Research on Abnormal Analysis Method of Monitoring Data for Hydropower Units Based on Graph Neural Network

In response to the problem that the amount of state monitoring data for hydroelectric generating units is gradually increasing while the data quality is poor, a method for abnormal data detection of state monitoring data for hydroelectric generating units based on graph neural networks is proposed. Firstly, a graph topology data is established for the input data and abnormal data is detected by combining with graph neural networks. Taking 1088 pieces of the deflection data of the guide bearing of a hydropower station's state monitoring system as an example, 100 pieces of random data are inserted at the same time interval. Through the comparison of clustering performance and time performance with conventional DBScan, K-means, and OCSVM algorithms, the proposed method has the highest recognition accuracy, the best data cleaning effect, and can significantly reduce the computing time.