Reinforcement Learning-Assisted Multi-Stage Evolutionary Constrained Multi-Objective optimization

Multi-stage evolutionary algorithms, which divide the population evolution process into multiple stages, have demonstrated their effectiveness in constrained multi-objective optimization. The evolution stages are often manually arranged before optimization for existing algorithms, which significantly affects the performance in solving various problems. To address this issue, a reinforcement learning based stage determination strategy is proposed. Specifically, the evolution process is divided into numerous stages, and reinforcement learning determines the evolution stages during optimization. Besides, an evolutionary algorithm with two candidate evolution stages is also developed to verify the effectiveness of the proposed strategy. Experimental results on benchmark and real-world problems have revealed the superiority of the proposed algorithm over the state-of-the-art algorithms.