Constructing a Multidimensional Abnormal Data Feature Library under Clustering Algorithm and Improving the Quality of Electric Power Data

In the massive data generated by the power system, abnormal data may lead to unstable system operation and even cause major faults. Firstly, the historical operation data of power system were preprocessed, including data cleaning, data conversion and feature extraction. Then, the K-means clustering algorithm was applied to analyze the data to construct the abnormal data feature library. Then it compared the performance of two algorithms, DBSCAN (density-based spatial clustering of applications with noise) and Hierarchical Clustering algorithm (hereinafter referred to as Hierarchical algorithm) in terms of timeliness, anomaly detection rate, data integration and other evaluation indicators, and evaluated their effects in improving power data quality. The research results show that the timeliness of multidimensional abnormal data feature database based on K-means clustering algorithm was only 109ms, and its anomaly detection rate and data integration degree can reach 95% and 89% respectively. It can effectively improve the quality of power data and become a powerful tool for power system anomaly detection because of its excellent performance in data integration degree.