Please use this identifier to cite or link to this item: http://dx.doi.org/10.18419/opus-9153
Authors: Fernández, Betsaida
Title: Automated calibration for numerical models of riverflow
Other Titles: Automatische Kalibrierung für numerische Flussmodelle
Issue Date: 2016
metadata.ubs.publikation.typ: Abschlussarbeit (Master)
metadata.ubs.publikation.seiten: xii, 118
URI: http://elib.uni-stuttgart.de/handle/11682/9170
http://nbn-resolving.de/urn:nbn:de:bsz:93-opus-ds-91708
http://dx.doi.org/10.18419/opus-9153
Abstract: Calibration of numerical models is fundamental since the beginning of all types of hydro system modeling, to approximate the parameters that can mimic the overall system behavior. Thus, an assessment of different deterministic and stochastic optimization methods is undertaken to compare their robustness, computational feasibility, and global search capacity. Also, the uncertainty of the most suitable methods is analyzed. These optimization methods minimize the objective function that comprises synthetic measurements and simulated data. Synthetic measurement data replace the observed data set to guarantee an existing parameter solution. The input data for the objective function derivate from a hydro-morphological dynamics numerical model which represents an 180-degree bend channel. The hydro- morphological numerical model shows a high level of ill-posedness in the mathematical problem. The minimization of the objective function by different candidate methods for optimization indicates a failure in some of the gradient-based methods as Newton Conjugated and BFGS. Others reveal partial convergence, such as Nelder-Mead, Polak und Ribieri, L-BFGS-B, Truncated Newton Conjugated, and Trust-Region Newton Conjugated Gradient. Further ones indicate parameter solutions that range outside the physical limits, such as Levenberg-Marquardt and LeastSquareRoot. Moreover, there is a significant computational demand for genetic optimization methods, such as Differential Evolution and Basin-Hopping, as well as for Brute Force methods. The Deterministic Sequential Least Square Programming and the scholastic Bayes Inference theory methods present the optimal optimization results.
Appears in Collections:02 Fakultät Bau- und Umweltingenieurwissenschaften

Files in This Item:
File Description SizeFormat 
BF_Automated_Calibration.pdfMaster Thesis8,93 MBAdobe PDFView/Open
AutomatedCalibrationPoster.pdfPoster1,31 MBAdobe PDFView/Open


Items in OPUS are protected by copyright, with all rights reserved, unless otherwise indicated.