Multimodal Optimization by Evolution Strategies with Repelling Subpopulations

This work presents a niching method based on the concept of repelling subpopulations for multimodal optimization. It utilizes several existing concepts and techniques in order to develop a new multimodal optimization algorithm that does not make any of specific assumptions on the shape, size, and distribution of minima. In the proposed method, several subpopulations explore the search space in parallel. Offspring of weaker subpopulations are forced to maintain a distance from the fitter subpopulations and the previously identified niches. This defines taboo regions to hinder the exploration of the same regions of the search space and previously identified niches. The size of each taboo region is adapted independently so that the method can handle challenges of minima with dissimilar basin sizes and irregular distribution. The local shape of a basin is approximated by the distribution of the subpopulation members converging to that basin. The proposed niching strategy is incorporated into the state-of-the-art evolution strategies, and the resulting method is compared with some of the most successful multimodal optimization methods on composite test problems. A comparison of numerical results demonstrates the superiority of our proposed method.