A novel multi-fidelity modeling method with double adaptive selection of kernel and learning functions—Application to spaceborne deployable antennas

The multi-fidelity surrogate (MFS) model that integrates high-fidelity and low-fidelity data is a promising and powerful tool for tackling complex modeling problem. There are two essential issues during its construction: sampling and kernel. The traditional MFS modeling methods mostly rely on the artificial intervention to determine training sample and kernel. However, this is a highly experience dependent task, further hindering the application of the MFS model in practice. Therefore, to tackle this challenge and get rid of artificial intervention in the modeling process, this paper proposes a novel multi-fidelity modeling method which can realize the adaptive selection of both the learning function and kernel function based on recursive Gaussian process. The significant advantage of this novel method is that it combines learning functions and kernel functions for adaptive selection according to the previous training performance, rather than random selection and repeated experiments. This double adaptive selection strategy not only avoids the artificial intervention but also improves the modeling accuracy. Finally, several case studies, including an engineering application to spaceborne deployable antennas, are conducted to demonstrate the effectiveness of the proposed method. All the results demonstrated that the proposed MFS modeling method can avoid artificial selection process and provide more accurate prediction results.