Magnetic Field Distribution Prediction of Wireless Power Transfer Based on Machine Learning

Wireless power transfer (WPT) is becoming increasingly critical for electronic devices and electric vehicles, making it essential to predict the magnetic field distribution during the design process. However, traditional methods of predicting WPT magnetic field distribution suffer from numerous drawbacks, including time-consuming computations, inaccurate models, and imprecise results. With the rise of Artificial Intelligence (AI), its ability to deliver quick computations and control makes it highly attractive for WPT parameter optimization. Therefore, this paper proposes a method of predicting the magnetic field distribution of WPT based on machine learning (ML). The U-Net was adapted to train the coil output parameter under the input parameters of the designed coil. The output parameter is the spatial magnetic field distribution map, which is useful in identifying safe design parameters within the safe magnetic field threshold. Our training results reveal that our model takes an average of about 3.2 ms with a normalized image prediction error of 0.0034 to calculate the results, compared to an average computation time of approximately 23.74 s by COMSOL.