Machine-Learning-Assisted Wireless Power Transfer for Electric Vehicle Charging Under Coil Misalignment
Abstract
Recent advances in wireless power transfer (WPT) have enabled contactless charging of electric vehicles (EVs), but coil misalignment and load variation still cause significant efficiency degradation and DC-link voltage fluctuations. This paper proposes a lightweight deep neural network (DNN) that predicts optimal compensation parameters in a series-series inductive WPT system to maximize powertransfer efficiency while regulating the output voltage under lateral and angular misalignment. The DNN is trained on a dataset generated from a physics-based circuit model over a wide range of misalignment and load conditions. At run time, the trained model provides real-time predictions of compensation capacitance adjustment and inverter switching frequency, which are applied through a simple supervisory controller. MATLAB/Simulink simulations show that the proposed ML-assisted controller improves system efficiency by up to 11.8% and reduces output-voltage ripple by 46.3% compared to a conventional fixed-parameter design under severe misalignment scenarios. The proposed method requires only basic sensing of coil currents and voltages and can be implemented on low-cost embedded platforms, making it suitable for practical EV WPT systems.