Experimental Validation and Model-Based Optimization of Regenerative Braking in Light Electric Vehicles Using IT2FL and DDPG-RL

Electric vehicles (EVs) face range limitations primarily due to battery capacity, and increasing battery size introduces weight, efficiency, and performance trade-offs. Regenerative braking offers a solution by recovering energy deceleration, but Light Electric Vehicles (LEVs) such as e-Trikes present challenges due to low mass, limited energy storage, and strict current limits. This study experimentally evaluated regenerative braking on the LEV platform under unloaded and loaded conditions. The collected data validated a simulation model with low mean absolute errors (0.19 V, 4.31 A, and 0.10 Ah), ensuring high fidelity for control strategy testing. Two intelligent energy recovery strategies, Interval Type-2 Fuzzy Logic (IT2FL) and Deep Deterministic Policy Gradient Reinforcement Learning (DDPG RL) were implemented and compared to a baseline (original) controller. Results show that DDPG RL achieved the highest state of charge retention (98.91%), outperforming IT2FL and the baseline (both 98.35%). Kernel Density Estimation confirmed that DDPG RL maintained optimal power density (3000–4000 W) over a wider speed range (<20 km/h), unlike other methods limited to mid-speed operation. These findings highlight DDPG RL as a promising approach for enhancing LEV energy efficiency, battery protection, and regenerative braking performance.