On-Policy Reinforcement Learning-based Energy Management Strategy for Fuel Cell-Powered Ship with Fuzzy SARSA

This paper proposes an improved novel energy management strategy for fuel cell-powered ships, based on the Fuzzy SARSA algorithm. By combining fuzzy logic with on-policy SARSA learning, the method effectively handles continuous states and actions, achieves fast convergence and adaptive real-time power distribution between fuel cells and batteries. The algorithm transforms continuous state variables into fuzzy features using Gaussian membership functions, and obtains continuous action from fuzzy action by weighted average method. Simulation results for a 14-day mission profile demonstrate fuel consumption reduction and battery SOC maintenance compared to rule-based strategies. Specifically, the total effective hydrogen consumption is reduced from 621.04 kg to 522.34 kg, representing a 14.66% reduction in fuel usage.