Long Short-Term Memory-Based Anomaly Detection for State Estimation in Smart Grid Considering Anomalous Database

To achieve real-time state estimation in smart grids, Phasor Measurement Units (PMUs) have been extensively utilized. However, statistics show that PMU data can also contain various types of anomalies. In response to these challenges, anomaly detection has become a key area of research. Many existing studies on anomaly detection rely on databases, but they often assume that the historical data used are free of anomalies. In practice, however, data may be contaminated due to factors such as system detection errors, cyberattacks. Using such contaminated data in studies can significantly increase the vulnerability of the smart grid. Manual inspection of data on a case-by-case basis is both time-consuming and costly. This paper proposes a method for detection of anomalies against state estimation system considering anomalous database. A subset of the database is selected and a hybrid approach combining model-based and data-driven anomaly detection methods is implemented through residual-based analysis and subsequence correlation in both temporal and spatial dimensions of the data series. The data is assessed in terms of its adaptability to the smart grid and its pattern correlation in terms of anomalies identification. After identifying the anomalies, a powerful long short-term memory (LSTM) network is trained. Once the trained model is implemented in remaining parts of database, anomalies are also detected. The proposed method is applied to the IEEE 39-bus network. The results presented demonstrate its ability to successfully detect anomalies in comparison with support vector machine and decision tree algorithms.