Problem

The increasing frequency and severity of power outages, driven by aging infrastructure and extreme weather events, pose significant economic and societal challenges, with annual costs in the U.S. estimated at tens of billions of dollars. Despite ongoing research, a lack of standardized definitions and metrics hinders accurate resilience assessment. Existing studies often fall short in comprehensively exploring advanced assessment frameworks, integrating artificial intelligence and machine learning (AI/ML) techniques, and linking resilience efforts with crucial policy, regulatory, and socio-economic factors. Furthermore, the critical interdependencies between power systems and other essential infrastructures, coupled with diverse community vulnerabilities, necessitate a more holistic, community-centric approach to resilience, a gap this review aims to address.

Method

This paper conducts a comprehensive review of community-centric power system resilience, structured around several key themes. It first delves into resilience assessment approaches, broadly categorizing them into detailed engineering (micro-level) and data-driven (macro-level) methods. Detailed engineering assessments focus on fragility curves, which model the probability of individual component failure under various stress factors. Data-driven approaches, conversely, analyze aggregated system performance using frameworks like the resilience triangle, which quantifies performance degradation and recovery, and the more advanced resilience trapezoid, which accounts for prolonged degraded states. The review then examines the intricate interdependencies between power system resilience and community resilience, considering socioeconomic factors like social vulnerability and the critical reliance of other infrastructures (e.g., water, telecommunications, transportation) on continuous power supply. It synthesizes state-of-the-art strategies for enhancing resilience, including system hardening, strategic resource allocation (such as distributed generators and mobile power sources), and network reconfiguration techniques. A significant section is dedicated to the integration of AI/ML methods, covering traditional ML for outage prediction, neural networks, reinforcement learning for stability control and restoration, and emerging transformer-based generative models for advanced analysis and scenario generation. Finally, the paper addresses the techno-legal dimensions of resilient power systems, contrasting regulatory landscapes in the European Union (e.g., GDPR, NIS2, EU AI Act) and the United States (e.g., NERC CIP, NIST CSF, HIPAA), and exploring their implications for data handling, cybersecurity, and governance in resilience planning.

Results

The review highlights that both engineering-based (fragility curves) and data-driven (resilience triangle and trapezoid) methods provide quantitative frameworks for assessing power system resilience, each with specific strengths and limitations regarding data requirements and comprehensiveness. A key finding is the profound interdependence between power system resilience and community resilience, with socio-economic factors such as social vulnerability, age, health status, and racial/ethnic demographics significantly impacting outage exposure, duration, and recovery trajectories. Cascading failures across interconnected infrastructures (water, gas, transport, telecommunications) underscore the necessity of a system-of-systems perspective in resilience planning. Resilience hubs are identified as a critical community-centric strategy, offering essential services during emergencies and providing grid-supporting functions. Various resilience enhancement strategies, including system hardening, optimized resource allocation (e.g., distributed energy resources, mobile power sources), and intelligent network reconfiguration, are shown to be effective. The paper demonstrates that AI and ML techniques—ranging from predictive analytics for outage forecasting to reinforcement learning for real-time control and advanced generative models for scenario simulation—are pivotal for improving the analysis, prediction, and management of power system behavior during extreme events. Furthermore, the review reveals distinct yet parallel techno-legal frameworks in the EU and US that impose strict requirements on data privacy, cybersecurity, and AI governance, directly influencing the design and operation of resilient power systems.

Implications

The findings of this review carry significant implications for advancing community-centric power system resilience. Future research must prioritize the development of more comprehensive failure models that accurately capture the spatiotemporal dynamics of natural disasters and their grid impacts. Enhanced data utilization, including addressing data scarcity through improved collection and integration practices, alongside the use of synthetic networks to overcome confidentiality barriers, is crucial. The application of generative AI for creating realistic extreme weather scenarios can significantly improve anticipatory planning. Integrating AI/ML with stochastic methods promises more precise modeling of complex, nonlinear system behaviors. Beyond technical considerations, the review underscores that resilience must be approached holistically, integrating social vulnerability, interdependency modeling, and robust techno-legal frameworks into planning and operational strategies. This includes aligning resilience indices with compliance-oriented indicators to ensure accountability for operators and interdependent critical infrastructures. The insights gained are essential for developing more adaptive, secure, and socially equitable power grids capable of effectively mitigating and recovering from high-impact, low-probability disruptive events.