Kriging model with modified nugget effect for random simulation with heterogeneous variances

Random simulation with heterogeneous variance is a common phenomenon in real-world simulation cases. Most of the current meta-models assume homogeneous variances and hence are unable to provide satisfactory estimations of these models. This article presents a new kriging model with modified nugget effect, which accounts for heterogeneous variances in simulations. The new model can reduce the influence of noise with heterogeneous variance by incorporating local variance information into the model. A theoretical explanation and numerical experiment will be shown to provide clear illustration of this new modified model.