Improving the Reliability of Distributed Energy Systems based on Graph Neural Networks

Predicting the technical condition of distributed energy systems is a pressing issue in the energy industry and should be addressed through systems analysis and intelligent processing of structured data. Current analysis methods typically ignore the graph-based nature of distributed energy systems and work only with tabular data. A graph model of an electric power system accounts for network topology and internode dependencies in accordance with Kirchhoff’s laws. This article describes a comparative study of machine learning methods for tabular data and graph neural networks (GNNs) for forecasting node failure. IEEE test power systems were used as experimental data, specifically network topology and node parameters such as active and reactive load, voltage magnitude and angle. The results confirm the high effectiveness of graph approaches for predictive diagnostics, monitoring, and operational management of distributed energy systems.