A Dynamic Multi-Method Evolutionary Algorithm for Multi-Objective Optimization with Learning Strategy

Multi-Objective evolutionary algorithms (MOEAs) are popular for solving complex optimization problems with multiple objectives and they have gained much success in benchmark competitions as well as real-world applications. Most MOEAs are effective for specific types of problems, leading to a lack of generality. This paper presents a dynamic MOEA employing a learning strategy to adaptively allocate the appropriate optimizer for the given problem. Three key issues are considered by the designed learning strategy, including environmental information collection, information-reward conversion, and dynamic method selection. In the dynamic selection process, a set of candidate algorithms with various features is required, which improves the universality of our method. Extensive experimental comparisons have been done based on the benchmark functions WFG1–WFG9, as well as a real-world problem, which shows that the overall performance of the proposed method is superior to the state-of-the-art peer algorithms.