A Multiobjective Evolutionary Algorithm Based on Decomposition and Preselection

The preselection aims to choose promising offspring solutions from a candidate set in evolutionary algorithms. Usually the preselection process is based on the real or estimated objective values, which might be expensive. It is arguable that the preselection is doing classification in nature, which requires to know a solution is good or not instead of knowing how good it is. In this paper we apply a classification based preselection (CPS) to a multiobjective evolutionary algorithm based on decomposition (MOEA/D). In each generation, a set of candidate solutions are generated for each subproblem and only a good one is chosen as the offspring by the CPS. The modified MOEA/D, denoted as MOEA/D-CPS, is applied to a set of test instances, and the experimental results suggest that the CPS can successfully improve the performance of MOEA/D.