2016 Intl. Conference on Advances in Computing, Communications and Informatics (ICACCI), Sept.

21-24, 2016, Jaipur, India

Fast Convergent Biogeography based Optimization

Algorithm
Priya Sharmaa , Harish Sharmab and Nirmala Sharmac
Rajasthan Technical University, Kota, India
Email: a [email protected], b [email protected] , c [email protected]

Abstract—Biogeography Based Optimization (BBO) Algorithm produced to solve distinct optimizing problems.
is a stochastic and population-based evolutionary search tech- In this paper, to achieve the fast convergence and maintains a
nique modeled on the theory of biogeography. Like other proper balance between the intensification and diversification
population-based evolutionary algorithms, BBO also suffers from
the problem of slow convergence. Therefore, in this article, a new capability of BBO algorithm, a modified variant of BBO
variant of BBO algorithm is introduced, namely Fast Convergent is studies, named as Fast Convergent Biogeography based
Biogeography based Optimization (FCBBO) Algorithm. In the Optimization (FCBBO) Algorithm. In FCBBO, to explore
proposed algorithm, a fitness based position update process is the solution search field, the individuals have to take large
introduced. In the proposed fitness based process, step size of step size where as to exploit the search space the step size
the solutions are controlled through a probability which is a
function of fitness. Further, Mutualism-relationship is introduced is less comparative to avoid the situation of skipping the
in migration process which modifies individuals by evaluating the optima. Further, we also introduced a Mutualism-relationship
difference among the finest solution and the mean of two random in migration process which imitates the interactive nature seen
individuals. This modification helps to improve the exploration between individuals. In this candidate solutions generating by
capability of the proposed algorithm. The developed algorithm evaluating the difference among the finest individual and the
compared with BBO and two other algorithms, namely Differen-
tial Evolution (DE) Algorithm and Particle Swarm Optimization mean of two individuals which gives it an advantage in ex-
(PSO) Algorithm with the experiments over 12 test problems. ploring new areas. This strategy maintains the proper balance
Obtained results confirm the competitive performance of the between the intensification and diversification properties of the
proposed algorithm. algorithm and enhances the efficiency of BBO algorithm also.
Index Terms—Meta-heuristic Optimization Techniques; Evo- Section II of this paper gives the brief overview of BBO
lutionary Algorithms; Heuristic search algorithm; Mutualism-
relationship Algorithm and the modified algorithm portrayed in Section
III. Performance of FCBBO is tested with several numerical
I. I NTRODUCTION benchmark functions and presented under Section IV. Finally,
Section V gives a summary and conclude the work.
Nature Inspired Algorithms (NIA) [1] are a significant
source for motivations to both grow intelligent systems and II. OVERVIEW OF BBO ALGORITHM
provide solutions to challenging engineering optimization
problems. Various evolutionary methods have been evolved BBO [9] is an evolutionary search technique that describes
to deal with difficult engineering optimization problems. Evo- the dispersion of bio-diversity of biological species through
lutionary processes influenced by the natural selection and space and time. The Scientific model of biogeography de-
model of the biological evolution process. Evolutionary [2] scribes Speciation (evolution of new species), Migration of
algorithms optimize a function by repetitively developing species (animal, birds and plants) among island (habitat)
candidate solution in terms of given measure of goodness. and Extinction of species. In BBO [9], each individual is
BBO is an evolutionary search algorithm that is modeled on considered to be analogs to an island and is recognized by
the science of biogeography that contains sharing of biological habitat (island) suitability index (HSI). Habitat (solution) that
species through space and time. are friendly to life are known with high HSI. HSI corresponds
Research globally in the field of BBO has produced new to the goodness of a BBO solution. Facets that correlate
optimization methods that have proven greater to primary with HSI include rainfall, land area, temperature, vegetation
approaches. Approximately, entire metaheuristic methods [3, diversity, topographic diversity and others. Facets that identify
4, 5, 6, 7] share the identical properties such as the natural habitability are known as suitability index variable (SIVs).
phenomenon inspires they all, use random variables and they Figure 1 demonstrates a representation of species profuse in
have several attributes. Every metaheuristic method has dif- a particular island.
ferent assets on reliability, efficiency, robustness and accuracy In terms of habitability, SIVs can be taken as independent
in a noisy atmosphere and unique problem domain. Accord- variables and HSI as dependent variables of the island respec-
ing to ”no-free-lunch” theorem [8], it is impossible for any tively. Habitat that is supported by a large number of species is
metaheuristic method to handle entire optimization problems called High-HSI habitat and habitat with only a few species is
optimally. Thus, new metaheuristic methods are continuously called Low-HSI habitat. High-HSI solutions give their facets

Authorized licensed use limited to: Rajasthan Technical University. Downloaded on May 01,2025 at 06:18:03 UTC from IEEE Xplore. Restrictions apply.
978-1-5090-2029-4/16/$31.00 @2016 IEEE 782
 2016 Intl. Conference on Advances in Computing, Communications and Informatics (ICACCI), Sept. 21-24, 2016, Jaipur, India

here α is the user defined migration operator and k is the

dimension of the solution.
B. Mutation
Mutation is accountable for keeping the variety of solutions
in BBO algorithm. Mutation provides a chance to improve the
goodness of solutions for low-HSI candidate solutions, and
for high-HSI solutions, it can enhance the quality of solutions
even more than solutions already have.
After, each generation we apply elitism in BBO for pre-
serving the best solutions from present generation to future
generation. Elitism prevents the best solution from being
polluted by migration or mutation of the future generation.
From the above discussion in Section II, procedure of the
BBO is represented in Algorithm 1.

Algorithm 1 BBO Algorithm

creating a randomly dispersed set of solutions H1, H2,
...,Hn;
Fig. 1: Species Model [9] Calculate goodness (HSI) of every solutions;
while termination criteria met do
Sort the solutions from best to worst;
Calculate λ and μ for each solutions based on HSI ;
to low-HSI solutions and new facets are accepted by low- Apply migration procedure;
HSI solutions from high-HSI solutions. High-HSI solutions Renew the probability for each solutions;
emigrate their facets to low-HSI solutions. Apply mutation procedure;
The BBO algorithm starts by creating a randomly dispersed Calculate goodness of every solutions;
population of habitats. Then evaluate the emigration rate (μ) Apply elitism for maintaining best solutions;
and immigration rate (λ) for every individual. end while

ki III. FAST C ONVERGENT B IOGEOGRAPHY BASED

λi = I(1 − ) (1)
n O PTIMIZATION A LGORITHM
ki
μi = )
E( (2) In the evolutionary optimization algorithms, exploration
n and exploitation are the two vital processes. The exploration
Here, E is highest possible μ, I is highest possible λ and ki consists of probing a much larger portion of the search region
is goodness rank of ith individual after sorting goodness of with the desire of finding other promising solutions that is
ith individual. yet to be refined. While the exploitation consists of probing a
After initialisation of the BBO population, it needs the limited region of the search space with the desire of inspiring
repeated iterations of the two procedures, named as migration a promising solution that we already have in hand. For
procedures and another one is mutation procedures. Each achieving superior performance in optimization algorithms,
procedure is portrayed as follows. both exploration and exploitation properties should be well
A. Migration coordinated with each other.
Like other evolutionary algorithms, BBO also suffers from
Migration procedure is probabilistically sharing of facets the problem of stagnation. To avoid issues of stagnation in
between habitats by taking advantage of the immigration local optima, it is necessary that initially algorithm explore
rate (λi ) and emigration rate (μj ). The migration operator the search space while exploits in later iterations. From the
(α) is accountable for sharing the facets between individuals equation 3 of basic BBO, it is clear that immigrating solution
to modify goodness. The immigrating solution is selected accepts facets only from emigrating solution. Therefore, in this
according to the probability of (λi ) and emigrating solution paper, we modify the migration equation of the individuals as
is selected according to the probability of (μj ) of solutions. follows:
Based on these probabilities algorithm decide which of the
solution feature(SIV) of immigrating solution(Hi ) is needed IntelligenceF actor
to be modified. Then immigrating solution(Hi ) SIV is replaced   
by emigrating solution’s (Hj ) SIV. new(Hik ) = (prob(i) ∗ Hik ) + (1.1 − prob(i))
Mutualism−relationship
(4)
  
new(Hik ) = Hik + α(Hjk − Hik ) (3) ∗ (HBestSolk − M V ∗ BF )

Authorized licensed use limited to: Rajasthan Technical University. Downloaded on May 01,2025 at 06:18:03 UTC from IEEE Xplore. Restrictions apply.
783
 2016 Intl. Conference on Advances in Computing, Communications and Informatics (ICACCI), Sept. 21-24, 2016, Jaipur, India

Here, MV is a M utual−V ector that represent the relationship Algorithm 2 FCBBO Algorithm
characteristic between two random individuals Hik and Hjk Generate a random set of solutions H1, H2, ...,Hn;
which is calculated using equation 5. Calculate HSI values;
Hik + Hjk while the termination criteria is not met do
M V (M utual − V ector) = (5) MIGRATION
2
BF is called Benef it − F actor that is random either 1 or Sort the solutions from best to worst;
2 and HBestSolk is the finest solution in all individuals. The Compute λ and μ for each solutions based on HSI ;
probi is a probability which is a function of individual’s fitness For any solution;
and calculated by equation 6. For any SIV (k) (solution dimension);
Elect solution Hi with probability ∝ λi ;
0.9 × f it(i) if Hi is selected then
probi = + 0.1 (6)
maxf it(i) Elect habitat Hj with probability ∝ μj ;
The proposed strategy follows the approach that the probability if Hj is elected then
will be high for superior fit individuals while the probability Calculate MV (M utual−vector) by using equation
is low for inferior fit individuals. 5;
Here, f it(i) is the fitness of ith individual. Fitness is new(Hik ) = (prob(i) ∗ Hik ) + (1.1 − prob(i)) ∗
calculated using the objective value (fi ) as equation 7. (HBestSolk − M V ∗ BF );
 end if
1
, if fi ≥ 0
individual fitness (f it(i)) = 1+fi (7) end if
1 + abs(fi ), if fi < 0 MUTATION
In equation 4, the term (1.1 − probi ) is an intelligence factor Elect Hi (SIV ) based on mi ;
(IF). The value of IF will be low for high fit solutions (due if Hi (SIV ) is elected then
to high value of probi ), hence the step size will be low. So, Replace Hi (SIV ) with a randomly created SIV;
the individuals having a higher probability (higher fitness) will end if
move slowly and exploit the search space in its neighbourhood. Recompute HSI values;
While for the fewer probability individuals, the value of IF will end while
be high (due to low value of probi ) so as the step size.
Further, the term (HBestSolk − M V ∗ BF ) of equation 4
shows that through MV and BF, the solution going to update is B. Parameter setting
force to move towards the best solution found so far. As shown To test the efficiency of FCBBO, a comparative experiment
in equation 5, to avoid premature convergence, an average is carried out among FCBBO, BBO, DE [10] and PSO
of ith solution and its neighbourhood solution is taken in [11] over considered optimization test functions. Following
calculation of MV. Further, the BF component helps to speed experimental parameters are adopted:
up the convergence of the solution i.e. BF represents the level • Highest immigration rate (I)=1
of benefit to the individual. • Highest emigration rate (E)=1
It is expected that the true solution should be nearby to the • pMutation=0.01
superior fit individuals and if step sizes of superior individuals • Elitism parameter=2
will be large then may be a chance of skipping global optima • Number of population=50
due to greater step size. Hence, the step size is lower for • Total number of iteration=5000
superior fit individuals which are accountable for intensifica- • Total number of runs=30
tion and step size are higher for inferior fit individuals which • Parameter selection settings for the algorithms BBO [9],
are accountable for diversification. Therefore in the modified DE [10] and PSO [11] are imitated from their elementary
work, the superior fit individuals exploit the solution search research papers.
space, and inferior fit individuals explore the search field.
The proposed migration process is describe in Algorithm 2: C. Results comparison
Table II shows the experimental results of the FCBBO,
BBO, DE and PSO algorithms and also gives a description
IV. R ESULTS
of the standard deviation (SD), mean error (M E), average
In this section, a fair comparative study will be carried out number of function evaluations (AF E) and success rate (SR).
to establish the competitiveness of the proposed FCBBO in Results in Table II represent that maximum time FCBBO
the field of swarm intelligence based algorithms. better performs in terms of reliability, efficiency as well as
A. Test problems accuracy in comparison to the BBO, DE and PSO. Based on
boxplots and acceleration rate (AR) some more analysis has
To analyze the validity of the FCBBO algorithm, 12 distinct
been performed for results of FCBBO, BBO, DE and PSO
global numerical optimisation functions (f1 to f12 ) are used
algorithms.
here, displayed in Table I.

Authorized licensed use limited to: Rajasthan Technical University. Downloaded on May 01,2025 at 06:18:03 UTC from IEEE Xplore. Restrictions apply.
784
 TABLE I: TP: Test problems, OF: Objective Function, SR: Search Range, OV: Optimum Value, D: Dimension, AE: Acceptable Error

TP OF SR OV D AE

Dekkers and Aarts f1 (x) = 105 x21 + x22 − (x21 + x22 )2 + 10−5 (x21 + x22 )4 [−20, 20] f (0, 15) = f (0, −15) = −24777 2 5.0E − 01

D
Moved axis parallel f2 (x) = i=1 5i × x2i [−5.12, 5.12] f (x) = 0; x(i) = 5 × i, i = 1 : D 30 1.0E − 15
hyper-ellipsoid

D 2
Rastrigin f3 (x) = 10D + i=1 [xi − 10 cos(2πxi )] [−5.12, 5.12] f (0) = 0 30 1.0E − 05

D
Ackley f4 (x) = −20 + e + exp(− 0.2
D i=1 xi 3 ) [−1, 1] f (0) = 0 30 1.0E − 05

D 
Cosine Mixture f5 (x) = i=1 xi 2 − 0.1( Di=1 cos 5πxi ) + 0.1D [−1, 1] f (0) = −D × 0.1 30 1.0E − 05

D D

785
Schewel f6 (x) = i=1 |xi | + i=1 |xi | [−10, 10] f (0) = 0 30 1.0E − 05
  
sin2 (100x2 2
D−1 i +xi+1 )−0.5
Pathological function f7 (x) = i=1 0.5 + 1+0.001(x2 2 2 [−100, 100] f (0) = 0 30 1.0E − 01
i −2xi xi+1 +xi+1 )

D
Sum of different pow- f8 (x) = i=1 |xi |i+1 [−1, 1] f (0) = 0 30 1.0E − 05
ers

Beale f9 (x) = [1.5−x1 (1−x2 )]2 +[2.25−x1 (1−x22 )]2 +[2.625−x1 (1−x32 )]2 [−4.5, 4.5] f (3, 0.5) = 0 2 1.0E − 05

Brannin’s Function f10 (x) = a(x2 − bx21 + cx1 − d)2 + e(1 − f ) cos x1 + e [−5, 0], [10, 15] f (0) = 0.3979 2 1.0E − 04

Six-Hump-Camel- f11 (x) = (4 − 2.1x21 + x41 /3)x21 + x1 x2 + (−4 + 4x22 )x22 [−5, 5] f (−0.0898, 0.7126) = −1.0316 2 1.0E − 05
Back

2
Easom’s function f12 (x) = −cosx1 cosx2 e((−(x1 −Π) −(x2 −Π)2 )) [−10, 10] f (π, π) = −1 2 1.0E − 13
2016 Intl. Conference on Advances in Computing, Communications and Informatics (ICACCI), Sept. 21-24, 2016, Jaipur, India

Authorized licensed use limited to: Rajasthan Technical University. Downloaded on May 01,2025 at 06:18:03 UTC from IEEE Xplore. Restrictions apply.
 2016 Intl. Conference on Advances in Computing, Communications and Informatics (ICACCI), Sept. 21-24, 2016, Jaipur, India

Besides, boxplots [1] analysis is carried out for comparing TABLE II: Comparative result of TP, TP: Test Problem
the examined algorithms in the form of combined performance TP Algorithm SD ME AFE SR
though it can efficiently depict the empirical dispersion of data FCBBO 7.16E+03 5.55E+03 681.67 30
graphically. The boxplots for FCBBO, BBO, DE and PSO BBO 6.51E+03 4.52E+03 1493.33 30
f1 DE 5.94E+03 5.41E+03 895.00 30
are exhibited in Figure 2. The results reveal that interquartile PSO 1.17E+03 1.32E+03 910.00 30
range and medians of FCBBO are relatively less. Further, FCBBO 1.09E-16 8.65E-16 19896.67 30
BBO 2.81E-16 6.13E-16 147703.33 30
f2 DE 6.21E-17 9.08E-16 58980.00 30
x 10
4 PSO 1.86E-16 8.34E-16 20020.00 30
20 FCBBO 1.31E-06 8.23E-06 5203.33 30
BBO 1.12E+01 3.93E+01 200000.00 0
f3 DE 4.02E+00 1.08E+00 187071.67 19
10 PSO 1.49E+01 5.28E+01 200000.00 0
FCBBO 8.03E-07 9.01E-06 9411.67 30
BBO 1.28E-06 8.61E-06 60383.33 30
0 f4 DE 4.28E-07 9.49E-06 42043.33 30
PSO 8.26E-01 7.79E-01 107183.33 15
FCBBO 3.03E-07 9.68E-06 10673.33 30
FCBBO BBO DE PSO BBO 1.97E-01 2.36E-01 156776.67 7
f5 DE 7.96E-07 9.02E-06 21763.33 30
PSO 4.28E-01 9.95E-01 200000.00 0
Fig. 2: Boxplots graphs (Average Number of Function FCBBO 1.10E-06 9.00E-06 10236.67 30
Evaluation) BBO 9.51E-07 9.09E-06 48673.33 30
f6 DE 4.87E-07 9.58E-06 37901.67 30
PSO 4.47E-02 1.32E-02 158420.00 7
FCBBO 6.65E-02 1.33E-01 140671.67 19
all examined algorithms are correspondingly compared by BBO 7.39E-02 2.00E-01 198528.33 1
giving weighted importance to the M E, SR and AF E. The f7 DE 6.51E-01 1.43E+01 200000.00 0
PSO 5.99E-01 1.36E+00 200000.00 0
performance indices which is described in [1] is measurement FCBBO 4.98E-07 9.42E-06 4838.33 30
for the comparison. The resultant values of P I [1] for the BBO 5.66E-06 8.56E-06 42401.67 27
f8 DE 2.28E-06 7.26E-06 7665.00 30
FCBBO, BBO, DE and PSO are computed and equivalent per- PSO 2.04E-06 7.49E-06 5243.33 30
formance indices graphs are shown in Figure 3. The weighted FCBBO 3.18E-06 5.34E-06 12980.00 30
BBO 2.22E-01 5.92E-02 189628.33 2
importance of AFE, SR and ME are characterized in Figures f9 DE 3.01E-06 4.54E-06 2933.33 30
3(a), 3(b) and 3(c) subsequently. In these graphs, horizontal PSO 2.96E-06 4.86E-06 1393.33 30
FCBBO 2.86E-05 6.01E-05 16231.67 30
axis represents the weights and vertical axis expresses the P I. BBO 1.87E-05 7.40E-05 54626.67 30
f10 DE 4.10E-05 6.55E+04 200000.00 0
PSO 3.18E-05 3.43E-05 1146.67 30
0.8
FCBBO 1.03E-05 7.64E-06 1878.33 30
PErformance Index

BBO 3.74E-01 2.45E-01 76745.00 21

0.6 f11 DE 4.10E-05 6.55E+04 200000.00 0
FCBBO
PSO 9.83E-06 1.44E-05 1263.33 30
0.4 FCBBO 1.79E-01 3.33E-02 79840.00 29
BBO
DE BBO 4.12E-05 1.92E-05 186843.33 2
0.2 PSO f12 DE 2.70E-14 4.98E-14 5938.33 30
PSO 2.79E-14 3.61E-14 2423.33 30
0 0.2 0.4 0.6 0.8 1
Success Rate
0.8
Performance Index

0.6
outperforms on the considered test problems as compare to
0.4 FCBBO BBO, DE and PSO.
BBO
DE AFEs are tested to analyzed the examined algorithms for
0.2 PSO the convergence speed (CS) and lesser AFEs denotes the
0 0.2 0.4 0.6 0.8 1
Average Number of Function Evaluation faster convergence. Acceleration Rate (AR) [12] is used to
0.8 analyze CS and is calculated as in equation 8 for all examined
Performance Index

algorithms.
0.6

0.4 FCBBO
BBO AR = AF EALG /AF EF CBBO (8)
DE
0.2 PSO
0 0.2 0.4 0.6 0.8 1 where ALG ∈ BBO, DE, P SO and AR > 1 means that
Mean Error
FCBBO is speedy than the other considered algorithms. To
Fig. 3: PIs for test functions; (a) for weighted importance to examine the AR of the developed algorithm, as compared to
SR, (b) for weighted importance to AF E and (c) for the BBO, DE and PSO, results of Table II are analyzed and
weighted importance to M E. the value of AR is evaluated using equation 8. Table III shows
a clear comparison among FCBBO-BBO, FCBBO-DE, and
It is clear from Figure 3 that P I of FCBBO are superior FCBBO-PSO in terms of AR. It is clear from Table III that
to the other examined algorithms in each case i.e. FCBBO FCBBO converges faster than the other examined algorithms.

Authorized licensed use limited to: Rajasthan Technical University. Downloaded on May 01,2025 at 06:18:03 UTC from IEEE Xplore. Restrictions apply.
786
 2016 Intl. Conference on Advances in Computing, Communications and Informatics (ICACCI), Sept. 21-24, 2016, Jaipur, India

TABLE III: Test Problems:TP, Acceleration Rate (AR) of [7] Dan Simon, Mehmet Ergezer, and Dawei Du. Population
F CBBO compare to the conventional BBO, DE and P SO distributions in biogeography-based optimization algo-
TP BBO DE PSO rithms with elitism. In Systems, Man and Cybernetics,
f1 2.1907 1.3130 1.3350 2009. SMC 2009. IEEE International Conference on,
f2 7.4235 2.9643 1.0062 pages 991–996. IEEE, 2009.
f3 38.4369 35.9523 38.4369 [8] Travis C Service. A no free lunch theorem for multi-
f4 6.4158 4.4672 11.3883 objective optimization. Information Processing Letters,
f5 14.6886 2.0390 18.7383 110(21):917–923, 2010.
f6 4.7548 3.7025 15.4757 [9] Dan Simon. Biogeography-based optimization. Evolu-
f7 1.4113 1.4218 1.4218
tionary Computation, IEEE Transactions on, 12(6):702–
f8 8.7637 1.5842 1.0837
713, 2008.
f9 14.6093 0.2260 0.1073
[10] Kenneth Price, Rainer M Storn, and Jouni A Lampinen.
Differential evolution: a practical approach to global
f10 3.3654 12.3216 0.0706
optimization. Springer Science & Business Media, 2006.
f11 40.8580 106.4774 0.6726
[11] M. Clerc. Particle swarm optimization. Wiley-ISTE,
f12 2.3402 0.0744 0.0304
2006.
[12] Harish Sharma, Jagdish Chand Bansal, KV Arya, and
Xin-She Yang. Lévy flight artificial bee colony algorithm.
V. C ONCLUSION International Journal of Systems Science, pages 1–19,
2015.
In this paper, fitness based position update process is
developed and incorporated with the Biogeography based
Optimization (BBO) algorithm. In the proposed position up-
date process, the step sizes of the individuals are decided
on the basis of fitness. The proposed strategy is named as
Fast Convergent Biogeography based Optimization (FCBBO)
Algorithm. The competitiveness of the proposed strategy is
established through various statistical analyses over 12 well
known complex optimization problems while comparing with
DE and PSO. In future, the proposed strategy may be applied
to solve some complex continuous real world optimization
problem.

R EFERENCES
[1] Jagdish Chand Bansal, Harish Sharma, KV Arya, and
Atulya Nagar. Memetic search in artificial bee colony
algorithm. Soft Computing, 17(10):1911–1928, 2013.
[2] Kit-Sang Tang, Kim-Fung Man, Sam Kwong, and Qun
He. Genetic algorithms and their applications. Signal
Processing Magazine, IEEE, 13(6):22–37, 1996.
[3] Xiangtao Li, Jinyan Wang, Junping Zhou, and Minghao
Yin. A perturb biogeography based optimization with
mutation for global numerical optimization. Applied
Mathematics and Computation, 218(2):598–609, 2011.
[4] Haiping Ma and Dan Simon. Blended biogeography-
based optimization for constrained optimization. Engi-
neering Applications of Artificial Intelligence, 24(3):517–
525, 2011.
[5] Dan Simon, Mehmet Ergezer, Dawei Du, and Rick
Rarick. Markov models for biogeography-based opti-
mization. Systems, Man, and Cybernetics, Part B: Cy-
bernetics, IEEE Transactions on, 41(1):299–306, 2011.
[6] Wenyin Gong, Zhihua Cai, Charles X Ling, and Hui
Li. A real-coded biogeography-based optimization
with mutation. Applied Mathematics and Computation,
216(9):2749–2758, 2010.

Authorized licensed use limited to: Rajasthan Technical University. Downloaded on May 01,2025 at 06:18:03 UTC from IEEE Xplore. Restrictions apply.
787