Nozhati, Saeed, authorEllingwood, Bruce R., advisorMahmoud, Hussam N., advisorChong, Edwin K. P., committee membervan de Lindt, John W., committee member2019-09-102019-09-102019https://hdl.handle.net/10217/197318This dissertation introduces approximate dynamic programming (ADP) techniques to identify near-optimal recovery strategies following extreme natural hazards. The proposed techniques are intended to support policymakers, community stakeholders, and public or private entities to manage the restoration of critical infrastructure of a community following disasters. The computation of optimal scheduling schemes in this study employs the rollout algorithm, which provides an effective computational tool for optimization problems dealing with real-world large-scale networks and communities. The Markov decision process (MDP)-based optimization approach incorporates different sources of uncertainties to compute the restoration policies. The fusion of the proposed rollout method with metaheuristic algorithms and optimal learning techniques to overcome the computational intractability of large-scale, multi-state communities is probed in detail. Different risk attitudes of policymakers, which include risk-neutral and riskaverse attitudes in community recovery management, are taken into account. The context for the proposed framework is provided by objectives related to minimizing foodinsecurity issues and impacts within a small community in California following an extreme earthquake. Probabilistic food security metrics, including food availability, accessibility, and affordability, are defined and quantified to provide risk-informed decision support to policymakers in the aftermath of an extreme natural hazard. The proposed ADP-based approach then is applied to identify practical policy interventions to hasten the recovery of food systems and reduce the adverse impacts of food insecurity on a community. All proposed methods in this study are applied on a testbed community modeled after Gilroy, California, United States, which is impacted by earthquakes on the San Andreas Fault. Different infrastructure systems, along with their spatial distributions, are modeled as part of the evaluation of the restoration of food security within that community. The methods introduced are completely independent of the initial condition of a community following disasters and type of community (network) simulation. They treat the built environment like a black box, which means the simulation and consideration of any arbitrary network and/or sector of a community do not affect the applicability and quality of the framework. Therefore, the proposed methodologies are believed to be adaptable to any infrastructure systems, hazards, and policymakers' preferences.born digitaldoctoral dissertationsengCopyright and other restrictions may apply. User is responsible for compliance with all applicable laws. For information about copyright law, please see https://libguides.colostate.edu/copyright.community resilienceMarkov decision processrolloutfood securityapproximate dynamic programmingoptimal recovery managementText