Optimizing regenerative braking in light electric vehicles using deep deterministic policy gradient reinforcement learning

The low mass, limited motor capacity, and small battery size of light electric vehicles (LEVs) constrain the regenerative energy recovery process, limiting the driving range of these vehicles and in turn their widespread use. An effective regenerative braking control strategy is required to maximize energy recovery while maintaining brake stability. This paper presents the modeling and experimental validation of three regenerative braking control strategies for LEVs: a baseline (original) controller, an interval type-2 fuzzy logic controller, and a deep deterministic policy gradient reinforcement learning (DDPG-RL) controller. Under the worldwide harmonized light vehicles test cycle (WLTC) Class 1 driving cycle, the DDPG-RL controller achieved the best performance, yielding the lowest energy consumption of 1.99 kWh, highest regenerative energy contribution of 18.15 %, and highest energy efficiency of 12.59 km/kWh, corresponding to a 15.4 % increase in driving range over the baseline (original) controller. A kernel density estimation analysis also revealed that DDPG-RL exhibited the most consistent and intense regenerative power distribution, particularly in the 20–40 km/h range, which is typical for urban driving. The baseline model was experimentally validated to ensure the power flow representation accuracy. The results revealed a mean absolute error of 0.17 % in the battery state of charge and a final deviation of 0.33 %, thus verifying the reliability of the comparative evaluation. These results validate the DDPG-RL strategy as a highly effective approach for maximizing energy recovery, reducing consumption, and extending the driving range, thus being a potential solution for the sustainable optimization of LEVs.