"Power System Multi-source Events Dataset"

"This dataset provides graph-structured time-series measurements for power system multi-source event detection, localization, and classification (ED-L-C). Generated from the IEEE 34 Node Test Feeder via electromagnetic transient simulation, it contains 7,500 samples, each consisting of 120 Hz PMU recordings across 34 nodes with 9 feature channels (three-phase voltage magnitudes, current magnitudes, and power factor cosines). The samples cover 10 distinct fault event classes with varying concurrent event multiplicities, simulating complex scenarios where multiple events occur simultaneously or in rapid succession at spatially distributed locations. Each sample is annotated with event-level labels that jointly encode the spatial location (node or edge) and event class of each concurrent event within a single label structure, enabling simultaneous inference of all three subtasks. The dataset is designed to support end-to-end AI models for joint ED-L-C learning, eliminating the need for sequential subtask decomposition."