AC Loss Calculation of High Temperature Superconducting Coils Based on a Surrogate Model

Abstract

AC losses have a significant impact on the design and operation of superconducting power devices. Therefore, fast and accurate estimation of AC losses is essential. However, calculating AC losses in high-temperature superconducting (HTS) coils often requires considerable computational resources. To address this, a time-series surrogate model based on a convolutional neural network (CNN) is proposed. The model is trained using sample points X (e.g., HTS coil parameters and current profiles) and corresponding AC loss responses Y obtained from finite element analysis (FEA). It integrates a self-attention hybrid convolution module and a current change perception mechanism to extract deep temporal features, and employs an adaptive threshold to enhance open-set recognition and prediction robustness. The model's accuracy is validated against COMSOL simulation results, demonstrating that it significantly improves computational efficiency while maintaining high prediction accuracy.