Highly precise modelling of resonant coils based on deep neural networks for wireless power transfer systems

Abstract

To address the issue of precise coil design in Wireless Power Transfer (WPT) systems, this paper conducts a theoretical analysis of the coil's self-inductance and mutual inductance. By combining theoretical calculations with Finite Element Analysis (FEA) simulation values, a model for the precisely predicting coil’s self-inductance and mutual inductance is obtained. This model can derive the coil’s electrical parameters based on its physical parameters, such as the radius of the litz wire, coil radius, and the magnetic permeability of the ferrite grating plate. Taking a spiral circular coil as an example, the coil is first modeled in a three-dimensional coordinate system, and theoretical formulas for calculating the self-inductance and mutual inductance with a ferrite plate under various offset conditions are derived using the mirror method to obtain corresponding theoretical calculation values. Additionally, considering the data-fitting characteristics of AI algorithms, ANN and transfer learning is chosen to combine theoretical calculation data, FEA simulation data, and experimental measurement data, yielding a more accurate calculation model. Finally, the predictive performance of the three proposed deep learning models is compared, allowing for the calculation of the coil’s self-inductance and mutual inductance under given coil parameter conditions. The results show that, compared to the model SM, the accuracy of our proposed Hybrid SCM model is only 1.38\% lower than that of SM, while the accuracy of the Hybrid SCM model with transfer learning is 0.67\% higher than that of SM.