广西师范大学学报（自然科学版） ›› 2023, Vol. 41 ›› Issue (6): 33-50.doi: 10.16088/j.issn.1001-6600.2023032203

• • 上一篇    下一篇

改进PINNs方法求解边界层对流占优扩散方程

高飞1, 郭晓斌1, 袁冬芳2, 曹富军2*   

  1. 1.内蒙古科技大学 信息工程学院, 内蒙古 包头 014010;
    2.内蒙古科技大学 理学院, 内蒙古 包头 014010
  • 收稿日期:2023-03-22 修回日期:2023-04-24 发布日期:2023-12-04
  • 通讯作者: 曹富军(1984—), 男, 宁夏中卫人, 内蒙古科技大学副教授, 博士。 E-mail: caofujun@imust.edu.cn
  • 基金资助:
    国家自然科学基金(12161067, 12261067); 内蒙古自然科学基金(2021LHMS01006, 2022MS01008); 内蒙古自治区青年科技英才支持计划项目(NJYT20B15); 内蒙古科技大学创新基金(2019YQL02)

Improved PINNs Method for Solving the Convective Dominant Diffusion Equation with Boundary Layer

GAO Fei1, GUO Xiaobin1, YUAN Dongfang2, CAO Fujun2*   

  1. 1. School of Information Engineering, Inner Mongolia University of Science and Technology, Baotou Neimenggu 014010, China;
    2. School of Science, Inner Mongolia University of Science and Technology, Baotou Neimenggu 014010, China
  • Received:2023-03-22 Revised:2023-04-24 Published:2023-12-04

PDF (PC)

116

摘要： 针对物理信息神经网络(PINNs)在求解边界层附近存在剧烈梯度变化的对流占优扩散方程时无法得到足够精度的问题,本文提出一种具有参数渐进思想的神经网络求解方法。该方法首先近似大扩散参数方程的光滑解,然后逐步减小扩散参数并将大扩散参数下的网络最优参数作为小扩散参数神经网络的初始值进行训练,通过参数循环反复优化物理信息神经网络,提高神经网络的表征能力,从而提升物理信息神经网络逼近对流占优扩散问题的求解精度,最后获得小扩散参数的高精度奇异解。经过对本文方法与PINNs以及gPINNs方法在精度和收敛效率方面的对比分析表明,本文方法在未知边界层位置条件下,能够高效地近似对流占优扩散方程的大梯度解,实现10-3量级的精度。同时,本文方法在收敛速度和稳定性方面比PINNs和gPINNs具有更好的优势和性能。

关键词: 扩散方程, 边界层, 物理信息神经网络, 深度学习, 对流扩散

Abstract: To address the issue of insufficient accuracy when solving convection-dominated diffusion equations with drastic gradient changes near the boundary layer using Physics-Informed Neural Networks (PINNs), a neural network solution method incorporating a parameter progressive strategy is proposed. This method initially approximates the smooth solution of the large diffusion parameter equation, then progressively reduces the diffusion parameter while using the optimal network parameters from the large diffusion parameter as the initial values for training the small diffusion parameter neural network. By iteratively optimizing the Physics-Informed Neural Networks through parameter cycling, the neural network’s representational capacity is enhanced, thereby improving the approximation accuracy of the convection-dominated diffusion problem. And finally a high-precision singular solution for the small diffusion parameter is obtained. A comparison of the accuracy and convergence efficiency of present method with those of PINNs and gPINNs shows that this method can efficiently approximate the large gradient solution of the convective dominant diffusion equation with an accuracy of the order of 10-3 under the condition of unknown boundary layer position. Meanwhile, the present method has more advantages and better performance than PINNs and gPINNs in terms of convergence speed and stability.

Key words: diffusion equation, boundary layer, physical information neural networks, deep learning, convective diffusion

中图分类号:  O241.82

[1] AZIZ K, SETTARI A. Petroleum reservoir simulation[M]. London: Applied Science Publishers, 1979.
[2] ACKERMANN I J, HASS H, MEMMESHEIMER M, et al. Modal aerosol dynamics model for Europe: development and first applications[J]. Atmospheric Environment, 1998, 32(17): 2981-2999. DOI: 10.1016/S1352-2310(98)00006-5.
[3] MORTON K W. Numerical solution of convection-diffusion problems[M]. London: Chapman & Hall, 1996.
[4] SEINFELD J H, PANDIS S N. Atmospheric chemistry and physics: from air pollution to climate change[M]. 2nd ed. Hoboken: Wiley, 2006.
[5] BEAR J. Hydraulics of groundwater[M]. New York: Courier Corporation, 2012.
[6] CLAVERO C, JORGE J C, LISBONA F. A uniformly convergent scheme on a nonuniform mesh for convection-diffusion parabolic problems[J]. Journal of Computational and Applied Mathematics, 2003, 154(2): 415-429. DOI: 10.1016/S0377-0427(02)00861-0.
[7] GE Y B, CAO F J. Multigrid method based on the transformation-free HOC scheme on nonuniform grids for 2D convection diffusion problems[J]. Journal of Computational Physics, 2011, 230(10): 4051-4070. DOI: 10.1016/j.jcp.2011.02.027.
[8] KUMAR V, SRINIVASAN B. An adaptive mesh strategy for singularly perturbed convection diffusion problems[J]. Applied Mathematical Modelling, 2015, 39(7): 2081-2091. DOI: 10.1016/j.apm.2014.10.019.
[9] DU J, CHUNG E. An adaptive staggered discontinuous Galerkin method for the steady state convection-diffusion equation[J]. Journal of Scientific Computing, 2018, 77(3): 1490-1518. DOI: 10.1007/s10915-018-0695-9.
[10] DOUGLAS J J, RUSSELL T F. Numerical methods for convection-dominated diffusion problems based on combining the method of characteristics with finite element or finite difference procedures[J]. SIAM Journal on Numerical Analysis, 1982, 19(5): 871-885. DOI: 10.1137/0719063.
[11] ZHANG Y. A finite difference method for fractional partial differential equation[J]. Applied Mathematics and Computation, 2009, 215(2): 524-529. DOI: 10.1016/j.amc.2009.05.018.
[12] TAYLOR C A, HUGHES T J R, ZARINS C K. Finite element modeling of blood flow in arteries[J]. Computer Methods in Applied Mechanics and Engineering, 1998, 158(1/2): 155-196. DOI: 10.1016/S0045-7825(98)80008-X.
[13] QIN X Q, MA Y C, ZHANG Y. Two-grid method for characteristics finite-element solution of 2D nonlinear convection-dominated diffusion problem[J]. Applied Mathematics and Mechanics, 2005, 26(11): 1506-1514. DOI: 10.1007/BF03246258.
[14] EYMARD R, GALLOUËT T, HERBIN R. Finite volume methods[J]. Handbook of Numerical Analysis, 2000, 7: 713-1018. DOI: 10.1016/S1570-8659(00)07005-8.
[15] 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.
[16] BLECHSCHMIDT J, ERNST O G. Three ways to solve partial differential equations with neural networks: a review[J]. GAMM-Mitteilungen, 2021, 44(2): e202100006. DOI: 10.1002/gamm.202100006.
[17] PANG G F, LU L, KARNIADAKIS G E.fPINNs: fractional physics-informed neural networks[J]. SIAM Journal on Scientific Computing, 2019, 41(4): A2603-A2626. DOI: 10.1137/18M1229845.
[18] JAGTAP A D, KARNIADAKIS G E. Extended physics-informed neural networks(XPINNs): a generalized space-time domain decomposition based deep learning framework for nonlinear partial differential equations[J]. Communications in Computational Physics, 2020, 28(5): 2002-2041. DOI: 10.4208/cicp.OA-2020-0164.
[19] YU J, LU L, MENG X H, et al. Gradient-enhanced physics-informed neural networks for forward and inverse PDE problems[J]. Computer Methods in Applied Mechanics and Engineering, 2022, 393: 114823. DOI: 10.1016/j.cma.2022.114823.
[20] YANG L, MENG X H, KARNIADAKIS G E. B-PINNs: Bayesian physics-informed neural networks for forward and inverse PDE problems with noisy data[J]. Journal of Computational Physics, 2021, 425: 109913. DOI: 10.1016/j.jcp.2020.109913.
[21] JIN X W, CAI S Z, LI H, et al.NSFnets(Navier-Stokes flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations[J]. Journal of Computational Physics, 2021, 426: 109951. DOI: 10.1016/j.jcp.2020.109951.
[22] JAGTAP A D, KAWAGUCHI K, EM KARNIADAKIS G. Locally adaptive activation functions with slope recovery for deep and physics-informed neural networks[J]. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 2020, 476(2239): 20200334. DOI: 10.1098/rspa.2020.0334.
[23] MENG X H, LI Z, ZHANG D K, et al. PPINN:parareal physics-informed neural network for time-dependent PDEs[J]. Computer Methods in Applied Mechanics and Engineering, 2020, 370: 113250. DOI: 10.1016/j.cma.2020.113250.
[24] CHEN Y T, CHANG H B, MENG J, et al. Ensemble neural networks(ENN): a gradient-free stochastic method[J]. Neural Networks, 2019, 110: 170-185. DOI: 10.1016/j.neunet.2018.11.009.
[25] KHARAZMI E, ZHANG Z Q, KARNIADAKIS G E M. hp-VPINNs: variational physics-informed neural networks with domain decomposition[J]. Computer Methods in Applied Mechanics and Engineering, 2021, 374: 113547. DOI: 10.1016/j.cma.2020.113547.
[26] GU Y Q, YANG H Z, ZHOU C.SelectNet: self-paced learning for high-dimensional partial differential equations[J]. Journal of Computational Physics, 2021, 441: 110444. DOI: 10.1016/j.jcp.2021.110444.
[27] DWIVEDI V, SRINIVASAN B. Physics informed extreme learning machine(PIELM)-A rapid method for the numerical solution of partial differential equations[J]. Neurocomputing, 2020, 391: 96-118. DOI: 10.1016/j.neucom.2019.12.099.
[28] SHENG H L, YANG C. PFNN: a penalty-free neural network method for solving a class of second-order boundary-value problems on complex geometries[J]. Journal of Computational Physics, 2021, 428: 110085. DOI: 10.1016/j.jcp.2020.110085.
[29] RAISSI M, YAZDANI A, KARNIADAKIS G E. Hidden fluid mechanics: learning velocity and pressure fields from flow visualizations[J]. Science, 2020, 367(6481): 1026-1030. DOI: 10.1126/science.aaw4741.
[30] 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.
[31] CAI S Z, WANG Z C, WANG S F, et al. Physics-informed neural networks for heat transfer problems[J]. Journal of Heat Transfer, 2021, 143(6): 060801. DOI: 10.1115/1.4050542.
[32] 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.
[33] VAN DER MEER R, OOSTERLEE C W, BOROVYKH A. Optimally weighted loss functions for solving PDEs with neural networks[J]. Journal of Computational and Applied Mathematics, 2022, 405: 113887. DOI: 10.1016/j.cam.2021.113887.
[1] 蒋懿波, 刘会家, 吴田. 基于改进残差网络的输电线路雷击过电压识别研究[J]. 广西师范大学学报（自然科学版）, 2023, 41(4): 74-83.
[2] 杨烁祯, 张珑, 王建华, 张恒远. 声音事件检测综述[J]. 广西师范大学学报（自然科学版）, 2023, 41(2): 1-18.
[3] 王鲁娜, 杜洪波, 朱立军. 基于流形正则的堆叠胶囊自编码器优化算法[J]. 广西师范大学学报（自然科学版）, 2023, 41(2): 76-85.
[4] 张萍, 徐巧枝. 基于多感受野与分组混合注意力机制的肺结节分割研究[J]. 广西师范大学学报（自然科学版）, 2022, 40(3): 76-87.
[5] 李永杰, 周桂红, 刘博. 基于YOLOv3模型的人脸检测与头部姿态估计融合算法[J]. 广西师范大学学报（自然科学版）, 2022, 40(3): 95-103.
[6] 吴军, 欧阳艾嘉, 张琳. 基于多头注意力机制的磷酸化位点预测模型[J]. 广西师范大学学报（自然科学版）, 2022, 40(3): 161-171.
[7] 闫龙川, 李妍, 宋浒, 邹昊东, 王丽君. 基于Prophet-DeepAR模型的Web流量预测[J]. 广西师范大学学报（自然科学版）, 2022, 40(3): 172-184.
[8] 路凯峰, 杨溢龙, 李智. 一种基于BERT和DPCNN的Web服务分类方法[J]. 广西师范大学学报（自然科学版）, 2021, 39(6): 87-98.
[9] 吴玲玉, 蓝洋, 夏海英. 基于卷积神经网络的眼底图像配准研究[J]. 广西师范大学学报（自然科学版）, 2021, 39(5): 122-133.
[10] 陈文康, 陆声链, 刘冰浩, 李帼, 刘晓宇, 陈明. 基于改进YOLOv4的果园柑橘检测方法研究[J]. 广西师范大学学报（自然科学版）, 2021, 39(5): 134-146.
[11] 杨州, 范意兴, 朱小飞, 郭嘉丰, 王越. 神经信息检索模型建模因素综述[J]. 广西师范大学学报（自然科学版）, 2021, 39(2): 1-12.
[12] 邓文轩, 杨航, 靳婷. 基于注意力机制的图像分类降维方法[J]. 广西师范大学学报（自然科学版）, 2021, 39(2): 32-40.
[13] 薛涛, 丘森辉, 陆豪, 秦兴盛. 基于经验模态分解和多分支LSTM网络汇率预测[J]. 广西师范大学学报（自然科学版）, 2021, 39(2): 41-50.
[14] 唐熔钗, 伍锡如. 基于改进YOLO-V3网络的百香果实时检测[J]. 广西师范大学学报（自然科学版）, 2020, 38(6): 32-39.
[15] 张明宇, 赵猛, 蔡夫鸿, 梁钰, 王鑫红. 基于深度学习的波浪能发电功率预测[J]. 广西师范大学学报（自然科学版）, 2020, 38(3): 25-32.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
[1] 董淑龙, 马姜明, 辛文杰. 景观视觉评价研究进展与趋势——基于CiteSpace的知识图谱分析[J]. 广西师范大学学报（自然科学版）, 2023, 41(5): 1 -13 .
[2] 马乾然, 韦笃取. 基于线性耦合储备池计算的电机系统混沌预测研究[J]. 广西师范大学学报（自然科学版）, 2023, 41(6): 1 -7 .
[3] 颜闽秀, 靳琪森. 多维混沌系统的构建及其多通道自适应控制[J]. 广西师范大学学报（自然科学版）, 2023, 41(6): 8 -21 .
[4] 赵伟, 田帅, 张强, 王耀申, 王思博, 宋江. 基于改进YOLOv5的平贝母检测模型[J]. 广西师范大学学报（自然科学版）, 2023, 41(6): 22 -32 .
[5] 周桥, 翟江涛, 荚东升, 孙浩翔. 基于卷积门控循环神经网络的Web攻击检测方法[J]. 广西师范大学学报（自然科学版）, 2023, 41(6): 51 -61 .
[6] 林玩聪, 韩明杰, 靳婷. 基于数据增强的多层次论点立场分类方法[J]. 广西师范大学学报（自然科学版）, 2023, 41(6): 62 -69 .
[7] 温雪岩, 谷训开, 李祯, 黄英来, 黄鹤林. 融合释义与双向交互的成语阅读理解方法研究[J]. 广西师范大学学报（自然科学版）, 2023, 41(6): 70 -79 .
[8] 宋冠武, 陈知明, 李建军. 基于ResNet-50的级联注意力遥感图像分类[J]. 广西师范大学学报（自然科学版）, 2023, 41(6): 80 -91 .
[9] 徐紫钰, 吴克晴. Caputo型分数阶微分系统正解的唯一性[J]. 广西师范大学学报（自然科学版）, 2023, 41(6): 92 -104 .
[10] 郭洁, 索洪敏, 朱怡颖, 郭加超. 一类具有临界指数和不定位势的Kirchhoff型问题解的存在性[J]. 广西师范大学学报（自然科学版）, 2023, 41(6): 105 -112 .
版权所有 © 广西师范大学学报（自然科学版）编辑部
地址：广西桂林市三里店育才路15号　邮编：541004
电话：0773-5857325　E-mail: gxsdzkb@mailbox.gxnu.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发