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

2025-04-17

AI Summary

Current AI agents deployed in high-risk real-time congestion management of power systems lack quantitative assessment methods for robustness and resilience. Method: We propose the first decoupled evaluation paradigm that separately models environmental states and observation perturbations, implemented within the Grid2Op digital twin platform. Customized perturbation agents simulate natural and adversarial input disturbances—without altering the underlying system state—and formally define and quantify robustness (stability plus reward impact) and resilience (capacity to recover after performance degradation). Results: Experiments identify critical vulnerabilities of RL policies under voltage limit violations and line tripping events; the framework improves agent robustness by 37% and reduces average recovery time by 52%, thereby filling a methodological gap in regulatory AI assessment frameworks—such as the EU AI Act—for high-risk AI systems.

Technology Category

• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness
• Multiagent Systems: Adversarial Agents
• Intelligent Robots: State Estimation

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

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.

Problem

• Defining robustness and resilience for AI in congestion management
• Quantifying AI performance under natural and adversarial disruptions
• Assessing stability, reward impact, and recovery in AI systems

Innovation

• Framework evaluates robustness and resilience quantitatively
• Grid2Op simulates disruptions without altering environment
• Metrics measure stability, reward impact, and recovery