February 23, 2024
Conference Paper

Model Agnostic Bayesian Framework for Online Anomaly/Event Detection in PMU Data


Phasor measurement units (PMU) are integral to the modernization and automation plan of the electric power industry. A PMU data signature contains system-level events (e.g., faults, generation/load change, etc.) and any measurement/device-related errors. Therefore, the reliable and resilient operation of power systems is equivalent to the quality of the PMU data and the situation awareness provided by its data signature. Despite recent progress, current state-of-the-art methods are not fool-proof and have certain limitations tracing an error/abnormality to sensor sub-components and grid systems. This is because of technical challenges imposed by the scarcity of the labeled information, loss of data quality, and non-stationarity of data. In this paper, we consider the online PMU data stream as an output of a stochastic process and pose the anomaly/event detection as a changepoint detection problem dealing with detecting parameter changes in the underlying stochastic processes. The proposed model-agnostic framework relies on: (a) feature extraction utilizing the minimum volume enclosing ellipsoids (MVEE) method from raw PMU observations and (b) a Bayesian framework of changepoint detection. The validity of the proposed methodology is discussed through numerical experiments on real-world utility-scale PMU data.

Published: February 23, 2024


Hossain R., K. Mahapatra, and J.P. Ogle. 2023. Model Agnostic Bayesian Framework for Online Anomaly/Event Detection in PMU Data. In Proceedings of the IEEE Power & Energy Society General Meeting (PESGM 2023), July 16-20, 2023, Orlando, FL, 1-5. Piscataway, New Jersey:IEEE. PNNL-SA-179850. doi:10.1109/PESGM52003.2023.10252365