Fault Diagnosis Method for Compensation Capacitor and Tuning Area Based on Graph Convolutional Network

To further explore the compensation capacitor and tuning area state characteristics contained in the dynamic monitoring data of jointless track circuit, a fault diagnosis method for compensation capacitor and tuning area based on graph convolutional network is proposed by combining SSA-VMD algorithm and graph convolutional network. The SSA-VMD algorithm is used to decompose the shunt current envelope curves under different states of the compensation capacitor and tuning area, and the energy values of the decomposed eigenmode components are calculated using a normalization formula to form a shunt current feature vector as a fault dataset. The shunt current feature vectors under different fault states are used as node features of a fully connected graph, and a 3-layer graph convolutional network is used to extract the node features of the fully connected graph. Finally, the extracted node features are classified and identified using a fully connected layer and a softmax layer. The simulation results show that the average detection accuracy rate of the model for 11 types of compensation capacitors and tuning zone faults reaches 96.97%, the average recall rate reaches 96.32%, the average $F 1$ value reaches 95.68, and the average false alarm rate is only 3.68%. The research results provide a more accurate and efficient method for fault diagnosis of compensation capacitors and tuning zones in noninsulated track circuits.