Dual-space History Information-driven Dynamic Multimodal Evolutionary Algorithm

When solving dynamic multimodal optimization problems (DMMOPs) with evolutionary algorithms, researchers often respond to environment change by designing dynamic response mechanisms, and the use of information from the historical environment greatly affects the performance of the mechanisms. However, existing research tends to find connections between environments only from the perspective of decision space or objective space, which leads to underutilization of historical information. To solve these issues, a dual-space history information-driven dynamic multimodal evolutionary algorithm is proposed in this paper. Specifically, the algorithm identifies the similar historical environments for the new environment based on the similarity between the decision space and the objective space simultaneously, and migrates the approximate global optimal solutions found in these similar environments to the new environment. These individuals are utilized to generate a partial initial population, which is focused on convergence. Furthermore, the algorithm improves the performance of the random reinitialization strategy based on the information in the most neighboring environment. The excellent individuals from that environment are utilized to assist in generating the remaining initial individuals, which balances the diversity and convergence of the population. The effectiveness of proposed algorithm is demonstrated on 24 benchmark DMMOPs compared with five state-of-the-art methods.