[Paper Review] Topology-Aware Spatio-Temporal Graph Transformer for Predicting Smart Grid Failures

A. Lê, Phat K. Huynh+4 more|arXiv (Cornell University)|Jan 6, 2026

Smart Grid Security and Resilience0 citations

TL;DR

Introduces a topology-aware spatio-temporal graph transformer (ST-GT) that integrates physical transmission topology, static descriptors, and PMU sequences to predict smart grid failures with perfect recall and strong F1 scores, outperforming XGBoost baselines.

ABSTRACT

Smart grid infrastructure needs improved resilience and preventive maintenance through more accurate predictions. Current methodologies lack accurate representation of spatio-temporal-causal interdependencies and class imbalance in failure prediction tasks. This study introduces a Topology-Aware Spatio-Temporal Graph Transformer (ST-GT) architecture that overcomes existing limitations by using three main innovations: (1) directly incorporating physical transmission network topology into the transformer attention mechanism to identify spatial failure propagation patterns; (2) unified processing of static topological descriptors and (3) temporal Phasor Measurement Units (PMU) sequences in an end-to-end framework. The ST-GT model exhibited outstanding performance in five-fold cross-validation across 10 substations, attaining perfect recall (1.000 $\pm$ 0.001) and an F1-score of 0.858 $\pm$ 0.009, markedly surpassing XGBoost baselines (0.683 accuracy/F1). Perfect recall guarantees that no critical failures are overlooked, which is essential for grid safety; however, it may lead to an increase in false alarm rates. This framework integrates temporal dynamics modeling with spatial graph awareness for critical infrastructure monitoring. It offers interpretable insights into failure propagation pathways and enhances maintenance strategies. Future research focuses on developing cost-weighted loss functions for precision-recall trade-off enhancement, implementing real-time monitoring systems with uncertainty quantification, and creating cost-sensitive frameworks balancing false alarm expenditures with failure consequences. The methodology success suggests its potential for wider application in critical infrastructure areas requiring spatial temporal failure prediction.

Motivation & Objective

• Motivate improved resilience and preventive maintenance for smart grids through accurate failure prediction.
• Address limitations of existing methods in representing spatio-temporal-causal interdependencies and class imbalance.
• Develop a unified end-to-end model that fuses topology, static descriptors, and temporal PMU data.
• Provide interpretable insights into failure propagation pathways to inform maintenance strategies.

Proposed method

• Directly incorporate physical transmission network topology into the transformer attention mechanism to identify spatial failure propagation patterns.
• Unified processing of static topological descriptors and temporal PMU sequences within an end-to-end framework.
• Employ a spatio-temporal graph transformer (ST-GT) architecture to model dynamics across 10 substations with five-fold cross-validation.
• Analyze performance against baselines (e.g., XGBoost) using metrics such as recall and F1-score.
• Discuss potential extensions including cost-weighted loss, real-time monitoring with uncertainty quantification, and cost-sensitive frameworks.

Experimental results

Research questions

• RQ1 How can topology be integrated into a transformer to capture spatial propagation patterns of grid failures?
• RQ2 Can a unified model leverage static topological descriptors and temporal PMU data to improve failure prediction?
• RQ3 What is the performance of the proposed ST-GT approach compared to traditional baselines on smart grid failure tasks?

Key findings

• ST-GT achieves recall of 1.000 ± 0.001 and F1-score of 0.858 ± 0.009 in five-fold CV across 10 substations.
• The model substantially surpasses XGBoost baselines (0.683 accuracy/F1).
• The approach provides interpretable insights into failure propagation pathways and supports preventive maintenance decisions.
• Perfect recall ensures no critical failures are missed, with potential trade-offs in false alarm rates.
• Future work includes cost-weighted loss, uncertainty-aware real-time monitoring, and cost-sensitive deployment.