A Hybrid Analytical‐Machine Learning Method for Mutual Inductance Calculation of Rectangular Coils With Bar Ferrite Cores

For wireless power transfer (WPT) systems, mutual inductance is universally acknowledged as a key parameter, with its precise calculation constituting the essential theoretical foundation for system design and optimization. For magnetic couplers integrated with bar ferrites, a structured arrangement design is adopted to balance magnetic field stability, heat dissipation efficiency, and engineering adaptability. This design not only addresses the issues of poor heat dissipation and high manufacturing cost associated with monolithic ferrite cores but also overcomes the scenario limitations of annular and other conventional structures, thus serving as an optimal solution that reconciles performance and practicality in WPT and electromagnetic induction systems. Although traditional finite element analysis (FEA) delivers high computational accuracy, it suffers from exorbitant computational cost, making it difficult to meet the demands of rapid design iteration in engineering practice. In this paper, a surrogate model based on the Light Gradient Boosting Machine learning algorithm is presented for rapidly and accurately predicting the mutual inductance coefficient. Innovatively, the analytical solutions for the ferrite‐free working condition are incorporated as physical prior features, which are combined with the geometric parameters (width and thickness) of bar ferrites and the relative positions of coils to construct a high‐efficiency feature set. Training data covering various structural configurations is generated via a computational approach, including diverse combinations of coil turns and different spatial offset conditions. The results demonstrate that the proposed LightGBM model achieves a mean prediction error of less than 1%, while reducing the computational time for multi‐sample cases from the minute scale of FEA to the millisecond scale—representing an improvement of more than three orders of magnitude in computational efficiency. Further analysis of the prediction results reveals the critical regulatory effects of the width and thickness of bar ferrites on the baseline mutual inductance value. The proposed prediction framework exhibits high adaptability to the engineering demands for fast and precise solution of mutual inductance in dynamic scenarios such as WPT systems.