A multi-fidelity surrogate model with sparse neural kernel

Li Xinran; Lian Feng; Jia Jiaxing

doi:10.7638/kqdlxxb-2024.0180

English Version Journal Login

您当前的位置：

首页 >

文章列表页 >

A multi-fidelity surrogate model with sparse neural kernel

更新时间：2026-05-28

- A multi-fidelity surrogate model with sparse neural kernel
- Vol. 44, Issue 2, Pages: 37-49(2026)
- 作者机构：
  
  西安交通大学电子与信息学部自动化科学与工程学院,西安,710049
- 作者简介：
- 基金信息：
- DOI：10.7638/kqdlxxb-2024.0180
  CLC： TP391;V211.3
- Published：2026
- 稿件说明：
移动端阅览
李欣然, 连峰, 贾家兴. 基于稀疏神经核的多保真度代理模型[J]. 空气动力学学报, 2026, 44(2): 37−49.

Li Xinran, Lian Feng, Jia Jiaxing. A multi-fidelity surrogate model with sparse neural kernel[J]. 2026, 44(2): 37-49.
李欣然, 连峰, 贾家兴. 基于稀疏神经核的多保真度代理模型[J]. 空气动力学学报, 2026, 44(2): 37−49. DOI： 10.7638/kqdlxxb-2024.0180

Li Xinran, Lian Feng, Jia Jiaxing. A multi-fidelity surrogate model with sparse neural kernel[J]. 2026, 44(2): 37-49. DOI： 10.7638/kqdlxxb-2024.0180.

摘要

在航空航天技术领域，代理模型发挥着关键作用。为平衡代理模型的训练成本与预测精度，需要发展能够有效挖掘多保真度数据间潜在相关性的建模方法。针对现有模型依赖预设核函数、缺乏数据自适应性的问题，提出了一种基于稀疏混合专家神经核（mixture of experts neural kernel

MoENK）的多保真度代理模型。MoENK通过线性混合和乘积混合基本单元构造新核函数，选择性屏蔽中间结果以过滤噪声，并应用于多任务高斯过程中。将该方法应用于3个函数示例和2个翼型算例中，结果表明该方法的预测精度有较大提升，尤其在NACA0012翼型阻力系数的预测中，相较于次佳方法LR-MFS，RMSE和MAE分别降低了40.42%和44.70%。证实了所提出的MoENK核函数能够不依赖预设核函数进行自适应预测，具有良好的泛化能力和鲁棒性，为工程系统的代理模型构建提供了新的工具。

Abstract

Surrogate models play a crucial role in aerospace technology. To offer a trade-off between training costs and model accuracy

we introduce multi-fidelity surrogate models based on Gaussian processes. These models leverage data with varying levels of fidelity

to enhance predictive performance while simultaneously minimizing computational expenses. By integrating information from both high-fidelity and low-fidelity datasets

multi-fidelity models can achieve higher accuracy without the need for extensive high-fidelity data

which is often costly and time-consuming to obtain. However

existing multi-fidelity GP models predominantly rely on empirically predetermined kernel functions. These kernel functions are typically selected based on prior experience and general assumptions about the data

without considering the specific characteristics and underlying structures of the datasets at hand. As a result

these kernels are not tailored to the data. To overcome these challenges

this paper introduced a novel approach to constructing kernel functions that are specifically designed to uncover and exploit the latent correlations between datasets of different fidelities. To address these issues

we first proposd the mixture of experts neural kernel (MoENK) which construct a more powerful kernel based on some basic kernels. In MoENK

we used two primary components: MoE-Linear and MoE-Product to selectively mask intermediate results

thereby effectively filtering out noise which may introduce negative transfer. This was then applied to multi-task Gaussian processes

which allows for the sharing of information across different fidelity levels

thereby improving the robustness and accuracy of the surrogate models. To rigorously evaluate the performance of the proposed method

we conducted a series of experiments using three benchmark function examples and two airfoil cases. It exhibited particularly notable advantages in predicting the high-dimensional drag coefficient scenario of the NACA0012 airfoil

reducing RMSE and MAE by 40.42% and 44.70%

respectively

compared to the suboptimal method

LR-MFS. These results confirm that the proposed MoENK kernel function can adaptively predict without relying on predefined kernels

offering strong generalization capabilities and robustness. This provides a new tool for constructing surrogate models in engineering systems.

关键词

Keywords

references

Huang Jihong, Su Honglian, Zhao Xinhua. Aerodynamic coefficient prediction of airfoil using BP neural network[J]. Advances in Aeronautical Science and Engineering, 2010, 1(1): 36−39 (in Chinese). doi: 10.3969/j.issn.1674-8190.2010.01.009

Liu Xin. Simulation of airfoil plunging aerodynamic parameter prediction based on neural network[J]. Computer Simulation, 2015, 32(12): 67−71 (in Chinese). doi: 10.3969/j.issn.1006-9348.2015.12.014

Kennedy M C, O'hagan A. Predicting the output from a complex computer code when fast approximations are available[J]. Biometrika, 2000, 87(1): 1−13. doi: 10.1093/biomet/87.1.1

Han Z H, Görtz S. Hierarchical kriging model for variable-fidelity surrogate modeling[J]. AIAA journal, 2012, 50(9): 1885−1896. doi: 10.2514/1.J051354

Morris M D, Mitchell T J, Ylvisaker D. Bayesian design and analysis of computer experiments: use of derivatives in surface prediction[J]. Technometrics, 1993, 35(3): 243−255. doi: 10.1080/00401706.1993.10485320

Ulaganathan S, Couckuyt I, Dhaene T, et al. Performance study of gradient-enhanced Kriging[J]. Engineering with computers, 2016, 32(1): 15−34. doi: 10.1007/s00366-015-0397-y

Xiao M, Zhang G, Breitkopf P, et al. Extended co-Kriging interpolation method based on multi-fidelity data[J]. Applied Mathematics and Computation, 2018, 323: 120−131. doi: 10.1016/j.amc.2017.10.055

Zhang L, Wu Y, Jiang P, et al. A multi-fidelity surrogate modeling approach for incorporating multiple non-hierarchical low-fidelity data[J]. Advanced Engineering Informatics, 2022, 51: 101430. doi: 10.1016/j.aei.2021.101430

Ning Chenjia, Wang Xu, Wang Wenzheng, et al. Data association and fusion aerodynamic modeling method based on efficient sampling[J]. Acta Aerodynamica Sinica, 2022, 40(5): 39−49 (in Chinese). doi: 10.7638/kqdlxxb-2021.0425

Wang Xu, Ning Chenjia, Wang Wenzheng, et al. Intelligent fusion method of multi-source aerodynamic data for flight tests[J]. Acta Aerodynamica Sinica, 2023, 41(2): 12−20 (in Chinese). doi: 10.7638/kqdlxxb-2021.0428

Lin Feng, Hai Chunlong, Mei Liquan. Multi-source aerodynamic data fusion modeling with XGBoost[J]. Acta Aerodynamica Sinica, 2024, 42(7): 27−34 (in Chinese). doi: 10.7638/kqdlxxb-2023.0066

Jin Feiteng, Shen Haidong, Ye Yiqiao, et al. Adaptive point-added Co-RBF multi-fidelity surrogate modeling for hypersonic vehicles[J]. Journal of Astronautics, 2023, 44(12): 1850−1859 (in Chinese). doi: 10.3873/j.issn.1000-1328.2023.12.007

Adjei R A, Zheng X, Lou F, et al. Multifidelity optimization under uncertainty for robust design of a micro-turbofan turbine stage[J]. Journal of Engineering for Gas Turbines and Power, 2022, 144(10): 101006. doi: 10.1115/1.4055231

Xu C, Han Z, Zhang K, et al. A novel multi-fidelity cokriging model assisted by multiple non-hierarchical low-fidelity datasets[J]. Structural and Multidisciplinary Optimization, 2024, 67(4): 52. doi: 10.1007/s00158-024-03744-8

Sun S, Zhang G, Wang C, et al. Differentiable compositional kernel learning for Gaussian processes[C]//International Conference on Machine Learning. 2018: 4828-4837. doi: 10.48550/arXiv.1806.04326

Bonilla E V, Chai K, Williams C. Multi-task Gaussian process prediction[J]. Advances in neural information processing systems, 2007, 20.

Jacobs R A, Jordan M I, Nowlan S J, et al. Adaptive mixtures of local experts[J]. Neural computation, 1991, 3(1): 79−87. doi: 10.1162/neco.1991.3.1.79

Shazeer N, Mirhoseini A, Maziarz K, et al. Outrageously large neural networks: the sparsely-gated mixture-of-experts layer[J]. arXiv preprint arXiv: 1701.06538, 2017. doi: 10.48550/arXiv.1701.06538

Aronszajn N. Theory of reproducing kernels[J]. Transactions of the American mathematical society, 1950, 68(3): 337−404. doi: 10.2307/1990404

Duvenaud D, Lloyd J, Grosse R, et al. Structure discovery in nonparametric regression through compositional kernel search[C]// International Conference on Machine Learning. 2013: 1166-1174. doi: 10.1007/978-3-642-05290-3_42

Lloyd J, Duvenaud D, Grosse R, et al. Automatic construction and natural-language description of nonparametric regression models [C]//Proceedings of the AAAI Conference on Artificial Intelligence: vol. 28: 1. 2014.

Fernández-Godino M G, Dubreuil S, Bartoli N, et al. Linear regression-based multifidelity surrogate for disturbance amplification in multiphase explosion[J]. Structural and Multidisciplinary Optimization, 2019, 60(6): 2205−2220. doi: 10.1007/s00158-019-02387-4

Leen G, Peltonen J, Kaski S. Focused multi-task learning in a Gaussian process framework[J]. Machine learning, 2012, 89(1): 157−182. doi: 10.1007/s10994-012-5302-y

Kingma D P, Ba J. Adam: A method for stochastic optimization[J]. arXiv preprint arXiv: 1412.6980, 2014. doi: 10.48550/arXiv.1412.6980

Gardner J, Pleiss G, Weinberger K Q, et al. Gpytorch: Blackbox matrix-matrix gaussian process inference with gpu acceleration[J]. Advances in neural information processing systems, 2018, 31. doi: 10.48550/arXiv.1809.11165

Gregory N, O'reilly C. Low-speed aerodynamic characteristics of NACA 0012 aerofoil section, including the effects of upper-surface roughness simulating hoar frost[J]. 1970. doi: 10.1002/chin.199248228

Schmitt V, Charpin F. Pressure distributions on the ONERA M6 wing at transonic Mach numbers: AGARD AR 138[R]. Experimental Data Base for Computer Program Assessment, 1979. https://api.semanticscholar.org/CorpusID:117002952.

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Active control of supersonic jet based on multiple microjets

Aerodynamic and wind energy harvesting integrative design for UAVs

Flow field prediction and reconstruction of aircraft based on flexible smart skin

Data-driven rapid prediction of surface heat flux for lifting-body aircraft

Deep neural network-based reduced-order modeling of high-speed airfoil flow field

Related Author

Arun Kumar Perumal

Ning Yinghao

Xue Yuling

Fan Dewei

Luo Yang

Yang Wenqing

Song Bifeng

An Weigang

Related Institution

Department of Aerospace Engineering, Indian Institute of Technology, Kanpur, Kanpur

Zhengzhou Tobacco Research Institute of CNTC, Zhengzhou

Hangzhou International Innovation Institute, Beihang University, Hangzhou

School of Aeronautics, Northwestern Polytechnical University, Xi'an

National Key Laboratory of Aircraft Configuration Design, Xi'an

AI问答

⁰