ReinforceAdapt: multi-objective approach for environmental adaptation method

Multi-objective optimization (MOO) underpins numerous real-world applications requiring the simultaneous optimization of conflicting objectives. Evolutionary Algorithms (EAs), particularly Multi-Objective Evolutionary Algorithms (MOEAs), have demonstrated exceptional capabilities in approximating Pareto-optimal solutions. However, their reliance on static operator configurations often results in suboptimal performance in dynamic and high-dimensional search spaces. To address this limitation, we propose a novel framework that integrates Deep Reinforcement Learning (DRL) with MOEAs, enabling adaptive and context-aware operator selection. Our methodology formulates operators as actions and solution states within a reinforcement learning paradigm. By leveraging Q-learning, the framework dynamically evaluates and selects operators that balances exploration and exploitation to optimise convergence and diversity. The implementation incorporates modular optimization, adaptive credit assignment, and decomposition-based subproblem partitioning which ensures scalability across diverse problem domains. Experimental evaluations on benchmark suites reveal that the proposed DRL-MOEA framework achieves superior performance, significantly improving Inverted Generational Distance (IGD) metrics while enhancing Pareto front diversity compared to state-of-the-art approaches. The results shows the framework's robustness and adaptability, establishing it as a powerful tool for addressing the challenges of multi-objective optimization in complex and dynamic environments.