A self-adaptive differential evolution with local search applied to multimodal optimization

Abstract

The main difficulty encountered by population-based approaches in multimodal problems is their loss of diversity while converging to an optimum. Also, it is known that parameters play a big role in the performance of metaheuristics. Hence, in this paper two variations of the NCDE algorithm for multimodal optimization are proposed. The first version applies the jDE self-adaptive mechanism for parameter tuning along with the neighborhood mutation and crowding strategies, called NCjDE. The second version adds to the first the Hooke-Jeeves direct search at the end of the optimization process, called NCjDE-HJ. The proposed algorithm is compared in terms of peak ratio with three other state-of-the-art algorithms and results obtained show that the proposed variations are competitive for multimodal problems.