Core Loss Modeling of Magnetic Components Using a Data-Driven Method

Research on loss characteristics of magnetic components is a critical topic in power conversion technology. To address the significant discrepancies between existing core loss models and practical application requirements, this paper establishes a high-precision core loss evaluation model based on eXtreme Gradient Boosting. The model uses temperature, material, and excitation waveform as decision variables, with minimizing core loss as the objective. The study first evaluates the model through confusion matrices while conducting single-factor analyses of temperature, excitation waveform, and material using boxplots and investigating synergistic interactions between single factors. Subsequently, key metrics including coefficient of determination, mean squared error, and mean absolute error are employed for comprehensive regression assessment. Residual plots are utilized to examine the fit between predicted and actual values, demonstrating the model's high accuracy, strong applicability, and practical relevance, thereby validating its superiority. Finally, the paper explores maximizing magnetic energy transmission by solving the optimization problem using simulated annealing algorithms to determine conditions that achieve both minimum core loss and maximum transmission efficiency, extending the research to multi-objective optimization.