LSTM Based Hybrid Neural Network for PMU Data Forecasting and Anomaly Detection

The electric power grid is subject to several vulnerabilities, which have an adverse impact on the reliability of the system. Some of the vulnerabilities that the grid faces are anomalous events that escalate to the failure of the power system. System monitoring is thus, critical during these events; immediate control of these events is only possible with real-time situational awareness of the grid conditions. Phasor Measurement Unit (PMU) data provides real-time measurements of electrical quantities across the power system, and the high sampling rate of which further provides accurate visibility of the system’s conditions. This paper proposes a Long Short-Term Memory based Hybrid Neural Network (LSTM-HNN) to forecast and detect anomalies in PMU data with the goal of maintaining the reliability of the grid. The proposed LSTM-HNN model consists of the combination of an LSTM Neural Network and an LSTM autoencoder. The performance of the proposed hybrid model is validated through error metrics and accuracy measures against baseline models. The experimental simulation results demonstrate that the proposed LSTM-HNN model for PMU data forecasting and PMU anomaly detection provides more accurate analysis of the system conditions.