基于ELECTRE-III法的高维多目标调和进化算法
2020-09-02易高明耿秀荣
易高明 耿秀荣



摘 要:针对基于Pareto支配的低维多目标进化算法在求解3个以上目标的高维多目标时出现收敛压力不足等问题,将调和模型中面向排序的ELECTRE-III引入高维多目标进化方法中,提出一种新的锦标赛选择算子。该算子包含两层操作,分别是快速非支配分层操作和同一非劣层中的赋值级别高于关系排序操作。将这种赋值级别高于关系构造的ELECRE-III排序法嵌入NSGA-II中并应用于高维多目标进化个体的优劣排序。对典型高维测试集WFG函数进行仿真实验,验证该高维多目标调和进化算法的有效性。
关键词:ELECTRE III方法;多目标进化算法;高维多目标;锦标赛选择;赋值级别高于关系
DOI:10. 11907/rjdk. 201488 开放科学(资源服务)标识码(OSID):
中图分类号:TP312文献标识码:A 文章编号:1672-7800(2020)008-0089-06
Abstract: Aiming at the insufficient convergence pressure of low-dimensional and multi-objective evolutionary algorithms based on Pareto domination in solving three or more high-dimensional and multi-objective evolutionary algorithms, we introduce the ranking-oriented ELECTRE-III in the harmonic model into the high-dimensional and multi-objective evolutionary methods, and propose a new tournament selection operator. The operator consists of two layers of operations. The assignment level of fast non-dominated hierarchical operation and the same non-inferior layer is higher than that of relational sorting operation. The ELECRE-III ranking method with higher assignment level than relation construction is embedded in NSGA-II and applied to the ranking of evolutionary individuals in high-dimensional multi-objective problems. The simulation results of typical high-dimensional test set WFG function verify the effectiveness of the proposed harmonic evolutionary algorithm.
Key Words: ELECTRE III method;multi-objective evolution algorithm;many-objectives;competition selection;valued outranking relation
0 引言
多目標进化算法可有效求解2~3个目标优化问题,然而当目标个数超过3时,这类问题就变成高维多目标问题[1]。传统基于Pareto占优机制的多目标进化优化方法选择非支配解的压力极大降低,导致逼近真实Pareto前沿所需进化个体数量呈指数级增加。因此,基于绝对Pareto机制占优的多目标进化方法难以有效解决高维多目标问题[2]。
为了解决高维多目标优化问题,Ahmed [3]提出格支配占优的高维多目标进化算法,这是一类非常典型的细粒度Pareto占优方法。同时,研究者们开始关注第二级MOEs选择算子,即多样性保持算子,寄希望于在这个第二级算子上实施选择压力。Deb等[4]在NSGA-II基础上提出其改进版本NSGA-III,替换了NSGA-II中的拥挤距离算子;Ren等[5]提出对……
