Machine Learning-Based Design of Load-Independent WPT Systems

This paper proposes a machine learning (ML)–based design for a load-independent (LI) wireless power transfer (WPT) system. The fundamental concept of the proposed design strategy is to solve an optimization problem for achieving high-frequency LI operation by full numerical computations. An evaluation function for optimization is formulated, taking into account output voltage regulation against load variations and high power-delivery efficiency. The developed optimization software provides the best parameter set of coupling coil parameters and circuit-component values. In addition, the ML-based optimization finds the optimal parameter set from parameters outside the range that it had considered to be common sense, which is an unexpected but valuable result. The WPT system, designed through the proposed method, achieved ideal LI operation in the experiments. The quantitative agreements between experimental and numerical waveforms substantiate the validity of the proposed design.