Research on Iron Core Loss Prediction Based on Data-Driven and Machine Learning Algorithms

This paper focuses on the performance optimization of magnetic components, and discusses the key issues such as the construction of a data-driven core loss prediction model. A data-driven core loss prediction model is constructed, which integrates a variety of machine learning algorithms, such as random forest, gradient lift tree, K-nearest neighbor regression, decision tree regression and artificial neural network, and improves the stability and accuracy of prediction through weighted prediction model. Finally, with the core loss prediction model as the objective function, particle swarm optimization algorithm was used to determine the optimization conditions of the magnetic components, including the optimal material, temperature, frequency, waveform and magnetic flux density peak, and the overall performance optimization of the magnetic components was realized.