Feature-Based Pmu Event Classification Under Variable Pmu Participation and Overlapping Events

Phasor Measurement Units (PMUs) stream time-synchronized and high-resolution measurements from the grid. These measurements enable the development of data-driven techniques for event detection and classification. Events occur frequently in power systems, and accurately detecting and classifying them improves grid reliability and stability. Events can be detected by varying numbers of PMUs. For example, system-wide frequency disturbances may be recorded by many PMUs across the grid, while only a few may capture localized faults. Additionally, the duration of the event can vary depending on the type of event. These variations introduce challenges for standard classification models that require uniformly sized input data. Moreover, multiple events may coincide, which increases classification complexity. Standard classifiers assign each instance to the class with the highest predicted probability, whereas overlapping events may exhibit comparable probabilities across multiple classes. In this study, to handle data size variability, we extract a wide range of features, including basic statistics, envelope areas over sliding windows, and various time–frequency domain characteristics. These features are computed across all PMUs within each event, concatenated into a vector, and used to train machine learning models, including Random Forest, XGBoost, and Multilayer Perceptron. To account for overlapping events, a probabilistic post-processing step is applied. For a given data instance, if multiple predicted class probabilities exceed 30% and the differences between them are less than 10%, the event is assigned to multiple classes. Experimental results on real-world PMU event data with five manually labeled classes demonstrate the superior performance of the Random Forest classifier, which achieves an accuracy of 95% and confirms the effectiveness of the extracted features. Applying the probabilistic post-processing step increases the overall accuracy by approximately 3%.