.png){kind=icon}
.png){kind=icon}
Return
A dynamic multi-objective optimization evolutionary algorithm based on multi-modal feature fusion and entropy-driven reinforcement learning
.png){kind=icon}
PRE
AI
DOI：10.1016/j.swevo.2025.102212.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Dynamic Multi-objective Optimization Problems (DMOPs), where objective functions, constraints, or decision variables change over time, present significant challenges to maintaining both convergence and diversity during the optimization process. Effectively identifying and tracking the optimal solution while balancing the convergence and diversity of the solution set remains the core challenge faced by evolutionary algorithms. To address these challenges, we propose a Dynamic Multi-objective Optimization Evolutionary Algorithm (DMOEA) based on multi-modal feature fusion and Reinforcement Learning (RL). Firstly, a multi-modal feature fusion strategy was designed that integrates Pareto front distribution, decision variable variation, crowding distance, and centroid shift to accurately detect environmental changes and classify their severity. Secondly, a distribution entropy-driven RL strategy is used to dynamically identify diversity-oriented decision variables, and tailored uniformization strategy is applied to increase the diversity. Finally, an adaptive bidirectional search strategy is designed that can perform fine-grained searches on non diverse decision variables in two directions to enhance convergence without sacrificing diversity. Extensive experiments on dynamic test functions demonstrate that our method significantly improves population adaptability, diversity maintenance, and convergence accuracy compared to state-of-the-art DMOEA, offering a promising direction for real-time dynamic optimization in complex environments.
Journal
IF:
8.5
Papers:
2.2K
Citations:
1.0W
Organization
No organization information available