Abstract Engineering computer codes are often computationallyexpensive. To lighten this load, we exploit newcovariance kernels to replace computationally expensivecodes with surrogate models. For input spaces with largedimensions, using the Kriging model in the standard wayis computationally expensive because a large covariancematrix must be inverted several times to estimate the parametersof the model. We address this issue herein byconstructing a covariance kernel that depends on onlya few parameters. The new kernel is constructed basedon information obtained from the Partial Least Squaresmethod. Promising results are obtained for numerical exampleswith up to 100 dimensions, and significant computationalgain is obtained while maintaining sufficientaccuracy.
Publié le : 2015-11-24
Classification:
partial least squares,
surrogate model,
kriging,
[MATH.MATH-OC]Mathematics [math]/Optimization and Control [math.OC],
[MATH.MATH-ST]Mathematics [math]/Statistics [math.ST],
[STAT.ME]Statistics [stat]/Methodology [stat.ME]
@article{hal-01232938,
author = {Bouhlel, Mohamed-Amine and Bartoli, Nathalie and Otsmane, Abdelkader and Morlier, Joseph},
title = {Improving kriging surrogates of high-dimensional design models by Partial Least Squares dimension reduction},
journal = {HAL},
volume = {2015},
number = {0},
year = {2015},
language = {en},
url = {http://dml.mathdoc.fr/item/hal-01232938}
}
Bouhlel, Mohamed-Amine; Bartoli, Nathalie; Otsmane, Abdelkader; Morlier, Joseph. Improving kriging surrogates of high-dimensional design models by Partial Least Squares dimension reduction. HAL, Tome 2015 (2015) no. 0, . http://gdmltest.u-ga.fr/item/hal-01232938/