A multi-fidelity surrogate model for non-hierarchical data and its application in sensitivity analysis

The multi-fidelity (MF) surrogate model combines the information of high fidelity (HF) data and low fidelity (LF) data to realize the trade-off between computational cost and prediction accuracy, which has attracted wide attention in recent years. In particular, sensitivity analysis based on a surrogate model can achieve computational acceleration and cost savings. This paper focuses on non-hierarchical multi-fidelity data to propose a multi-fidelity approximation modeling method by taking advantage of the PC-Kriging model’s high accuracy and wider applicability. Specifically, we use the linear combination of LF PC-Kriging models as the trend term of the Kriging model to construct the multi-fidelity surrogate model of the system. Furthermore, we evaluated the sensitivity indices based on this surrogate model. The numerical tests showed that this method can effectively identify the importance of variables and maintain its accuracy.