Anomaly Detection of Structural Monitoring Points for High-speed Railway Ballastless Track Based on Parallel Graph Convolution Neural Network

Aiming at the abnormal monitoring points in the structural health monitoring of high-speed railway ballastless track caused by local structural damage or sensor failures during service, a parallel graph convolution neural network model was established and applied to the anomaly detection of structural monitoring points of high-speed railway ballastless track. Based on the monitoring data of the early initial state of the structure, the parallel graph convolution neural network was trained to obtain the spatial correlations among various monitoring points in the initial state of the structure. The parallel graph convolution neural network was then used to predict the monitoring data of ballastless track monitoring points in service, to realize the identification of abnormal monitoring points. In addition, for data with significant drift, the prediction results can be corrected based on directed graph analysis. The method was also applied to the long-term monitoring data of ballastless track structure of high-speed railways, while the presented model was used to identify abnormal monitoring points.