Reinforcement Learning-Guided Multi-Environmental Selection Strategy for Constrained Multi-Objective Evolutionary Optimization

For constrained multi-objective optimization problems (CMOPs), this paper introduces a multi-environmental selection strategy guided by reinforcement learning. The proposed method dynamically assesses the feasibility, convergence, and diversity of the population with respect to both decision and objective space distributions. It incorporates three interaction-force-driven selection mechanisms and adaptively identifies optimal strategies through reinforcement learning. Experimental outcomes indicate that this approach outperforms existing algorithms in managing complex constraints, handling discontinuous feasible regions, and addressing circuit testing challenges, thereby confirming its efficacy and robustness.