Hybrid TD3-PI Control for Wireless Charging of Light Electric Vehicles

Abstract

This article proposes a deep reinforcement learning (DRL)-assisted adaptive control strategy for wireless power transfer (WPT) in light electric vehicle (LEV) charging. A phase-shifted full-bridge (PSFB) converter with series–series (S–S) compensation is considered, where the primary-side controller regulates the secondary charging current. Conventional proportional–integral (PI) control, tuned around nominal coupling, suffers performance degradation and even loss of convergence under coil misalignment and coupling variation. To address this, a twin delayed deep deterministic policy gradient (TD3) agent is used to adapt the PI gains online based on secondary-side feedback. The agent is trained in a Simulink environment and deployed in inference-only mode on a TI C2000 F28379D microcontroller, occupying approximately 25% of on-chip memory, demonstrating the feasibility of DRL-based adaptive control on resource-constrained embedded platforms. Experimental results show that the proposed TD3-assisted adaptive PI achieves constant-current (CC) regulation with significantly faster convergence compared to fixed-gain PI, settling time is reduced from 8.65 to 5.95 ms under aligned conditions, from 17.00 to 11.05 ms under lateral misalignment, and from 26.00 to 14.82 ms under angular misalignment, corresponding to improvements of about 35% and 43%. The faster transient response lowers stress and losses in the power switches. These results validate that DRL-enabled adaptive control enhances robustness of WPT-based LEV charging under realistic misalignment scenarios while remaining implementable on industrial-grade digital controllers.