Graph Neural Network-Based Framework for Intelligent Monitoring and Anomaly Detection in Power Equipment Using Multi-Source Data

The rapid advancement of smart grids and the Industrial Internet of Things (IIoT) has generated massive multi-source, heterogeneous data from power systems, including real-time sensor readings, operational logs, and maintenance records. To address the challenges posed by temporal dependencies, spatial correlations, and inherent noise in such data, this study proposes a novel Graph Neural Networks (GNNs)-based anomaly detection framework for power equipment. The framework integrates multi-source data into a unified graph representation, where nodes correspond to equipment, and edges encode their relationships. A tailored GNN architecture is developed to capture spatial and temporal dependencies through graph convolution and sequential modeling, incorporating advanced techniques such as attention mechanisms. Experimental validation on real-world datasets demonstrates the model's superior performance, achieving 94.5% accuracy, 92.3% recall, and 93.4% F1-score, significantly outperforming traditional machine learning methods and baseline GNNs. These results highlight the effectiveness and robustness of the proposed approach in detecting rare and subtle anomalies in complex power systems, establishing its potential for scalable, real-time monitoring and predictive maintenance.