A Novel Centrality-Driven Machine Learning Approach for Clustering Critical Nodes in Cyber-Physical Power Systems

Modern power distribution systems are increasingly integrated with information and communication technologies (ICT), forming complex cyber–physical power systems (CPPSs). Ensuring the resilience of these interdependent systems requires identifying and clustering components whose failure could critically impact both the cyber and physical layers. This paper introduces a novel centrality-driven machine learning framework that, for the first time, combines multiple complex network centrality metrics—degree, closeness, and betweenness—with K-means clustering to identify and cluster critical nodes in CPPSs. By using these centrality metrics as multidimensional feature vectors, the proposed approach captures complementary aspects of node importance and enhances the interpretability and accuracy of clustering results compared with traditional single-metric analyses. The proposed methodology is validated on a real-world CPPS in Northeastern Italy, modeled using graph theory, demonstrating its effectiveness in identifying node clusters whose failure could severely impact system performance. The method provides a scalable and transparent analytical tool to support resilience assessment, resource prioritization, and infrastructure protection planning in modern CPPSs.