MOEA/D with the online agglomerative clustering based self-adaptive mating restriction strategy

Abstract

In MOEA/D-DE, the appropriate value of the mating restriction probability varies with the evolutionary process. Furthermore, different subproblems have been solved in different degree during the evolution, so different subproblems have distinct requirements for exploitation and exploration. Additionally, MOEA/D-DE defines the neighborhood according to the distance between the weight vectors. However, the individuals corresponding to the neighbor subproblems may distribute far away in the decision space, which will affect the performance of exploitation. Accordingly, this paper proposes a MOEA/D with the online agglomerative clustering based self-adaptive mating restriction strategy (MOEA/D-OMR). MOEA/D-OMR utilizes the online agglomerative clustering algorithm to extract the neighborhood information in the decision space. The mating pool is then constructed by the neighbor population or the whole population based on the mating restriction probability. What is more, a separate mating restriction probability is assigned to each subproblem. The mating restriction probability is updated at each generation by the survival length, which is the number of generations that the solution has survived over the last certain period of time. Experimental results show that MOEA/D-OMR has a better performance than the comparison algorithms.