348 0

Recommendations on the types and parameters of radial basis functions used for metamodel-based sequential global optimization

Title
Recommendations on the types and parameters of radial basis functions used for metamodel-based sequential global optimization
Author
WANG PEIDONG
Advisor(s)
최동훈
Issue Date
2017-08
Publisher
한양대학교
Degree
Master
Abstract
In recent years, sequential approximate global optimization (SAGO) algorithms have been widely used in mathematical and engineering test problems. SAGO finds the global optimum by sequentially optimizing a design problem using meta-models and a sampling of design points. Because SAGO requires a relatively small number of function evaluations, it is especially appropriate for mathematical problems requiring high efficiency. In this study, we adopted a radial basis function (RBF) as a meta-model and used the Constrained Optimization By RAdial basis function interpolation (COBRA) SAGO method proposed by Regis (2014). Selecting appropriate RBF types and associated parameters (if any) is difficult and requires different choices from problem to problem. To cope with this intrinsic difficulty, we chose seven mathematical problems of various kinds, assessed the performance of various RBFs for each example problem, and generated recommendations on the types and associated parameters (if any) of radial basis functions with generally good performance for most of the example problems. Gaussian, multiquadric, cubic, and thin plate spline RBFs were examined along with a wide range of associated parameter values (if any). The number of initial sample points was varied between NDV+1 and 50*NDV, where NDV denotes the number of design variables, to explore the influence of NEXP on performance. Analysis of these test results indicated the types and parameters of RBFs expected to perform well in SAGOs relating to a number of problems.
URI
http://hdl.handle.net/20.500.11754/33486http://hanyang.dcollection.net/common/orgView/200000431045
Appears in Collections:
GRADUATE SCHOOL[S](대학원) > MECHANICAL CONVERGENCE ENGINEERING(융합기계공학과) > Theses (Ph.D.)
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