A Magnetic Core Loss Model Based on Physics-Informed Neural Network With Cross-Attention

Due to the unclear loss mechanisms inherent in soft magnetic materials, loss modeling often becomes a bottleneck in the analysis of power electronic systems. On one hand, losses significantly affect the overall efficiency; on the other hand, miniaturization caused by high-frequency operation makes the temperature rise of high-frequency magnetic components more sensitive to losses, rendering thermal reliability analysis critically important. However, the applicability of existing loss models deteriorates under complex operating conditions due to their high sensitivity to such conditions. This article proposes an adaptive loss model that enhances the learning capability and operating condition adaptability of the physical loss model through a cross-attention mechanism, achieving an average error of 2.8% and a maximum error of 12.3% on the test set. Furthermore, the accuracy of the proposed model is validated through thermal analysis, yielding a relative error of 1.7%.