A Method to Balance the Exploratory and Exploitative Capabilities of Decomposition-based Multi-objective Evolutionary Algorithm
Minh Tran Binh, Long Nguyen, Kien Thai Trung · 2023
In the multi-objective evolutionary algorithms (MOEAs) research area, well-designed algorithms strive to achieve the final population or solution set that satisfies both convergence and diversity. To achieve this goal, powerful algorithms aim to ensure a balance between depth and breadth search in the search progress. However, maintaining an equilibrium between the exploratory and exploitative ability of the evolutionary process is a particularly challenging issue in MOEAs. Our approach involves assessing the search process's trajectory by analyzing how convergence and diversity change over time. Based on this evaluation, seek techniques to adjust the evolution to balance convergence and diversity. By doing so, we enhance the exploration and exploitation capabilities of the evolutionary process. This research applies the approach above to improve the balance between the exploration and the exploitation of decomposition-based MOEAs. We combined variation correlation information of some convergence and diversity measures in the evolutionary process with a boundary limit adjustment method when selecting the neighborhood of subproblems. Experiments were performed on two typical benchmark sets (ZDTs and UFs) with two famous quality evaluation metrics (GD and IGD). The results showed that our proposed approach enhances the algorithm's efficiency in maintaining a balanced ability to explore and exploit the population.