Optimization Methods for Improving the Resilience of Civil Infrastructure Systems Subject to Natural Hazards

Modeling of civil infrastructure systems subject to hazards and disruptions, and subsequent optimization of their performance under such scenarios, is crucial to protect critical lifeline services. This research seeks to develop methods for discovering cost-effective, incremental, stage-wise investment strategies to improve long-term resilience and performance of complex, interdependent systems against future, unknown hazards. An adaptable modeling framework for the civil infrastructure system is developed to represent system components, quantify the services they provide, and the interdependencies among components. Resilience metrics are created that quantify various aspects of system--wide resilience based on the status of components, allowing consideration of system--wide performance after a natural hazard. An illustration of the metrics conducted on realistic data from a coastal community subject to hurricanes shows that the approach can give an assessment of performance in various scenarios. A mathematical programming approach is used to optimally select component investments and improvements according to the serviceability resilience metric.;The introduced metrics and modeling framework are then extended to a stage-wise approach, where investment decisions are made incrementally, as hazard events occur. A stochastic integer programming model for improving resilience is presented for this incremental, stage--wise approach to making decisions. The separation of decisions into multiple stages and consideration of an increased number of hazard scenarios necessitates the implementation of a customized Benders decomposition algorithm for certain resilience metrics. The results indicate that specialized algorithms or meta-heuristics are necessary to find good solutions.;The computational challenges of particular metrics and the desire to simultaneously optimize multiple measures of resilience lead to the development of an evolutionary multiobjective optimization algorithm for diverse solutions, EMODS, to find a population of potential solutions that achieve two goals. The first goal is to discover Pareto optimal candidates to capture the trade--offs among different measures of resilience. The second goal is to locate alternative solutions that are near--Pareto optimal (i.e., within a given relaxation of Pareto optimality) that are diverse and distinct in the decision space to better handle unmodeled problem characteristics. The notion of solution diversity is defined for multiobjective problems using the concept of substitutability. Various diversity metrics are defined that are necessary for quantifying the different aspects of solution diversity in a multiobjective context. EMODS is compared against other diversity-enhancing EMO algorithms and the results demonstrate that the algorithm is competitive with, and frequently improves on, the other algorithms in terms of the quality of nondominated frontiers and diversity metrics, defined here and in other papers.;EMODS is adapted to apply to problems with discrete decision variables for the civil infrastructure system resilience problem. Experimental runs are conducted for EMODS to optimize multiple resilience metrics of a civil infrastructure system based on the realistic data from the coastal community. The final solution sets obtained by EMODS for two- and four-objective problems are compared with those obtained by the NSGA-II algorithm and a random search. EMODS performs competitively in terms of quality of its nondominated frontiers and outperforms the other methods in regards to the chosen diversity metrics. Additionally, the robustness of the solutions sets obtained by the various methods are analyzed by evaluating final solution sets on a set of alternative hazard scenarios, simulating the real-world situation of prediction errors and unmodeled objectives. The final solution sets obtained by EMODS are shown to exhibit high diversity among substitutable solutions that are near--nondominated and nondominated.