基于IGD^(+)指标的两阶段选择高维多目标进化算法  

IGD^(+)indicator based many-objective evolutionary algorithm with two stage selection

作　　者：张伟[1] 刘建昌[1] 刘圆超 郑恬子 杨婉婷 ZHANG Wei;LIU Jian-changy;LIU Yuan-chao;ZHENG Tian-zi;YANG Wan-ting(College of Information Science and Engineering,Northestern University,Shenyang Liaoning 110819,China)

机构地区：[1]东北大学信息科学与工程学院,辽宁沈阳110819

出　　处：《控制理论与应用》2023年第5期801-816,共16页Control Theory & Applications

基　　金：国家自然科学基金项目(61773106)资助。

摘　　要：针对在高维空间下多目标进化算法难以维持种群收敛性和多样性平衡的问题,本文提出一个基于IGD^(+)指标的两阶段选择高维多目标进化算法(MaOEA–ITS).在第1阶段,算法基于IGD^(+)指标选择收敛性良好的精英个体,其所需的参考点通过引入切割平面截距法构建.在第2阶段,MaOEA–ITS使用模糊c均值算法对参考向量进行聚类,聚类后的参考向量引导种群分解策略对剩余个体进行环境选择,从而维持种群的多样性.另外,为了保护能够提高种群多样性的极值解,本文提出一个参考点分布自适应策略.最后,通过仿真实验来验证MaOEA–ITS的有效性和优越性.In order to solve the problem that the multi-objective evolutionary algorithm is difficult to balance between the population convergence and diversity in high-dimensional space,in this paper,an IGD^(+)indicator based many-objective evolutionary algorithm with two stage selection(MaOEA–ITS)is proposed.In the primary stage,the proposed algorithm adopts IGD^(+)indicator as selection criterion to select individuals with favourable convergence,and the reference points required are constructed by introducing the intercepts way of cutting plane.In the second stage,the MaOEA–ITS uses a fuzzy c-means algorithm to cluster reference vectors.Then,reference vectors clustered guide population partition strategy to select remaining individuals of population,thereby maintaining the population diversity.In addition,for protecting the extreme solutions that enables to improve population diversity,a reference point distribution based on the adaptive strategy is proposed.Finally,simulation experiments are used to verify the effectiveness and superiority of MaOEA–ITS.

关 键 词：高维多目标优化 IGD^(+)指标 两阶段选择策略 参考点分布自适应策略 种群分解策略 进化算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]