Energy Management for Electrified Tracked Vehicles via Improved Soft Actor-Critic Algorithm

Energy management strategies (EMSs) play a crucial role in determining the power distribution of hybrid electric vehicles (HEVs). Recently, deep reinforcement learning (DRL) has emerged as a prominent approach for developing EMSs. This paper proposes a DRL-based EMS for a series hybrid electric tracked vehicle (SHETV) equipped with an enginegenerator set (EGS) and a battery pack to improve fuel economy. Firstly, an improved soft actor-critic (SAC) algorithm is formulated by integrating the prioritized experience replay (PER) mechanism. Subsequently, an EMS is developed based on the improved SAC algorithm. Then, a comprehensive evaluation is conducted to assess the superiority of proposed EMS. Extensive comparisons reveal that the proposed EMS achieves a 31.82% faster convergence speed and enhances fuel economy by 4.28% compared to the EMS based on the standard SAC algorithm.