Ensemble crowding differential evolution with neighborhood mutation for multimodal optimization

Many optimization problems possess multiple global solutions or comparably fit local solutions. These multimodal optimization problems require the identification of not just one global optimal, but also multiple compatible solutions. Differential Evolution (DE) has been demonstrated to be highly effective for solving single-objective unimodal problems, but its loss of diversity over the course of evolution prevents it from locating multiple compatible solutions. Our proposed method combines the diversity maintenance of niching and neighborhood mutation techniques with the versatility of ensemble parameters for DE to enhance the exploitation of individual peaks on difficult multi-modal problems. Greedy local mutation strategy and crossover are shown to have improved the performance of the neighborhood crowding DE (NCDE) in our experiment with 14 common multimodal benchmark functions.