Precise Coil Inductance Prediction in WPT Systems: A Transfer Learning Approach

Abstract

Precise and efficient coil inductance calculations in wireless power transfer (WPT) systems are crucial for accelerating the coil design process and optimizing system performances. However, commonly used finite element simulations are quite time-consuming, while traditional analytical inductance models not only suffer from complex computations and low accuracies but also require intricate mathematical models for describing various coil structures. To address these issues, a unified coil structure model is proposed in this article for the first time to capture the inherent structural consistency between the widely used circular and rectangular coils, eliminating the heavy burden of developing unique structure models for different coils. Based on this unified coil structure model, a transfer learning-improved feedforward neural network (FNN) is developed to precisely and efficiently predict the self-inductance and mutual inductance of both circular and rectangular coils under varied misalignments in WPT systems. The transfer learning-improved FNN allows for efficient model transfer without requiring extensive data acquisition and lengthy model retraining while maintaining high prediction accuracy for different coils under varied misalignments. Finally, the accuracy and generalization of the proposed transfer learning-improved FNN are validated experimentally, indicating a significant peak mean accuracy increment of 85.52% compared to the previous analytical inductance model.