Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On the Definition of Robustness and Resilience of AI Agents for Real-time Congestion Management

Timothy Tjhay, Ricardo J. Bessa, Jose Paulos

TL;DR

The paper addresses the lack of quantitative methods for assessing robustness and resilience of RL agents in real-time congestion management under the EU AI Act. It introduces a framework that uses Grid2Op and three perturbation agents—Random Perturbation Agent, Gradient Estimation Perturbation Agent, and RL-based Perturbation Agent—to simulate natural and adversarial input disruptions, accompanied by robust and resilient performance metrics. A case study on the IEEE-14 bus system with Grid2Op demonstrates the framework’s ability to identify vulnerabilities and compare how different perturbation strategies affect stability, reward, and recovery, highlighting that RL-based perturbations can expose deeper weaknesses than random or gradient-based perturbations. The work provides a practical methodology for evaluating AI-assisted grid operations at test-time, with implications for standardization, safety certification, and improvement of AI robustness and resilience in critical infrastructure.

Abstract

The European Union's Artificial Intelligence (AI) Act defines robustness, resilience, and security requirements for high-risk sectors but lacks detailed methodologies for assessment. This paper introduces a novel framework for quantitatively evaluating the robustness and resilience of reinforcement learning agents in congestion management. Using the AI-friendly digital environment Grid2Op, perturbation agents simulate natural and adversarial disruptions by perturbing the input of AI systems without altering the actual state of the environment, enabling the assessment of AI performance under various scenarios. Robustness is measured through stability and reward impact metrics, while resilience quantifies recovery from performance degradation. The results demonstrate the framework's effectiveness in identifying vulnerabilities and improving AI robustness and resilience for critical applications.

On the Definition of Robustness and Resilience of AI Agents for Real-time Congestion Management

TL;DR

The paper addresses the lack of quantitative methods for assessing robustness and resilience of RL agents in real-time congestion management under the EU AI Act. It introduces a framework that uses Grid2Op and three perturbation agents—Random Perturbation Agent, Gradient Estimation Perturbation Agent, and RL-based Perturbation Agent—to simulate natural and adversarial input disruptions, accompanied by robust and resilient performance metrics. A case study on the IEEE-14 bus system with Grid2Op demonstrates the framework’s ability to identify vulnerabilities and compare how different perturbation strategies affect stability, reward, and recovery, highlighting that RL-based perturbations can expose deeper weaknesses than random or gradient-based perturbations. The work provides a practical methodology for evaluating AI-assisted grid operations at test-time, with implications for standardization, safety certification, and improvement of AI robustness and resilience in critical infrastructure.

Abstract

The European Union's Artificial Intelligence (AI) Act defines robustness, resilience, and security requirements for high-risk sectors but lacks detailed methodologies for assessment. This paper introduces a novel framework for quantitatively evaluating the robustness and resilience of reinforcement learning agents in congestion management. Using the AI-friendly digital environment Grid2Op, perturbation agents simulate natural and adversarial disruptions by perturbing the input of AI systems without altering the actual state of the environment, enabling the assessment of AI performance under various scenarios. Robustness is measured through stability and reward impact metrics, while resilience quantifies recovery from performance degradation. The results demonstrate the framework's effectiveness in identifying vulnerabilities and improving AI robustness and resilience for critical applications.

Paper Structure

This paper contains 15 sections, 14 equations, 5 figures, 3 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Congestion Management Use Case
  3. Assessment Methodology
  4. Perturbation Agents
  5. Random perturbation agent (RPA)
  6. Gradient estimation perturbation agent (GEPA)
  7. RL-based perturbation agent (RLPA)
  8. Robustness Metrics
  9. Resilience Metrics
  10. Numerical Results
  11. Case-study
  12. Analysis
  13. Robustness
  14. Resilience
  15. Conclusions

Figures (5)

  • Figure 1: Degradation and restorative state during the testing of the AI system
  • Figure 2: Robustness metrics as a percentage of the unperturbed performance of the AI system
  • Figure 3: Visualization of values in the grid that are vulnerable to perturbation
  • Figure 4: Example of the reward obtained in each step between step 2450 and 3200 of an episode in an environment with the GEPA
  • Figure 5: Example of cosine similarity to unperturbed state in each step between steps 350 and 750 of an episode in the environment with the RLPA