Joint Optimization of Power Generation and Voyage Scheduling in Ship Power System Based on Operating Scene Clustering and Multitask Deep Reinforcement Learning

The variability in sailing environment and load demand, along with seasonal fluctuations in port electricity prices, presents significant challenges for operating ship power systems (SPSs) in all-electric ships (AESs). However, traditional single-task deep reinforcement learning (DRL) struggles to cope with the high randomness of SPS operation scenarios. This article proposes a novel real-time joint optimization method for power generation and voyage scheduling in SPSs that considers operating scene clustering and multitask DRL (MTDRL). A unified triplet is used to describe operational uncertainty, and the SPS operation scenes are clustered. Then, the importance weighted actor-learner architecture (IMPALA) combined with random network distillation (RND) mechanism, what is called IMPALA-RND algorithm, is applied to minimize operational costs by adjusting power generation and voyage scheduling. The proposed method can achieve differentiated learning for clustered multitask operational scenes and enhance the agent’s capability to explore unknown states. A case study is analyzed based on historical operational datasets of four-DG SPS. Numerical results verify the superiority and real-time performance of the proposed algorithm for joint optimization in multitask operation scenarios.