基于重构邻域策略的分解多目标进化算法  

Decomposition Multi-objective Evolutionary Algorithm Based on Rrefactoring Neighborhood Strategy

作　　者：曹鸽 郭海东[1,2] 王丽萍[2,3] 徐梦娜 CAO Ge;GUO Hai-dong;WANG Li-ping;XU Meng-na(College of Business Administration,Zhejiang University of Technology,Hangzhou 310023,China;Institute for Information Intelligence and Decision Optimization,Zhejiang University of Technology,Hangzhou 310023,China;College of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 310023,China)

机构地区：[1]浙江工业大学经贸管理学院,杭州310023 [2]浙江工业大学信息智能与决策优化研究所,杭州310023 [3]浙江工业大学计算机科学与技术学院,杭州310023

出　　处：《小型微型计算机系统》2020年第3期477-484,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61472366,61379077)资助;浙江省自然科学基金项目(LZ13F020002,LY17F020022)资助.

摘　　要：MOEA/D算法的每个子问题都从邻域中选取父代解进行交叉变异,邻域结构在整个进化过程中维持不变,在一定程度上限制了父代选择的范围,算法在搜索后期会出现种群退化、收敛速度放缓等问题.为解决以上问题,本文提出了重构邻域策略改善父代解集质量来提升种群逼近Pareto前沿面的速度.改进算法MOEA/D-RNS改变了原始框架下的父代选择方式,从子问题邻域和精英解组成的新邻域集合中选择父代.其中,精英解分两步筛选确定,利用更新策略筛选出预备精英解集,再通过计算每个预备精英的潜力值,并根据潜力值排序来选择精英解集.在不改变邻域规模情况下,改善了父代解集质量,同时提升了解集的多样性,加快了种群收敛速度.在2至10维DTLZ1-4系列测试问题上对算法进行性能测试,实验结果表明,本算法能有效平衡算法收敛性与多样性.For each subproblems of MOEA/D,the parent solutions are selected from the neighborhood for reproducting,the neighborhood structure remains unchanged throughout the evolution process. To a certain extent,the scope of mating selection is limited,and the algorithm will have problems such as population degradation and slow convergence in the later stage of the search. In order to solve the above problems,this paper proposes a refactoring neighborhood strategy to promote the quality of the parent solutions to improve speed of the population approaching the Pareto frontier. MOEA/D-RNS changes the mating selection method in the original framework,and selects parent solutions from the new neighborhood set consisting of the subproblem neighborhood and the elite solutions set,among them,the elite solutions are determined by two-step screening,and the updated strategy is used to screen out the preliminary elite solution set,and then the Potentiality value of each reserve elites is calculated,the elite solutions sets are selected according to the Potentiality value order. Without changing the neighborhood size,the quality of the parents solution set is improved. Furthermore,the diversity of the set is improved and the speed of population convergence is accelerated. The performance of the algorithm is tested on DTLZ series benchmarks with 2-10 objectives,the experimental results show that the proposed algorithm can effectively balance the convergence and diversity of the algorithm.

关 键 词：多目标优化 重构邻域策略 父代选择 潜力值 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]