Large-Scale Multi-Objective Evolutionary Algorithm Based on Multi-Group Co-Evolution

Multi-objective evolutionary algorithms currently face scalability challenges and struggle to maintain a balance between convergence and distribution in addressing large-scale multi-objective optimization problems. To address this limitation and more comprehensively explore the exponentially growing decision space, we introduce a large-scale multi-objective evolutionary algorithm based on multi-group co-evolution. This algorithm initiates by creating two independent subpopulations. The first subpopulation employs a multi-strategy, multi-objective evolutionary algorithm grounded in dominance relationships to preserve diversity within the decision space. Simultaneously, the second subpopulation utilizes a decomposition-based multi-objective evolutionary algorithm with virtual target vectors to facilitate a rapid convergence toward the target solution. As the populations progress to a specific stage, they are merged, and outstanding individuals are selected based on an enhanced hypervolume parameter. Subsequently, the populations undergo multiple loops with a reset based on predefined rules. Throughout different evolutionary stages, the sizes of the two populations dynamically adjust to maintain a balance between population convergence and distribution. To validate the effectiveness of our proposed algorithm, we compare it with five advanced large-scale multi-objective evolutionary algorithms and evaluate its performance on benchmark test problems LSMOPI to LSMOP9. The experimental results clearly demonstrate its superiority in solving large-scale multi-objective problems.