| Title |
Latent Embedding based Open-Set Classification of Power System Events via Graph Attention Network |
| Authors |
이주석(Juseok Lee) ; 박천규(Cheonkyu Park) ; 김도인(Do-In Kim) |
| DOI |
https://doi.org/10.5370/KIEE.2026.75.9.2083 |
| Keywords |
Power System Events; Open-set Recognition(OSR); Graph Attention Networks(GAT); Phasor Measurement Units(PMU); Situational Awareness. |
| Abstract |
This study proposes a robust open-set classification framework for power system events using Graph Attention Networks (GAT). To address limited Phasor Measurement Unit (PMU) deployment, we implement a physics-informed graph reduction strategy for computational efficiency. Transient features are extracted via Discrete Wavelet Transform (DWT) to capture multi-resolution signatures from voltage and frequency measurements. The primary contribution is a multi-stage discrimination mechanism that overcomes the limitations of conventional closed-set models. By integrating synthetic steady-state data during training and utilizing L2-normalized latent embeddings with class-specific distance thresholds, the framework effectively identifies unseen disturbances. Simulation results on the IEEE 68-bus system demonstrate a reliable solution for situational awareness in evolving power grid environments. |