Adaptive energy scheduling strategy for port logistics systems: A dual-consolidation continual reinforcement learning approach

The utilization of clean energy and microgrids offers significant opportunities for the green transition of port logistics systems. However, the dual uncertainties in energy generation and consumption pose challenges for typical rule- or model-based energy scheduling methods. This study investigates the dynamic interactions among clean energy generation, energy storage, the main grid, and energy consumption within a port microgrid, and proposes a dual-consolidation continual reinforcement learning (DC-CRL) framework for adaptive energy scheduling in real time. The framework incorporates elastic weight consolidation to protect core parameters of the value function and policy distillation to regularize policy outputs, mitigating catastrophic forgetting and enhancing long-term adaptability. Meanwhile, a rolling reward window is introduced to improve decision quality by encouraging globally optimal behavior, and a dynamic action space is employed to ensure physical feasibility, allowing the agent to adjust its actions in real time according to the state of the storage device. To verify the performance of the proposed approach, we construct key baselines, including a rule-based allocation strategy and a system-optimal strategy with perfect foresight. Experimental results based on Shanghai Yangshan Port demonstrate that DC-CRL achieves a 5.90 % improvement in economic and environmental performance and a 15.0 % enhancement in convergence speed compared with the baseline methods. The case study further provides wind–solar configuration recommendations to support intelligent microgrid control and green port development.