Methodology for Machine Learning Anomaly Detection in Phasor Measurement Unit Data

This paper presents a methodology for the development and benchmarking of Machine Learning (ML) algorithms for Anomaly Detection (AD) in Phasor Measurement Unit (PMU) Data. A rigorous method, which is presently lacking in literature, is presented for obtaining and processing PMU data, and for ML AD algorithm development, training, and performance benchmarking. Anomalies in power system frequency, observed through PMU data, are investigated in this work. ML was used to develop three AD algorithms under three different feature transformations: Fano Factor, Matrix Profile, and Z-Score. Benchmarking results show all algorithms are suitable for real-time applications and the Fano Factor algorithm is highest performing. Additionally, datasets used in this work have been made publicly available to replicate the results in this work, and further development of other PMU based AD.