A diversity enhanced multiobjective particle swarm optimization

Abstract Multiobjective particle swarm optimizations (MOPSOs) are confronted with convergence difficulty as well as diversity deviation, due to combined learning orientations and premature phenomenons. Numerous adaptations of MOPSO have been introduced around the elite definition and leader selection in previous studies. Meanwhile the unique leader-oriented updating which reflects some properties of evolving, may provide control assistance under particular conditions. However, repetition and inefficient works on leader determination exist, and seldomly have studies taken PSO’s evolve rhythms into consideration to adjust the optimize strategy adaptively. In view of the above problems, and aim to balance the convergence and diversity during searching procedure, a novel diversity enhanced multiobjective particle swarm optimization (DEMPSO) is proposed in this paper. The novel method mainly focuses on the following innovations. First, a simplified leader-oriented formulation in PSO updating is introduced. Second, through taking full advantages of the PSO learning mechanism and extracting particles velocity information, novel intersection measurement for elite definition and novel decision variable analysis method for diversity enhancement are proposed. Third, an adaptive two-fold leader selection strategy is presented. The experimental results on benchmark test instances illustrate that DEMPSO outperforms other PSO-cored algorithms, and greatly improves the diversity maintain ability in high-dimensional objective spaces in comparison with some state-of-the-art decomposition-based and dominated-based evolutionary algorithms.