Regenerative Braking System with Smart Energy Recovery

á Norði, Bjartur Ragnarsson ; Spiliotopoulos, Konstantinos

4. term

Electro-Mechanical System Design, Master

2026

2026-01-02

177

This thesis develops and evaluates a Reinforcement Learning (RL) controller for regenerative braking in a battery-electric vehicle (BEV). The RL agent operates as a torque-split controller, allocating braking demand between regenerative and friction systems under the same physical constraints as a rule-based baseline. A simplified HPPC-inspired State-of-Power (SoP) estimator provides a dynamic battery charging limit, enabling the agent to learn time-varying power constraints. A Deep Deterministic Policy Gradient (DDPG) agent is trained within a Simulink environment, where an EV drivetrain is modeled. The reward function encourages effective use of the SoP-limited regenerative capability while penalizing constraint violations and abrupt control actions. Across various drive cycles and state-of-charge levels, the RL controller consistently demonstrates strong performance, despite the highly restricted training data and time. The results highlight RL’s potential as a foundation for future adaptive energy-management strategies.

Regenerative braking ; Battery electric vehicle ; Torque-split contorl ; reinforcement learning ; Deep Deterministic Policy Gradient ; Simulink simulation ; State of Power estimation ; HPPC-based battery model ; energy recovery optimization ; safety-constrained control

Download
View record in AAU Student Projects