Parallel multi-strategy evolutionary algorithm using massage passing interface for many-objective optimization

Abstract

Evolutionary multi-objective optimization (EMO) algorithms have become prevalent and obtained a great success for solving two- or three-objective problems. However, with the number of objectives increases, most of the algorithms cannot perform well due to the expansion of the objective space. Therefore, there is an urgent need for improving EMO algorithms to handle many-objective (four or more objectives) optimization problems (MaOPs). To this end, this paper proposes a parallel multi-strategy evolutionary algorithm (PMEA) to make full use of the advantages of different selection strategies. Specially, PMEA maintains three populations in parallel to select individuals based on three strategies as decomposition-based approach, indicator-based approach, and shift-based density estimation approach. PMEA uses message passing interface (MPI) to share the information after the selection of the three strategies, so that the advantages of diverse approaches can be utilized. In this way, PMEA can explore the objective space more thoroughly and thus achieve more promising performance. We evaluated PMEA on two frequently used MaOP suites and compared the results with several state-of-the-art many-objective peer algorithms. Numerical results demonstrate that PMEA can achieve a statistically superior performance, or at least highly competitive performance on most of the problems instances. © 2016 IEEE.