摘要
针对传统的基于严格Pareto支配的多目标进化算法在计算3个以上目标的高维问题时出现收敛性不足的问题,以淘汰选择法ELECTRE-Ⅰ为构造基础,提出了一个改进的优势个体比较选择算子。该锦标赛规则可以有效克服Pareto选择压力不足的问题,与一般的锦标赛选择算子不同,该算子借助于面向最小优势子集选择的ELECTRE-Ⅰ法来找到同一层级非劣集中的优势个体,从而对每一层的非支配进化个体生成一个最小优势子集,进而给出进化个体中的优劣排序新规则,最后在典型高维测试集DTLZ函数上进行仿真实验。实验结果表明,该算法比经典多目标进化算法NSGA-Ⅱ、NSGA-Ⅲ、α-NSGA-Ⅱ和HV-NSGA-Ⅱ在求解某些特定问题集上具有更好的多样性与收敛性。
Aiming at the problem of insufficient convergence of the traditional multi-objective evolutionary algorithm based on strict Pareto domination in calculating the optimal solution of high-dimensional problems with more than three objectives, an improved dominant individual comparative selection operator is proposed based on the elimination selection method ELECTRE-Ⅰ.The championship rule can effectively overcome the problem of insufficient Pareto selection pressure.Different from the general tournament selection operator, the operator uses the ELECTRE-Ⅰ method for the selection of the minimum dominant subset to find the dominant individuals in the non inferior set at the same level, so as to generate a minimum dominant subset for the non dominated evolutionary individuals at each level, and then give the new rules for ranking the advantages and disadvantages of the evolutionary individuals.Finally, the simulation experiment is carried out on the DTLZ function of the typical high-dimensional test set.Experimental results show that the algorithm has better diversity and convergence than the classical multi-objective evolutionary algorithms NSGA-Ⅱ,NSGA-Ⅲ,-NSGA-Ⅱ and HV-NSGA-Ⅱ.
作者
易高明
张永闯
施武祖
赵彬
YI Gaoming;ZHANG Yongchuang;SHI Wuzu;ZHAO Bin(School of Science,Guilin University of Aerospace Technology,Guilin Guangxi 541000,China;College of Business,Guilin University of Aerospace Technology,Guilin Guangxi 541000,China;School of Computer Science&Artificial Intelligence,Lanzhou Institute of Technology,Lanzhou 730050,China;College of Computer and Information Engineering,Tianjin Normol University,Tianjin 300380,China)
出处
《兰州工业学院学报》
2022年第4期78-84,共7页
Journal of Lanzhou Institute of Technology
基金
2020年桂林航天工业学院教学团队建设项目基金(2020JXTD16)
2022年桂林航天工业学院科研基金(XJ22KT17)。
关键词
多目标进化
高维问题
淘汰选择法
锦标赛规则
最小优势子集
multi-objective evolution
high dimensional problems
elimination selection method
tournament rules
minimum dominant subset
作者简介
易高明(1991-),男,安徽宣城人,助教,硕士;通讯作者:施武祖(1983-),男,甘肃永登人,工程师,助教。
