Automated Detection and Classification of Critical Power System Events using ML on PMU Data

The increasing amount of data from phasor measurement units (PMUs) available in power systems around the world lately has raised the need for automated, real-time event assessment tools that can help operators extract relevant actionable information in real-time. This paper investigates the automated detection and classification of line faults, fundamental frequency deviations, and oscillation events using PMU data and machine learning (ML) methods. The study advances previous research by implementing a variety of ML algorithms and investigating the effects of feature engineering, classifier selection, and dataset sources on algorithm performance. Using both field-recorded and simulated data sources, it emphasizes the relevance of data quality and dataset balance for algorithm training purposes. The results underline importance of the well-labeled dataset, effective feature engineering, and balanced data model training using field and simulated data to achieve automated and reliable event detection and classification using ML.