Adaptive candidate estimation-assisted multi-objective particle swarm optimization

The selection of global best (Gbest) exerts a high influence on the searching performance of multi-objective particle swarm optimization algorithm (MOPSO). The candidates of MOPSO in external archive are always estimated to select Gbest. However, in most estimation methods, the candidates are considered as the Gbest in a fixed way, which is difficult to adapt to varying evolutionary requirements for balance between convergence and diversity of MOPSO. To deal with this problem, an adaptive candidate estimation-assisted MOPSO (ACE-MOPSO) is proposed in this paper. First, the evolutionary state information, including both the global dominance information and global distribution information of non-dominated solutions, is introduced to describe the evolutionary states to extract the evolutionary requirements. Second, an adaptive candidate estimation method, based on two evaluation distances, is developed to select the excellent leader for balancing convergence and diversity during the dynamic evolutionary process. Third, a leader mutation strategy, using the elite local search (ELS), is devised to select Gbest to improve the searching ability of ACE-MOPSO. Fourth, the convergence analysis is given to prove the theoretical validity of ACE-MOPSO. Finally, this proposed algorithm is compared with popular algorithms on twenty-four benchmark functions. The results demonstrate that ACE-MOPSO has advanced performance in both convergence and diversity.