Gaussian Processes with Derivatives (GPD) model has been widely used as a surrogate model in engineering to mitigate computational costs. However, most existing GPD models do not consider multi-fidelity derivatives and the relationship between high-fidelity (HF) and low-fidelity (LF) models. In this paper, we propose a novel GPD model called multi-fidelity Gaussian processes with derivatives (MFGPD), in which the posterior mean of the LF GPD replaces the prior mean of the HF GPD. The relationship between the HF and LF models is depicted by an unknown function. To demonstrate the performance of MFGPD, we provide two analytical examples and an aerodynamic coefficient modeling case, as well as comparisons to two other surrogate models. Results show that MFGPD outperforms the other models in both global and local prediction, indicating great potential for engineering applications.
Multi-fidelity Gaussian Processes with Derivatives for Aerodynamic Data Fusion
18.07.2023
887647 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Multi-Fidelity for MDO Using Gaussian Processes
Springer Verlag | 2020
|MULTI-FIDELITY GAUSSIAN PROCESS MODEL FOR CFD AND WIND TUNNEL DATA FUSION
TIBKAT | 2021
|