一种融合邻域搜索的多策略差分进化算法  

A Multi-strategy Differential Evolution Algorithm Combined with Neighborhood Search

作　　者：孙灿 周新宇[1] 王明文[1] Sun Can;Zhou Xinyu;Wang Mingwen(School of Computer and Information Engineering,Jiangxi Normal University,Nanchang 330022,China)

机构地区：[1]江西师范大学计算机信息工程学院,江西南昌330022

出　　处：《系统仿真学报》2020年第6期1071-1084,共14页Journal of System Simulation

基　　金：国家自然科学基金(61603163,61966019,61876074);江西省自然科学基金(20192BAB207030)。

摘　　要：设计多策略差分进化算法的难点在于选择何种变异策略以及如何分配这些策略。提出一种融合邻域搜索的多策略差分进化算法,根据个体适应度值将种群分为3个子种群,每个子种群分别采用不同的变异策略和参数值,使得各子种群的搜索能力可互补,有助于平衡整个种群的勘探和开采能力。同时,对适应度值最好的子种群采用邻域搜索操作,充分挖掘优质个体可能包含的有益信息用于指导搜索。在34个测试函数上实验,与包含7种差分进化算法在内的12种进化算法进行对比,结果表明该算法在大多数函数上取得了更好性能。The difficulties of designing a multi-strategy differential evolution(DE)algorithm are how to select the mutation strategies and allocate these strategies.A multi-strategy DE algorithm combined with the neighborhood search operator is proposed.The population is divided into three subpopulations according to the fitness values,and each subpopulation employs a different mutation strategy and parameter settings to complement the search ability,to balance the exploration and exploitation ability of the whole population.The subpopulation with the best fitness values employs the neighborhood search operator to exploit possible benefit information to guide the search.Extensive experiments are carried out on 34 test functions to compare with 12 different evolutionary algorithms,which include the 7 DE algorithms.The results show that the algorithm can perform better on most test functions.

关 键 词：差分进化 多策略 邻域搜索 勘探能力 开采能力 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]