基于低阶导数物理信息神经网络的流动和传热反演问题研究

张贤; 胡春; 崔永赫; 王秋旺; 赵存陆

doi:10.7638/kqdlxxb-2024.0015

欢迎来到能源电力期刊网！

English Version 期刊登录

您当前的位置：

首页 >

文章列表页 >

基于低阶导数物理信息神经网络的流动和传热反演问题研究

研究论文 | 更新时间：2026-05-28

- 基于低阶导数物理信息神经网络的流动和传热反演问题研究
- Study on inverse problems of flow and heat transfer using low-order derivative physics-informed neural network
- 空气动力学学报 2025年43卷第1期页码：34-43
- 作者机构：
  
  1. 国核电力规划设计研究院有限公司,北京,100095
  2. 西安交通大学热流科学与工程教育部重点实验室,西安,710049
- 作者简介：
  
  张贤(1981—)，男，正高级工程师，研究方向：核电厂常规岛、智慧核电研发设计. E-mail：zhangxian@snpdri.com
  赵存陆*，男，教授，博士生导师，研究方向：流动与传热研究. E-mail：mclzhao@xjtu.edu.cn
- 基金信息：
  
  国家自然科学基金（51976157，51721004）
- DOI：10.7638/kqdlxxb-2024.0015
  中图分类号： TK124
- 纸质出版：2025
- 稿件说明：
移动端阅览
张贤, 胡春, 崔永赫, 等. 基于低阶导数物理信息神经网络的流动和传热反演问题研究[J]. 空气动力学学报, 2025, 43(1): 34−43.

ZHANG Xian, HU Chun, CUI Yonghe, et al. Study on inverse problems of flow and heat transfer using low-order derivative physics-informed neural network[J]. 2025, 43(1): 34-43.
张贤, 胡春, 崔永赫, 等. 基于低阶导数物理信息神经网络的流动和传热反演问题研究[J]. 空气动力学学报, 2025, 43(1): 34−43. DOI： 10.7638/kqdlxxb-2024.0015

ZHANG Xian, HU Chun, CUI Yonghe, et al. Study on inverse problems of flow and heat transfer using low-order derivative physics-informed neural network[J]. 2025, 43(1): 34-43. DOI： 10.7638/kqdlxxb-2024.0015.

摘要

求解空气动力学领域中流动和传热反演问题对于飞行器设计和飞行环境控制至关重要。然而，传统数值方法在处理这类问题时，往往面临计算复杂性和数据依赖性的挑战。为解决此问题，基于物理信息神经网络（PINN），本文构建了低阶导数物理信息神经网络（LPINN），其特点在于仅需有限的实验测量数据，即可有效地解决流动和传热的反演问题。为验证LPINN在反演问题上的应用效果，选择了两种典型的二维流动及传热案例（泊肃叶流动和顶盖驱动方腔流动）进行研究。结果表明，在缺乏明确边界条件的前提下，LPINN能利用稀疏的实验数据，准确预测整个计算区域内的流场和温度场，并能够较为精确地确定控制方程中的雷诺数和佩克莱数。对随机取点、均匀取点和基于先验知识取点3种不同的实验观测点选取方案进行了对比分析，发现基于先验知识取点方案在保证高反演精度的同时，其所需的实验观测点数量最少，这对提高反演问题的求解效率具有积极意义。此外，LPINN在处理反演问题时展现出对实验数据中噪声的高度鲁棒性。

Abstract

Solving inverse problems of flow and heat transfer in aerodynamics is crucial for aircraft design and flight environment control. However

traditional numerical methods often encounter challenges related to computational complexity and data dependency when addressing such problems. To tackle these issues

based on the physics-informed neural network (PINN) framework

we present a low-order derivative physics-informed neural network (LPINN)

which can effectively solve inverse problems in flow and heat transfer using only a limited amount of experimental measurement data. Two typical two-dimensional cases

namely Poiseuille flow and lid-driven cavity flow

are selected to comprehensively evaluate the effectiveness and reliability of LPINN in solving inverse problems. Results indicate that

without explicit boundary conditions

LPINN can accurately predict the flow and temperature fields within the entire computational domain using sparse observation data and can also precisely determine the unknown Reynolds and Péclet numbers in the governing equations. Comparisons of three observation point selection schemes—random

uniform

and prior-knowledge-based—reveal that the prior-knowledge-based scheme requires the fewest observation points to achieve high inversion accuracy

thereby enhancing the efficiency of solving inverse problems. Additionally

LPINN exhibits strong robustness against noise in experimental data.

关键词

Keywords

references

KUTZ J N . Deep learning in fluid dynamics[J ] . Journal of Fluid Mechanics, 2017 , 814 : 1 − 4 .

GOODFELLOW I, BENGIO Y, COURVILLE A, et al. Deep learning: Adaptive computation and machine learning series[M]. Cambridge, MA: The MIT Press, 2016.

LING J L, JONES R, TEMPLETON J . Machine learning strategies for systems with invariance properties[J ] . Journal of Computational Physics, 2016 , 318 : 22 − 35 .

RAISSI M, PERDIKARIS P, KARNIADAKIS G E . Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equa tions[J ] . Journal of Computational Physics, 2019 , 378 : 686 − 707 .

BAI X D, WANG Y, ZHANG W . Applying physics informed neural network for flow data assimilation[J ] . Journal of Hydrodynamics, 2020 , 32 ( 6 ): 1050 − 1058 .

MAO Z P, JAGTAP A D, KARNIADAKIS G E . Physics-informed neural networks for high-speed flows[J ] . Computer Methods in Applied Mechanics and Engineering, 2020 , 360 : 112789 .

LAUBSCHER R . Simulation of multi-species flow and heat transfer using physics-informed neural networks[J ] . Physics of Fluids, 2021 , 33 ( 8 ): 087101 .

ARZANI A, WANG J X, D’SOUZA R M . Uncovering near-wall blood flow from sparse data with physics-informed neural networks[J ] . Physics of Fluids, 2021 , 33 ( 7 ): 071905 .

LU Z B, LI Y M, HE C, et al . Integrating physics-informed neural networks with partitioned coupling strategy for modeling conjugate heat transfer[J ] . CIESC Journal, 2022 , 73 ( 12 ): 5483 − 5493 (in Chinese).

LU L, PESTOURIE R, YAO W J, et al . Physics-informed neural networks with hard constraints for inverse design[J ] . SIAM Journal on Scientific Computing, 2021 , 43 ( 6 ): B1105 − B1132 .

LIU X, FENG W H, LIAN F, et al . Aerodynamic data fusion method based on physics-informed neural network[J ] . Acta Aerodynamica Sinica, 2023 , 41 ( 8 ): 87 − 96 (in Chinese).

FU J Q, ZHONG B W, ZHONG Y Q, et al . A physics informed neural network based method for aircraft aerodynamic parameter identification[J ] . Acta Aerodynamica Sinica, 2023 , 41 ( 9 ): 30 − 37 (in Chinese).

CUI Y H, ZHANG W Y, YAN H L, et al . “Hard” boundary low-order derivative physics informed neural networks: A fluid flow solver[J ] . Journal of Xi'an Jiaotong University, 2022 , 56 ( 9 ): 123 − 133 (in Chinese).

BAYDIN A G, PEARLMUTTER B A, RADUL A A, et al . Automatic differentiation in machine learning: A survey[J ] . Journal of Machine Learning Research, 2018 , 18 : 1 − 43 .

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于物理信息神经网络的气动数据融合方法

风扇叶片颗粒侵蚀特性的多因素分析与模拟

大展弦比飞行器突风响应数值计算

离心通风机降噪研究进展

基于神经网络与遗传算法的热气防冰结构优化