User Preference-Adaptive Battery Fast Charging Strategy Considering Temperature Rising Based on Transfer Reinforcement Learning

Battery safe fast-charging is the key technique to promote the large-scale popularization of electric vehicles. However, fast-charging control is a multiphysics-constrained optimization problem with complex coupled dynamics. The heat generation rate and energy loss are critical for battery thermal safety and efficiency. This article thus formulates the charging optimization objective function as a weighted sum of charging time, temperature rising and energy loss, and proposes a deep reinforcement learning (DRL) framework with a deterministic policy gradient method to cope with continuous sets of states and actions. First, an electrothermal coupled model for a wide temperature range from 0°C to 55°C is established. Second, a DRL-based optimizer is proposed to solve the multi-objective problem. The charging time, energy loss, and temperature rising are formulated as the rewards. Third, the proposed method is validated through experiments conducted on 30 Ah LiFePO₄ cells and the effects of different penalty weights are analyzed. Finally, a transfer learning strategy is employed to make the methods adaptive to meet the diverse needs of different users with a small amount of training. The effectiveness, superiority, and adaptability of the proposed algorithm are validated using the comparative results with rule-based strategies.