Machine Learning-Based Excitation Waveform Classification Model for Magnetic Components

Magnetic components play a crucial role in modern power electronic systems, with their performance directly affecting system stability and efficiency. This paper proposes a machine learning-based excitation waveform classification model that integrates time-domain, frequency-domain, and shape feature variables. Six classification algorithms—Support Vector Machine (SVM), k-Nearest Neighbors (k-NN), Multilayer Perceptron (MLP), Random Forest (RF), XGBoost, and LightGBM—are employed to construct the classification model. The experiment adopts a five-fold cross-validation method, and classification performance is evaluated using coefficient of determination ($\mathbf{R}^{\mathbf{2}}$), mean squared error (MSE), and mean absolute error (MAE). The results show that the Random Forest model performs the best across all three metrics, achieving the highest fitting accuracy ($\mathbf{R}^{\mathbf{2}} =0.9743$) and the lowest prediction error ($\mathrm{MSE}=0.0047, \mathrm{MAE}=$ 0.0524), making it the most suitable classification model for this task. This study not only enhances the automatic recognition capability of excitation waveforms but also provides new insights for core loss modeling, offering significant theoretical and engineering application value.