Online Anomaly Detection Using Machine Learning and HPC for Power System Synchrophasor Measurements
- BATTELLE (PACIFIC NW LAB)
This paper presents an online early anomaly detection framework for phasor measurement units (PMUs) to monitor the power system dynamics and help prevent blackouts using machine learning approaches. Dynamical machine learning solutions including state space model and Kalman filter are conducted in this study to learn the nonlinear and nonstationary PMU measurements and accurately predict system behaviors in real-time. The anomalies can be detected within seconds by comparing the predicted system behaviors with the real system observations. The method proposed in this framework trains PMU data with a given time window (e.g., 5 seconds) using a dynamic nonlinear model, and then predicts system behaviors during the following time window. High prediction accuracy is achieved by applying the dynamic nonlinear model to the real-world PMU measurements – the anomalies detected are successfully validated given the recorded real-world events. High-performance-computing (HPC) techniques are utilized to further reduce computational time to provide real time power system situational awareness.
- Research Organization:
- Pacific Northwest National Lab. (PNNL), Richland, WA (United States)
- Sponsoring Organization:
- USDOE
- DOE Contract Number:
- AC05-76RL01830
- OSTI ID:
- 1492435
- Report Number(s):
- PNNL-SA-131920
- Resource Relation:
- Conference: 2018 IEEE International Conference on Probabilistic Methods Applied to Power Systems (PMAPS), June 24-28, 2018, Boise, ID
- Country of Publication:
- United States
- Language:
- English
Similar Records
Pattern Mining and Anomaly Detection based on Power System Synchrophasor Measurements
Anomaly Detection, Localization and Classification using Drifting Synchrophasor Data Streams