Real-Time Anomaly Detection in Distribution Grids Using Long Short Term Memory Network

The massive amount of data generated by smart meters provides opportunities to better monitor and control power grids in real-time. However, making use of such enormous amounts of data can be a challenge. Sensor measurements collected in power systems can be anomalous, reflecting sensor malfunctions, power system disturbances, or any other problems that may cause abnormal readings. With the rapid deployment of distributed energy resources, traditional methods for protecting the grid, which relies on emergency load tripping through relay actions, have limited performance. For example, relay actions may be delayed. In addition, they cannot detect anomalies that are within the required voltage range. To address these shortcomings, this article proposes a data-driven framework based on a long-short-term memory network model to directly detect anomalies in distribution systems using voltage magnitude measurements. The proposed solution is validated against known anomalies using data from a real distribution grid. The simulation results show highly accurate anomaly identification.