Surrogate-assisted Automatic Niching PSO for Expensive Multimodal Optimization Problems

Addressing expensive multimodal optimization problems (EMMOPs) faces two challenges: costly objective function evaluation and difficult multi-optima localization. Current EMMOP-focused studies remain limited. This paper proposes a Surrogate-Assisted Automatic Niching Particle Swarm Optimization (SANPSO)to address these hurdles. First, mean-shift clustering divides the entire population without requiring any parameters. The subswarm is then guided to explore the entire space using a surrogate-assisted dimensional learning strategy. Furthermore, the model management strategy selects the promising particles, which are then precisely evaluated. Finally, for each promising subswarm, a local search based on trust region is used to improve the solution accuracy even further. The results obtained from simulations demonstrate that the proposed SANPSO algorithm can more efficiently deal with EMMOPs when compared to other multimodal algorithms.