A multi-objective optimization approach for regenerative braking control in electric vehicles using MPE-SAC algorithm

Regenerative braking technology has been extensively promoted to increase the driving range of electric vehicles and satisfy the desire for more environmentally friendly transportation. This study focuses on independently driven front-and-rear-axle vehicles, proposing a Munchausen Prioritized Experience Soft Actor-Critic (MPE-SAC) based regenerative braking control strategy (RBCS) to optimize energy recovery during braking. The proposed RBCS ensures vehicle safety and incorporates battery life degradation as a multi-objective optimization goal, mitigating the adverse impact of high braking currents on battery longevity. The MPE-SAC-based RBCS integrates Prioritized Experience Replay (PER), Emphasizing Recent Experience (ERE), and Munchausen reinforcement learning into the SAC framework, resulting in faster convergence, improved control effectiveness, and greater adaptability. Simulation results show that the RBCS increases regenerative braking rewards by 8.57 %, 2.99 %, 1.45 %, and 0.71 % over rule-based, DDPG, TD3, and SAC methods, achieving 99.28 % of the dynamic programming (DP) algorithm's performance. Additionally, the contributions of PER, ERE, and the Munchausen reinforcement learning algorithms to the performance enhancements of the MPE-SAC-based regenerative braking control system were validated through ablation analysis, and the algorithm's real-time capability is confirmed through the hardware-in-the-loop test.