A parallel adaptive sampling strategy to accelerate the sampling process during the modeling of a Kriging metamodel

Higher modeling efficiency is an important goal for the modeling of a Kriging (KG) metamodel, and the sampling approach affects the modeling efficiency directly. Considering the effect of the employed correlation model on prediction accuracy of a KG model, a multiple KG models based parallel adaptive sampling strategy (MKPAS) is proposed using the combination forecasting method, in which the added new points in the sampling process are determined using multiple KG models with different correlation models. The effectiveness of the proposed approach is verified by two low dimensional benchmark functions as well as a high dimensional one. And an engineering application is also used to demonstrate the effectiveness of the proposed MKPAS approach. The results show that the proposed approach can improve the modeling efficiency of a KG model significantly compared with other ordinary sampling approaches.