Reliable infrastructure systems are essential for community resilience against natural hazards and extreme events. Recent events such as Hurricane Beryl in 2024, Turkey-Syria earthquakes in 2023, Winter Storm Uri in 2021, Hurricane Florence in 2018, and Hurricane Matthew in 2016, have highlighted the vulnerability of electric power networks (EPNs) to such disasters. This vulnerability underscores the need for robust performance assessments, which play a pivotal role in guiding decisions related to the design, operation, maintenance, upgrade, and repair of EPNs.

Reliability and resilience are key performance indicators for critical infrastructure systems such as electric power grids. Electric power systems are considered critical due to their crucial role in maintaining the normal function of communities, interdependence on other infrastructure systems, and broader national stability. The computational methods used to estimate these performance metrics require realistic and accurate data, appropriate digital models for the computational problems, and high computational power along with suitable approximation techniques. While prior work has focused on the use of synthetic data and surrogate models to overcome these challenges, there is significant room for improvement, particularly for applications in community-level analyses. This dissertation aims to address these challenges through methodological advancements and algorithmic improvements for assessing the reliability of EPNs, especially for small communities with limited access to real-world data. By focusing on methodological innovations informed by practical constraints, this research seeks to provide more accurate and efficient tools for infrastructure analysis for the upkeep of real world infrastructure systems.

This dissertation introduces significant advancements in the assessment and enhancement of electric power networks (EPNs). First, it proposes an integrated modeling technique that accounts for both the power transmission system and power distribution system in building-level power availability analyses, and thus better assess impacts on community members. We aim to address the limited availability of real power grid data by creating equivalenced synthetic grid models. Power transmission and distribution performance assessments remain largely siloed today, but require integration to accurately assess community-level impacts. To integrate analysis of the power transmission network and the power distribution network in community-level reliability and resilience assessment, we use power availability from the power transmission network analysis as tractable boundary conditions for the power distribution network analysis. Secondly, this dissertation proposes a novel recursive method designed to compute the expected functional reliability of radial infrastructure systems. Our technique is computationally efficient, enabling accurate reliability assessments under various operational scenarios and providing crucial insights into system behavior both during routine operations and following disruptive events. As our technique uses an application agnostic formulation, it can be applied to any infrastructure system with radial topology such as power distribution systems, water distribution systems, renewable energy farms, etc. Finally, this work addresses the high computational demands of traditional reliability analysis by developing graph neural network (GNN) models within an augmented artificial intelligence framework. Our proposed GNN model efficiently approximates all-terminal reliability, drastically reducing computational costs while maintaining accuracy in connectivity reliability assessments. Together, these methodologies offer a comprehensive suite of tools for assessing and improving the reliability of critical infrastructure in the face of natural hazards and other extreme conditions. Our methods and models are validated using several case studies with real communities such as Lumberton, North Carolina, and Galveston Island, Texas, alongside numerous synthetic benchmarks.

The adoption of these integrated modeling, recursive reliability assessment, and GNN based all terminal reliability approximation techniques holds promise for improving decision support for community-level infrastructure resilience planning and intervention. By addressing data scarcity, reducing computational cost, and integrating practical constraints, these methods will contribute to more effective infrastructure reliability assessments and facilitate integration of socio-techno-economic models for improving the performance of systems during and after extreme events. Ultimately, this research aims to empower small communities to plan for, withstand, and recover from extreme events.