A Multivariate Interpolation and Regression Enhanced Kriging Surrogate Model

21st AIAA Computational Fluid Dynamics Conference, San Diego, California, June 2013

https://arc.aiaa.org/doi/abs/10.2514/6.2013-2964

We present a Kriging surrogate model that is enhanced with a Multivariate Interpolation and Regression (MIR) through a dynamic training point selection. We propose an adaptive training point selection strategy where MIR is used as a local surrogate model that guides the construction of the global Kriging surrogate model. The quality of the resulting MIR enhanced Kriging surrogate model is demonstrated for two-, five- and nine-dimensional analytic test functions. The results indicate that the model performs better than currently available Kriging surrogates as well as a previously enhanced Kriging surrogate that uses Dutch Intrapolation as a local surrogate model. Preliminary results of using variable-fidelity data in the construction of the MIR enhanced Kriging surrogate model are also presented and show promise.

Figure: The proposed adaptive framework for training point selection.