Core Loss Prediction Using a PC-GBT Model Based on Permutation Feature Importance and Grid Search Optimization

This study aims to predict core loss in magnetic materials using a PC-GBT model based on Permutation Feature Importance and Grid Search Optimization. The research analyzes the impact of three key factors—temperature, excitation waveform, and core material—on core loss, including both their individual and paired interactive effects, and evaluates the significance of each factor. Descriptive statistical analysis, distribution characteristics analysis, and univariate analysis were conducted to establish the relationships between core loss and these factors. The results indicate a positive correlation between core loss and both excitation waveform and material, while a negative correlation is observed with temperature. A two-factor analysis further explores the synergistic effects of temperature, material, and excitation waveform on core loss. Additionally, flux density data was extracted, and potential features were identified with reference to the core loss separation model and two modified Steinmetz equations. Benchmark model comparisons reveal that the Gradient Boosted Trees (GBT) model outperforms Support Vector Machines (SVM) in terms of prediction accuracy and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> values. The development of the PC-GBT model includes Permutation Feature Importance analysis and Grid Search Optimization to identify the optimal hyperparameter configuration. Ultimately, the PC-GBT model demonstrates a lower mean error and an improved R2 value compared to the GBT model, confirming the effectiveness of the PC-GBT model.