Insulation Condition Recognition of Cable Terminals for High-Speed Trains Using Multi-Dimensional Image Modeling and SwinCNNFusion

Accurately assessing the severity of cable terminal defects is crucial for ensuring the stable operation of high-speed trains. However, the partial discharge (PD) characteristics of the same type of defect show a high degree of similarity in different severity levels, which brings difficulties to the accurate identification of insulation defects. To address this issue, this paper proposes a new method for defect detection and severity identification. Firstly, the original PD signal is converted into an image by using Multi-scale Differential Synchro-squeezing Transformation (MDSST) and Dynamic Phase Coupling Matrix (DPCM). Then, the SwinCNNFusion model - a parallel fusion model integrating the Swin Transformer and the Convolutional Neural Network (CNN)- is used for classification. This PD signal processing technology simultaneously embeds the time-frequency information and phase information of different orders of differences of the original PD signal into the generated image. This reduces the similarity between classes and significantly improves the distinguishability of the transformed representation. Meanwhile, the SwinCNNFusion model is designed to accurately identify and integrate the multi-dimensional modeling information derived from PD signals. The experimental results show that the detection accuracy of this method is 96.78%, which is superior to several existing methods.