Graph Convolutional Network Based Power System Risk Assessment

In modern electric power systems, accurate and fast risk assessment is of paramount importance. Traditional methods, relying on iterative computations, have a substantial computational demand, hindering their efficiency in real-time evaluations. This paper introduces a novel risk assessment approach using Graph Convolutional Networks (GCNs), which leverages the graph structure of power system data, thereby enhancing the efficiency of risk assessment and further bolstering the power grid's risk analysis and early warning capabilities. This study utilized the IEEE RTS-79 system to generate representative samples, designed experiments to fine-tune model hyperparameters via grid search, and subsequently trained the model. A comparative analysis was then carried out among various traditional machine learning methods and deep learning models. The results indicate that GCNs exhibit exceptional generalization capabilities, meeting the demands for real-time risk assessment in modern power systems.