Advances and Applications in Statistics
Volume 48, Issue 1, Pages 1 - 31
(January 2016) http://dx.doi.org/10.17654/AS048010001 |
|
AN EFFICIENT METHOD FOR GLOBAL OPTIMIZATION OF BLACK-BOX FUNCTIONS USING ONE-DIMENSIONAL INTERPOLATION AND REDUCED DESIGN SPACE
Saleem Z. Ramadan
|
Abstract: In this paper, a simple and efficient metamodel-based global optimization method for black-box functions using one-dimensional interpolation and reduced design space is proposed. The method transfers the n-dimensional black-box function into one-dimensional function where interpolation is used to generate cheap points. Moreover, the proposed method employs the hypersphere strategy to narrow down the search space. The proposed method does not require prior tuning step or the use of any other algorithms like sampling algorithms as it is a self tuning and standalone algorithm. Results from eight benchmark problems showed that this method is robust as the variances and ranges of the best values found in the experimentation were reasonably small. Moreover, the results showed that this method is efficient in terms of computational time and memory consumption as it deals with one-dimensional rather than n-dimensional interpolation. |
Keywords and phrases: global optimization, black-box functions, interpolation, metamodeling, continuous variable optimization, discrete variable optimization. |
|
Number of Downloads: 405 | Number of Views: 948 |
|