Contaminant source monitoring and characterization in water distribution systems

Recent anxiety surrounding the security of the nation's critical water infrastructure has increased interest in monitoring and characterization problems in water distribution system contexts. This dissertation addresses the difficulties associated with solving source identification inverse problems in water distribution systems due to ill-conditioning. Specifically, novel solution and monitoring design methods are investigated. A description of the source identification problem is developed and it is shown that the problem can be formulated as a discrete linear inverse problem. Such problems are well understood and powerful tools exist for their analysis and solution. Regularization is a technique for stabilizing the solution of ill-conditioned inverse problems. Typically, an inverse problem is regularized by incorporating additional information into the problem prior to solution. The form of the problem is modified and an approximate solution is sought. The effect of regularization methods on the source identification solution is investigated. It is concluded that regularizing for sparse solutions is most meaningful for the contaminant source identification problems in water distribution systems.;A novel simulation optimization based solution approach for environmental monitoring and characterization problems is also investigated. The approach utilizes global search heuristics such as evolutionary algorithms as opposed to classical gradient based algorithms. Simulation optimization requires many evaluations of the simulation model as the search progresses making them computationally intensive. Evolutionary algorithms are amenable to parallelization and in this work they are combined with the computing power of computational grids making the solution approach tractable. A general framework for parallel evolutionary algorithms is developed with the specific intent of solving environmental monitoring and characterization problems. The solution and computational performance achieved using the framework were studied for representative environmental characterization problems. Results indicate that significant raw performance improvements are possible using the approach and that global search techniques identify high quality solutions for the characterization problems studied.;The structure of the errors associated with an inverse problem solution are a function of monitoring observations. Optimal inverse experiment design is investigated as an approach for improving solution quality. The approach involves the selection of monitoring locations that are best suited to the generation of a well-conditioned source identification inverse problem. The monitoring design problem is formulated as a non-linear combinatorial optimization problem and solved using the optimization framework developed previously. The monitoring designs generated exhibit an optimal substructure that may be exploited to develop more efficient methods of solution. An analysis is conducted to evaluate the source inversion performance of an optimized monitoring network relative to networks designed using different methods. The results of the analysis demonstrate conclusively that when the source identification problem is underdetermined the number of monitoring sensors installed in the network is more important than the method used to locate them.