Beta 1

Title Optimization using Surrogate Modelling
Author Christensen, Max la Cour
Eskildsen, Klaus Langgren
Supervisor Nielsen, Hans Bruun (Scientific Computing, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Institution Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
Thesis level Bachelor thesis
Year 2010
Abstract The subject of this thesis is unconstrained and derivative-free optimization using surrogate modelling. Unconstrained in the sense that the domain of the optimization parameters is unbounded. The goal of this thesis is to combine the good properties of conventional unconstrained optimization algorithms with surrogate modelling. This is done to reduce the number of function evaluations when optimizing a function. In this relation, the computational time of the optimization algorithms themselves is unimportant. The article of Murat Kuhlahci et al. [4] is the inspiration to the various algorithms developed during the work of this thesis. A discussion of the ideas in the article is given, and new ideas are introduced to improve efficiency and robustness. These ideas are incorporated into a new algorithm called SuMOpti. The SuMOpti algorithm is an unconstrained and derivative-free optimization algorithm that does not require prior knowledge of the function being optimized. The cornerstone of the SuMOpti algorithm and other algorithms presented in this thesis is the surrogate modelling tool provided in the Matlab Kriging toolbox DACE. A performance analysis is carried out to see how the SuMOpti algorithm compares to a recognised optimization algorithm. To study possible ways of improving the SuMOpti algorithm, alternative approaches are introduced. These approaches utilize an error estimator of the surrogate models to improve the surrogate models in certain areas of interest. Based on the experiences from the alternative approaches, an attempt is made to improve the efficiency of the SuMOpti algorithm. In addition to testing the algorithms on well-known test functions, the oil reservoir simulator ECLIPSE is also used as a test case.
Imprint Technical University of Denmark (DTU) : Kgs. Lyngby, Denmark
Series IMM-B.Sc.-2010-18
Original PDF bac10_18.pdf (5.07 MB)
Admin Creation date: 2010-06-25    Update date: 2010-06-25    Source: dtu    ID: 264110    Original MXD