A reinforcement learning based real-time energy management method for mobile microgrid considering photovoltaic uncertainty

• An adaptive control technology based on DQN is proposed under the uncertainty of on-board PV power. • The Markov decision model in real-time energy management of new energy ship is set innovatively. • On the basis of satisfying power demand, the operating cost of new energy ships is reduced vastly. With the ever-increasing awareness of worldwide greenhouse gas emissions, traditional diesel-driven ships are gradually being replaced by renewable energy ships. Zero-carbon power sources, such as photovoltaic (PV) power generation, are progressively integrated into electric ships. However, the uncertainty associated with onboard PV generation has become a critical factor limiting effective energy management on alternative energy ships. To address this issue, this paper proposes a real-time energy management optimization method based on reinforcement learning, specifically tailored to handle PV uncertainty and dynamic load variations during navigation. The proposed algorithm optimizes the energy flow between the onboard diesel generator and the energy storage system in real-time, aiming to minimize fuel consumption and enhance operational stability. Real-world shipboard microgrid data is utilized to perform case studies. Simulation results indicate that fuel consumption under the proposed approach is only 90.32% and 94.57% of that in scenarios without PV systems and traditional robust optimization methods, respectively. Moreover, the method effectively stabilizes the state of charge within a safe operational range of [0.2, 0.8], which is helpful for energy storage lifespan.