.png){kind=icon}
.png){kind=icon}
Return
Physics-Informed Multi-Scale Adaptive Graph Learning for Unified Event Detection, Localization, and Classification in Power System
.png){kind=icon}
PRE
AI
W
J
杨
X
DOI：10.1109/tii.2026.3682628.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This article proposes a physics-informed multi-scale adaptive graph neural network (PI-MSA) for rapid event detection, localization, and classification (ED-L-C) of single and multi-source coupling events in power systems to prevent large-scale failures. Prior work decomposes ED-L-C tasks into separate classification subtasks. Instead, this work frames ED-L-C as a unified regression problem through a spatial-aware and task-coupled three-dimensional target formulation, enabling one-shot inference of all abnormal events and their class probabilities at both node and edge levels. By incorporating physical grid topology and shortest-path admittance encoding through parallel multi-scale networks, this unified one-stage pipeline enables joint optimization and learning of shared graph features as well as progressive inter-task dependencies. Experimental validation on five power systems of varying scales and configurations demonstrates that PI-MSA achieves over 75% mean average precision and 90% localization accuracy across various scenarios, outperforming traditional methods by over 40% and exceeding state-of-the-art methods by at least 5%. The model maintains robust and accurate event analysis under topology changes and reduced observability conditions while achieving computational efficiency at 225 frames per second, enabling real-time monitoring and analysis for modern power systems.
Keywords:
Event detection localization and classification
modern power system
multi-source events
multi-scale adaptive graph network
physics-informed
Journal
IF:
9.9
Papers:
8.3K
Citations:
6.0W
Organization
Cited Papers