Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robustness Analysis of AI Models in Critical Energy Systems

Pantelis Dogoulis, Matthieu Jimenez, Salah Ghamizi, Maxime Cordy, Yves Le Traon

TL;DR

This study reveals substantial robustness gaps in AI-based power-flow predictors when tested under the $N-1$ security criterion. By combining a graph-theoretic analysis of node connectivity with extensive experiments on IEEE 14 and IEEE 118 grids, it shows that line disconnections can cause 10–100× increases in prediction error, particularly for high-degree nodes. The authors demonstrate that including $N-1$ scenarios in training (mixed training) markedly improves robustness, suggesting practical scenario-aware data generation as a key direction. The work highlights the need to integrate topology changes into model design for reliable deployment in critical energy systems and points to future work on topology-aware sampling and more complex disturbances.

Abstract

This paper analyzes the robustness of state-of-the-art AI-based models for power grid operations under the $N-1$ security criterion. While these models perform well in regular grid settings, our results highlight a significant loss in accuracy following the disconnection of a line.%under this security criterion. Using graph theory-based analysis, we demonstrate the impact of node connectivity on this loss. Our findings emphasize the need for practical scenario considerations in developing AI methodologies for critical infrastructure.

Robustness Analysis of AI Models in Critical Energy Systems

TL;DR

This study reveals substantial robustness gaps in AI-based power-flow predictors when tested under the security criterion. By combining a graph-theoretic analysis of node connectivity with extensive experiments on IEEE 14 and IEEE 118 grids, it shows that line disconnections can cause 10–100× increases in prediction error, particularly for high-degree nodes. The authors demonstrate that including scenarios in training (mixed training) markedly improves robustness, suggesting practical scenario-aware data generation as a key direction. The work highlights the need to integrate topology changes into model design for reliable deployment in critical energy systems and points to future work on topology-aware sampling and more complex disturbances.

Abstract

This paper analyzes the robustness of state-of-the-art AI-based models for power grid operations under the security criterion. While these models perform well in regular grid settings, our results highlight a significant loss in accuracy following the disconnection of a line.%under this security criterion. Using graph theory-based analysis, we demonstrate the impact of node connectivity on this loss. Our findings emphasize the need for practical scenario considerations in developing AI methodologies for critical infrastructure.
Paper Structure (16 sections, 6 equations, 2 figures, 3 tables)

This paper contains 16 sections, 6 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Problem Description
  4. Newton-Raphson method
  5. Related Work
  6. Experimental Setup
  7. Datasets
  8. Models
  9. Evaluation Protocol
  10. Metric
  11. Implementation Details
  12. Results
  13. $N-1$ case
  14. Graph Analysis
  15. Mix Training
  16. ...and 1 more sections

Figures (2)

  • Figure 1: Example of a toy power grid featuring two generators and three loads. In the upper section of the image, an overflow in the transmission lines is depicted. To address this issue, the overloaded line is disconnected from the grid causing a topological change and necessitating a recalculation of the grid's state.
  • Figure 2: Visual comparison of the model's prediction in the $N$ case and three different $N-1$ cases for the IEEE 14 dataset. In (a), the standard topology is shown. In (b), the line from bus 1 to bus 2 is disconnected. In (c), the line from bus 1 to bus 3 is disconnected. In (d), the line from bus 1 to bus 4 is disconnected.