A Deep Reinforcement Learning-Based Energy-Aware Open Shop Scheduling Method

Energy-aware scheduling is essential for achieving operational efficiency in modern smart manufacturing systems under dynamic production conditions. This research tackles the flexible operation ordering and complex state space of the Energy-Aware Open Shop Scheduling Problem (EOSSP). The EOSSP remains highly challenging due to its combinatorial complexity nature and the need to jointly optimize productivity, energy usage, and delivery reliability. This study presents a Deep Reinforcement Learning (DRL)-based framework that reformulates the EOSSP as a multi-objective Markov Decision Process. The DRL agent learns adaptive scheduling policies through experience replay and target network stabilization, ensuring robust convergence and improved sample efficiency. Results emphasize a fundamental trade-off between solution quality and computational efficiency. Exact method and traditional heuristics, while effective for small instances, face scalability and adaptability limitations under dynamic and uncertain production conditions. Numerical results demonstrated the proposed method consistently maintains a near-optimal gap within 5% of global benchmarks with statistical significance (p < 0.05). The weight preferences study confirmed that even minor variations can cause a significant shift in the solution set along the Pareto Front. Increasing specific weight beyond threshold led to solutions that exhibited an over-optimization bias toward the dedicated objective at the unnecessary expense of other objectives. This research bridges deep reinforcement learning and sustainable manufacturing. Providing a scalable EOSSP solution for high-throughput environments such as semiconductor fabs or data centers.