OPTIMIZATION FOR CALIBRATION OF WATER RESOURCES SYSTEMS INCLUDING NEW PARALLEL GLOBAL ALGORITHMS AND APPLICATIONS TO HYDRODYNAMICS AND WATER QUALITY LAKE PDE MODELS

This thesis introduces efficient parallel optimization algorithms for computationally expensive optimization problems and applies them to water resources issues. A new parallel surrogate global optimization algorithm PODS is developed and successfully applied to the calibration of two computationally expensive 3D hydrodynamic lake models (5 hours per simulation). We also introduce another new parallel algorithm GOPS to efficiently use many processors (up to 128 processors in parallel) for optimization problems with a high dimensional decision vector. GOPS shows superior performance than prior methods on benchmark problems and an expensive water quality model calibration problem with 21 decision variables. The impact of memory hardware performance on calibration efficiency is also investigated. The calibration’s efficiency can be improved by as much as 20% with proper affinity settings compared with the default setting. The algorithms and methodologies proposed in this study are general-purpose and could be applied to other mathematically similar applications.

Author dc:creator

• XIA WEI

• Lake hydrodynamics,Water quality models,Global surrogate optimization,Parallel optimization,Parameter calibration,High-performance computing