Enhanced goal direction and remodeling of population distributions multimodal multi-objective evolutionary algorithm
Hangyu Liu, Shaobo Deng, Keyun Li, Hui Shi, Min Li · Expert Systems with Applications · 2025
In recent years, multimodal multi-objective optimization problems (MMOPs) have become a prominent research focus in the field of computational evolution, increasing the demands on the performance of multimodal multi-objective evolutionary algorithms (MMOEAs). To evaluate the merits of MMOEA, it is usually necessary to satisfy the following three key criteria: (1) good convergence, (2) the ability to find more equivalent Pareto solutions (PSs), and (3) a uniform population distribution in the decision space and objective space. However, most current algorithms fail to satisfy all the above criteria simultaneously when facing challenges such as search tasks of varying difficulty and uneven allocation of computational resources. To address these challenges, this paper proposes an enhanced goal direction and remodeling of population distributions multimodal multi-objective evolutionary algorithm (MMOEA-EGR). The algorithm dynamically selects evolutionary stages through reinforcement learning, flexibly guiding the population to evolve under the guidance of the objectives of each stage, thus promoting efficient collaboration among the stages in the whole optimization process. Meanwhile, the algorithm adopts a remodeling population distribution strategy to enhance the evolutionary efficiency while optimizing the diversity of the decision space. Experimental results show that MMOEA-EGR outperforms several mainstream multimodal multi-objective evolutionary algorithms on several MMOPs standard test sets.