Optimization of Core Loss Model and Correction of Steinmetz Equation Based on Support Vector Machine and Bayesian Parameter Estimation

With the development of power electronics technology, the study of core loss model to improve the efficiency of magnetic components has become the key to optimize the performance of power conversion systems. Based on this, this paper adopts the support vector machine (SVM) model, Bayesian parameter estimation model, etc. to optimize the modeling of core loss and magnetic energy transfer. In order to solve the prediction problem of excitation waveform, the highdimensional complex data are first preprocessed, and the isolated forest algorithm is used to identify and process the outliers to ensure the accuracy and validity of the data. Subsequently, the magnetic flux density was feature extracted based on distribution features and shape features, respectively, and the SVM model was used to classify and identify the excitation waveforms. In order to solve the problem of fitting correction for the Steinmetz equation, five correction methods for the Steinmetz equation are proposed in this paper, including linear correction, exponential correction and inverse correction. The Bayesian parameter estimation method was used to accurately estimate the parameters of each correction equation, to optimize the parameters of different correction models, and to assess the fitting effect of each model based on several model evaluation indexes. The results show that the inverse modified Steinmetz equation performs the best, with an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}}$</tex> value of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 9 1 5 2}$</tex> and a MAPE of 40.52%, which improves the prediction performance by about 2% compared with the original equation, and significantly improves the prediction accuracy and applicability of the model under different working conditions.