Please wait a minute...
浙江大学学报(理学版)  2023, Vol. 50 Issue (1): 56-62    DOI: 10.3785/j.issn.1008-9497.2023.01.009
数学与计算机科学     
双曲型方程激波捕捉的物理信息神经网络(PINN)算法
郑素佩,靳放(),封建湖,林云云
长安大学 理学院,陕西 西安 710064
PINN-type algorithm for shock capturing of hyperbolic equations
Supei ZHENG,Fang JIN(),Jianhu FENG,Yunyun LIN
School of Science,Chang'an University,Xi'an 710064,China
 全文: PDF(1323 KB)   HTML( 7 )
摘要:

双曲型方程的数值求解算法研究一直是偏微分方程研究的热点,其中,双曲型方程的间断捕捉是难点。受物理信息神经网络(physics-informed neural networks,PINN)启发,构造了改进的PINN算法,近似求解双曲型方程的间断问题。将坐标构造的数据集作为神经网络的输入,将PINN算法中的损失函数作为训练输出值与参考解(基于细网格的熵相容格式数据)或准确解的误差值,通过网络优化,最小化损失函数,得到最优网络参数。最后用数值算例验证了算法的可行性,数值结果表明,本文算法能捕捉激波,分辨率高,且未产生伪振荡。

关键词: 双曲守恒律方程网络预测物理信息神经网络(PINN)激波捕捉    
Abstract:

The numerical solution of hyperbolic equation is a well-know hot topic in the field of numerical solution of partial differential equation, among which the discontinuous capturing of hyperbolic equation is always a difficult problem. Inspired by physical-informed neural networks (PINN), this paper presents a PINN-type algorithm to approximately solve discontinuity problem of hyperbolic equations. It takes the data set constructed by coordinate as the input of neural network. The loss function in PINN algorithm is converted to the error between the output value of the training network and the reference solution (entropy compatible format data based on the fine grid) or the exact solution. Then the loss function is minimized by network optimization to obtain the optimal network parameters. Finally, some numerical examples are demonstrated to verify the feasibility of the proposed algorithm. The numerical results show that the proposed algorithm can capture shock waves, and it has high resolution, without nonphysical oscillations.

Key words: hyperbolic conservation laws equation    network prediction    PINN    shock capturing
收稿日期: 2021-11-12 出版日期: 2023-01-13
CLC:  O 241.8  
基金资助: 国家自然科学基金资助项目(11971075);陕西省自然科学基金青年项目(2020JQ-338)
通讯作者: 靳放     E-mail: jinfang53@163.com
作者简介: 郑素佩(1978—),ORCID:https://orcid.org/0000- 0003-2502-6998,女,博士,副教授,主要从事微分方程数值算法研究.
服务  
把本文推荐给朋友
加入引用管理器
E-mail Alert
RSS
作者相关文章  
郑素佩
靳放
封建湖
林云云

引用本文:

郑素佩, 靳放, 封建湖, 林云云. 双曲型方程激波捕捉的物理信息神经网络(PINN)算法[J]. 浙江大学学报(理学版), 2023, 50(1): 56-62.

Supei ZHENG, Fang JIN, Jianhu FENG, Yunyun LIN. PINN-type algorithm for shock capturing of hyperbolic equations. Journal of Zhejiang University (Science Edition), 2023, 50(1): 56-62.

链接本文:

https://www.zjujournals.com/sci/CN/10.3785/j.issn.1008-9497.2023.01.009        https://www.zjujournals.com/sci/CN/Y2023/V50/I1/56

图1  改进的PINN算法的网络结构
图2  无黏Burgers方程训练集、测试集预测值与精确解的比较
图3  无黏Burgers方程压缩波训练集、测试集预测值与精确解的比较
图4  线性对流方程压缩波训练集、测试集预测值与精确解的比较
图5  Euler方程压缩波训练集、测试集预测值与精确解的比较
图6  二维无黏Burgers方程训练集、测试集预测值与精确解的比较
1 LIU X D, OSHER S, CHAN T. Weighted essentially non-oscillatory schemes[J]. Journal of Computational Physics, 1994, 115(1): 200-212. DOI:10.1006/jcph.1994.1187
doi: 10.1006/jcph.1994.1187
2 ISMAIL F, ROE P L. Affordable, entropy-consistent Euler flux functions II: Entropy production at shocks[J]. Journal of Computational Physics, 2009, 228(15): 5410-5436. DOI:10.1016/j.jcp. 2009.04.021
doi: 10.1016/j.jcp. 2009.04.021
3 王令, 郑素佩. 基于移动网格的熵稳定格式求解浅水波方程[J]. 水动力学研究与进展(A辑), 2020, 35(2): 188-193. DOI:10.16076/j.cnki.cjhd.2020. 02.006
WANG L, ZHENG S P. Solving shallow water wave equation based on moving grid entropy stable scheme[J]. Chinese Journal of Hydrodynamics(A), 2020, 35(2): 188-193. DOI:10.16076/j.cnki.cjhd.2020.02.006
doi: 10.16076/j.cnki.cjhd.2020.02.006
4 程晓晗, 聂玉峰, 蔡力, 等. 基于移动网格的熵稳定格式[J]. 计算物理, 2017, 34(2): 175-182. DOI:10. 19596/j.cnki.1001-246x.2017.02.007
CHENG X H, NIE Y F, CAI L, et al. Entropy stable scheme based on moving meshes for hyperbolic conservation laws[J]. Chinese Journal of Computational Physics, 2017, 34(2): 175-182. DOI:10.19596/j.cnki.1001-246x.2017.02.007
doi: 10.19596/j.cnki.1001-246x.2017.02.007
5 TANG H Z, TANG T. Adaptive mesh methods for one- and two-dimensional hyperbolic conservation laws[J]. SIAM Journal on Numerical Analysis,2003,41(2): 487-515. DOI:10.1137/S003614290138437X
doi: 10.1137/S003614290138437X
6 HAN J Q, ARNULF J, E W N. Solving high-dimensional partial differential equations using deep learning[J]. Proceedings of the National Academy of Sciences, 2018, 115(34): 8505-8510. DOI:10. 1073/pnas.1718942115
doi: 10. 1073/pnas.1718942115
7 SIRIGNANO J, SPILIOPOULOS K. DGM: A deep learning algorithm for solving partial differential equations[J]. Journal of Computational Physics, 2018, 375: 1339-1364. DOI:10.1016/j.jcp.2018. 08.029
doi: 10.1016/j.jcp.2018. 08.029
8 SUN L N, GAO H, PAN S W, et al. Surrogate modeling for fluid flows based on physics-constrained deep learning without simulation data[J]. Computer Methods in Applied Mechanics and Engineering, 2020, 361: 112732. DOI:10.1016/j.cma.2019. 112732
doi: 10.1016/j.cma.2019. 112732
9 黄婕, 张丰, 杜震洪, 等. 基于RNN-CNN集成深度学习模型的PM2.5小时浓度预测[J]. 浙江大学学报(理学版), 2019, 46(3): 370-379. DOI:10.3785/j.issn.1008-9497.2019.03.016
HUANG J, ZHANG F, DU Z H, et al. Hourly concentration prediction of PM2.5 based on RNN-CNN ensemble deep learning model[J]. Journal of Zhejiang University(Science Edition), 2019, 46(3): 370-379. DOI:10.3785/j.issn.1008-9497. 2019.03.016
doi: 10.3785/j.issn.1008-9497. 2019.03.016
10 HORNIK K, STINCHCOMBE M, WHITE H. Universal approximation of an unknown mapping and its derivatives using multilayer feedforward networks[J]. Neural Networks, 1990, 3(5): 551-560. DOI:10.1016/0893-6080(90)90005-6
doi: 10.1016/0893-6080(90)90005-6
11 ALLAN P. Approximation theory of the MLP model in neural networks[J]. Acta Numerica, 1999, 8: 143-195. DOI:10.1017/S0962492900002919
doi: 10.1017/S0962492900002919
12 CHEN T, CHEN H. Universal approximation to nonlinear operators by neural networks with arbitrary activation functions and its application to dynamical systems[J]. IEEE Transactions on Neural Networks, 1995, 6(4): 911-917. DOI:10.1109/72.392253
doi: 10.1109/72.392253
13 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 equations[J]. Journal of Computational Physics, 2019, 378: 686-707. DOI:10.1016/j.jcp.2018.10.045
doi: 10.1016/j.jcp.2018.10.045
14 JAGTAP A D, KAWAGUCHI K, KARNIADAKIS G E. Adaptive activation functions accelerate convergence in deep and physics-informed neural networks[J]. Journal of Computational Physics, 2020, 404: 109136. DOI:10.1016/j.jcp.2019.109136
doi: 10.1016/j.jcp.2019.109136
15 LU L, MENG X H, MAO Z P, et al. DeepXDE: A deep learning library for solving differential equations[J]. SIAM Review, 2021, 63(1): 208-228. DOI:10.1137/19M1274067
doi: 10.1137/19M1274067
16 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. DOI:10. 1016/j.cma.2019.112789
doi: 10. 1016/j.cma.2019.112789
[1] 郑素佩,赵青宇,封建湖. 基于WENO重构保号的四阶熵稳定格式[J]. 浙江大学学报(理学版), 2022, 49(3): 329-335.