A Transient Switching Modeling Method for IGBT Based on Transfer Learning

This paper proposes an IGBT transient switching modeling approach based on transfer learning, designed to enhance modeling efficiency and accuracy while reducing reliance on extensive data and complex parameter extraction. The study uses two IGBT modules with similar specifications as case examples. Data under various operating conditions were collected using the industrial simulation tool L Tspice, resulting in a source dataset and a target dataset, with the target dataset comprising only 13.43% of the size of the source dataset. Initially, a Multi-Layer Perceptron (MLP) neural network is pre-trained on the source dataset to capture the fundamental switching behaviors of the IGBTs. Subsequently, the pre-trained model is fine-tuned through transfer learning to adapt it to the specific target IGBT module using the target dataset. The model's performance is evaluated using the relative root mean square error (RMSE). The results demonstrate that this method can accurately replicate IGBT transient switching characteristics with limited data, offering an efficient and robust solution for IGBT transient modeling in power electronic systems.