Differential Evolution with Cluster-Based External Archive and Local Search for Multimodal Optimization
Gabriel Dominico, Mateus Boiani, Rafael Stubs Parpinelli · 2019
One of the main objectives in multimodal optimization is to find multiple optima solutions in a search space. Hence, population-based metaheuristics are suitable for this class of problems but their loss of diversity while converging may become a problem when tracking multiple optima. In this paper, we propose the use of a cluster-based external archive maintenance strategy along with the jDE algorithm, namely NCjDE-HJar. The DBSCAN algorithm is employed to group candidate solutions in an external archive representing multiple peaks that will feed the Hooke-Jeeves local search algorithm. Also, the Michalewicz mutation strategy is applied to refine the solutions found by the jDE algorithm. The proposed approach is compared with five state-of-the-art algorithms in terms of peak ratio and the results obtained show that the proposed modifications favor the finding more multimodal peaks.