MULTI-VIEW GRAPH CONVOLUTIONAL NEURAL NETWORKS FOR ABNORMAL MEASUREMENT DATA IDENTIFICATION IN POWER SYSTEM

This paper introduces a Multi-View Graph Convolutional Network (MVGCN) for anomaly measurement data recognition in power systems.Designed to address challenges in transient anomaly detection, the model enhances feature extraction adaptability and captures complex transient patterns.The architecture integrates graph convolutional operations with multi-view learning, incorporating diverse structural representations of power networks.An attention-based feature aggregation mechanism dynamically fuses multi-view information during graph convolution, prioritizing salient features while suppressing noise.This significantly improves the model's representational capacity and robustness against data irregularities common in grid environments.Evaluations on ENZYMES and PROTEINS datasets-selected for their structural relevancedemonstrate MVGCN's advantages over state-of-the-art baselines.The model more accurately identifies correlations between anomalies and sensor nodes, enabling precise root-cause diagnosis.It also achieves notably higher computational efficiency, substantially reducing training and inference runtime through optimized attention fusion and parallel view processing.These capabilities position MVGCN as a scalable solution for real-time anomaly detection in dynamic smart grids, enhancing grid resilience and operational reliability through rapid, interpretable diagnostics.