Differential evolution with adaptive subpopulation size for multimodal optimization problems

Multimodal optimization problems arise commonly in real-world applications. The main difficulty is to find all the optimal solutions. Niching techniques are considered as the effective methods for locating multiple optimal solutions. However, niching-based algorithms introduce additional parameters and are sensitive to parameters. To solve this problem, a niching technique based on adaptive subpopulation size is proposed. In addition, due to the different sizes and shapes of the basins of attraction, the convergence situation of algorithms is different in different search regions. An optimal solution recognition technique is proposed, which is used to remove the subpopulations that have converged to the optimal solutions. Experimental results show the proposed algorithm has a superior or at least competitive performance.