Curriculum-Transfer-Learning-Based Physics-Informed Neural Networks for Simulating Long-Term-Evolution Convection-Diffusion Behaviors on Curved Surfaces

MIN Jian; FU Zhoujia; GUO Yuan

doi:10.21656/1000-0887.440320

Volume 45 Issue 9

Sep. 2024

Turn off MathJax

Article Contents

Article Navigation > Applied Mathematics and Mechanics > 2024 > 45(9): 1212-1223

MIN Jian, FU Zhoujia, GUO Yuan. Curriculum-Transfer-Learning-Based Physics-Informed Neural Networks for Simulating Long-Term-Evolution Convection-Diffusion Behaviors on Curved Surfaces[J]. Applied Mathematics and Mechanics, 2024, 45(9): 1212-1223. doi: 10.21656/1000-0887.440320

Citation:

PDF( 3389 KB)

Curriculum-Transfer-Learning-Based Physics-Informed Neural Networks for Simulating Long-Term-Evolution Convection-Diffusion Behaviors on Curved Surfaces

doi: 10.21656/1000-0887.440320

Center for Numerical Simulation Software in Engineering and Sciences, College of Mechanics and Engineering Science, Hohai University, Nanjing 211100, P.R.China

Received Date: 2023-10-25
Rev Recd Date: 2024-02-22
Publish Date: 2024-09-01

Abstract

Abstract

Physics-informed neural networks (PINNs) encode prior physical knowledge into neural networks, alleviating the need for extensive data volume within the network. However, for long-term problems involving time-dependent partial differential equations, the traditional PINN exhibits poor stability and struggles to obtain effective solutions. To address this challenge, a novel physics-informed neural network based on curriculum learning and transfer learning (CTL-PINN) was introduced. The main idea of this method is to transform the problem of long-term course simulation into multiple short-term course simulation problems within this time domain. Under the concept of curriculum learning, and step by step from simpleness to difficulty, the scope of the time domain to be solved was gradually expanded by training the PINN within small time quanta. Furthermore, the transfer learning method was adopted to transfer across the time domain based on the curriculum learning, and the PINN was gradually employed for solution, thus to achieve long-term simulation of convection-diffusion behaviors on curved surfaces. The CTL-PINN was combined with the extrinsic surface operator processing technology to simulate long-term convection-diffusion behaviors on complex surfaces, and the effectiveness and robustness of the improved physics-informed neural network were verified through multiple numerical examples.
- physics-informed neural network,
- curriculum learning,
- transfer learning,
- convection-diffusion,
- surface,
- long-term evolution

(Contributed by FU Zhoujia, M.AMM Youth Editorial Board)

FullText(HTML)

References(27)

References

[1]	SAYEVAND K, MACHADO J T, MASTI I. Analysis of dual Bernstein operators in the solution of the fractional convection-diffusion equation arising in underground water pollution[J]. Journal of Computational and Applied Mathematics, 2022, 399 : 113729. doi: 10.1016/j.cam.2021.113729
[2]	NAZEMIFARD N, MASLIYAH J H, BHATTACHARJEE S. Particle deposition onto charge heterogeneous surfaces: convection-diffusion-migration model[J]. Langmuir, 2006, 22 (24): 9879-9893. doi: 10.1021/la061702q
[3]	BARREIRA R, ELLIOTT C M, MADZVAMUSE A. The surface finite element method for pattern formation on evolving biological surfaces[J]. Journal of Mathematical Biology, 2011, 63 (6): 1095-1119. doi: 10.1007/s00285-011-0401-0
[4]	殷雅俊. 曲面物理和力学[J]. 力学与实践, 2011, 33 (6): 1-8. YIN Yajun. Physics and mechanics on curved surfaces[J]. Mechanics in Engineering, 2011, 33 (6): 1-8. (in Chinese)
[5]	CASULLI V. A semi-implicit finite difference method for non-hydrostatic, free-surface flows[J]. International Journal for Numerical Methods in Fluids, 1999, 30 (4): 425-440. doi: 10.1002/(SICI)1097-0363(19990630)30:4<425::AID-FLD847>3.0.CO;2-D
[6]	DZIUK G, ELLIOTT C M. Finite elements on evolving surfaces[J]. IMA Journal of Numerical Analysis, 2006, 27 (2): 262-292.
[7]	张一鸣, 王雪雅, 冯春, 等. 连续体的有限单元法课程理论与上机课程建设[J]. 高教学刊, 2022, 8 (16): 88-91. ZHANG Yiming, WANG Xueya, FENG Chun, et al. Continuum finite element method curriculum theory and computer-based curriculum construction[J]. Journal of Higher Education, 2022, 8 (16): 88-91. (in Chinese)
[8]	CALHOUN D A, HELZEL C. A finite volume method for solving parabolic equations on logically Cartesian curved surface meshes[J]. SIAM Journal on Scientific Computing, 2010, 31 (6): 4066-4099. doi: 10.1137/08073322X
[9]	TANG Z, FU Z, CHEN M, et al. An efficient collocation method for long-time simulation of heat and mass transport on evolving surfaces[J]. Journal of Computational Physics, 2022, 463 : 111310. doi: 10.1016/j.jcp.2022.111310
[10]	KARNIADAKIS G E, KEVREKIDIS I G, LU L, et al. Physics-informed machine learning[J]. Nature Reviews Physics, 2021, 3 (6): 422-440. doi: 10.1038/s42254-021-00314-5
[11]	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
[12]	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
[13]	MAO Z P, MENG X H. Physics-informed neural networks with residual/gradient-based adaptive sampling methods for solving partial differential equations with sharp solutions[J]. Applied Mathematics and Mechanics (English Edition), 2023, 44 (7): 1069-1084. doi: 10.1007/s10483-023-2994-7
[14]	CAI S Z, MAO Z P, WANG Z C, et al. Physics-informed neural networks (PINNs) for fluid mechanics: a review[J]. Acta Mechanica Sinica, 2021, 37 (12): 1727-1738. doi: 10.1007/s10409-021-01148-1
[15]	林云云, 郑素佩, 封建湖, 等. 间断问题扩散正则化的PINN反问题求解算法[J]. 应用数学和力学, 2023, 44 (1): 112-122. LIN Yunyun, ZHENG Supei, FENG Jianhu, et al. Diffusive regularization inverse PINN solutions to discontinuous problems[J]. Applied Mathematics and Mechanics, 2023, 44 (1): 112-122. (in Chinese)
[16]	BELLMAN R. Dynamic programming[J]. Science, 1966, 153 (3731): 34-37. doi: 10.1126/science.153.3731.34
[17]	FANG Z, ZHAN J. A physics-informed neural network framework for PDEs on 3D surfaces: time independent problems[J]. IEEE Access, 2020, 8 : 26328-26335. doi: 10.1109/ACCESS.2019.2963390
[18]	TANG Z C, FU Z J, REUTSKIY S. An extrinsic approach based on physics-informed neural networks for PDEs on surfaces[J]. Mathematics, 2022, 10 (16): 2861. doi: 10.3390/math10162861
[19]	汤卓超, 傅卓佳. 基于物理信息的神经网络求解曲面上对流扩散方程[J]. 计算力学学报, 2023, 40 (2): 216-222. TANG Zhuochao, FU Zhuojia. Physics-informed neural networks for solving convection-diffusion equations on surfaces[J]. Chinese Journal of Computational Mechanics, 2023, 40 (2): 216-222. (in Chinese)
[20]	KRISHNAPRIYAN A, GHOLAMI A, ZHE S, et al. Characterizing possible failure modes in physics-informed neural networks[J]. Advances in Neural Information Processing Systems, 2021, 34 : 26548-26560.
[21]	PENWARDEN M, JAGTAP A D, ZHE S, et al. A unified scalable framework for causal sweeping strategies for physics-informed neural networks (PINNs) and their temporal decompositions[J]. March Learning, 2023, 493 : 112464. .
[22]	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.
[23]	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.
[24]	GUO J, YAO Y, WANG H, et al. Pre-training strategy for solving evolution equations based on physics-informed neural networks[J]. Journal of Computational Physics, 2023, 489 : 112258.
[25]	GOSWAMI S, ANITESCU C, CHAKRABORTY S, et al. Transfer learning enhanced physics informed neural network for phase-field modeling of fracture[J]. Theoretical and Applied Fracture Mechanics, 2020, 106 : 102447.
[26]	MVNZER M, BARD C. A curriculum-training-based strategy for distributing collocation points during physics-informed neural network training[R/OL]. 2022[2024-02-22]. https://arxiv.org/pdf/2211.11396.
[27]	LOH W L. On Latin hypercube sampling[J]. The Annals of Statistics, 1996, 24 (5): 2058-2080.