Integrating Equation-Based Methods with Random Forest Regression for Improved Accuracy of Magnetic Core Loss Modeling

Magnetic components are responsible for a significant portion of the losses in switched-mode power converters. Predicting these losses is vital to a successful design process. Core losses, specifically, are hard to predict, and many standard modeling techniques are often inaccurate. This paper presents a novel approach that integrates an existing, equation-based, core loss algorithm based on Steinmetz parameters with a simple random forest regression model to provide increased accuracy in core loss predictions without incurring significant computational costs. The regression model uses the equation-based model as a starting point and attempts to predict and correct the error through a multiplicative correction coefficient. This hybrid approach is trained and validated using the MagNet database of experimentally measured core loss data that includes a wide variety of materials and operating conditions [1]. Results are presented showing that the hybrid model reduces errors to an average of about one-fifth of what they were when using conventional equation-based techniques.