Security Simulation Analysis of Magnetic Core Loss Prediction Model based on KD-Tree Federated XGBoost Prediction

With the wide application of power electronics, core loss modelling of magnetic components has become a key issue. Traditional core loss models have many limitations, and this paper proposes a federal XGBoost prediction model based on KD-Tree for core loss prediction. Firstly, common empirical core loss calculation models, such as Steinmetz equation, are introduced, and an attempt is made to correct the MSE equation with the quadratic function of temperature, while the XGBoost algorithm is introduced. By constructing a KD-Tree to select similar characteristic materials for model training, the model performance is improved by combining the idea of federated learning. The experimental results show that the corrected model exhibits advantages in the comparison of different correction methods, and the federal XGBoost core loss prediction model based on KD-Tree reaches R² of 0.9909 and MAPE of 15.8893, which has good Prediction effect, and the frequency, the peak flux density and the material type have significant effects on the core loss. This study helps to optimise the core material selection and production process, and provides strong support for related design and decision-making.