Dynamic Crowding Distance?A New Diversity Maintenance Strategy for MOEAs

Abstract

In multi-objective evolutionary algorithms (MOEAs), the diversity of Pareto front (PF) is significant. For good diversity can provide more reasonable choices to decision-makers. The diversity of PF includes the span and the uniformity. In this paper, we proposed a dynamic crowding distance (DCD) based diversity maintenance strategy (DMS) (DCD-DMS), in which individualpsilas DCD are computed based on the difference degree between the crowding distances of different objectives. The proposed strategy computes individualspsila DCD dynamically during the process of population maintenance. Through experiments on 9 test problems, the results demonstrate that DCD can improve diversity at a high level compared with two popular MOEAs: NSGA-II and epsiv-MOEA.