Optimization of Regenerative Braking Control Strategy for Dual-Motor Electric Vehicles Based on Deep Reinforcement Learning

In order to maximize the braking energy recovery of dual motor-driven electric vehicles to guarantee braking safety and comfort, this paper proposes a Deep Reinforcement Learning (DRL) braking energy recovery strategy with an improved prioritized experience replay and exploration mechanism. Firstly, the energy recovery process of the electric vehicle is analyzed, and a mathematical model reflecting the dual-motor braking energy conversion and constraint characteristics is established. Secondly, an energy recovery decision-making framework is established based on the twin delayed deep deterministic policy gradient (TD3) algorithm, and an improved prioritized experience replay policy is designed to address the problem of inefficient empirical sampling in the traditional TD3. Finally, in order to change the imbalance between exploration and utilization that exists in the deterministic policy approach, an exploration policy that is automatically adjusted with the training process is introduced to enhance the algorithm’s decision-making ability in complex environments. The effectiveness of the proposed strategy was verified using a Matlab/Simulink simulation model under China Light Vehicle Test Cycle (CLTC) and World Light Vehicle Test Cycle (WLTC) operating conditions. The results showed that the proposed method exhibits higher efficiency in braking energy recovery when compared to other algorithms, with energy recovery efficiencies of 35.18% and 25.79% under two operating conditions, which both met the braking comfort and safety targets.