A Physics‐Guided Data‐Driven Collaborative Method for Accurate Prediction of Mutual Inductance Between Rectangular Coils at Arbitrary Positions With Magnetic Media

Abstract

The fast and accurate calculation of mutual inductance between rectangular coils at arbitrary positions with magnetic media is considered a critical bottleneck in the optimal design of wireless power transfer (WPT) systems. However, conventional finite element methods suffer from high computational cost, and analytical models are difficult to formulate and implement. To address this issue, a data‐driven image equivalent method (DDIEM) is proposed in this paper, the core of which lies in the construction of a physics‐guided data‐driven collaborative framework. In this method, the modeling dimensionality is first reduced through the utilization of an improved image method. Subsequently, an analytical expression of mutual inductance at arbitrary positions based on the magnetic vector potential is incorporated to establish a rigorous physical backbone. On this basis, a light gradient boosting machine (LightGBM) machine learning algorithm is introduced to accurately capture the high‐dimensional nonlinear mapping relationships induced by bounded magnetic media, thereby enabling precise prediction of coil mutual inductance. Compared with purely data‐driven models, the developed physics‐guided data‐driven collaborative model is characterized by higher prediction accuracy and stronger generalization capability. Validation through finite element simulations and experimental measurements shows that both prediction and experimental errors are within 5.27%. Moreover, the computational speed is improved by at least 30 times compared with finite element simulations, and batch prediction based on CSV files is supported, thereby demonstrating the rapidity and accuracy of the proposed method.