TY - GEN
T1 - Combined kriging surrogate model for efficient global optimization using the optimal weighting method
AU - Appriou, Tanguy
AU - Shimoyama, Koji
N1 - Publisher Copyright:
© 2020 Owner/Author.
PY - 2020/7/8
Y1 - 2020/7/8
N2 - When solving design optimization problems using evolutionary algorithms, the optimization process can be computationally expensive. To accelerate the optimization process, ordinary Kriging (OK) surrogate models are often used with the efficient global optimization (EGO) framework. However, in some cases the EGO framework can lead to a globally inaccurate OK surrogate model when many sample points are close to each other. One way to tackle this issue is to use a regression OK model instead of an interpolation OK model. In this paper, we propose an interpolation method which solve the issue by combining a local and a global OK model fitted to different set of the sample points. This paper describes the optimal weighting method used to combine the different Kriging models and compares the performance of the new method to interpolation and regression OK for the modified Branin test function. We find that when many sample points exist close to each other, the combined Kriging method outperform both the interpolation and the regression OK.
AB - When solving design optimization problems using evolutionary algorithms, the optimization process can be computationally expensive. To accelerate the optimization process, ordinary Kriging (OK) surrogate models are often used with the efficient global optimization (EGO) framework. However, in some cases the EGO framework can lead to a globally inaccurate OK surrogate model when many sample points are close to each other. One way to tackle this issue is to use a regression OK model instead of an interpolation OK model. In this paper, we propose an interpolation method which solve the issue by combining a local and a global OK model fitted to different set of the sample points. This paper describes the optimal weighting method used to combine the different Kriging models and compares the performance of the new method to interpolation and regression OK for the modified Branin test function. We find that when many sample points exist close to each other, the combined Kriging method outperform both the interpolation and the regression OK.
KW - Combined kriging
KW - Efficient global optimization (EGO)
KW - Kriging surrogate model
KW - Optimal weighting method
UR - http://www.scopus.com/inward/record.url?scp=85089733307&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85089733307&partnerID=8YFLogxK
U2 - 10.1145/3377929.3398163
DO - 10.1145/3377929.3398163
M3 - Conference contribution
AN - SCOPUS:85089733307
T3 - GECCO 2020 Companion - Proceedings of the 2020 Genetic and Evolutionary Computation Conference Companion
SP - 29
EP - 30
BT - GECCO 2020 Companion - Proceedings of the 2020 Genetic and Evolutionary Computation Conference Companion
PB - Association for Computing Machinery, Inc
T2 - 2020 Genetic and Evolutionary Computation Conference, GECCO 2020
Y2 - 8 July 2020 through 12 July 2020
ER -