Resumen de A hybrid variable-fidelity global approximation modelling method combining tuned radial basis function base and kriging correction
Jun Zheng, Xinyu Shao, Liang Gao, Ping Jiang, Zilong Li
Currently, approximation models have been widely adopted to replace the actual computationally expensive experiments or simulations in engineering design optimisation. However, it is not easy to create an accurate approximation model, especially when the computation-intensive high-fidelity (HF) sample is strictly limited to a small size. To ease this problem, a hybrid variable-fidelity global approximation modelling method (hybrid VF) combining tuned radial basis function (RBF) base and kriging correction is proposed in this paper. In the proposed hybrid VF approach, information from both HF and low-fidelity (LF) models are employed. The LF model is first approximated by RBF, and then this RBF LF approximation is pulled to the HF by the linear tuning model to gain a base surrogate. After that, this tuned RBF base is modified by a kriging-based linear correction function. Some numerical and engineering examples are demonstrated to validate the efficiency of the proposed hybrid VF method. The results show that the proposed hybrid VF can achieve a much more accurate approximation than other metamodelling methods in the small HF sample situation.
