A dual-population paradigm for evolutionary multiobjective optimization

Abstract

Convergence and diversity are two basic issues in evolutionary multiobjective optimization (EMO). However, it is far from trivial to address them simultaneously, especially when tackling problems with complicated Pareto-optimal sets. This paper presents a dual-population paradigm (DPP) that uses two separate and co-evolving populations to deal with convergence and diversity simultaneously. These two populations are respectively maintained by Pareto- and decomposition-based techniques, which arguably have complementary effects in selection. In particular, the so called Pareto-based archive is assumed to maintain a population with competitive selection pressure towards the Pareto-optimal front, while the so called decomposition-based archive is assumed to preserve a population with satisfied diversity in the objective space. In addition, we develop a restricted mating selection mechanism to coordinate the interaction between these two populations. DPP paves an avenue to integrate Pareto- and decomposition-based techniques in a single paradigm. A series of comprehensive experiments is conducted on seventeen benchmark problems with distinct characteristics and complicated Pareto-optimal sets. Empirical results fully demonstrate the effectiveness and competitiveness of the proposed algorithm.