Evolutionary multiobjective optimization with clustering-based self-adaptive mating restriction strategy

Abstract

Mating restriction plays a key role in MOEAs, while clustering is an effective method to discover the similarities between individuals and therefore can assist the mating restriction. What is more, it is inappropriate to set the same mating restriction strategy for all individuals as solutions are very different between clusters. This paper proposes a multiobjective evolutionary algorithm with clustering-based self-adaptive mating restriction strategy (SRMMEA). In SRMMEA, k-means algorithm is used to cluster the population. With a certain probability, mating parents are selected from the clusters or the whole population for exploitation and exploration, respectively. To better balance the exploration and exploitation, different mating restriction probabilities are assigned to solutions in different clusters. Moreover, the mating restriction probability is updated at each generation according to the number of newly generated individuals in each cluster. SRMMEA is compared with some state-of-the-art multiobjective evolutionary methods on a number of test instances. Experimental results demonstrate SRMMEA’s superiority over other comparison algorithms.