基于动态多种群机制的增强花授粉算法  

Enhanced flower pollination algorithm by adopting dynamic multi-group mechanism

作　　者：李大海 凌继源 王振东 Li Dahai;Ling Jiyuan;Wang Zhendong(School of Information Engineering,Jiangxi University of Science&Technology,Ganzhou Jiangxi 341000,China)

机构地区：[1]江西理工大学信息工程学院,江西赣州341000

出　　处：《计算机应用研究》2024年第12期3671-3678,共8页Application Research of Computers

基　　金：国家自然科学基金资助项目(61563019,615620237);江西理工大学校级资助项目(205200100013)。

摘　　要：针对花授粉算法易陷入局部最优、收敛精度不足和过早收敛的问题,提出一种基于动态多种群机制的增强花授粉算法(DMEFPA)。首先,DMEFPA使用一种融合个体适应度值和相对距离的方法挑选中心个体,使选出的个体既保持较高质量又保持在搜索空间的分布广泛,再将剩余个体划分到距离最近的中心个体构成多种群,随后依据概率来考虑是否接受种群状态变化。其次,各子群通过随机顺序动态构成环拓扑进行个体迁移,以增强种群多样性避免陷入局部最优。最后,通过改进局部搜索策略,以完善对解空间的探索。选用CEC2017测试函数集中的12个函数作为性能基准函数,将DMEFPA和其他5个改进算法:SCFPA、HLFPA、WOFPA、AMSSA、SHSSA进行评测,并对改进策略进行了消融实验。基于实验结果的Friedman检验表明,在改进策略的共同作用下,DMEFPA能获取最优的性能,且全局收敛性能较为稳定。Aiming at overcoming drawbacks of lower accuracy and being trapped easily to local optimums of the flower pollination algorithm(FPA),this paper proposed an enhanced flower pollination algorithm based on dynamic multi-subgroup mechanism(DMEFPA).Firstly,DMEFPA selected the central individuals using a method that combined individual fitness values and relative distances,ensuring that the selected individuals maintain high quality while being widely distributed in the search space.Then,it assigned the remaining individuals to the closest central individuals to form multiple subpopulations,and considered the acceptance of population state changes based on probability.Secondly,each subpopulation dynamically formed a ring topology in a random order for individual migration to enhance population diversity and avoid local optima.Finally,the algorithm improved local search strategies to refine the exploration of the solution space.12 functions from CEC2017 test suit were selected as the benchmark to evaluate the performance of DMEFPA and other 5 algorithms:SCFPA,HLFPA,WOFPA,AMSSA and SHSSA.Friedman test based on experimental results show that DMEFPA can achieve the supreme performance.Ablation experiments were also conducted to verify effectiveness of proposed improvement strategies.Experimental results illustrate that DMEFPA can achieve the outstanding performance and with stable convergence under the joint of all improvement strategies.

关 键 词：花授粉算法 多种群 动态拓扑 个体迁移 局部搜索策略 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]