Multi-fidelity information fusion with hierarchical surrogate guided by feature mapping

Abstract

Multi-fidelity information fusion has attracted increasing attention in the recent for its promising in engineering design and optimization. However, most of the existing fusion methods are sensitive to the correlation between low-fidelity (LF) and high-fidelity (HF) information, which could be further improved in terms of approximation performance. To address this problem, we develop a feature mapping-based hierarchical surrogate model for the fusion of LF and HF information. Compared to most current models, the presented model has a distinct hierarchical structure, where the LF model is first built through radial basis function. Then, a mapping relationship is established to describe the correlation between the transitional prediction and the HF prediction, and the mapping relationship is optimized using the Rayleigh quotient. Finally, a surrogate model is developed that can fuse information with different fidelities. To verify the effectiveness of the developed model, widely used numerical problems are analyzed, and five state-of-the-art methods are selected as comparative surrogates. Moreover, one engineering case is investigated to illustrate the capability of the proposed model in support of complex engineering design. Experiments demonstrate that our method outperforms other methods in 97.436% of the thirteen test problems under three different quantitative evaluations.