An Anomaly Detection Method Based on GCN and Correlation of High Dimensional Sensor Data in Power Grid System

Monitoring data or sensor data could reflect the working situation of power grid system at a fine-grained level. Specifically, when an anomaly event happened, some variations will appear and propagate between these interrelated sensor data. However, their latent relationship are complex and difficult to capture. To address this challenge, we propose a data-driven anomaly detection method, which performs real-time correlation analysis of sensor data and implements anomaly detection at runtime. Firstly, the method adopts the correlation coefficient calculation methods to obtain the time-varying correlation between sensed data. Additionally, graph is applied to represent the relationship between them. The edges of the graph are labeled with the degree of correlation and the nodes are marked with some statistical characteristics of the original sensor data. Moreover, an anomaly detection algorithm based on graph convolution network is implemented. The effectiveness of this approach is verified based on real power grid datasets.