Container Ship Fleet Scheduling Based on Reinforcement Learning Considering Carbon Emissions

Against the backdrop of the continuous development of ship informatization, joint scheduling for the entire fleet within a region brings numerous advantages. The optimization of scheduling problems for such a regional fleet, in addition to considering the number of orders completed, reducing operational costs, and further reducing carbon emissions, is also a key research point to address the increasingly severe climate change. This study establishes maritime scheduling strategies for container transport fleets considering energy management. It simulates a shipping company’s operations to meet freight demands among multiple ports. Utilizing reinforcement learning (RL) to choose the optimal scheduling strategy for each individual ship, the study ultimately derives the optimal operational plan for the shipping company. During the process of completing each navigation, every ship will attempt to control its voyage speed for reducing carbon emissions and operating costs. Double deep Q‐learning (DDQN) is used to improve the performance of the RL algorithm, and an additional Q Rank network is used to reduce the action and state space. Ultimately, this paper validates the superiority of the model using a case study that includes multiple ports and ships.