Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Physics-Informed Graph Neural Jump ODEs for Cascading Failure Prediction in Power Grids

Birva Sevak, Shrenik Jadhav, Van-Hai Bui

Abstract

Cascading failures in power grids pose severe risks to infrastructure reliability, yet real-time prediction of their progression remains an open challenge. Physics-based simulators require minutes to hours per scenario, while existing graph neural network approaches treat cascading failures as static classification tasks, ignoring temporal evolution and physical laws. This paper proposes Physics-Informed Graph Neural Jump ODEs (PI-GN-JODE), combining an edge-conditioned graph neural network encoder, a Neural ODE for continuous power redistribution, a jump process handler for discrete relay trips, and Kirchhoff-based physics regularization. The model simultaneously predicts edge and node failure probabilities, severity classification, and demand not served, while an autoregressive extension enables round-by-round temporal cascade prediction. Evaluated on the IEEE 24-bus and 118-bus systems with 20,000 scenarios each, PI-GN-JODE achieves a Precision--Recall Area Under the Curve of 0.991 for edge failure detection, 0.973 for node failure detection, and a coefficient of determination of 0.951 for demand-not-served regression on the 118-bus system, outperforming a standard graph convolutional network baseline (0.948, 0.925, and 0.912, respectively). Ablation studies reveal that the four components function synergistically, with the physics-informed loss alone contributing +9.2 points to demand-not-served regression. Performance improves when scaling to larger grids, and the architecture achieves the highest balanced accuracy (0.996) on the PowerGraph benchmark using data from a different simulation framework.

Physics-Informed Graph Neural Jump ODEs for Cascading Failure Prediction in Power Grids

Abstract

Cascading failures in power grids pose severe risks to infrastructure reliability, yet real-time prediction of their progression remains an open challenge. Physics-based simulators require minutes to hours per scenario, while existing graph neural network approaches treat cascading failures as static classification tasks, ignoring temporal evolution and physical laws. This paper proposes Physics-Informed Graph Neural Jump ODEs (PI-GN-JODE), combining an edge-conditioned graph neural network encoder, a Neural ODE for continuous power redistribution, a jump process handler for discrete relay trips, and Kirchhoff-based physics regularization. The model simultaneously predicts edge and node failure probabilities, severity classification, and demand not served, while an autoregressive extension enables round-by-round temporal cascade prediction. Evaluated on the IEEE 24-bus and 118-bus systems with 20,000 scenarios each, PI-GN-JODE achieves a Precision--Recall Area Under the Curve of 0.991 for edge failure detection, 0.973 for node failure detection, and a coefficient of determination of 0.951 for demand-not-served regression on the 118-bus system, outperforming a standard graph convolutional network baseline (0.948, 0.925, and 0.912, respectively). Ablation studies reveal that the four components function synergistically, with the physics-informed loss alone contributing +9.2 points to demand-not-served regression. Performance improves when scaling to larger grids, and the architecture achieves the highest balanced accuracy (0.996) on the PowerGraph benchmark using data from a different simulation framework.
Paper Structure (25 sections, 19 equations, 6 figures, 9 tables)

This paper contains 25 sections, 19 equations, 6 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Problem Formulation
  4. Cascade Simulation and Dataset Generation
  5. Input Representation
  6. PI-GN-JODE Architecture
  7. Edge-Conditioned GNN Encoder
  8. Neural ODE for Continuous Dynamics
  9. Jump Process Handler
  10. Prediction Heads and Physics Regularization
  11. Multi-Round Autoregressive Extension
  12. Training and Evaluation
  13. Loss Function and Optimization
  14. Evaluation Metrics
  15. Ablation and Baseline Design
  16. ...and 10 more sections

Figures (6)

  • Figure 1: PI-GN-JODE architecture. The input graph is processed by a GNN encoder, neural ODE, jump handler, and physics-regularized multi-task decoders to generate four predictions.
  • Figure 2: Multi-round training on the IEEE 24-bus system over 75 epochs. (a) Loss with best epoch at 42. (b) Edge and node PR-AUC. (c) DNS $R^2$, severity F1, and teacher-forcing ratio.
  • Figure 3: Component build-up on IEEE 118-bus. (a) Edge PR-AUC and (b) DNS $R^2$ improve overall, with a brief dip after the Neural ODE that is recovered by the Jump Handler and Physics Loss.
  • Figure 4: One-shot performance comparison between the IEEE 24-bus and IEEE 118-bus systems. Performance improves on all metrics for the larger grid, with the largest gains in DNS $R^2$ and node PR-AUC.
  • Figure 5: Performance comparison on the PowerGraph IEEE 24-bus benchmark varbella2024powergraph. (a) Binary balanced accuracy for cascade detection. (b) DNS mean squared error on a log scale. PI-GN-JODE achieves the highest accuracy with the second-lowest DNS error.
  • ...and 1 more figures