基于参考点拥挤度改进的NSGAⅢ算法  

Improved NSGAⅢ algorithm based on congestion of reference points

作　　者：庞善天 陈基漓[1] 谢晓兰[1] PANG Shan-tian;CHEN Ji-li;XIE Xiao-lan(College of Information Science and Engineering,Guilin University of Technology,Guilin 541004,China)

机构地区：[1]桂林理工大学信息科学与工程学院

出　　处：《计算机工程与设计》2019年第6期1626-1633,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61762031)

摘　　要：当NSGAIII算法中的参考点小生境数比较小时,与该参考点相关联的所有成员都同时进入下一代,未考虑拥挤程度,降低了种群的多样性,减缓了种群的收敛速度。针对这一问题,提出一种个体选择策略,每次从参考点集合选取个体进入下一代之前,先计算集合中种群成员相互之间的拥挤度,当拥挤距离小于所设定阈值时,随机删除其中任意一个个体,使其不参与下一代的进化。在PlatEMO平台上与原算法进行实验对比,实验结果表明,改进算法使Pareto解集的收敛性和分布性方面得到了改善。When the Niche number of reference poins in the NSGAIII arithmetic is small, all the members associated with the re- ference point will enter next generation at the same time without considering the degree of congestion, the diversity of population is then reduced and the convergence rate of population is slowed down. To solve this problem, a strategy of individual selection was proposed. Before selecting individuals from the reference point set to enter the next generation, the crowding degree among the members of the population in the set was calculated, and when the crowding distance was under the setting threshold, any one of these individuals was randomly deleted, avoiding it participating in the next generation evolution. Comparing with the original algorithm on PlatEMO platform, experimental results show that the improved algorithm improves the convergence and distribution of the Pareto solution set.

关 键 词：多目标优化 第三代非支配排序遗传算法 参考点 小生境数 拥挤度 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]