Improved DBSCAN-based Data Anomaly Detection Approach for Battery Energy Storage Stations

Abstract

In battery energy storage stations (BESSs), the power conversion system (PCS) as the interface between the battery and the power grid is responsible for battery charging and discharging control and grid connection. Any anomaly in the data of a PCS will threaten the security of the BESS. It is difficult to detect anomalies in real-time data because of the large scale, chaos, and small deviations between normal and abnormal values. In this paper, the density-based clustering algorithm DBSCAN is used for data anomaly detection. However, the traditional DBSCAN has a limitation in that it has difficulty in the parameter selection and the parameter is strongly correlated to the detection accuracy. To address this issue, we propose a parameter self-selection-based improved DBSCAN model for detecting PCS anomalies in BESSs. The detection is achieved by mining the correlations between data sets and combining them with the DBSCAN algorithm, and the model is updated in real time based on the normal data of the PCSs. The proposed method is further validated using a comparative experiment based on real-world BESS data.