Edge-level explainable graph neural networks with network centric features for transmission line failure prediction in power grids

Abstract

Cascading outages in high-voltage power grids pose a severe risk, causing blackouts with global economic losses estimated at $\approx \$100$ billion annually. These outages disrupt economic activity and impact energy efficiency and sustainability goals by necessitating less efficient backup generation and hindering the integration of renewable energy sources. This paper introduces a pioneering, explainable Graph Neural Network (GNN) framework for edge-level transmission line failure prediction, addressing a critical gap in grid resilience analytics. Our work presents two key innovations: first, it is the first framework of its kind to be systematically validated across four standard power grid benchmarks (IEEE-24, 39, 118, and the UK grid), demonstrating robust generalization. Second, it advances an interdisciplinary approach by uniquely integrating network science principles with deep learning, augmenting traditional electrical data with topological descriptors like betweenness centrality and load ratio. This fusion enhances the predictive power of three GNN architectures: GINE, GAT, and EdgeAwareGC. On the IEEE-24 test system, this integration boosts macro-F1 scores from 0.498 to 0.871 for EdgeAwareGC and from 0.335 to 0.859 for GINE. This scalability and effectiveness are further demonstrated on the highly imbalanced IEEE-118 network, where EdgeAwareGC achieves a strong F1 score of 0.553. Explainability techniques such as gradient-based attribution and GNNExplainer uncover key physical and topological predictors, providing actionable guidance for grid operators. Our findings inform grid modernization policies, supporting initiatives like the U.S. Department of Energy’s Grid Resilience and Innovation Partnerships (GRIP) program. This work highlights the potential of GNNs to substantially improve power system reliability, contributing to more sustainable and efficient energy infrastructure, and aligning with global efforts toward decarbonization and enhanced energy security.