A Method for Node Importance Evaluation in Power Systems Based on Graph Neural Networks

Modern power systems are increasingly complex and require accurate evaluation of node importance to ensure grid stability and efficient resource allocation. Traditional methods relying on basic topological metrics often fail to capture dynamic interdependencies among grid components. In this paper, we propose a novel Graph Neural Network (GNN)-based approach that leverages self-supervised learning to integrate topological structure and multi-dimensional node attributes for real-time node importance evaluation without labeled data. Experimental results on real-world power grid data show that our method significantly outperforms traditional approaches in rank accuracy, precision, recall, and F1 score, highlighting its potential for smart grid applications.