Temperature-Frequency-Dependent Magnetic Core Loss Prediction Under Triangular-Wave Excitation Based on iGSE and dFDB-TLSABC

Abstract

Abstract

Accurate magnetic core loss prediction is important for efficiency evaluation and thermal design in high-frequency power electronic converters. Under varying temperature and frequency conditions, core loss exhibits strong nonlinear behavior, making conventional fixed-parameter Steinmetz-type models difficult to maintain stable accuracy. To address this problem, this work proposes a temperature–frequency-dependent core loss prediction method under triangular-wave excitation by combining the improved Generalized Steinmetz Equation (iGSE), dynamic Fitness–Distance Balance Teaching-Learning-Studying-Based Artificial Bee Colony (dFDB-TLSABC), and a gated recurrent unit (GRU) network. The dFDB-TLSABC algorithm identifies iGSE parameters under different operating conditions, and GRU maps temperature and frequency to the corresponding parameters for dynamic loss calculation. Based on triangular-wave data for N87, 3C90, and 77 ferrites from the MagNet database, the proposed model reduces the average prediction error by at least 61.87% and 61.60% relative to the conventional iGSE model and the Improved Steinmetz Equation (ISE) model, respectively. Compared with Long Short-Term Memory Network (LSTM), Random Forest (RF), Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), and polynomial regression (PR) models, the reductions are 5.91%, 28.50%, 32.26%, 3.80%, and 2.64%, respectively. These results indicate improved prediction accuracy under temperature–frequency variations.