Automatic design of hybrid optimizers based on particle swarm optimization and differential evolution

Hybridization is termed as an efficient strategy to improve the performance of many optimizers, and the design of efficient hybrids for difficult optimization problems remains a challenging field. In this study, an automatic method is proposed to generate various hybrid optimizers based on particle swarm optimization and differential evolution for a given problem. Based on the unified framework for hybridization proposed in our recent work, the proposed method can provide plenty of hybrid optimizers for a specific problem with given parent optimizers. Besides, an approach which can provide the unified representation of different kinds of hybridization levels is developed. Moreover, all the generated hybrids are evaluated on the CEC2017 benchmark problems, and the results of the statistic analysis and the Wilcoxon's rank-sum tests demonstrate that the hybrid optimizer utilizing the hybridization strategy <C, P, P, D >performs obviously better than the others on the test functions.