Abstract
This paper studies robust event classification using imperfect real-worldphasor measurement unit (PMU) data. By analyzing the real-world PMU data, wefind it is challenging to directly use this dataset for event classifiers dueto the low data quality observed in PMU measurements and event logs. To addressthese challenges, we develop a novel machine learning framework for trainingrobust event classifiers, which consists of three main steps: datapreprocessing, fine-grained event data extraction, and feature engineering.Specifically, the data preprocessing step addresses the data quality issues ofPMU measurements (e.g., bad data and missing data); in the fine-grained eventdata extraction step, a model-free event detection method is developed toaccurately localize the events from the inaccurate event timestamps in theevent logs; and the feature engineering step constructs the event featuresbased on the patterns of different event types, in order to improve theperformance and the interpretability of the event classifiers. Based on theproposed framework, we develop a workflow for event classification using thereal-world PMU data streaming into the system in real-time. Using the proposedframework, robust event classifiers can be efficiently trained based on manyoff-the-shelf lightweight machine learning models. Numerical experiments usingthe real-world dataset from the Western Interconnection of the U.S powertransmission grid show that the event classifiers trained under the proposedframework can achieve high classification accuracy while being robust againstlow-quality data.