Simulation of Power Grid Resilience: An AI Approach to Assessing Vulnerability and Aiding Real- Time Response

Abstract

Modern society's profound dependence on a continuous supply of electricity makes power grid

resilience a matter of critical national importance. The increasing frequency and intensity of

extreme weather events, coupled with aging infrastructure and evolving cyber threats, expose the

vulnerabilities of conventional power systems. Traditional resilience assessment methods, such as

N-k contingency analysis, are often static, computationally intensive, and inadequate for capturing

the complex dynamics of cascading failures in real time. This paper proposes a novel, integrated

Artificial Intelligence (AI) framework to enhance power grid resilience through proactive

vulnerability assessment and intelligent real-time response guidance. The methodology employs a

two-stage approach. First, a Graph Neural Network (GNN) is developed to model the power grid as

a complex network, learning topological and electrical features to accurately identify critical

components and predict the propagation paths of cascading failures under duress. Second, a Deep

Reinforcement Learning (DRL) agent is trained to formulate optimal, adaptive restoration

strategies following a disruption. The DRL agent utilizes the vulnerability insights from the GNN to

prioritize actions, aiming to minimize restoration time and the amount of unserved energy. The

framework's efficacy is validated through high-fidelity simulations on a standard IEEE test grid

under various simulated extreme event scenarios. The results demonstrate that the GNN model

significantly outperforms traditional methods in identifying non-obvious, high-impact

vulnerabilities. Furthermore, the AI-guided restoration strategy substantially reduces system

recovery time and energy loss compared to conventional heuristic-based response protocols. This

research underscores the transformative potential of AI to shift grid management from a reactive

to a proactive and predictive paradigm, offering a powerful new toolkit for operators to plan for

and respond to large-scale disturbances.