Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Trustworthiness Layer for Foundation Models in Power Systems: Application for N-k Contingency Assessment

Antonio Alcántara, Spyros Chatzivasileiadis

TL;DR

The paper tackles the challenge of trustworthy, real-time contingency analysis in power systems by turning GridFM outputs into statistically valid uncertainty intervals using a Stratified Conformal Prediction layer. It couples a physics-informed fine-tuning regime with a stratified calibration scheme that adapts prediction intervals to contingency severity and grid element, ensuring safe, AC-faithful state reconstruction. On IEEE-24 and IEEE-118 systems, the approach yields recall values exceeding $0.9$ for congestion screening, delivers up to $2$–$3$x precision gains over DC Power Flow, and achieves up to $18\times$ faster inference than AC Power Flow while providing full AC state. Importantly, the method generalizes to unseen high-order contingencies up to $N$-$5$ when trained on lower-order outages, highlighting practical impact for real-time grid security with quantified uncertainty.

Abstract

This work introduces for the first time, to our knowledge, a trustworthiness layer for foundation models in power systems. Using stratified conformal prediction, we devise adaptive, statistically valid confidence bounds for each output of a foundation model. For regression, this allows users to obtain an uncertainty estimate for each output; for screening, it supports conservative decisions that minimize false negatives. We demonstrate our method by enhancing GridFM, the first open-source Foundation Model for power systems, with statistically valid prediction intervals instead of heuristic error margins. We apply it for N-k contingency assessment, a combinatorial NP-Hard problem. We show that trustworthy GridFM can offer richer and more accurate information than DC Power Flow, having 2x-3x higher precision, while running up to 18x faster than AC Power Flow for systems up to 118 buses. Moving a step further, we also examine the ability of trustworthy GridFM to generalize to unseen high-order contingencies: through a rigorous analysis, we assess how a model trained on N-1 or N-2 outages extrapolates to unseen contingencies up to N-5.

Trustworthiness Layer for Foundation Models in Power Systems: Application for N-k Contingency Assessment

TL;DR

The paper tackles the challenge of trustworthy, real-time contingency analysis in power systems by turning GridFM outputs into statistically valid uncertainty intervals using a Stratified Conformal Prediction layer. It couples a physics-informed fine-tuning regime with a stratified calibration scheme that adapts prediction intervals to contingency severity and grid element, ensuring safe, AC-faithful state reconstruction. On IEEE-24 and IEEE-118 systems, the approach yields recall values exceeding for congestion screening, delivers up to –x precision gains over DC Power Flow, and achieves up to faster inference than AC Power Flow while providing full AC state. Importantly, the method generalizes to unseen high-order contingencies up to - when trained on lower-order outages, highlighting practical impact for real-time grid security with quantified uncertainty.

Abstract

This work introduces for the first time, to our knowledge, a trustworthiness layer for foundation models in power systems. Using stratified conformal prediction, we devise adaptive, statistically valid confidence bounds for each output of a foundation model. For regression, this allows users to obtain an uncertainty estimate for each output; for screening, it supports conservative decisions that minimize false negatives. We demonstrate our method by enhancing GridFM, the first open-source Foundation Model for power systems, with statistically valid prediction intervals instead of heuristic error margins. We apply it for N-k contingency assessment, a combinatorial NP-Hard problem. We show that trustworthy GridFM can offer richer and more accurate information than DC Power Flow, having 2x-3x higher precision, while running up to 18x faster than AC Power Flow for systems up to 118 buses. Moving a step further, we also examine the ability of trustworthy GridFM to generalize to unseen high-order contingencies: through a rigorous analysis, we assess how a model trained on N-1 or N-2 outages extrapolates to unseen contingencies up to N-5.
Paper Structure (27 sections, 10 equations, 2 figures, 7 tables)

This paper contains 27 sections, 10 equations, 2 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Foundation Models for Power Systems
  4. Machine Learning for Contingency Analysis
  5. Uncertainty Quantification
  6. Problem Formulation
  7. State Prediction
  8. Physics-Consistent Line Flow Calculation
  9. Congestion Screening
  10. The Standard Approach: DC Power Flow
  11. Methodology: Calibrated and Reliable Foundation Models
  12. The GridFM Architecture
  13. Fine-tuning Strategy
  14. Bounding with Conformal Prediction
  15. General Formulation
  16. ...and 12 more sections

Figures (2)

  • Figure 1: Recall for line congestion screening under different contingency levels on IEEE-118 system. Results for the different GridFM models, with and without conformalization.
  • Figure 2: Precision–Recall trade-off for GridFM on the IEEE-118 system (overall metrics). Dots indicate point estimates; squares indicate 90%/95% conformal settings; the gray cross is the DCPF baseline. Lines connect each model family across settings.