A multi-stage competitive swarm optimization algorithm for solving large-scale multi-objective optimization problems

Hundreds or thousands of decision variables are involved in large-scale multi-objective optimization problems (LSMOPs), which may include scheduling and artificial intelligence. Solving LSMOPs presents formidable obstacles due to the exponential expansion of search volume for solutions and the catastrophic expansion of local optimum during the evaluation process, which are attributed to the increasing count of decision variables. This article presents a two-stage competitive swarm optimization algorithm to tackle LSMOPs. In the first stage, the proposed method designs a fuzzy search strategy for loser particles and an adaptive dual-directional sampling strategy for winner particles to efficiently explore the entire space. During the subsequent phase, a novel update learning tactic is developed for the loser particles, integrating the global optimum to direct the updated trajectory of the loser particles and facilitate faster algorithm convergence. To validate the said method's efficacy, extensive empirical studies utilizing the LSMOPs benchmark problems were undertaken to compare it with five contemporary algorithms. According to the outcomes, the method surpasses the compared algorithms regarding HV and IGD alike.