Statistical surrogate model based sampling criterion for stochastic global optimization of problems with constraints

Title
Statistical surrogate model based sampling criterion for stochastic global optimization of problems with constraints
Author
이태희
Keywords
Constrained global optimization; Metamodel-based design optimization; Kriging surrogate model; Stochastic global optimization
Issue Date
2015-04
Publisher
KOREAN SOC MECHANICAL ENGINEERS
Citation
JOURNAL OF MECHANICAL SCIENCE AND TECHNOLOGY, v. 29, NO 4, Page. 1421-1427
Abstract
Sequential surrogate model-based global optimization algorithms, such as super-EGO, have been developed to increase the efficiency of commonly used global optimization technique as well as to ensure the accuracy of optimization. However, earlier studies have drawbacks because there are three phases in the optimization loop and empirical parameters. We propose a united sampling criterion to simplify the algorithm and to achieve the global optimum of problems with constraints without any empirical parameters. It is able to select the points located in a feasible region with high model uncertainty as well as the points along the boundary of constraint at the lowest objective value. The mean squared error determines which criterion is more dominant among the infill sampling criterion and boundary sampling criterion. Also, the method guarantees the accuracy of the surrogate model because the sample points are not located within extremely small regions like super-EGO. The performance of the proposed method, such as the solvability of a problem, convergence properties, and efficiency, are validated through nonlinear numerical examples with disconnected feasible regions.
URI
http://link.springer.com/article/10.1007%2Fs12206-015-0313-9http://hdl.handle.net/20.500.11754/24020
ISSN
1738-494X; 1976-3824
DOI
10.1007/s12206-015-0313-9
Appears in Collections:
COLLEGE OF ENGINEERING[S](공과대학) > AUTOMOTIVE ENGINEERING(미래자동차공학과) > Articles
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

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

BROWSE