The emerging class of waveform measurement units (WMUs) can enhance event situational awareness in distribution grids by better characterizing the signatures of events and reporting higher sampling rates compared to traditional measurement units. This study aims to improve the classification and localization of power quality events and transient disturbances in active distribution grids by integrating time-frequency characterization of WMU data and a grid-aware learning algorithm. A time-frequency decomposition technique with a moving window is proposed to extract complex frequencies and residues from a limited set of available WMU data. This endeavor seeks to characterize the unique signatures of events occurring in various locations. Extracted features are then used as nodal graph signals in a grid-aware autoregressive moving average (ARMA) graph convolution to classify event type and location. Despite the scarcity of waveform measurements, the proposed grid-aware model captures the spatial relationship between derived signatures of measured signals at different nodes. Through our numerical studies, the proposed strategy improves the classification and localization of events such as capacitor bank switching, abrupt load changes, generation outages, and various types of faults (e.g., arcing high impedance faults), when compared to state-of-the-art models. The proposed method is validated under various scenarios, including noisy data, different measurement configurations and reporting rates, diverse operational conditions, and unbalanced feeders.
