Objective In the artificial bee colony (ABC) algorithm, employed bees search the entire search space while onlooker bees concentrate their efforts near high-quality food sources. From the perspective of exploration and exploitation, employed bees primarily handle exploration, whereas onlooker bees are responsible for exploitation. However, the basic ABC algorithm prioritizes exploration during the search process and performs poorly in exploitation, leading to slow convergence speed and low solution accuracy. Therefore, this study proposes an artificial bee colony algorithm based on the division between exploration and exploitation (called ABC_DEE), which consists of three stages: employed bees for exploration, onlooker bees for exploitation, and scout bees for supplementation. Methods In the exploration stage, excessive reliance on random solution information biased the process toward random search. Therefore, a solution-search equation guided by diverse elites was designed for employed bees, along with the introduction of a breadth-first search strategy to enhance exploration. In the exploitation stage, overutilization of optimal solution information led to premature convergence. Hence, a solution-search equation guided by objective-oriented elites was designed for onlooker bees, using a depth-first search strategy to strengthen exploitation. Considering that the random initialization method of scout bees discarded the previous search experience, and the search equation of employed and onlooker bees was singular, a neighborhood search equation was designed for scout bees. This equation considered the optimal solution based on objective value, the optimal solution based on diversity, and the previous search experience. Results and Discussions The performance of ABC_DEE was assessed on both the CEC2021 test set and the esophageal cancer prediction problem, with comparisons made against six ABC variants. In numerical optimization problems with D=10, ABC_DEE outperformed ABC, GABC, REABC, ENABC, NSABC, and RNSABC on 49, 46, 51, 61, 52, and 48 functions, respectively. In contrast, ABC_DEE performed worse than these algorithms on 26, 26, 24, 12, 22, and 23 functions, respectively. The Friedman test results indicated that ABC_DEE ranked first. The Wilcoxon test results indicated that there was no significant difference between ABC_DEE and GABC, but ABC_DEE significantly outperformed the other algorithms. When D=20, ABC_DEE was superior to the comparison algorithms on at least 47 functions and inferior on at most 26 functions. The results of both the Friedman test and the Wilcoxon test were consistent with those when D=10. In addition, ABC_DEE exhibited superiority in terms of convergence speed and time efficiency. Regarding the optimization problem of the esophageal cancer prediction model based on Kernel Extreme Learning Machine (KELM), in terms of accuracy, ABC_DEE‒KELM achieved 85.83%, which was 3.00% higher than the second-ranked RNSABC‒KELM and 15.72% higher than ABC‒KELM. Regarding sensitivity, ABC_DEE‒KELM reached 91.98%, outperforming the second-ranked ENABC‒KELM by 0.61% and ABC‒KELM by 20.14%. In terms of specificity, ABC_DEE‒KELM attained 79.31%, exceeding the second-ranked RNSABC‒KELM by 0.70% and ABC‒KELM by 15.03%. Regarding the F1 score, ABC_DEE‒KELM achieved 85.49%, showing a lead of 2.54% over the second-ranked RNSABC‒KELM and 16.23% over ABC‒KELM. Conclusions The classical ABC suffers from an imbalance between exploration and exploitation. Regulating the relationship between exploration and exploitation is an effective method for improving the performance of ABC. This study proposes a novel ABC based on the separation of exploration and exploitation without altering the original ABC framework. Under the "base vector + perturbation" search mode, the proposed approach strengthens the division of labor between employed bees and onlooker bees in terms of exploration and exploitation through a dual-elite-guided strategy based on diversity and objective values. Simultaneously, a search equation is formulated for scout bees that incorporates both search experience and diversified optimal solution information. The proposed algorithm is compared to six other ABC algorithms across 80 benchmark functions, with its superiority evaluated in terms of solution quality, non-parametric tests, convergence speed, and time efficiency. In addition, its effectiveness in practical optimization problems is validated using esophageal cancer prediction as an example. Future research can be approached from two perspectives. First, the proposed algorithm will be applied to address more practical problems such as path planning and image processing. Second, by exploring strategies such as neighborhood topology and inertia weights to fine-tune the balance between exploration and exploitation in ABC.
上述改进方法均提高了ABC算法的性能,但学者们仍在不懈地进行改进研究。针对人工蜂群算法探索强、开采弱的问题,本文提出一种基于探采分工的人工蜂群算法(artificial bee colony algorithm based on the division between exploration and exploitation,ABC_DEE)。本文算法令雇佣蜂执行探索,跟随蜂实施开采,侦察蜂进行补充。在探索和开采阶段,分别利用多样性和目标值评价解质量,进而设计基于相应精英解引导的搜索方程,同时配合广度优先搜索策略和深度优先搜索策略强化探索和开采。令侦察蜂采用邻域搜索算子代替随机搜索,以弥补雇佣蜂和跟随蜂的搜索方程形式单一的不足。为验证ABC_DEE算法的性能,将其用于函数优化和食管癌预测模型优化,并与其他改进ABC算法进行对比。
AkayB, KarabogaD, GorkemliB,et al.A survey on the Artificial Bee Colony algorithm variants for binary,integer and mixed integer programming problems[J].Applied Soft Computing,2021,106:107351. doi:10.1016/j.asoc.2021.107351
[2]
ShuJian, LiRuirui, XiongTao,et al.Link prediction based on learning automaton and firefly algorithm[J].Advanced Engineering Sciences,2021,53(2):133‒140. doi:10.15961/j.jsuese.202000781
WangYingcong, SuiChengcheng, LiuChi,et al.Chicken swarm optimization with an enhanced exploration‒exploitation tradeoff and its application[J].Soft Computing,2023,27(12):8013‒8028. doi:10.1007/s00500-023-07990-8
[5]
YuXiuwu, QinXiaokun, LiuYong,et al.DV‒Hop localization algorithm optimized based on global artificial fish swarm algorithm[J].Advanced Engineering Sciences,2022,54(4):228‒234.
CuiYibing, HuWei, RahmaniA.A reinforcement learning based artificial bee colony algorithm with application in robot path planning[J].Expert Systems with Applications,2022,203:117389. doi:10.1016/j.eswa.2022.117389
[10]
JawadF K J, OzturkC, WangDansheng,et al.Sizing and layout optimization of truss structures with artificial bee colony algorithm[J].Structures,2021,30:546‒559. doi:10.1016/j.istruc.2021.01.016
ZhuGuopu, KwongS.Gbest-guided artificial bee colony algorithm for numerical function optimization[J].Applied Mathematics and Computation,2010,217(7):3166‒3173. doi:10.1016/j.amc.2010.08.049
[13]
KongDepeng, ChangTianqing, DaiWenjun,et al.An improved artificial bee colony algorithm based on the ranking selection and the elite guidance[J].Control and Decision,2019,34(4):781‒786.
MengHongyun, WeiBingke.An improved artificial bee colony algorithm based on elite solution and random individual neighborhood information[J].Control and Decision,2020,35(9):2169‒2174. doi:10.13195/j.kzyjc.2018.1797
WangHui, WangWenjun, XiaoSongyi,et al.Improving artificial bee colony algorithm using a new neighborhood selection mechanism[J].Information Sciences,2020,527:227‒240. doi:10.1016/j.ins.2020.03.064
[18]
YeTingyu, WangWenjun, WangHui,et al.Artificial bee colony algorithm with efficient search strategy based on random neighborhood structure[J].Knowledge‒Based Systems,2022,241:108306. doi:10.1016/j.knosys.2022.108306
[19]
WangZhigang, WangMinggang.Multi-search strategy of artificial bee colony algorithm based on symbolic function[J].Control and Decision,2016,31(11):2037‒2044.
HakliH, KiranM S.An improved artificial bee colony algorithm for balancing local and global search behaviors in continuous optimization[J].International Journal of Machine Learning and Cybernetics,2020,11(9):2051‒2076. doi:10.1007/s13042-020-01094-7
DuZhenxin, LiuGuangzhong, HanDezhi,et al.Artificial bee colony algorithm with global and unbiased search strategy[J].Acta Electronica Sinica,2018,46(2):308‒314.
SongXiaoyu, ZhaoMing, YanQifeng,et al.A high-efficiency adaptive artificial bee colony algorithm using two strategies for continuous optimization[J].Swarm and Evolutionary Computation,2019,50:100549. doi:10.1016/j.swevo.2019.06.006
ImanianN, ShiriM E, MoradiP.Velocity based artificial bee colony algorithm for high dimensional continuous optimization problems[J].Engineering Applications of Artificial Intelligence,2014,36:148‒163. doi:10.1016/j.engappai.2014.07.012
[36]
MohamedA, HadiA, MohamedA,et al.Problem definitions and evaluation criteria for the CEC 2021 special session and competition on single objective bound constrained numerical optimization[R].Singapore:Nanyang Technological University,2020. doi:10.1145/3377929.3398187
[37]
ZhouXinyu, WangHui, WangMingwen,et al.Enhancing the modified artificial bee colony algorithm with neighborhood search[J].Soft Computing,2017,21(10):2733‒2743. doi:10.1007/s00500-015-1977-x