International Journal of Computational Intelligence Systems

Volume 12, Issue 2, 2019, Pages 597 - 612

Time-Based Dance Scheduling for Artificial Bee Colony Algorithm and Its Variants

Authors
Selcuk Aslan*
Department of Computer Engineering, Ondokuz Mayıs University, Samsun, Turkey
Corresponding Author
Selcuk Aslan
Received 19 December 2018, Accepted 12 April 2019, Available Online 11 May 2019.
DOI
10.2991/ijcis.d.190425.001How to use a DOI?
Keywords
ABC algorithm; Employed bees; Time-based dance scheduling
Abstract

Artificial Bee Colony (ABC) algorithm inspired by the intelligent source search and consumption characteristics of the real honey bees is one of the most powerful optimization techniques. Although the existing behaviours of the honey bees in standard ABC algorithm are capable of producing optimal or near optimal solutions for the vast majority of the problems, there are still some intelligent characteristics that are not modeled yet such as decision-making mechanism used by employed bees to determine when the dancing will be completed. In this study, a mechanism that adjusts the dancing durations of the employed bees according to the fitness values of the associated food sources is proposed and integrated to the standard ABC algorithm and its well-known variants. Experimental studies on a set of complex continuous numerical problems showed that the newly proposed dance scheduling approach significantly improved the search and consumption characteristics of the artificial bees of the standard ABC algorithm and its some variants.

Copyright
© 2019 The Authors. Published by Atlantis Press SARL.
Open Access
This is an open access article distributed under the CC BY-NC 4.0 license (http://creativecommons.org/licenses/by-nc/4.0/).

1. INTRODUCTION

The intelligent characteristics such as division of labor, self-organization, multiple interactions of the some species especially living together as swarms or colonies in nature have lead to emerge a new research area called swarm intelligence. Researchers have tried to model the mentioned characteristics of the agents or individuals and proposed various swarm intelligence–based optimization techniques over the last two decades [1,2]. Among these swarm intelligence–based techniques, Artificial Bee Colony (ABC) algorithm gained a special position when the robust and flexible bee phases, good balance between exploitation and exploration operations, easily implementable and configurable steps, and finally requiring less control parameters were considered [35].

The studies that focus on improving the performance of the ABC algorithm can be classified into three major groups as illustrated in Figure 1. In the first group of these studies, the mathematical models used in bee phases of the ABC algorithm are changed with information provided by other solutions or modified with well-known techniques. Tsai et al. changed the candidate generation schema of the onlooker bees of the standard ABC algorithm by utilizing the Newtonian Law of universal gravitation and introduced the intersection ABC (IABC) algorithm [6]. In the proposed ABC algorithm, an analogy between food sources and masses is formed and a gravitational force calculated between two masses is used as a coefficient [6]. Zhu and Kwong also modified the candidate generation schema of the onlooker bees and proposed gbest-guided ABC (GABC) algorithm [7]. In GABC algorithm, a candidate solution was produced by utilizing the information of the best food source in addition to the information of a randomly determined neighbor [7]. Akay and Karaboga used a probabilistic selection schema in which the decision whether a parameter will be changed or not is made by comparing a random value with a new control parameter called MR [8]. Experimental studies with the proposed ABC algorithm on solving high-dimensional numeric and constrained engineering problems showed that perturbing more than one parameters can increase the solution qualities [8]. Gao et al. introduced two new ABC algorithm variants called ABC/best/1 and ABC/best/2. ABC/best/1 and ABC/best/2 algorithms are based on generating candidate solutions by the guidance of the best solution. Gao et al. also proposed a new ABC algorithm for which an employed or an onlooker bee produces new solutions by using two randomly selected neighbor solutions rather than utilizing only one randomly determined solution [9]. In other study, they combined advantageous sides of different search and multi-population techniques into together and introduced ILABC algorithm [10]. For ILABC algorithm, the whole colony is divided into subcolonies and employed bee phase of the ABC algorithm is powered by the best solution in the corresponding subcolony while onlooker bee phase of the ABC algorithm is directed with the global best solution of the whole colony [10].

Abro and Saleh changed the scout bee phase of the original ABC algorithm by proposing a new scouting characteristics that are based on receiving information of the best food source found so far rather than receiving information of a randomly determined one [11]. Tran et al. introduced a new ABC algorithm–based clustering technique called enhanced ABC algorithm/K-means for short EABCK [12]. In EABCK, the search equations used by employed and onlooker bee phases are changed with the additions of the arithmetic mutation operator and the information of the global best solution [12]. Hu et al. introduced a novel communication model abbreviated as HCM for ABC algorithm [13]. In the HCM model, the whole colony is divided into groups. The group of HCM contains one or more species that can be in communication with each other according to the island model [13]. Each group at one level up is related with a community. All of the species can set their information through groups to the communities [13]. He and Ma proposed a new ABC algorithm in which the numbers of the randomized Halton sequence is used [14]. A series of experiments with Halton sequence–based ABC algorithm demonstrated that Halton sequence–based initialization increases the convergence speed and accuracy of the solutions compared to the opposition-based, chaotic, and pure randomized initializations [14]. Luo et al. changed the probabilistic search mechanism of the standard ABC algorithm with a deterministic selection procedure in which the best food source is consumed by onlookers until the completion of the onlooker bee phase [15].

Figure 1

A chart for the studies about the Artificial Bee Colony (ABC) algorithm.

The second group of studies is devoted to the hybridization of the ABC algorithm with other intelligent methods. Baykasoglu et al. modified ABC algorithm with shift neighborhood search and greedy randomized adaptive search techniques and tested its performance on solving generalized assigned problem [16]. Kang et al. used ABC algorithm with Nelder–Mean Simplex method and proposed Hybrid Simplex ABC (HSABC) algorithm [17]. Kang et al. also integrated Hookee Jeeves pattern search approach to the ABC algorithm and introduced Hookee Jeeves ABC (HJABC) algorithm [18]. Ozturk and Karaboga combined ABC algorithm with Levenberg–Marquardt algorithm for training neural networks [19]. Duan et al. proposed a new hybrid-solving technique for continuous problems by utilizing ABC and Quantum Evolutionary algorithms [20]. Lavanya and Srinivasan hybridized ABC and genetic algorithm (GA) and tested mentioned hybrid optimization algorithm on solving training problem related with the neural networks [21]. Badem et al. combined ABC and limited memory-based BFGS algorithms and used their new variant for training deep neural networks [22]. Badem et al. also investigated the performance of their hybridized ABC algorithm on solving numerical benchmark problems [23].

Finally, the third group of the studies can be directly related with the parallelized implementations of the ABC algorithm. Narasimhan parallelized ABC algorithm by dividing the whole colony into equally sized subcolonies running simultaneously [24]. Banharnsakun et al. proposed a parallel ABC algorithm for distributed memory-based architectures [25]. Subotic et al. utilized the computation power of the multicore systems by running multiple colonies for ABC algorithm [26,27]. In the multiple colony-based parallelization, after completion of a predetermined number of cycles, the best food sources between colonies were exchanged [26,27]. Parpinelli et al. investigated parallelized ABC algorithm with different work schemas including master-slave, multi-hive with migration, and hybrid hierarchical [28]. Basturk and Akay analyzed coarse-grained model-based parallel ABC algorithm for solving numerical optimization problems and training neural networks [29]. Karaboga and Aslan proposed a new emigrant creation strategy for parallelized implementation of the ABC algorithm [30].

For the vast majority of the studies about ABC algorithm, it can be generalized that intelligent behaviours modeled in the standard ABC algorithm are found enough and improved versions of the ABC algorithm are based on integrating useful properties of the other techniques to the existing workflow of the algorithm. However, some intelligent characteristics of the employed foragers that are not modeled or tried to be managed by the randomized operations should be the first choice for further improving the performance of the ABC algorithm. In this study, one of the most important forager characteristics related with the determination of the fitness-based dancing durations is modeled and integrated to the standard ABC algorithm and its commonly used variants. Experimental studies on a set of benchmark problems showed that ABC algorithm and its variants for which the selection of the food sources being consumed is managed by probabilistic operations are significantly improved in terms of solution qualities and convergence speeds. The rest of the paper is organized as follows: In Section 2, fundamental steps of the ABC algorithm and its variants are introduced. Properties of the newly introduced fitness-based dancing mechanism is given in Section 3. Experimental studies with different control parameters and comparisons between algorithms are presented in Section 4. Finally, in Section 5, conclusion and future works are summarized.

2. STANDARD ABC ALGORITHM

In real honey bee colonies, the search operations for food sources are successfully managed by three group of bees called employed, onlooker, and scout bees and two different behaviour modes related to the consumption and abandonment of a source. Employed bees are responsible for finding food sources around the previously visited ones and carrying nectars to the hive [3134]. When an employed bee turns back to the hive, she shares information about the utilized food source with the onlooker bees waiting on the dance area. The information shared by the employed bees with dances have an important effect on the source selection procedures of the onlookers. If a food source is rich in terms of nectar amount and worth at consuming by other foragers, the employed bee associated with this food source stays longer on the dance area and continues dancing for attracting more onlooker bees to the mentioned food source compared to other ones. Finally, the last group of bees, scout bees, searches food sources randomly without using information provided by employed bees and the total number of scouts is roughly 510% of the whole colony [3134].

By considering the properties of the honey bees in foraging, Karaboga proposed a new swarm intelligence–based optimization technique called ABC algorithm. In ABC algorithm, the position of a food source corresponds to a possible solution of the problem being optimized and the nectar amount of the food source is directly related to the quality of the solution. With the discovery of the initial solutions or food sources by scout bees, ABC algorithm tries to improve these solutions with cyclic-iterative employed, onlooker and scout bee phases until satisfying the previously determined termination criteria [3134].

2.1. Generating Initial Solutions

When solving a D-dimensional optimization problem with ABC algorithm, the jth parameter of the ith food source in the set with SN solutions can be randomly determined between xjmin and xjmax bounds as in the Equation (1) given below [3134].

xi,j=xjmin+rand0,1xjmaxxjmin

In Equation (1), xi,j corresponds to the jth parameter of the ith food source showed by xi and rand0,1 is a uniformly distributed random number between 0 and 1 [3134].

2.2. Sending Employed and Onlooker Bees to Food Sources

The standard implementation of the ABC algorithm relates an employed bee with only one food source, in other words, the number of food sources is equal to the number of employed bees. As mentioned before employed bees are responsible for finding new food sources within the neighborhood of the memorized ones. After completion of source assignment, an employed bee determines a candidate food source in ABC algorithm by utilizing the Equation (2).

vi,j=xi,j+ϕxi,jxk,j

In Equation (2), vi,j is the randomly chosen jth parameter of the vi food source whose parameters are same with the xi food source except the jth parameter. xk,j is the jth parameter of the randomly determined xk food source. Although the k index is randomly chosen from the 1,2,,SN set, it should be noted that the k and i indexes must be different [3134]. Finally, ϕ is a random number between 1 and +1.

Assume that the fitness value of the vi food source is showed as fit(vi) . For a minimization problem, the value of the fitvi is calculated using the Equation (3) over the objective function value of the same food source given as objvi. If the fitness value of the vi food source is better than the fitness value of the xi, xi food source is replaced with vi and a trial counter showed by triali is set to zero [3134]. Otherwise, the xi food source is still memorized by its employed bee and the triali is incremented one to indicate that the food source is not improved in the current cycle iteration [3134].

fitvi=1/1+objvi;objvi>01+|objvi|;objvi0

When all of the employed bees turn back to the hive, they share information about the memorized food sources with onlooker bees. For recruiting more onlookers to the qualified sources, ABC algorithm determines selection probabilities for each food source according to the Equation (4) given below. As seen from the Equation (4), the selection probability of the xi food source is directly proportional with the fitness value of the same food source. If a food source is chosen by an onlooker, this food source is also consumed by the onlooker bee as is done by an employed [3134].

pxi=fitxij=1SNfitxj

Algorithm 1: Fundamental steps of the ABC algorithm

1: Initialization:

2: Generate SN initial food sources by using Equation (1).

3: Assign value to maximum fitness evaluations, MFE, and limit.

4: Set evaluation counter, evalCounter, to SN.

5: Repeat

6: //Employed bee phase

7: for i1SN do

8: if evalCounter<MFE then

9: Generate xn around the xi by using Equation (2).

10: if fitxn>fitxi then

11: Change xi with xn

12: end if

13: evalCounterevalCounter+1

14: end if

15: end for

16: //Onlooker bee phase

17: sentBees0,c1

18: Find probability values of each food source by using Equation (4).

19: while sentBeesSN and evalCounter<MFE do

20: if pc>rand0,1 then

21: sentBeessentBees+1

22: Generate xn around the xc by using Equation (2).

23: if fitxn>fitxc then

24: Change xc with xn

25: end if

26: evalCounterevalCounter+1

27: end if

28: c=c+1   modSN

29: end while

30: //Scout bee phase

31: if evalCounter<MFE then

32: Determine the abandoned food source using limit values.

33: Generate a new source for abandoned food source by using Equation (1).

34: evalCounterevalCounter+1

35: end if

36: Until MFE is reached.

2.2.1. Abandoning food sources

In ABC algorithm, when the employed and onlooker bee phases are examined, it is clearly seen that both employed and onlooker bees consume existing food sources. However, if a food source is not worth consuming any more or it is exhausted completely, is should be changed with another food source. This type of mechanism in ABC algorithm is maintained by comparing trial counters with a specific control parameter called limit. In the scout bee phase, if there is a food source for which the trial counter exceeds the limit value at most, this food source is abandoned and the employed bee related with the abandoned food source becomes a scout for finding a new food source as described in Equation (1). Although ABC algorithm does not have a solid restriction about the values that can be assigned to the limit parameter, the colony size and number of parameters that are showed by CS=2×SN and D, respectively, can be used to determine the limit parameter as in the Equation (5).

limit=a×CS×Danda+

Initialization of the food sources and evaluating them with the employed, onlooker, and scout bee phases of the ABC algorithm can be analyzed more thoroughly from the Algorithm 1.

2.3. Different Candidate Generation Approaches in Variants of ABC Algorithm

The search equation used by employed and onlooker bee phases of the standard ABC algorithm is enough for different types of problems. However, some researches are also carried out for further improving convergence speed, qualities of the solutions, and solving capabilities of the algorithm. One of the most successful modifications in search equation of the standard ABC algorithm is made by Zhu and Kwong [7]. Zhu and Kwong proposed a new ABC algorithm called GABC algorithm. When the xi food source is chosen by an onlooker for consumption, candidate food source around the xi food source is produced by utilizing from the global best food source in addition to the randomly selected xk food source as in the Equation (6). In Equation (6), C is a nonnegative constant and experimental studies about the GABC algorithm showed that 1.5 is the appropriate value for C constant [7].

vi,j=xi,j+ϕxi,jxk,j+Cxbest,jxi,j

Another important modification to the search equation of the standard ABC algorithm is presented by Gao et al. Gao et al. proposed two ABC variants called ABC/best/1 and ABC/best/2 by utilizing the search equations in Equations (7) and (8), respectively [9]. For both Equations (7) and (8), xbest is the best food source found in the current population. r1,r2,r3, and r4 are different integers randomly selected from 1,2,,SN and they are not same with the index i [9].

vi,j=xbest,j+ϕxr1,jxr2,j
vi,j=xbest,j+ϕxr1,jxr2,j+ϕxr3,jxr4,j

The utilization approach from the best food source to direct the newly discovered solutions is handled by Luo et al. in a different manner [15]. When the candidate generation schema of the ABC algorithm is analyzed, it is clearly seen that the candidate solution is produced by modifying only one parameter and updating different dimensions of the problem still requires long evaluation times. By considering the mentioned effect of the single dimension-based update mechanism, Luo et al. decided to directly update best food source found so far with the selected food sources chosen by the onlookers as described in Equation (9) below for converge-onlookers ABC (COABC) algorithm [15].

vbest,j=xbest,j+ϕxbest,jxi,j

The global best solution found so far or the best solution found in the current evaluation is the most appropriate information source for directing search equations. However, if the global best food source cannot be improved after completion of a long evaluation period, the algorithm can get stack of a local optimum or the convergence to the global optimum is delayed because of the oscillations between the best and neighbor food sources. To overcome these possible disadvantages stemmed from the known oscillation phenomenon, Gao et al. generate a candidate food source by using two randomly determined solution as in the Equation (10) [35]. In Equation (10), r1 and r2 are different indexes taken from the set 1,2,,SN and they also must be different from the index i. Because of the similarity between the operation carried out by Equation (10) and crossover operator, Gao et al. named their ABC as CABC algorithm [35].

vi,j=xr1,j+ϕxr1,jxr2,j

In the vast majority of the studies, the coefficients used in the search equations are determined randomly between 1 and +1. However, the search equations used both standard and improved variants of the ABC algorithm can be changed by adding an extra coefficient that adjust the step size of the jumping between the original and utilized parameter values. Xiang and An changed the search equation of the ABC algorithm by using the fitness value of the food source on behalf of the randomly generated coefficient as described in the Equation (11) [36]. The ABC algorithm for which the Equation (11) is used in both employed and onlooker bee phases is named ERABC algorithm [36].

vi,j=xi,j+xi,jxk,jfitxi

When the search equations used by standard ABC algorithm and its mentioned variants are considered, it is clearly seen that one additional term is directly used without applying a multiplication, addition, or subtraction operation. In order to decrease the dependency of the changed parameter to the original value of it, Tran et al. modeled the arithmetic crossover method for the candidate generation procedure of the ABC algorithm as in the Equation (12) [12]. In Equation (12), xk1 and xk2 are randomly determined food sources by controlling that k1, k2, and i are different source indexes and xbest is the global best food source. For making a discrimination between standard ABC algorithm and the newly proposed ABC algorithm in which the employed and onlooker bees use Equation (12) is named as enhanced ABC for short EABC algorithm [12].

vi,j=ϕxi,jxk1,j+ϕxbest,jxk2,j

3. Time-Based Dance Scheduling for Employed Foragers

When the workflow of the employed, onlooker, and scout bee phases and used mathematical models for consuming, choosing, or abandoning sources are evaluated, it is clearly seen that there is a good balance between implementation flexibility of the ABC algorithm and intelligent behaviours of the real honey bees. Although ABC algorithm is capable of handing various optimization problems successfully with its default definitions because of the mentioned subtle balance, some important bee characteristics still need to be modeled and integrated to the standard the ABC algorithm.

One of the most important bee characteristics that are neither modeled by standard implementation of the ABC algorithm nor its variants is directly related with the dancing durations of the employed foragers. In the standard ABC algorithm, all of the employed bees start dancing with the completions of their source searching and they continue until all of the onlookers choose food sources. However, in a real honey bee colony, the existences of the employed foragers on the dance area are directly related with the qualities of the food sources being consumed. If a food source consumed by an employed bee is rich and the fitness value assigned to this food source is high compared to other food sources, the employed forager of the mentioned food source stay longer on the dance area for attracting more onlookers to this food source. While some of the employed foragers related with the qualified sources wait on the dance area and perform different dance figures, some of them related with the poor sources leave the dance area and fly without waiting onlooker bees. In Figure 2, how the number of employed bees is changing according to the time is illustrated for both artificial and real bees. The employed bees signed with red circles in Figure 2 associated with the qualified food sources. While the number of employed bees on the dance area decreases as time goes by in a real honey bee colony, standard ABC algorithm protects all of the employed bees on the dance area until the whole onlooker bees are sent.

Figure 2

Employed bees on the dance area for Artificial Bee Colony (ABC) (a)–(c) and real honey bee (d)–(f) colonies.

At the first sense, mentioned characteristics related with the employed bees might be thought as a routine and not be considered as an intelligent behaviour. But, determining dancing durations proportional with the fitness values of the memorized solutions can help sending more onlookers to the qualified solutions. The main idea lying behind in this study is directly based on the modeling of the dancing start and finish decisions made by employed bees. In order to model the mentioned characteristics of the employed bees, a new control mechanism called time-based dance scheduling, for short ts, is introduced. With the ts model, dancing duration of an employed bee is determined by considering the fitness value of the memorized food source and decremented by also considering the fitness value of the same source for each selection attempt in on looker bee phase. If the dancing duration of an employed forager becomes equal to zero, this employed bee leaves the dance area while others with qualified sources are still on the dance area.

In ts mechanism, initial dancing durations of the employed bees are determined as their indexes in the vector of fitness values sorted in ascending order. Assume that there are SN food sources consumed by SN different employed bees and xi food source is the best food source, the initial dancing duration for xi food source showed by timeit=0 is equal to SN due to the index of xi food source is SN in the vector of fitness values sorted in ascending order. After assigning initial dancing durations of employed bees, these durations should also be decreased for each selection attempt in the onlooker bee phase. To guarantee that qualified sources are introduced more longer compared to the other sources, dancing durations assigned to employed bees are multiplied by the corresponding selection probabilities in ts mechanism as described in Equation (13) rather than simply decrementing initial dancing durations one by one for each selection attempt showed by t=0,1,.

timeit+1=fitxij=1SNfitxj×timeit

As easily seen from the Equation (2) that is used to generate candidate solutions for standard ABC algorithm, one food source except than the selected one is required at least to satisfy the multiple interactions between solutions. In other words, there should be at least two different employed bees on the dance area. In ts mechanism, if the current time duration of an employed bee is equal to 0, she leaves the dance area. However, to make sure that there are at least two employed bees for onlookers, ts mechanism controls number of employed bees after decrementing the dancing durations. If the number of employed bees is less than 2 after updating dancing durations, ts mechanism protects the employed bees in the previous selection attempt until completion of the onlooker bee phase. Detailed steps of the ts mechanism and its integration to the onlooker bee phase of the ABC algorithm can be investigated by using the Algorithm 2.

Algorithm 2: Onlooker bee phase of ABC algorithm modified with ts mechanism

1: //Onlooker bee phase

2: dnDurations generate a 1×SN empty vector.

3: for i1SN do

4: dnDurations[i] assign initial dancing duration to the ith employed bee.

5: end for

6: sentBees0

7: updateCntr1

8: Find probability values of each food source by using Equation (4).

9: beesOnDnArea get number of employed bees on dance area using dnDurations.

10: while sentBeesSN and evalCounter<MFE do

11: c a random food source for which the dnDurationsc0.

12: if pc>rand0,1 then

13: sentBeessentBees+1

14: Generate xn around the xc by using Equation (2).

15: if fitxn>fitxc then

16: Change xc with xn

17: end if

18: evalCounterevalCounter+1

19: end if

20: if updateCntr==1 then

21: tempDnDurations a copy of dnDurations.

22: for i1SN do

23: Update dnDurationsi as described in Equation (13).

24: end for

25: beesOnDnArea get number of employed bees on dance area using dnDurations.

26: if beesOnDnArea<2 then

27: updateCntr0

28: for i1SN do

29: dnDurationsitempDnDurationsi

30: end for

31: end if

32: end if

33: c=c+1modSN

34: end while

4. Experimental Studies

In order to analyze the effects of the ts mechanism on solving capabilities of the ABC, GABC, ABC/best/1, ABC/best/2, COABC, CABC, ERABC, and EABC algorithms, 15 different bound constrained, single-objective, computationally expensive benchmark functions presented in a special session of Congress on Evolutionary Computation 2015 (CEC2015) are chosen. Each of the functions used in experiments are shifted, some of them are both shifted and rotated [37]. While three of fifteen benchmark functions are hybrid, three of fifteen benchmark functions are compositional [37]. For hybrid problems, the whole parameter set is divided into subsets and each subset is related with a specific subfunction. Assume that x is a parameter vector divided randomly into two subvectors named as x1 and x2. If the fx is defined using f1x1 and f2x2 specific functions as fx=f1x1+f2x2, it is said that fx is a hybrid function. Compositional functions can also be defined as the hybrid functions [37]. However, for compositional functions, the vector is not divided randomly and all the elements in the vectors are used for each specific subfunction. Assume that x is a vector and f1x and f2x are specific functions [37]. If the fx is defined using f1x and f2x specific functions as fx=f1x+f2x, it is said that fx is a compositional function [37]. For all test functions, lower and upper bounds of the parameters are determined as 100 and +100 [37]. Names of the benchmark functions, their base functions and reference values that are used as optimums can be seen in the Table 1 [37].

No Function Related Basic Functions Reference
f1 Rotated Bent Cigar function Bent Cigar function 100
f2 Rotated Discus function Discus function 200
f3 Shifted and Rotated Weierstrass function Weierstrass function 300
f4 Shifted and Rotated Schwefel's function Schwefel's function 400
f5 Shifted and Rotated Katsuura function Katsuura function 500
f6 Shifted and Rotated HappyCat function HappyCat function 600
f7 Shifted and Rotated HGBat function HGBat function 700
f8 Shifted and Rotated Expanded Griewank's plus Rosenbrock's function Griewank's function Rosenbrock's function 800
f9 Shifted and Rotated Expanded Scaffer's f6 function Expanded Scaffer's f6 function 900
f10 Hybrid function 1 (N = 3) Schwefel's function Rastrigin's function High-conditioned elliptic 1000
f11 Hybrid function 2 (N = 4) Griewank's function Weierstrass function Rosenbrock's function Scaffer's f6 function 1100
f12 Hybrid function 3 (N = 6) Katsuura function HappyCat function Griewank's function Rosenbrock's function Schwefel's function Ackley's function 1200
f13 Compositional function 1 (N= 4) Rosenbrock's function High-conditioned elliptic Bent Cigar function Discus function 1300
f14 Compositional function 2 (N = 3) Schwefel's function Rastrigin's function High-conditioned elliptic 1400
f15 Compositional function 3 (N = 5) HGBat function Rastrigin's function Schwefel's function Weiestrass function High-conditioned elliptic 1500
Table 1

Benchmark functions used in the experiments.

For all of the ABC algorithms, ABC, GABC, ABC/best/1, ABC/best/2, COABC, CABC, ERABC, EABC, and their ts mechanism based models named ts-ABC, ts-GABC, ts-ABC/best/1, ts-ABC/best/2, ts-COABC, ts-CABC, ts-ERABC, and ts-EABC algorithms, two different values including 30 and 50 are used as the colony size. Number of parameters is determined as 10 and 30 [37]. Maximum fitness evaluations (MFEs), are taken equal to 500 and 1,500 for 10 and 30 dimensional functions, respectively [37]. The limit parameter value of the ABC algorithms is calculated as the 0.5×CS×D and the C parameter of the GABC and ts-GABC algorithms is taken equal to 1.5. When the search equations used in ABC algorithm and its variants are considered, it should be noticed that the number of employed bees on dance area except the selected one varies between 1 and 4. While ABC and GABC algorithms require one different employed bee on the dance area rather than the selected food source, COABC and ERABC algorithms require at least one employed bee for generating a candidate solution. In addition to these, while ABC/best/1 and EABC algorithms require at least two employed bees on the dance area except the selected one, ABC/best/2 algorithm requires at least four employed bees on the dance area rather than the selected one for generating candidate solutions. For all of the ts-based ABC implementations, the mentioned limitations about the minimum number of employed bees on dance area are taken into account. Each of the benchmark functions are solved 20 different times with random seeds by ABC algorithms and mean best objective function values and standard deviations are recorded.

When the results given in the Tables 29 are investigated, it is clearly seen that ts mechanism improves the qualities of the final solutions for some of the ABC algorithms including standard ABC, GABC, ABC/best/1, ABC/best/2, CABC, and EABC. All of these algorithms select a food source or sources randomly for generating candidates around it as is done by real honey bees and also require selection of food source or sources to satisfy the multiple interaction between solutions. The main idea lying behind the ts mechanism is that the dancing durations of the employed bees should be adjusted according to the fitness values of the memorized food sources and the employed bees with more qualified food sources should be kept on the dance area as time goes by. With ts mechanism, the existence of an employed bee on dance area is proportional with its fitness value and choosability of a qualified food source by an onlooker is increased.

Although this mentioned realistic approach is suitable with the candidate generation schema of the standard ABC, GABC, ABC/best/1, ABC/best/2, CABC, and EABC algorithms, candidate generation schemas of the COABC and ERABC algorithms restrict the improving effect of the ts mechanism. In COABC algorithm, the fitness-based probabilistic selection characteristics of the onlookers are changed with a deterministic selection schema and all of the onlooker bees are sent to the global best food source for increasing the early convergence performance of the algorithm compared to its standard implementation. In other words, onlooker bees have no chance of selecting food sources for consuming and ts mechanism does not show its improving effect on the performance of the onlooker bees. For the ERABC algorithm, the fitness value of the food source used as a coefficient in the candidate production deteriorates the effect of the additional term if the fitness value is relatively close to zero. When the difficulties of the problems and small number of fitness evaluations are taken into account, multiplying the second term of the equation used for generating a candidate solution with the fitness value of the food source limits the contribution of the selected sources of the ts mechanism.

In order to analyze that how the ts mechanism contributes to the convergence performance of the ABC algorithm and its variants, the curves given in the Figure 3 below can be controlled. The convergence curves in Figure 3 are generated for ABC algorithms with 50 bees on solving 30 dimensional benchmark functions. When the evolutionary curves are examined, it can be generalized that the improving effect of the ts mechanism is not only visible in the qualities of the final solutions but also convergence speeds of the algorithms, especially for ABC, GABC, ABC/best/1, ABC/best/2, CABC, and EABC. For each evaluation in onlooker bee phase, the chance of consumption related with the qualified food sources is increased by the ts mechanism as in the case of real honey bee colonies and the convergence to the optimal value is accelerated compared to the default ABC implementations. However, it should be noticed that the COABC and ERABC algorithms change their convergence characteristics slowly with ts or show better performances compared to their ts mechanism–based counterparts. For COABC algorithm, because of the candidate food source is not generated within the neighborhood of the randomly selected solution, using ts mechanism does not accelerate the convergence speed of the algorithm. For the ERABC algorithm, the parameter being perturbed in candidate generation is usually not changed or simply reaches to the lower or upper bounds due to the chosen mathematical model. By considering this drawback on the candidate generation schema, long stagnation periods on the convergence curves for both ERABC and ts-ERABC algorithms can be explained.

Function ABC ts-ABC GABC ts-GABC ABC/best/1 ts-ABC/best/1 ABC/best/2 ts-ABC/best/2
f1 Mean 2.716658e+09 7.715196e+08 8.210307e+08 6.252916e+08 4.776608e+08 1.207570e+09 7.607792e+08 3.565431e+08
Std.Dev. 2.619777e+09 1.318637e+09 6.477353e+08 5.853795e+08 3.162419e+08 8.886205e+08 3.439542e+08 3.568213e+08
f2 Mean 3.870378e+04 3.900769e+04 4.790570e+04 3.855270e+04 4.100352e+04 3.692275e+04 4.605250e+04 4.583296e+04
Std.Dev. 1.313293e+04 1.407367e+04 1.806861e+04 1.567433e+04 1.429170e+04 9.994693e+03 2.342800e+04 1.460064e+04
f3 Mean 3.114588e+02 3.103083e+02 3.103419e+02 3.099561e+02 3.100924e+02 3.099762e+02 3.105299e+02 3.100281e+02
Std.Dev. 1.357763e+00 1.280487e+00 1.038200e+00 1.232346e+00 1.287297e+00 1.577455e+00 9.663669e-01 1.162901e+00
f4 Mean 1.521087e+03 1.062533e+03 1.439194e+03 9.610165e+02 1.433784e+03 1.102033e+03 1.545442e+03 1.031898e+03
Std.Dev. 2.335970e+02 2.680053e+02 2.979608e+02 2.266523e+02 2.346407e+02 2.695440e+02 2.282057e+02 1.590212e+02
f5 Mean 5.023907e+02 5.020619e+02 5.026821e+02 5.024273e+02 5.027842e+02 5.029346e+02 5.027086e+02 5.025562e+02
Std.Dev. 5.832669e-01 4.888150e-01 5.806247e-01 5.524283e-01 5.945906e-01 7.315304e-01 5.061681e-01 6.219925e-01
f6 Mean 6.037940e+02 6.024325e+02 6.024441e+02 6.011801e+02 6.016304e+02 6.020585e+02 6.019961e+02 6.012319e+02
Std.Dev. 9.912625e-01 9.683091e-01 7.219408e-01 3.097634e-01 5.553237e-01 9.076719e-01 6.358408e-01 3.686968e-01
f7 Mean 7.293699e+02 7.148945e+02 7.159116e+02 7.088647e+02 7.088151e+02 7.151470e+02 7.121670e+02 7.072477e+02
Std.Dev. 1.818570e+01 8.723010e+00 9.064312e+00 5.353068e+00 5.753663e+00 9.626231e+00 5.186575e+00 5.170760e+00
f8 Mean 1.527860e+03 8.457670e+02 9.631715e+02 9.074125e+02 8.555846e+02 1.620868e+03 9.139047e+02 8.279735e+02
Std.Dev. 1.120934e+03 5.161738e+01 2.787490e+02 1.906604e+02 5.269011e+01 1.702608e+03 1.927086e+02 4.906995e+01
f9 Mean 9.042098e+02 9.040828e+02 9.041019e+02 9.040767e+02 9.040568e+02 9.041801e+02 9.041138e+02 9.040719e+02
Std.Dev. 1.603778e-01 1.824507e-01 1.788973e-01 2.665699e-01 1.820721e-01 1.755280e-01 1.301141e-01 2.313002e-01
f10 Mean 1.178271e+06 1.003313e+06 6.124471e+05 5.362535e+05 3.990019e+05 7.659800e+05 3.458791e+05 5.988060e+05
Std.Dev. 2.817892e+06 2.684644e+06 4.638055e+05 4.807917e+05 5.088393e+05 9.471498e+05 2.519023e+05 7.526227e+05
f11 Mean 1.114830e+03 1.111651e+03 1.112123e+03 1.109447e+03 1.113348e+03 1.113243e+03 1.110513e+03 1.110851e+03
Std.Dev. 5.740322e+00 5.151412e+00 5.319018e+00 2.710290e+00 3.423248e+00 6.558813e+00 2.621291e+00 4.439402e+00
f12 Mean 1.510483e+03 1.444484e+03 1.496250e+03 1.434543e+03 1.475182e+03 1.453063e+03 1.501027e+03 1.443331e+03
Std.Dev. 1.213293e+02 9.774738e+01 7.500678e+01 1.003511e+02 8.097253e+01 8.700342e+01 1.073091e+02 9.680262e+01
f13 Mean 1.702227e+03 1.664968e+03 1.666147e+03 1.650612e+03 1.652384e+03 1.699363e+03 1.663966e+03 1.650791e+03
Std.Dev. 5.955833e+01 3.205841e+01 2.929865e+01 1.609759e+01 1.813559e+01 5.506221e+01 2.384845e+01 1.984467e+01
f14 Mean 1.621208e+03 1.613885e+03 1.615869e+03 1.613554e+03 1.614540e+03 1.615772e+03 1.616070e+03 1.614553e+03
Std.Dev. 9.655584e+00 6.287893e+00 5.387226e+00 5.484613e+00 5.274835e+00 6.809948e+00 6.126974e+00 4.859516e+00
f15 Mean 1.919081e+03 1.963648e+03 1.964664e+03 1.964024e+03 1.931967e+03 1.925884e+03 1.952774e+03 1.933972e+03
Std.Dev. 1.023101e+02 7.321057e+01 8.481114e+01 1.143008e+02 6.756013e+01 9.020307e+01 5.775669e+01 1.405099e+02

ABC, Artificial Bee Colony; GABC, gbest-guided ABC; Std.Dev., standard deviation.

Table 2

Comparison of ABC, GABC, ABC/best/1, ABC/best/2 algorithms, and their ts model–based variants with 30 bees on solving 10 dimensional problems.

Function COABC ts-COABC CABC ts-CABC ERABC ts-ERABC EABC ts-EABC
f1 Mean 1.154472e+08 1.276849e+09 6.557027e+08 2.228156e+08 2.062686e+10 2.170513e+10 1.403711e+09 1.125660e+09
Std.Dev. 1.212249e+08 1.662275e+09 4.389131e+08 1.610762e+08 7.444188e+09 5.705464e+09 9.416702e+08 6.692855e+08
f2 Mean 4.012159e+04 5.020721e+04 4.756039e+04 4.329411e+04 8.033285e+07 7.006532e+07 3.715573e+04 4.231731e+04
Std.Dev. 1.661216e+04 2.891368e+04 1.897498e+04 2.404596e+04 1.556228e+08 1.004056e+08 1.227938e+04 1.613730e+04
f3 Mean 3.097314e+02 3.095482e+02 3.107713e+02 3.100124e+02 3.149511e+02 3.157276e+02 3.109229e+02 3.100109e+02
Std.Dev. 1.632631e+00 1.959432e+00 1.010609e+00 1.134983e+00 1.605396e+00 1.564906e+00 1.186810e+00 1.102924e+00
f4 Mean 8.104725e+02 8.284283e+02 1.245628e+03 8.068427e+02 3.246849e+03 3.294964e+03 1.542388e+03 1.122538e+03
Std.Dev. 1.611879e+02 1.570734e+02 1.818175e+02 2.028760e+02 3.020462e+02 3.900027e+02 2.195640e+02 2.209403e+02
f5 Mean 5.018665e+02 5.021253e+02 5.025557e+02 5.026895e+02 5.055547e+02 5.053779e+02 5.025863e+02 5.025689e+02
Std.Dev. 5.199851e-01 5.943225e-01 5.337103e-01 6.756078e-01 1.427629e+00 1.329793e+00 5.788205e-01 7.541843e-01
f6 Mean 6.007468e+02 6.008256e+02 6.018007e+02 6.011450e+02 6.089245e+02 6.095337e+02 6.024926e+02 6.020531e+02
Std.Dev. 1.856181e-01 2.926697e-01 5.070661e-01 5.380740e-01 1.527262e+00 1.669186e+00 5.772861e-01 5.949175e-01
f7 Mean 7.008683e+02 7.028242e+02 7.123414e+02 7.043181e+02 8.284115e+02 8.248690e+02 7.162609e+02 7.114402e+02
Std.Dev. 4.393814e-01 4.878875e+00 5.541202e+00 2.546816e+00 4.130395e+01 3.897871e+01 7.975968e+00 3.844121e+00
f8 Mean 8.182980e+02 3.361056e+04 9.950345e+02 8.267994e+02 1.055673e+06 1.564280e+06 1.543894e+03 9.113788e+02
Std.Dev. 1.771768e+01 6.950828e+04 3.825697e+02 5.489403e+01 1.131215e+06 1.581205e+06 2.295640e+03 1.705232e+02
f9 Mean 9.039660e+02 9.039186e+02 9.041741e+02 9.040658e+02 9.046754e+02 9.045633e+02 9.041481e+02 9.040730e+02
Std.Dev. 2.430996e-01 2.720800e-01 1.494516e-01 2.145972e-01 1.554074e-01 2.285790e-01 1.710996e-01 2.080785e-01
f10 Mean 5.175479e+05 1.656484e+06 9.674845e+05 5.750111e+05 8.814672e+07 5.983723e+07 4.107140e+05 4.635940e+05
Std.Dev. 8.156218e+05 1.919514e+06 8.747779e+05 6.273505e+05 1.081411e+08 5.477027e+07 3.411544e+05 4.165436e+05
f11 Mean 1.110225e+03 1.112355e+03 1.112797e+03 1.112840e+03 1.255240e+03 1.291331e+03 1.114068e+03 1.114381e+03
Std.Dev. 2.777537e+00 5.544708e+00 3.487721e+00 3.620990e+00 6.653285e+01 1.526035e+02 4.790630e+00 5.114736e+00
f12 Mean 1.509919e+03 1.456713e+03 1.502446e+03 1.470657e+03 2.720915e+03 5.470793e+03 1.475801e+03 1.460404e+03
Std.Dev. 9.668761e+01 1.345258e+02 1.000146e+02 8.753230e+01 2.427922e+03 1.078440e+04 1.006706e+02 9.119755e+01
f13 Mean 1.660884e+03 1.666798e+03 1.675543e+03 1.649346e+03 3.057997e+03 2.707758e+03 1.683561e+03 1.681644e+03
Std.Dev. 2.524741e+01 4.395416e+01 3.379650e+01 1.346389e+01 7.418433e+02 6.490447e+02 3.064096e+01 4.294427e+01
f14 Mean 1.611896e+03 1.612063e+03 1.616420e+03 1.613649e+03 1.697243e+03 1.699400e+03 1.616794e+03 1.616111e+03
Std.Dev. 9.319448e+00 8.504274e+00 6.171834e+00 7.805774e+00 5.130750e+01 3.405484e+01 5.618579e+00 6.736457e+00
f15 Mean 1.912819e+03 1.981372e+03 1.992193e+03 1.944291e+03 2.251244e+03 2.247013e+03 1.970404e+03 1.898319e+03
Std.Dev. 1.480665e+02 1.082278e+02 4.979468e+01 6.677492e+01 1.077732e+02 1.023842e+02 6.179183e+01 9.863263e+01

ABC, Artificial Bee Colony; Std.Dev., standard deviation; COABC, converge-onlookers ABC.

Table 3

Comparison of COABC, CABC, ERABC, EABC algorithms, and their ts model–based variants with 30 bees on solving 10 dimensional problems.

Function ABC ts-ABC GABC ts-GABC ABC/best/1 ts-ABC/best/1 ABC/best/2 ts-ABC/best/2
f1 Mean 2.874022e+10 7.380010e+09 1.007479e+10 5.276032e+09 6.758463e+09 1.172245e+10 6.436318e+09 1.570801e+09
Std.Dev. 1.297193e+10 5.655872e+09 3.406875e+09 2.379473e+09 2.022907e+09 6.545935e+09 1.956905e+09 6.339895e+08
f2 Mean 1.086945e+05 1.119868e+05 1.221693e+05 1.188624e+05 1.100295e+05 1.153404e+05 1.132975e+05 1.130321e+05
Std.Dev. 1.484122e+04 1.982311e+04 2.172899e+04 1.680534e+04 1.335718e+04 1.660430e+04 1.911921e+04 1.763017e+04
f3 Mean 3.409215e+02 3.397359e+02 3.378629e+02 3.374135e+02 3.386195e+02 3.369604e+02 3.374306e+02 3.377273e+02
Std.Dev. 1.649650e+00 2.755678e+00 2.630754e+00 2.135595e+00 1.797164e+00 2.358775e+00 2.575609e+00 2.825846e+00
f4 Mean 4.898264e+03 2.580050e+03 5.064316e+03 2.894777e+03 4.500458e+03 2.876139e+03 4.889962e+03 3.313030e+03
Std.Dev. 6.113384e+02 5.291545e+02 4.783905e+02 3.890094e+02 4.181204e+02 3.933831e+02 3.874354e+02 6.045009e+02
f5 Mean 5.031803e+02 5.027337e+02 5.033885e+02 5.033083e+02 5.035661e+02 5.035613e+02 5.037563e+02 5.032828e+02
Std.Dev. 4.187038e-01 4.000041e-01 5.478987e-01 6.720153e-01 5.577199e-01 8.071338e-01 6.932659e-01 7.988223e-01
f6 Mean 6.048440e+02 6.031434e+02 6.023862e+02 6.008796e+02 6.014326e+02 6.021869e+02 6.014521e+02 6.009001e+02
Std.Dev. 5.489252e-01 8.100654e-01 4.980876e-01 3.014158e-01 5.737325e-01 9.341314e-01 5.414082e-01 2.784951e-01
f7 Mean 7.991826e+02 7.275427e+02 7.298575e+02 7.102252e+02 7.173405e+02 7.250216e+02 7.225398e+02 7.102095e+02
Std.Dev. 2.650585e+01 1.656431e+01 8.298526e+00 5.696513e+00 6.798589e+00 1.420327e+01 6.665316e+00 5.334576e+00
f8 Mean 4.287487e+05 1.378647e+05 4.368237e+05 5.983792e+04 2.035833e+05 6.545443e+05 2.925450e+05 9.330454e+04
Std.Dev. 7.306972e+05 3.014948e+05 4.998589e+05 1.291739e+05 2.376706e+05 8.537197e+05 3.429529e+05 1.902556e+05
f9 Mean 9.138568e+02 9.137548e+02 9.137144e+02 9.137025e+02 9.137695e+02 9.137312e+02 9.137964e+02 9.136917e+02
Std.Dev. 1.394280e-01 2.223189e-01 2.428242e-01 2.539421e-01 2.351620e-01 1.824119e-01 1.912461e-01 2.466965e-01
f10 Mean 2.127051e+07 2.022547e+07 1.848871e+07 1.583745e+07 2.234886e+07 2.067039e+07 1.987510e+07 1.298141e+07
Std.Dev. 1.253340e+07 1.477105e+07 1.245245e+07 8.948488e+06 1.368461e+07 1.135056e+07 1.141494e+07 9.495358e+06
f11 Mean 1.276733e+03 1.184721e+03 1.244496e+03 1.195046e+03 1.235538e+03 1.220965e+03 1.222708e+03 1.205167e+03
Std.Dev. 4.835406e+01 6.138788e+01 6.823799e+01 4.665858e+01 4.697590e+01 4.593123e+01 3.709698e+01 4.323351e+01
f12 Mean 2.375565e+03 2.247250e+03 2.324325e+03 2.172248e+03 2.213273e+03 2.306675e+03 2.302794e+03 2.247399e+03
Std.Dev. 2.687816e+02 2.582901e+02 2.949635e+02 2.971143e+02 2.226441e+02 3.301708e+02 1.775432e+02 1.907788e+02
f13 Mean 2.104460e+03 1.828667e+03 1.948367e+03 1.792256e+03 1.883473e+03 2.044204e+03 1.934404e+03 1.790321e+03
Std.Dev. 2.372981e+02 5.794128e+01 1.098594e+02 6.554537e+01 9.786156e+01 2.085624e+02 1.313962e+02 5.703170e+01
f14 Mean 1.751087e+03 1.702420e+03 1.711827e+03 1.700929e+03 1.709403e+03 1.702241e+03 1.707517e+03 1.699666e+03
Std.Dev. 5.320667e+01 2.527095e+01 3.303930e+01 3.102879e+01 2.464513e+01 2.616494e+01 2.327794e+01 2.261910e+01
f15 Mean 2.834826e+03 2.744596e+03 2.771290e+03 2.728688e+03 2.780323e+03 2.774281e+03 2.780808e+03 2.760362e+03
Std.Dev. 1.673755e+02 1.888101e+02 1.025995e+02 1.404296e+02 1.541621e+02 7.125748e+01 6.876194e+01 7.925066e+01

ABC, Artificial Bee Colony; GABC, gbest-guided ABC; Std.Dev., standard deviation.

Table 4

Comparison of ABC, GABC, ABC/best/1, ABC/best/2 algorithms, and their ts model–based variants with 50 bees on solving 10 dimensional problems.

Function COABC ts-COABC CABC ts-CABC ERABC ts-ERABC EABC ts-EABC
f1 Mean 3.164912e+08 2.432697e+09 4.213069e+09 1.662524e+09 1.126799e+11 1.203574e+11 1.381121e+10 7.677647e+09
Std.Dev. 2.656562e+08 2.901365e+09 1.362111e+09 1.980279e+09 1.674295e+10 1.604682e+10 4.524733e+09 2.034167e+09
f2 Mean 1.109747e+05 1.114874e+05 1.156213e+05 1.092496e+05 2.593265e+07 6.367834e+07 1.066621e+05 1.065415e+05
Std.Dev. 1.736156e+04 2.328571e+04 1.368680e+04 1.934016e+04 4.388028e+07 1.152567e+08 1.514714e+04 1.476639e+04
f3 Mean 3.382132e+02 3.388512e+02 3.387512e+02 3.368393e+02 3.514851e+02 3.504493e+02 3.394028e+02 3.376025e+02
Std.Dev. 3.233474e+00 3.235970e+00 2.260633e+00 2.321731e+00 2.252532e+00 1.907562e+00 1.608742e+00 2.334614e+00
f4 Mean 1.523955e+03 1.892381e+03 4.062849e+03 1.984803e+03 1.044373e+04 1.026075e+04 4.559737e+03 2.897692e+03
Std.Dev. 2.959567e+02 3.508363e+02 4.860837e+02 3.525632e+02 5.556362e+02 6.761046e+02 4.054989e+02 3.774576e+02
f5 Mean 5.024467e+02 5.025580e+02 5.037371e+02 5.038306e+02 5.077238e+02 5.068617e+02 5.037361e+02 5.035170e+02
Std.Dev. 6.704678e-01 7.032852e-01 5.744850e-01 5.592104e-01 1.125381e+00 1.324530e+00 6.015442e-01 6.824195e-01
f6 Mean 6.006283e+02 6.006307e+02 6.010139e+02 6.006387e+02 6.090965e+02 6.088120e+02 6.031234e+02 6.020081e+02
Std.Dev. 1.273650e-01 1.488749e-01 3.611400e-01 1.282884e-01 6.612400e-01 6.706021e-01 5.126429e-01 6.451348e-01
f7 Mean 7.006613e+02 7.006936e+02 7.112184e+02 7.027842e+02 9.580329e+02 9.653756e+02 7.319813e+02 7.198583e+02
Std.Dev. 5.442124e-01 8.342133e-01 4.705639e+00 1.780604e+00 4.258692e+01 5.020076e+01 9.941893e+00 4.318610e+00
f8 Mean 7.243095e+03 2.669283e+06 4.774664e+05 2.835704e+04 3.300064e+08 4.276526e+08 1.765466e+05 1.108185e+05
Std.Dev. 1.309429e+04 5.625904e+06 1.174563e+06 2.994867e+04 1.552979e+08 2.739110e+08 2.465664e+05 2.229498e+05
f9 Mean 9.137645e+02 9.136050e+02 9.138267e+02 9.137794e+02 9.145035e+02 9.145490e+02 9.137334e+02 9.137312e+02
Std.Dev. 2.382136e-01 3.080091e-01 1.970009e-01 2.153151e-01 1.896895e-01 1.687801e-01 2.212720e-01 1.845016e-01
f10 Mean 2.094852e+07 2.725651e+07 2.063197e+07 2.044892e+07 6.878707e+08 6.469510e+08 2.193600e+07 1.855870e+07
Std.Dev. 1.998104e+07 2.108055e+07 1.038327e+07 1.094937e+07 4.492558e+08 4.894871e+08 1.192082e+07 1.233614e+07
f11 Mean 1.166063e+03 1.175308e+03 1.238166e+03 1.213256e+03 2.762385e+03 2.429241e+03 1.228487e+03 1.192017e+03
Std.Dev. 5.333776e+01 6.377053e+01 4.197417e+01 5.559392e+01 6.295319e+02 3.544412e+02 4.505229e+01 3.053538e+01
f12 Mean 2.228561e+03 3.645217e+03 2.246239e+03 2.238476e+03 3.597975e+06 3.091981e+06 2.355412e+03 2.146743e+03
Std.Dev. 3.091278e+02 3.562596e+03 1.829921e+02 2.076445e+02 5.189896e+06 5.901716e+06 2.392137e+02 2.623970e+02
f13 Mean 1.720122e+03 1.809039e+03 1.903756e+03 1.779245e+03 4.922260e+03 4.740386e+03 2.012509e+03 1.921969e+03
Std.Dev. 4.716193e+01 9.806224e+01 1.280937e+02 6.685828e+01 1.031560e+03 1.150095e+03 1.151269e+02 8.730109e+01
f14 Mean 1.714285e+03 1.695189e+03 1.720198e+03 1.695523e+03 2.426968e+03 2.482275e+03 1.715202e+03 1.696582e+03
Std.Dev. 5.596615e+01 3.922859e+01 2.603988e+01 2.497509e+01 2.754998e+02 2.836083e+02 2.174420e+01 2.072144e+01
f15 Mean 2.828280e+03 2.926488e+03 2.651053e+03 2.689822e+03 4.590594e+03 4.363950e+03 2.686381e+03 2.657740e+03
Std.Dev. 1.896958e+02 1.211170e+02 1.387746e+02 1.613210e+02 1.155275e+03 1.079735e+03 1.762959e+02 2.281972e+02

ABC, Artificial Bee Colony; Std.Dev., standard deviation; COABC, converge-onlookers ABC.

Table 5

Comparison of COABC, CABC, ERABC, EABC algorithms, and their ts model–based variants with 50 bees on solving 10 dimensional problems.

Function ABC ts-ABC GABC ts-GABC ABC/best/1 ts-ABC/best/1 ABC/best/2 ts-ABC/best/2
f1 Mean 4.582446e+09 6.892993e+08 1.451181e+09 5.893642e+08 1.412891e+09 2.191833e+09 1.972129e+09 4.322524e+08
Std.Dev. 2.173263e+09 5.433192e+08 7.069864e+08 4.741935e+08 7.076209e+08 1.834647e+09 9.201705e+08 3.227833e+08
f2 Mean 3.807448e+04 4.238495e+04 4.594143e+04 4.554367e+04 4.577515e+04 4.175372e+04 4.645720e+04 4.478945e+04
Std.Dev. 1.104257e+04 2.058223e+04 1.822996e+04 1.571686e+04 1.473063e+04 1.085514e+04 1.509847e+04 1.553417e+04
f3 Mean 3.115898e+02 3.106794e+02 3.113097e+02 3.097030e+02 3.107839e+02 3.104673e+02 3.108889e+02 3.103613e+02
Std.Dev. 1.382279e+00 1.169919e+00 8.152745e-01 9.583828e-01 1.212920e+00 9.801397e-01 1.018289e+00 1.063953e+00
f4 Mean 1.883608e+03 1.155546e+03 1.820824e+03 1.034088e+03 1.723936e+03 1.081871e+03 1.745621e+03 1.162786e+03
Std.Dev. 3.379312e+02 1.993162e+02 3.348659e+02 2.644323e+02 2.646889e+02 2.244991e+02 2.637004e+02 2.814813e+02
f5 Mean 5.025922e+02 5.022320e+02 5.028552e+02 5.026105e+02 5.025398e+02 5.026179e+02 5.027329e+02 5.026148e+02
Std.Dev. 5.262078e-01 4.525835e-01 6.589051e-01 7.096314e-01 4.276504e-01 6.441077e-01 6.898570e-01 7.731720e-01
f6 Mean 6.045896e+02 6.031046e+02 6.030689e+02 6.021163e+02 6.026685e+02 6.028251e+02 6.028512e+02 6.021039e+02
Std.Dev. 1.186188e+00 9.298918e-01 9.960178e-01 7.919614e-01 6.840569e-01 9.591161e-01 7.737902e-01 5.986249e-01
f7 Mean 7.457595e+02 7.128856e+02 7.312647e+02 7.110933e+02 7.200655e+02 7.149002e+02 7.234851e+02 7.123453e+02
Std.Dev. 1.599424e+01 1.313267e+01 1.140030e+01 6.442564e+00 7.131862e+00 6.473222e+00 7.569151e+00 5.564067e+00
f8 Mean 1.944251e+03 1.769011e+03 1.089751e+03 1.101743e+03 9.431846e+02 1.496075e+03 1.840934e+03 9.874323e+02
Std.Dev. 1.764015e+03 2.338439e+03 2.876552e+02 7.758735e+02 2.141592e+02 9.353520e+02 1.555459e+03 3.302135e+02
f9 Mean 9.041371e+02 9.041069e+02 9.040866e+02 9.040379e+02 9.041134e+02 9.040055e+02 9.041394e+02 9.040439e+02
Std.Dev. 2.255586e-01 2.281802e-01 2.286015e-01 2.055990e-01 2.046119e-01 2.811609e-01 1.318425e-01 1.711723e-01
f10 Mean 9.284637e+05 4.287034e+05 8.749426e+05 5.026939e+05 6.226318e+05 7.465358e+05 6.009775e+05 3.728463e+05
Std.Dev. 1.836764e+06 6.061454e+05 9.191729e+05 7.967638e+05 5.241438e+05 9.227678e+05 7.952335e+05 4.772446e+05
f11 Mean 1.120480e+03 1.112678e+03 1.116592e+03 1.113383e+03 1.114031e+03 1.112003e+03 1.113704e+03 1.114241e+03
Std.Dev. 7.817952e+00 4.800240e+00 7.118440e+00 6.325912e+00 4.667764e+00 2.872470e+00 5.042692e+00 4.238452e+00
f12 Mean 1.505062e+03 1.472240e+03 1.515181e+03 1.432594e+03 1.452319e+03 1.451471e+03 1.469711e+03 1.453817e+03
Std.Dev. 8.845214e+01 1.332733e+02 1.048903e+02 9.842248e+01 8.809943e+01 1.024814e+02 1.211136e+02 9.492776e+01
f13 Mean 1.781767e+03 1.662031e+03 1.677124e+03 1.658571e+03 1.662533e+03 1.694396e+03 1.668999e+03 1.646848e+03
Std.Dev. 1.081087e+02 3.120635e+01 2.913494e+01 1.937843e+01 1.383230e+01 5.391510e+01 2.770616e+01 1.249953e+01
f14 Mean 1.621435e+03 1.616584e+03 1.617015e+03 1.614825e+03 1.617768e+03 1.619902e+03 1.617618e+03 1.615078e+03
Std.Dev. 6.479306e+00 7.630934e+00 4.342387e+00 6.012091e+00 7.233008e+00 6.402536e+00 5.857064e+00 6.048494e+00
f15 Mean 1.980294e+03 1.955122e+03 1.965033e+03 1.949007e+03 1.953426e+03 1.931952e+03 1.941749e+03 1.942885e+03
Std.Dev. 7.760823e+01 1.202862e+02 6.646707e+01 1.069406e+02 5.576135e+01 1.175118e+02 1.306081e+02 8.796490e+01

ABC, Artificial Bee Colony; GABC, gbest-guided ABC; Std.Dev., standard deviation.

Table 6

Comparison of ABC, GABC, ABC/best/1, ABC/best/2 algorithms, and their ts model–based variants with 30 bees on solving 30 dimensional problems.

Function COABC ts-COABC CABC ts-CABC ERABC ts-ERABC EABC ts-EABC
f1 Mean 1.205644e+08 1.296432e+09 1.941559e+09 4.160407e+08 1.640046e+10 1.691408e+10 2.624848e+09 1.261461e+09
Std.Dev. 2.105520e+08 1.693383e+09 9.536279e+08 3.331963e+08 4.794608e+09 6.207721e+09 1.410391e+09 8.817772e+08
f2 Mean 3.944165e+04 5.210309e+04 4.323893e+04 4.381957e+04 9.292209e+07 9.222668e+07 4.532516e+04 4.945074e+04
Std.Dev. 1.165973e+04 1.545058e+04 1.738220e+04 1.564388e+04 1.910761e+08 1.491740e+08 2.461431e+04 3.527727e+04
f3 Mean 3.099904e+02 3.101352e+02 3.110183e+02 3.099554e+02 3.150659e+02 3.150295e+02 3.112665e+02 3.102997e+02
Std.Dev. 1.807309e+00 1.509035e+00 1.261022e+00 1.296883e+00 1.305890e+00 8.293647e-01 1.156605e+00 1.172680e+00
f4 Mean 8.328717e+02 9.139263e+02 1.918158e+03 8.228753e+02 3.230524e+03 3.104760e+03 1.794611e+03 1.250781e+03
Std.Dev. 1.436631e+02 2.164556e+02 2.890025e+02 1.638514e+02 2.328735e+02 3.559319e+02 2.057491e+02 1.730000e+02
f5 Mean 5.019317e+02 5.019139e+02 5.027769e+02 5.027450e+02 5.054529e+02 5.046956e+02 5.027541e+02 5.025547e+02
Std.Dev. 7.204090e-01 5.565162e-01 7.252835e-01 7.828600e-01 1.645447e+00 1.085314e+00 5.898314e-01 5.853835e-01
f6 Mean 6.007950e+02 6.008018e+02 6.028213e+02 6.014781e+02 6.084839e+02 6.082277e+02 6.033995e+02 6.026396e+02
Std.Dev. 2.546493e-01 2.116735e-01 5.130346e-01 4.871717e-01 1.729182e+00 1.639006e+00 5.966775e-01 6.370124e-01
f7 Mean 7.019362e+02 7.035169e+02 7.232055e+02 7.080058e+02 8.155223e+02 8.240395e+02 7.293165e+02 7.130636e+02
Std.Dev. 2.667227e+00 6.851578e+00 7.831710e+00 5.391616e+00 2.862359e+01 2.926965e+01 1.160952e+01 5.624272e+00
f8 Mean 8.139696e+02 2.506534e+04 1.335845e+03 8.532294e+02 8.463249e+05 9.494432e+05 1.657164e+03 9.719318e+02
Std.Dev. 1.119454e+01 4.856979e+04 8.184892e+02 4.840300e+01 7.826209e+05 1.350060e+06 1.803051e+03 2.487026e+02
f9 Mean 9.040308e+02 9.040157e+02 9.041847e+02 9.041198e+02 9.045390e+02 9.044878e+02 9.040724e+02 9.040582e+02
Std.Dev. 2.414241e-01 3.566836e-01 1.304758e-01 1.404418e-01 2.094508e-01 1.738856e-01 2.040079e-01 1.984879e-01
f10 Mean 1.185435e+06 2.684726e+06 5.887457e+05 6.270639e+05 5.257448e+07 4.904696e+07 4.509143e+05 3.599065e+05
Std.Dev. 2.349779e+06 3.137358e+06 5.616428e+05 5.724910e+05 4.435668e+07 7.510672e+07 2.991768e+05 3.563709e+05
f11 Mean 1.110327e+03 1.109872e+03 1.121608e+03 1.111801e+03 1.229719e+03 1.241192e+03 1.117356e+03 1.114263e+03
Std.Dev. 4.553226e+00 3.798764e+00 7.774842e+00 4.477633e+00 6.497002e+01 7.542007e+01 5.856712e+00 4.606782e+00
f12 Mean 1.439473e+03 1.514181e+03 1.547522e+03 1.468066e+03 2.252120e+03 2.215557e+03 1.531650e+03 1.489349e+03
Std.Dev. 1.413108e+02 1.295462e+02 6.364329e+01 1.108035e+02 3.511586e+02 4.807802e+02 9.736181e+01 1.101213e+02
f13 Mean 1.661504e+03 1.667089e+03 1.691799e+03 1.656476e+03 2.375492e+03 2.326616e+03 1.698804e+03 1.700802e+03
Std.Dev. 3.882197e+01 3.641531e+01 3.419337e+01 4.073161e+01 4.133722e+02 3.816298e+02 3.818707e+01 3.281520e+01
f14 Mean 1.613656e+03 1.615363e+03 1.617743e+03 1.615597e+03 1.683274e+03 1.685478e+03 1.621198e+03 1.615555e+03
Std.Dev. 6.982122e+00 9.321286e+00 6.258022e+00 6.080017e+00 3.157819e+01 4.163759e+01 7.067241e+00 7.322561e+00
f15 Mean 1.964785e+03 1.955986e+03 1.988508e+03 1.949618e+03 2.231239e+03 2.191957e+03 1.926386e+03 1.880580e+03
Std.Dev. 5.441468e+01 1.134082e+02 7.875832e+01 7.146702e+01 1.139813e+02 8.762971e+01 1.505798e+02 1.375819e+02

ABC, Artificial Bee Colony; Std.Dev., standard deviation; COABC, converge-onlookers ABC.

Table 7

Comparison of COABC, CABC, ERABC, EABC algorithms and their ts model based–variants with 30 bees on solving 30 dimensional problems.

Function ABC ts-ABC GABC ts-GABC ABC/best/1 ts-ABC/best/1 ABC/best/2 ts-ABC/best/2
f1 Mean 5.401930e+10 1.016065e+10 1.928388e+10 6.612168e+09 1.580333e+10 1.601686e+10 1.827024e+10 2.605397e+09
Std.Dev. 1.159090e+10 6.767604e+09 4.397591e+09 4.930426e+09 4.501972e+09 7.060424e+09 4.403735e+09 1.146235e+09
f2 Mean 1.097091e+05 1.187948e+05 1.202809e+05 1.101371e+05 1.071584e+05 1.142787e+05 1.124367e+05 1.222727e+05
Std.Dev. 1.419203e+04 2.607008e+04 1.962019e+04 1.731523e+04 1.277990e+04 1.997548e+04 1.958941e+04 1.746471e+04
f3 Mean 3.417617e+02 3.392137e+02 3.395051e+02 3.374647e+02 3.398527e+02 3.383341e+02 3.396560e+02 3.373194e+02
Std.Dev. 1.654487e+00 2.178471e+00 2.079294e+00 2.131084e+00 2.412488e+00 2.276012e+00 1.169908e+00 2.750420e+00
f4 Mean 6.276495e+03 2.732561e+03 6.006557e+03 2.784096e+03 5.719822e+03 3.291666e+03 5.988274e+03 3.421316e+03
Std.Dev. 6.710218e+02 3.767996e+02 4.906094e+02 4.821994e+02 5.428894e+02 5.271669e+02 4.650399e+02 3.503447e+02
f5 Mean 5.033085e+02 5.033101e+02 5.036674e+02 5.034889e+02 5.037955e+02 5.037055e+02 5.038322e+02 5.035775e+02
Std.Dev. 6.569836e-01 5.756491e-01 4.720279e-01 5.575871e-01 5.177964e-01 6.483718e-01 6.067588e-01 4.535533e-01
f6 Mean 6.059541e+02 6.043150e+02 6.038948e+02 6.022756e+02 6.034313e+02 6.032336e+02 6.037581e+02 6.020349e+02
Std.Dev. 6.083780e-01 5.162583e-01 2.891287e-01 6.328638e-01 4.050946e-01 5.041688e-01 4.249729e-01 6.418903e-01
f7 Mean 8.280332e+02 7.265373e+02 7.674958e+02 7.134192e+02 7.471265e+02 7.393423e+02 7.558142e+02 7.149501e+02
Std.Dev. 2.711322e+01 1.264477e+01 1.061656e+01 6.111133e+00 1.085824e+01 1.294565e+01 1.282543e+01 6.187258e+00
f8 Mean 4.051586e+06 2.447719e+05 2.259768e+06 2.582121e+05 8.868484e+05 1.077119e+06 1.096711e+06 3.114737e+05
Std.Dev. 4.570396e+06 3.142059e+05 3.780407e+06 3.166477e+05 1.014774e+06 1.215732e+06 9.578954e+05 3.150931e+05
f9 Mean 9.138841e+02 9.138276e+02 9.137592e+02 9.136505e+02 9.136567e+02 9.136961e+02 9.137667e+02 9.137625e+02
Std.Dev. 1.967185e-01 1.720790e-01 1.847203e-01 2.675077e-01 2.257857e-01 2.123853e-01 1.658057e-01 1.593906e-01
f10 Mean 2.380438e+07 2.282666e+07 1.913100e+07 1.492403e+07 1.771557e+07 2.454621e+07 2.115008e+07 1.521697e+07
Std.Dev. 1.470553e+07 1.878189e+07 1.147958e+07 8.349137e+06 9.831280e+06 1.501224e+07 1.303823e+07 9.147938e+06
f11 Mean 1.341192e+03 1.215966e+03 1.286128e+03 1.203103e+03 1.278993e+03 1.229337e+03 1.284880e+03 1.198032e+03
Std.Dev. 8.150961e+01 6.135326e+01 4.476843e+01 5.488490e+01 4.359358e+01 4.696869e+01 4.644517e+01 4.526462e+01
f12 Mean 2.412364e+03 2.256578e+03 2.305637e+03 2.310203e+03 2.302388e+03 2.218442e+03 2.371829e+03 2.183720e+03
Std.Dev. 3.467679e+02 2.749184e+02 2.767168e+02 2.180678e+02 2.359488e+02 2.069351e+02 2.300865e+02 2.476781e+02
f13 Mean 2.419417e+03 1.907681e+03 2.118388e+03 1.808825e+03 2.019165e+03 2.030852e+03 2.049470e+03 1.815736e+03
Std.Dev. 2.079640e+02 1.475313e+02 2.027472e+02 6.872095e+01 1.297077e+02 1.730694e+02 1.253398e+02 5.285595e+01
f14 Mean 1.764708e+03 1.703526e+03 1.733239e+03 1.706430e+03 1.733157e+03 1.708865e+03 1.728673e+03 1.692752e+03
Std.Dev. 4.794642e+01 2.837500e+01 4.000809e+01 3.465256e+01 2.732741e+01 2.529661e+01 2.064841e+01 2.377258e+01
f15 Mean 2.932039e+03 2.845327e+03 2.796346e+03 2.771188e+03 2.807676e+03 2.749968e+03 2.769711e+03 2.769860e+03
Std.Dev. 9.098679e+01 8.219807e+01 9.463118e+01 8.818279e+01 1.351278e+02 1.445580e+02 1.339719e+02 8.896387e+01

ABC, Artificial Bee Colony; GABC, gbest-guided ABC; Std.Dev., standard deviation.

Table 8

Comparison of ABC, GABC, ABC/best/1, ABC/best/2 algorithms, and their ts model based–variants with 50 bees on solving 30 dimensional problems.

Function COABC ts-COABC CABC ts-CABC ERABC ts-ERABC EABC ts-EABC
f1 Mean 3.392230e+08 4.164658e+09 1.183611e+10 2.944819e+09 1.096107e+11 1.107493e+11 2.060155e+10 7.991217e+09
Std.Dev. 2.020635e+08 5.103633e+09 3.310679e+09 2.026062e+09 1.740786e+10 1.289122e+10 5.062198e+09 2.314629e+09
f2 Mean 1.164117e+05 1.281981e+05 1.105561e+05 1.101947e+05 1.061734e+07 1.666185e+07 1.056518e+05 1.014493e+05
Std.Dev. 1.505867e+04 1.571463e+04 1.455210e+04 1.605736e+04 1.578417e+07 2.073910e+07 1.569767e+04 1.910171e+04
f3 Mean 3.398018e+02 3.399267e+02 3.387373e+02 3.383513e+02 3.502569e+02 3.499533e+02 3.399393e+02 3.377565e+02
Std.Dev. 4.500009e+00 3.473378e+00 2.411661e+00 1.625555e+00 1.865920e+00 2.592209e+00 1.727220e+00 2.234824e+00
f4 Mean 1.542864e+03 1.827270e+03 5.505484e+03 2.176328e+03 1.011638e+04 1.011802e+04 5.677646e+03 2.716079e+03
Std.Dev. 2.790805e+02 4.285467e+02 4.310868e+02 3.991444e+02 5.483088e+02 5.934300e+02 4.973526e+02 2.829212e+02
f5 Mean 5.024106e+02 5.027114e+02 5.038783e+02 5.037789e+02 5.078405e+02 5.061605e+02 5.036095e+02 5.037457e+02
Std.Dev. 6.522530e-01 6.466203e-01 4.998308e-01 6.313289e-01 1.012284e+00 8.121373e-01 6.936673e-01 7.678846e-01
f6 Mean 6.006201e+02 6.006248e+02 6.032338e+02 6.014558e+02 6.083945e+02 6.088531e+02 6.040787e+02 6.032471e+02
Std.Dev. 1.049041e-01 1.544186e-01 3.615688e-01 6.146502e-01 1.127977e+00 6.806799e-01 4.411395e-01 5.302250e-01
f7 Mean 7.005874e+02 7.006926e+02 7.388427e+02 7.057290e+02 9.406802e+02 9.372345e+02 7.606808e+02 7.220232e+02
Std.Dev. 2.127087e-01 8.775633e-01 1.126668e+01 4.447253e+00 4.546680e+01 4.466095e+01 1.570908e+01 7.853288e+00
f8 Mean 1.047361e+04 4.221154e+06 9.479636e+05 1.441949e+05 2.995963e+08 3.642394e+08 1.023768e+06 1.784301e+05
Std.Dev. 1.527780e+04 5.896717e+06 1.336371e+06 1.528524e+05 1.787838e+08 2.094814e+08 9.555743e+05 1.473196e+05
f9 Mean 9.136145e+02 9.136088e+02 9.138261e+02 9.138501e+02 9.145886e+02 9.144542e+02 9.137967e+02 9.137944e+02
Std.Dev. 2.818030e-01 2.779025e-01 1.441430e-01 1.559745e-01 1.021574e-01 1.593844e-01 1.712227e-01 1.332187e-01
f10 Mean 1.406859e+07 3.224804e+07 2.557889e+07 2.003803e+07 5.719769e+08 4.844341e+08 2.156090e+07 1.849368e+07
Std.Dev. 9.306677e+06 2.857811e+07 1.251364e+07 1.158922e+07 2.940135e+08 2.386727e+08 1.138031e+07 1.210593e+07
f11 Mean 1.216180e+03 1.198985e+03 1.279237e+03 1.206475e+03 2.270559e+03 2.239188e+03 1.268710e+03 1.200591e+03
Std.Dev. 6.201865e+01 6.068841e+01 4.442197e+01 5.372697e+01 4.047155e+02 3.094176e+02 3.989952e+01 4.999593e+01
f12 Mean 2.088856e+03 4.173864e+03 2.554535e+03 2.180638e+03 1.138555e+06 1.625529e+06 2.312091e+03 2.280820e+03
Std.Dev. 3.430461e+02 6.317761e+03 2.502107e+02 2.427005e+02 1.441746e+06 1.956796e+06 3.656029e+02 2.016592e+02
f13 Mean 1.746439e+03 1.898094e+03 2.056320e+03 1.800538e+03 4.734815e+03 4.167272e+03 2.257822e+03 1.931461e+03
Std.Dev. 7.034381e+01 2.221552e+02 1.177981e+02 5.371752e+01 9.340347e+02 7.986044e+02 1.951653e+02 7.592999e+01
f14 Mean 1.688138e+03 1.709483e+03 1.738893e+03 1.706672e+03 2.360178e+03 2.265037e+03 1.739033e+03 1.706500e+03
Std.Dev. 4.148529e+01 3.715515e+01 3.027531e+01 2.675150e+01 2.041283e+02 2.288386e+02 3.135073e+01 2.027444e+01
f15 Mean 2.808342e+03 2.924411e+03 2.803138e+03 2.715771e+03 4.212530e+03 3.950307e+03 2.854667e+03 2.751790e+03
Std.Dev. 2.485736e+02 1.329088e+02 1.479178e+02 1.513812e+02 6.620187e+02 4.207504e+02 1.036900e+02 1.212787e+02

ABC, Artificial Bee Colony; Std.Dev., standard deviation; COABC, converge-onlookers ABC.

Table 9

Comparison of COABC, CABC, ERABC, EABC algorithms, and their ts model–based variants with 50 bees on solving 30 dimensional problems.

Figure 3

Convergence graphics of Artificial Bee Colony (ABC) algorithms for (l) functions.

Although the differences between mean best objective function values of the ABC algorithms and their ts mechanism–based variants are generally in favor of proposed models, a statistical test should also be used whether mentioned differences between variants are also meaningful or not. With this purpose, a commonly used statistical test called Wilcoxon signed rank test with the significance level 0.05 is used to analyze ABC algorithms with 30 bees on solving 30 dimensional benchmark problems and the results are given in Table 10. If the significance level showed by p is equal or less than 0.05, it is accepted that the difference between algorithms is significant in favor of one of the algorithms.

Function ABC vs ts-ABC GABC vs ts-GABC ABC/best/1 vs ts-ABC/best/1 ABC/best/2 vs ts-ABC/best/2
p-val. Sign. p-val. Sign. p-val. Sign. p-val. Sign.
f1 0.000089 ts-ABC 0.000254 ts-GABC 0.001713 ABC/best/1 0.000089 ts-ABC/best/2
f2 0.000001 ABC 0.736875 - 0.191334 - 0.822760 -
f3 0.135357 - 0.575486 - 0.012374 ts-ABC/best/1 0.601213 -
f4 0.000103 ts-ABC 0.000089 ts-GABC 0.000089 ts-ABC/best/1 0.000103 ts-ABC/best/2
f5 0.007189 ts-ABC 0.822760 - 0.940481 - 0.061953 -
f6 0.000089 ts-ABC 0.000103 ts-GABC 0.008968 ABC/best/1 0.000390 ts-ABC/best/2
f7 0.000120 ts-ABC 0.000103 ts-GABC 0.061953 - 0.000103 ts-ABC/best/2
f8 0.052222 - 0.000780 ts-GABC 0.167184 - 0.008034 ts-ABC/best/2
f9 0.145400 - 0.970220 - 0.390533 - 0.145400 -
f10 0.881293 - 0.654159 - 0.940481 - 0.073138 -
f11 0.000254 ts-ABC 0.027621 ts-GABC 0.278965 - 0.135357 -
f12 0.108427 - 0.145400 - 0.116888 - 0.217957 -
f13 0.000089 ts-ABC 0.000189 ts-GABC 0.002495 ABC/best/1 0.001162 ts-ABC/best/2
f14 0.003592 ts-ABC 0.247145 - 0.232226 - 0.204330 -
f15 0.040044 ts-ABC 0.262722 - 0.156004 - 0.247145 -
Function COABC vs ts-COABC CABC vs ts-CABC ERABC vs ts-ERABC EABC vs ts-EABC
p-val. Sign. p-val. Sign. p-val. Sign. p-val. Sign.
f1 0.001944 COABC 0.001162 ts-CABC 0.116888 - 0.000254 ts-EABC
f2 0.940481 - 0.455273 - 0.455273 - 0.970220 -
f3 0.627446 - 0.036561 ts-CABC 0.191334 - 0.025094 ts-EABC
f4 0.004550 - 0.000089 ts-CABC 0.455273 - 0.000089 ts-EABC
f5 0.654159 - 0.822760 - 0.046222 ts-ERABC 0.350656 -
f6 0.851925 - 0.000517 - 0.204330 - 0.000338 ts-EABC
f7 0.736875 - 0.000089 ts-CABC 0.940481 - 0.000681 ts-EABC
f8 0.000293 COABC 0.006425 ts-CABC 0.135357 - 0.079322 -
f9 0.125859 - 0.322467 - 0.350656 - 0.881293 -
f10 0.204330 - 0.940481 - 0.765198 - 0.331723 -
f11 0.313463 - 0.100458 - 0.047138 ts-ERABC 0.018675 ts-EABC
f12 0.001507 COABC 0.708905 - 0.822760 - 0.025094 ts-EABC
f13 0.001162 COABC 0.000189 ts-CABC 0.473271 - 0.040044 ts-EABC
f14 0.331723 - 0.008034 ts-CABC 0.455273 - 0.018675 ts-EABC
f15 0.056661 - 0.331723 - 0.313463 - 1.000000 -

ABC, Artificial Bee Colony; Std.Dev., standard deviation; COABC, converge-onlookers ABC; GABC, gbest-guided ABC.

Table 10

Statistical comparison between ABC, GABC, ABC/best/1, ABC/best/2, COABC, CABC, ERABC, EABC algorithms, and their ts mechanism–based variants.

When the test results given in Table 10 are investigated, it is easily seen that if there is a statistical significance between ABC algorithms and their ts mechanism–based counterparts, it is usually in favor of ts mechanism–based ABC variants. The small differences between mean best objective function values of ABC, GABC, ABC/best/2, CABC, EABC algorithms and their proposed variants are enough for statistical significances that are in favor of ts-based variants at least six or more benchmark functions. For ABC/best/1 and ERABC algorithms, there are two benchmark functions for which ts mechanism–based variants are also statistically significant. Only for COABC algorithm, ts mechanism cannot be capable of generating statistically significant differences between mean best objective function values. While the differences between COABC and ts-COABC algorithms are significant in favor of COABC algorithm for f1, f4, f8, f12 and f13 functions, there is no statistical significance between COABC and ts-COABC algorithms for f2, f3, f4, f5, f6, f7, f9, f10, f11, f14, and f15 functions.

A final comparison between ABC algorithms and their ts-based variants is made over the average execution times of them. The ts mechanism requires some operations for determining initial dancing durations and controlling whether an employed bee will stay on the dance area or not. Even though mentioned ts operations bring extra burden to the total execution times of the ABC algorithms, they significantly change the selection characteristics of the onlooker bees. As mentioned before, the food source selection procedure of the ABC algorithm and its variants is managed with a randomized model. In this randomized selection model, if a random number generated between 0 and +1 is less than the fitness value of the considered food source, this food source is chosen by an onlooker and a candidate is generated within the neighborhood of it. Although the selection operation is simple, fitness values of the food sources can be relatively close to 0 and generating a random number less than this value can require more trials. If the fitness values of the food sources are high, the selection of these sources can be completed quickly when the probability of generating a random number less than these fitness values is considered. With the ts mechanism, the employed bees on the dance area are generally related with the qualified food sources and the time spent for selecting them by onlookers is generally less than the time spent by the onlooker bees of conventional selection schema.

Because of the subtle balance between computational burden of the ts mechanism and positive contribution to the selection characteristics of the onlooker bees, the average execution times of the ABC algorithms and their ts-based implementations should be found relatively close to each other. When the average execution times in terms of seconds calculated over the 20 independent runs of the ABC algorithms with 30 bees on solving 30 dimensional problems in the Tables 11 and 12 are investigated, it can be seen that the computational burden stemmed from the ts mechanism does not produce a relatively high differences on the execution times compared to the original algorithms. The compute-intensive operations are not included by the steps of the ts mechanism and the time spent for determining initial dancing times and decrementing them until satisfying the constraints about the number of employed bees on dance area can be neglected when the time spent for other operations of the original implementations of the ABC algorithms.

Function ABC ts-ABC GABC ts-GABC ABC/best/1 ts-ABC/best/1 ABC/best/2 ts-ABC/best/2
f1 Mean 0.050131 0.046639 0.059183 0.052870 0.041121 0.037392 0.045853 0.041305
Std.Dev. 2.242718e-02 8.920995e-03 3.164774e-01 1.499793e-02 7.106844e-03 1.674738e-03 1.467020e-02 1.203549e-02
f2 Mean 0.040583 0.043533 0.054152 0.055728 0.038437 0.039688 0.049270 0.059717
Std.Dev. 2.850145e-02 2.139360e-02 2.760976e-02 2.285981e-02 2.228508e-03 2.313789e-03 1.697972e-02 2.257493e-02
f3 Mean 0.672528 0.605529 0.688751 0.580251 0.460302 0.470560 0.537115 0.538602
Std.Dev. 6.792684e-01 1.774436e-01 2.895477e-01 1.269505e-01 4.413920e-03 4.202001e-03 9.405997e-02 8.781438e-02
f4 Mean 0.051875 0.044763 0.041951 0.038930 0.039228 0.039782 0.048429 0.039847
Std.Dev. 3.532368e-01 1.707656e-02 1.160175e-02 2.430201e-03 2.091136e-03 2.446647e-03 3.055497e-02 2.147554e-02
f5 Mean 0.429912 0.444856 0.481442 0.434269 0.350375 0.343755 0.527405 0.474085
Std.Dev. 8.362566e-02 1.308049e-01 1.582201e-01 1.213690e-01 7.413277e-03 4.768584e-03 3.288734e-01 4.382081e-02
f6 Mean 0.048418 0.050962 0.055029 0.050854 0.045982 0.042839 0.040660 0.042495
Std.Dev. 2.142011e-02 3.196463e-01 2.933311e-01 1.921920e-01 9.299369e-03 1.319572e-03 2.099218e-01 1.992233e-02
f7 Mean 0.081208 0.089252 0.040867 0.040748 0.048316 0.040513 0.043427 0.048907
Std.Dev. 2.302384e-01 3.539231e-01 9.055348e-03 4.076368e-01 1.767417e-03 1.744643e-03 1.680284e-02 2.096029e-02
f8 Mean 0.058148 0.049594 0.054278 0.069050 0.035208 0.035991 0.092171 0.101456
Std.Dev. 2.455125e-01 5.855726e-03 2.699446e-02 4.720703e-01 1.964160e-03 2.021030e-03 3.344670e-01 3.038753e-01
f9 Mean 0.063104 0.068442 0.058388 0.060236 0.046402 0.037682 0.059763 0.052081
Std.Dev. 1.969610e-02 4.969772e-03 1.549435e-02 3.713047e-02 1.967376e-03 1.080150e-03 2.270765e-01 5.650666e-03
f10 Mean 0.105905 0.092958 0.068589 0.078626 0.048444 0.439376 0.058933 0.050908
Std.Dev. 4.294203e-01 9.130607e-02 1.245935e-01 5.811356e-01 2.310650e-03 1.915442e-03 2.997234e-02 1.240162e-02
f11 Mean 0.157163 0.156823 0.167664 0.166312 0.125320 0.127797 0.154117 0.132302
Std.Dev. 5.415421e-02 3.213212e-02 5.846262e-02 1.720524e-01 2.689619e-03 2.909550e-03 2.592682e-02 1.362776e-02
f12 Mean 0.091845 0.095219 0.102071 0.125829 0.072141 0.074464 0.130435 0.079509
Std.Dev. 1.658417e-02 2.953888e-02 2.345862e-02 2.097525e-01 2.735565e-03 2.963951e-03 5.064997e-02 9.973386e-03
f13 Mean 0.122911 0.114850 0.133987 0.109591 0.097796 0.097495 0.153980 0.173686
Std.Dev. 3.385718e-02 1.539274e-02 7.096232e-02 1.404100e-02 1.880880e-03 2.222095e-03 7.549713e-01 2.580564e-01
f14 Mean 0.149240 0.120973 0.122755 0.143693 0.091375 0.093722 0.107004 0.131532
Std.Dev. 6.229411e-02 1.193391e-02 5.597142e-01 5.455718e-02 2.973894e-03 1.123502e-02 2.603938e-02 3.791140e-01
f15 Mean 0.602323 0.642129 0.652188 0.678815 0.544688 0.549962 0.643163 0.779873
Std.Dev. 8.578987e-02 7.396672e-02 6.492040e-02 1.706552e-01 5.006217e-03 7.031245e-03 1.224565e-01 3.356357e-01

ABC, Artificial Bee Colony; Std.Dev., standard deviation; GABC, gbest-guided ABC.

Table 11

Average execution times of ABC, GABC, ABC/best/1, ABC/best/2 algorithms, and their ts mechanism–based variants.

Function COABC ts-COABC CABC ts-CABC ERABC ts-ERABC EABC ts-EABC
f1 Mean 0.065230 0.066156 0.048353 0.056149 0.046515 0.062363 0.042773 0.049179
Std.Dev. 2.297517e-02 1.652955e-01 2.492893e-03 1.890919e-02 2.339866e-03 2.313109e-02 1.500081e-02 2.101563e-03
f2 Mean 0.048287 0.048671 0.047731 0.049990 0.046675 0.051666 0.045216 0.049033
Std.Dev. 3.843406e-03 1.226867e-02 2.183988e-03 8.376007e-03 2.372530e-03 4.363600e-01 1.527630e-02 1.714590e-03
f3 Mean 0.642305 0.625817 0.464593 0.572149 0.467411 0.600064 0.655932 0.569024
Std.Dev. 2.020452e-01 1.927304e-01 4.733351e-03 3.458033e-01 3.271359e-02 1.275539e-01 3.551220e-01 4.200882e-03
f4 Mean 0.052991 0.052837 0.049458 0.046331 0.048568 0.053026 0.044320 0.049970
Std.Dev. 2.889452e-01 5.392231e-02 3.156447e-03 5.992266e-03 2.107408e-03 3.019394e-02 1.609946e-02 2.027419e-03
f5 Mean 0.486031 0.476229 0.554044 0.605130 0.547858 0.502536 0.494557 0.446165
Std.Dev. 1.748296e-01 1.440274e-01 4.264174e-03 1.997422e-01 3.356789e-03 2.664092e-01 5.522459e-02 3.541961e-03
f6 Mean 0.054199 0.057993 0.069152 0.077372 0.047787 0.057286 0.044491 0.049964
Std.Dev. 1.289931e-02 3.290299e-01 2.164248e-03 3.150003e-02 2.288113e-03 2.577948e-02 8.064113e-03 2.112660e-03
f7 Mean 0.058557 0.056617 0.049679 0.042025 0.039372 0.041644 0.036487 0.039817
Std.Dev. 2.450032e-01 5.599165e-03 2.488866e-03 1.175164e-02 2.025220e-03 1.466875e-02 8.675390e-03 2.522923e-03
f8 Mean 0.059404 0.061252 0.044772 0.053283 0.044144 0.053763 0.054268 0.047425
Std.Dev. 1.448049e-02 1.587708e-02 1.555508e-03 1.514707e-02 1.105658e-02 1.421387e-02 1.502297e-02 2.056765e-03
f9 Mean 0.042446 0.042979 0.045441 0.053175 0.044930 0.497018 0.067967 0.057086
Std.Dev. 5.893791e-03 5.309384e-02 1.960864e-03 3.008490e-02 1.662135e-03 7.127754e-01 3.304973e-02 1.971332e-03
f10 Mean 0.058155 0.062264 0.068542 0.063555 0.050538 0.065660 0.047023 0.050844
Std.Dev. 6.084131e-02 5.452598e-01 1.812356e-03 1.795279e-02 1.400168e-02 1.137388e-01 3.423352e-01 1.532782e-03
f11 Mean 0.144628 0.170425 0.124703 0.168718 0.124929 0.166217 0.128905 0.136100
Std.Dev. 4.635860e-02 4.743490e-02 2.834806e-03 6.327066e-02 3.438445e-03 5.829321e-02 4.551299e-01 2.603739e-03
f12 Mean 0.102036 0.072304 0.072711 0.190395 0.072076 0.100883 0.102415 0.073278
Std.Dev. 3.555396e-02 1.879055e-02 3.500502e-03 1.618380e-01 2.611370e-03 2.088056e-02 3.048965e-02 3.004425e-03
f13 Mean 0.183922 0.158835 0.096309 0.115690 0.095778 0.113198 0.134287 0.149578
Std.Dev. 5.134852e-01 4.997218e-02 1.729511e-03 1.395025e-02 2.479506e-03 5.187983e-02 4.515653e-02 3.588585e-03
f14 Mean 0.104493 0.099019 0.093205 0.113302 0.090379 0.117564 0.107687 0.090693
Std.Dev. 2.118410e-02 1.433848e-02 4.081801e-03 2.379353e-01 1.796991e-03 4.278472e-01 2.656273e-02 2.370891e-03
f15 Mean 0.687970 0.601772 0.557062 0.662383 0.552846 0.630580 0.611085 0.550082
Std.Dev. 1.955580e-01 5.276792e-02 3.366157e-03 3.250891e-01 4.767583e-02 8.352474e-02 2.777474e-01 8.321810e-03

ABC, Artificial Bee Colony; Std.Dev., standard deviation; COABC, converge-onlookers ABC.

Table 12

Average execution times of COABC, CABC, ERABC, EABC algorithms, and their ts mechanism–based variants.

5. CONCLUSION

In this study, dancing characteristics of the employed bees are modeled with a new approach called time-based dance scheduling for short ts mechanism. The ts mechanism is integrated to the standard ABC algorithm and its well-known variants. Experimental studies on a set of difficult benchmark problems showed that standard ABC algorithm and its variants especially those require selection of the food source being consumed and the neighbor food source or sources being utilized are improved with the proposed mechanism in terms of qualities of the solutions and convergence speeds.

The improvements with the ts mechanism also provide important information about the possible contributions of the unmodeled forager characteristics. Some bee characteristics that are not directly included in the standard ABC algorithm or related with the randomized operations can be integrated to the workflow of the algorithm for increasing the performance and reflecting the intelligent behaviours more completely. In future, ABC algorithm can be modified by modeling decision-making characteristics of the employed foragers that is used whether or not an employed forager is transitioned to the dance area or different timing mechanisms in which the start and finish times of dancing are determined according to the qualities of the memorized sources.

References

2.A.L. Bolaji, A.T. Khader, M.A. Al-Betar, and M.A. Awadallah, Artificial bee colony algorithm, its variants and applications: a survey, J. Theor. Appl. Inf. Technol., Vol. 47, No. 2013, pp. 434-459.
6.P.-W. TSai, J.-S. Pan, B.-Y. Liao, and S.-C. Chu, Enhanced artificial bee colony optimization, Int. J. Innovative Comput. Inf. Control, Vol. 5, 2009, pp. 5081-5092.
24.N. Narasimhan, Parallel aritificial bee colony algorithm, IEEE, in World Congress on Nature and Biologically Inspired Computing (Coimbatore, India), 2009, pp. 306-311.
26.M. Tuba, M. Subotic, and N. Stanarevic, Parallelization of the artificial bee colony algorithm, in Proceedings of the 11th WSEAS International Conference on Neural Networks and 11th WSEAS International Conference on Evolutionary Computing (Wisconsin, USA), 2010, pp. 191-196.
27.M. Tuba, M. Subotic, and N. Stanarevic, Different approaches in parallelization of the artificial bee colony algorithm, Int. J. Math. Models Methods Appl. Sci., Vol. 5, 2010, pp. 755-762.
37.Q. Chen, B. Liu, Q. Zhang, J.J. Liang, P.N. Suganthan, and B.Y. Qu, Problem definitions and evaluation criteria for CEC 2015 special session on bound constrained single-objective computationally expensive numerical optimization, 2014. Technical Report, Computational Intelligence Laboratory, Zhengzhou University, Zhengzhou, China and Technical Report, Nanyang Technological University https://al-roomi.org/multimedia/CEC_Database/CEC2015/RealParameterOptimization/ExpensiveOptimization/CEC2015_ExpensiveOptimization_TechnicalReport.pdf
Journal
International Journal of Computational Intelligence Systems
Volume-Issue
12 - 2
Pages
597 - 612
Publication Date
2019/05/11
ISSN (Online)
1875-6883
ISSN (Print)
1875-6891
DOI
10.2991/ijcis.d.190425.001How to use a DOI?
Copyright
© 2019 The Authors. Published by Atlantis Press SARL.
Open Access
This is an open access article distributed under the CC BY-NC 4.0 license (http://creativecommons.org/licenses/by-nc/4.0/).

Cite this article

TY  - JOUR
AU  - Selcuk Aslan
PY  - 2019
DA  - 2019/05/11
TI  - Time-Based Dance Scheduling for Artificial Bee Colony Algorithm and Its Variants
JO  - International Journal of Computational Intelligence Systems
SP  - 597
EP  - 612
VL  - 12
IS  - 2
SN  - 1875-6883
UR  - https://doi.org/10.2991/ijcis.d.190425.001
DO  - 10.2991/ijcis.d.190425.001
ID  - Aslan2019
ER  -