Robust Event Classification Using Imperfect Real-World PMU Data

This paper studies robust event classification using imperfect real-world phasor measurement unit (PMU) data. By analyzing the PMU data, we find it is challenging to directly use this dataset for classifiers due low data quality observed in measurements and logs. To address these challenges, develop a novel machine learning framework training classifiers, which consists of three main steps: preprocessing, fine-grained extraction, feature engineering. Specifically, preprocessing step addresses issues (e.g., bad missing data); extraction step, model-free detection method developed accurately localize events from inaccurate timestamps logs; engineering constructs features based on patterns different types, order improve performance interpretability classifiers. Based proposed framework, workflow streaming into system real time. Using can be efficiently trained many off-the-shelf lightweight models. Numerical experiments Western Interconnection U.S power transmission grid show that under achieve high accuracy while being against low-quality