo-Sanz, S., & Geem, Z. W.
(2013). A survey on applications of the harmony search algorithm. Engineering Applications of
Artificial Intelligence, 26, 1818– 1831. doi:10.1016/j.engappai.2013.05.008
 Journal of Experimental & Theoretical Artificial Intelligence 15

Mehrabian, A. R., & Lucas, C. (2006). A novel numerical optimization algorithm inspired from weed
colonization. Ecological Informatics, 1, 355– 366. doi:10.1016/j.ecoinf.2006.07.003
Pulliam, H. R. (1973). On the advantages of flocking. Journal of Theoretical Biology, 38, 419– 422. doi:10.
1016/0022-5193(73)90184-7
Pulliam, H. R., Pyke, G. H., Caraco, T., & Pulliam, H. R. (1982). The scanning behavior of juncos: A game-
theoretical approach. Journal of Theoretical Biology, 95, 89 – 103. doi:10.1016/0022-5193(82)
90289-2
Rezaee Jordehi, A. R., & Jasni, J. (2013). Parameter selection in particle swarm optimisation: A survey.
Journal of Experimental & Theoretical Artificial Intelligence, 25, 527– 542. doi:10.1080/0952813X.
2013.782348
Roberts, G. (2003). The group-size effect in non-feeding animals. Behavioral Processes, 63, 127– 128.
doi:10.1016/S0376-6357(03)00005-6
Shi, Y. H. (2011). Brain storm optimization algorithm. In Proceedings of 2th International Conference on
Swarm Intelligence (pp. 303– 309). Chongqing: IEEE Computational Intelligence Society.
Downloaded by [Central South University] at 20:30 07 December 2015

Sirot, E. (2006). Social information, antipredatory vigilance and flight in bird flocks. Animal behavior, 72,
373– 382. doi:10.1016/j.anbehav.2005.10.028
Smith, J. E. (2007). Coevolving Memetic Algorithms: A review and progress report. IEEE Transaction on
System, Man, and Cybernetics – Part B: Cybernetics, 37, 6– 17.
Sullivan, K. A. (1984). The advantages of social foraging in downy woodpeckers. Animal Behavior, 32,
16 – 22. doi:10.1016/S0003-3472(84)80319-X
Yang, X. S. (2010). A new metaheuristic bat-inspired algorithm. Nature Inspired Cooperative Strategies
for Optimization (NICSO 2010), Studies in Computational Intelligence, 284, 65 – 74.
Yang, X. S. (2014). Nature-inspired optimization algorithm. London: Elsevier.
Yang, X. S., & Deb, S. (2014). Cuckoo search: Recent advances and applications. Neural Computing &
Applications, 24, 169– 174. doi:10.1007/s00521-013-1367-1
Yang, X. S., Karamanoglu, M., & He, X. S. (2014). Flower pollination algorithm: A novel approach for
multiobjective optimization. Engineering Optimization, 46, 1222– 1237. doi:10.1080/0305215X.
2013.832237

View publication stats