Diagnosis of Current Transformers Based on Symmetrized Dot Pattern and Horizontal Enhanced Convolution

Abstract The stability of fibre optic current transformers (FOCTs), crucial for power metering systems, is paramount. This paper introduces an innovative fault diagnosis method to enhance FOCT diagnostic capabilities. Firstly, utilizing advanced AI and neural networks, the method employs Symmetric Point Pattern (SDP) for complex data fusion, alongside Principal Component Analysis (PCA) for feature reduction. Secondly, Horizontal Enhanced Convolution (HEC) identifies key characteristics, while residual architecture tackles gradient descent issues in traditional CNNs. Then, inspired by Feature Pyramid Networks (FPNs), a new architecture efficiently extracts multiscale features, preventing overfitting with Selective Pooling Prevention Filtering (SPPF). The developed SDP-MSFCNN model achieved a high diagnostic accuracy of 98.82%, surpassing the 90.72% of conventional SDP-CNN and the 96.61% of MA1-DCNN. This approach significantly enhances power system reliability and efficiency, and is ready for integration into high-voltage network fault diagnosis systems, improving the intelligence and safety of 110kV to 1000kV AC/DC transformers.