Neural-Network-Enhanced Adaptive MPC for Constant-Voltage Output in Dynamic Wireless Power Transfer Systems

This paper proposes an enhanced adaptive model predictive control (AMPC) scheme to regulate constant output voltage in dynamic wireless power transfer (DWPT) systems. DWPT technology is developed to alleviate range anxiety for electric vehicles (EVs), while wide fluctuations in output voltage caused by EV motion hinder its practical implementation. In this paper, an AMPC method is designed to mitigate model parameter mismatches resulting from variations in coil mutual inductance, ensuring a stable power supply during EV driving. To this end, an adaptive deviation compensation mechanism is established by incorporating a lightweight, online-trained radial basis function (RBF) neural network, enabling efficient compensation of nonlinear model mismatches between the system model and its physical implementation with minimal computational overhead. By integrating these real-time adjustments into the AMPC framework, the controller dynamically corrects parameter deviations and reoptimizes the control objective to generate the optimal control signal. Experimental results demonstrate that the proposed AMPC enhances dynamic response and improves disturbance rejection under diverse operational scenarios, validating its real-time feasibility and robustness in achieving constant output voltage control for EV-DWPT systems.