Intelligent Wireless EV Charging for Green Transportation: A Deep Reinforcement Learning Approach with Multi-Stage Current Battery Management

Abstract

Electric vehicle (EV) wireless power transfer (WPT) systems face two fundamental challenges that hinder their widespread adoption: sensitivity to coil misalignment and the need for battery-friendly fast charging protocols. This paper presents a comprehensive and integrated framework that addresses both challenges through a three-pronged approach combining advanced coil geometry optimization, multi-stage current method (MSCM) charging, and reinforcement learning (RL)-based adaptive control. First, a memetic algorithm hybridizing global exploration and local refinement is employed to design coil geometries that are inherently resilient to misalignment. The optimized coils maintain strong magnetic coupling under lateral displacements up to ±75 mm by strategically sizing the secondary coil’s outer diameter to be smaller than the primary coil’s, ensuring it remains within the optimal magnetic flux region. Second, an MSCM charging protocol is developed and optimized with the objective of balancing charging speed against battery thermal stability and state of health (SoH). The proposed strategy determines optimal current levels for each charging stage, reducing temperature rise compared to conventional CC-CV charging. Third, a novel Deep Q-Network (DQN) RL agent is implemented for real-time adaptive control of the WPT system. The RL controller dynamically adjusts phase shift in response to varying coupling conditions, load disturbances, and battery state, outperforming traditional PI controllers with 23% faster settling time and improved efficiency under dynamic misalignment scenarios. Finite element analysis (FEA) simulations validate the electromagnetic performance of the optimized coils, while an experimental prototype demonstrates the integrated system’s performance. Results show that the combined approach achieves 91.2% DC-DC efficiency under nominal conditions and maintains over 83% efficiency under lateral misalignments up to ±75 mm, fully complying with SAE J2954 alignment tolerance requirements. The MSCM charging protocol, guided by the memetic algorithm, limits battery temperature rise during a full charge cycle, while the RL controller ensures stable power delivery under real-world dynamic conditions. This work establishes a new paradigm for holistic WPT system design, demonstrating that synergistic optimization of magnetic structures, charging protocols, and intelligent control can simultaneously achieve misalignment resilience, fast charging, and adaptive robustness.