Two Stage Optimization Algorithm for Resource Leveling Problem

Resource leveling in shipyard mid-term scheduling is critical for operational efficiency, yet existing approaches face significant limitations. Constraint programming methods suffer from computational scalability issues, while Meta Heuristic algorithms show restricted search space exploration. This study proposes a two-stage optimization framework integrating heuristic sequencing with reinforcement learning. In Stage 1, an Enhanced Kahn's Algorithm generates efficient activity sequences by incorporating resource requirements and durations. In Stage 2, Q-Learning dynamically determines optimal start times through episodic learning, progressively minimizing resource variance. Experimental evaluations using actual shipyard mid-term scheduling data demonstrate that the proposed algorithm achieves superior performance in large-scale problems compared to constraint programming and Meta Heuristic approaches. Results indicate that heuristic-based sequencing provides substantially improved computational efficiency, while reinforcement learning achieves rapid convergence and superior outcomes. This research establishes the practical applicability of integrating heuristics with reinforcement learning for complex industrial scheduling problems.