Auto-Tuning Hyperparameters of Reinforcement Learning based Energy Management Strategy for Fuel Cell Tracked Vehicles via Bayesian Optimization

Energy management strategies (EMSs) are critical for optimizing energy efficiency in vehicles with hybrid energy storage systems. While deep reinforcement learning (DRL) has shown promise for EMS development, manual hyperparameter tuning remains labor-intensive and suboptimal. This study proposes an automated framework integrating Bayesian optimization (BO) with the soft actor-critic (SAC) algorithm for hyperparameter selection. Implemented on a fuel cell hybrid electric tracked vehicle model, the method demonstrates 66.67% faster convergence and 3.05% improved fuel economy versus manual tuning. The Bayesian-SAC synergy offers an efficient EMS solution, reducing engineering effort while enhancing both training performance and energy optimization.