P. Kossieris, Adaptation of evolutionary annealing-simplex algorithm for optimization of stochastic objective functions in water resource problems, Postgraduate Thesis, 209 pages, Department of Water Resources and Environmental Engineering – National Technical University of Athens, December 2013.
Water resource problems are characterized by the presence of multiple sources of uncertainty. The implementation of Monte Carlo simulation techniques within powerful optimization methods is required, in order to handle these uncertainties. In the framework of the present thesis we investigate how the various sources of uncertainty affect the optimization procedure as well as the various models. Furthermore, we investigate a modified version of the evolutionary annealing-simplex method in global optimization applications, where uncertainty is explicitly considered in terms of stochastic objective functions. We evaluate the algorithm against several benchmark functions, as well as in the stochastic calibration of a lumped rainfall-runoff model (Zygos). In this context, we examine different calibration criteria and different sources of uncertainty, in order to assess not only the robustness of the derived parameters but also the predictive capacity of the models. As one other problem that requires the combined use of optimization and simulation, we examine the applicability of a widely used rainfall model for the case of Athens. Taking advantage of the simulation and optimization functionalities of HyetosR package, we evaluate the performance of two versions of Bartlett-Lewis model in representing the convective and frontal rainfall of Athens. We demonstrate that although these models reproduce the essential statistical characteristics of rainfall at the hourly as well as daily time scales (mean, variance, autocorrelation structure), they fail to preserve important temporal properties, such as the duration and time distance of rainfall events.
Full text (7185 KB)