A variable-fidelity modeling method based on self-organizing maps spatial reduction

Abstract

Variable-fidelity (VF) approximation models are wildly used to replace computational expensive simulation models in complex engineering designs. In this paper, a design space reduction variable-fidelity metamodeling (DSR-VFM) approach is proposed. In the proposed DSR-VFM, addition scaling Kriging (ASK) is chosen as the approximation model and self-organizing maps (SOM) is adopt to reduce the design space and select the key areas. Then new sample points are selected though the maximum distance method within the key areas and added to the sample set to update the approximate model. A numerical case and the modeling of the drag coefficient of an aircraft are utilized to verify the applicability of the proposed approach.