Transfer Learning on Synchrophasor Data for Automatically Detecting and Classifying Critical Power System Events

The advent of phasor measurement units (PMUs) has significantly enhanced the resolution of data streams available for monitoring power system dynamics. Concurrently, this data's sheer volume and variability, compounded by data quality issues and event label scarcity, introduce substantial hurdles for manual analysis and conventional machine learning applications. In this paper, we introduce a transfer learning (TL) framework that: a) enhances simple classifiers to automatically detect and classify line faults, b) applies TL to modal features for robust detection of frequency events across large simulated and field-recorded datasets, and c) leverages a novel pixelated image-based representation of oscillation signatures to train neural networks for TL-driven oscillation detection. Furthermore, we test whether the TL models developed for each of these tasks on one dataset may be directly applied to other tasks using different datasets. This effort reveals the intricacies of automatically analyzing field-recorded PMU data guiding future research in automated power system event analysis using TL.