Feature-based Coevolution Algorithm for Multimodal Multi-objective Optimization

Abstract

When solving multimodal multi-objective optimization problems (MMOPs), it is important to maintain the diversity in the decision space. Since traditional Pareto-dominance-based multimodal multi-objective evolutionary algorithms (MMEAs) prioritize the convergence of individuals through Pareto dominated sorting, several well-distributed individuals might be dominated by other well-converged individuals during the optimization process of MMOPs. In order to solve this problem, we propose a coevolutionary algorithm that extracts and utilizes the features of multiple populations to preserve diversity. The proposed algorithm simultaneously explores different regions of the decision space through multi-population coevolution and controls the cross-learning behavior among sub-populations. The cross-learning is conducted through a feature weight vector which maintain the characteristics of multiple populations and preserve the diversity of the decision space. Meanwhile, the assignment and merging methods are designed to adaptively adjust the scales of sub-populations to speed up the convergence. The experimental results show that the introduction of the feature weight vector is effective in maintaining the distribution of the decision space, and the proposed algorithm is competitive with other representative MMEAs.