一种新角度支配关系的多目标进化算法

Multi-Objective Evolutionary Algorithm based ona Novel Angle Dominance Relation

为了兼顾多目标进化算法求解的收敛性和多样性,文中在分析解群支配以及多样性情况相关理论的基础上,借鉴网格支配和角度支配的优势,提出一种新的支配关系刻画解群在目标空间的分布状况,以较好地在收敛性和多样性间取得平衡;设计了一种新角度支配的多目标进化算法,该算法利用新角度支配关系增强选择压力,维持解群的多样性;比较了在NSGAII算法框架下,新提出的支配关系与原有两种支配关系的收敛性、多样性以及收敛速度。研究结果表明:文中算法在DTLZ及WFG基准测试集上获得的GD,IGD和Spacing指标性能均得到了提升,并且能够更快地获得高质量解集,故所提出的支配关系能够更有效地平衡收敛性和多样性。

The paper studies how to achieve the balance between the convergence and diversity of multi-objective evolutionary algorithms.Based on the theories of evolutionary populations’dominance relation and diversity,and drawing on the advantages of grid dominance relation and angle dominance relation,this paper presents a novel dominance relation to describe the distribution of evolutionary populations in the objective space,aiming at achieving a better balance between the convergence and diversity.A multi-objective evolutionary algorithm based on a novel angle dominance is developed.By using the novel angle dominance relation,it can improve the selection pressure and maintain the diversity of evolutionary populations.In the framework of NSGAII algorithm,the convergence,diversity and convergence rate of the newly proposed dominance relation are compared with those of the existing two dominance relations.The results show that the performance of the IGD,GD and Spacing obtained by the novel dominance relation on the DTLZ and WFG test suites has been improved,with high-quality solution sets obtained more quickly.It is concluded that the dominance relation proposed in this paper can balance convergence and diversity more effectively.