Adaptive regenerative and friction braking torque allocation using deep reinforcement learning for electric vehicles

The increasing adoption of electric vehicles (EVs) has accelerated the demand for intelligent regenerative braking strategies capable of improving braking energy recovery and reducing mechanical braking dependency under varying driving conditions. This paper proposes a deep reinforcement learning (DRL)-based personalized regenerative braking framework considering regenerative power limitation, battery state-of-charge (SOC) constraints, braking torque saturation, aerodynamic drag, rolling resistance, and emergency braking conditions. The proposed DRL controller adaptively allocates braking torque between regenerative and friction braking systems to maximize braking energy recovery while minimizing friction braking utilization. A reinforcement learning (RL)-based braking framework derived from existing literature is utilized as the baseline controller for comparative analysis. The proposed framework is implemented in MATLAB simulation environment under urban, city, and highway driving conditions using realistic EV parameters. The obtained results demonstrate that the proposed DRL framework significantly outperforms the RL-based controller. Regenerative braking energy recovery improvement varies from 106.77% to 185.60%, while friction braking reduction reaches up to 31.35%. In addition, the proposed framework improves battery SOC, brake pad operational life, and reduces brake service cost and particulate matter emissions under realistic electric vehicle operating conditions.