Core Loss Study of Magnetic Elements Based on Gradient Boosting Decision Tree and Data-Driven Magnetic Elements

Along with the development of power electronics and the third generation of power semiconductor technology, the power converter is developing towards high frequency, high power density and high reliability, and the magnetic components, as the key devices, have a significant impact on their loss characteristics. Winding loss can be accurately obtained by electromagnetic field finite element simulation, but the core loss is related to the operating frequency, magnetic flux density, excitation waveform and other factors, presenting nonlinear and interrelated, resulting in a large gap between the existing model and the actual demand. In this study, the core loss data of magnetic components under a given operating condition are measured, and mathematical modeling and algorithms are used to solve the problem of accurate core loss calculation. By determining the key indexes, a GBDT gradient boosting decision tree model is established, which is combined with the random forest algorithm to solve the problem. For different problems, the GBDT model is used to judge the excitation waveform, linear fitting is used to evaluate the correction effect of the equation, and the eigenvectors are optimized to improve the loss estimation accuracy. Finally, the core material loss model is successfully constructed to achieve accurate prediction of core loss under other working conditions, which provides powerful support for the design optimization of magnetic components.