HT-PeRCNN: Physics-encoded Recurrent Convolutional Neural Network Based on Hessian Trace Weight
(Information Service Center, Binnan Oil Production Plant of Shengli Oilfield Branch, China Petroleum and Chemical Co., Ltd., Binzhou 256600, China)
Abstract
In recent years, Physics-Informed Neural Networks (PINNs) have achieved significant progress as a Deep Learning-based approach for solving Partial Differential Equations (PDEs) in various fields. However, PINNs still suffer from problems of low training efficiency and slow inference speed. To address these problems, this study proposes an improved method of Physics-encoded Recurrent Convolutional Neural Network (PeRCNN) based on the Hessian trace weight, named HT-PeRCNN. The method uses the Hessian trace as a weighted factor to optimize the weighted distribution of loss function, enhancing model stability and extrapolation capability. Experimental results show that HT-PeRCNN improves solution accuracy by 50% compared to PeRCNN in multiple PDE-solving tasks.
Keywords
PDE; PINN; PeRCNN; Hessian trace weight
References
- [1] CAI S,MAO Z,WANG Z,et al. Physics-Informed Neural Networks (PINNs) for Fluid Mechanics: A Review [J].Acta Mechanica Sinica,2021,37(12):1727-1738.
- [2] CAI S,WANG Z,WANG S,et al. Physics-Informed Neural Networks for Heat Transfer Problems [J/OL].Journal of Heat Transfer,2021,143(6):060801(2021-04-21).https://doi.org/10.1115/1.4050542.
- [3] WOLFF T,CARRILLO H,MARTI L,et al. Assessing Physics-Informed Neural Networks in Ocean Modelling and Climate Change Applications [R/OL].(2021-05-07).https://iclr.cc/virtual/2021/workshop/2123.
- [4] ANDERSON J D,WENDT J. Computational Fluid Dynamics [M].Cham:Springer,1995.
- [5] FRIES T P,BELYTSCHKO T. The Extended/Generalized Finite Element Method: An Overview of the Method and Its Applications [J].International Journal for Numerical Methods in Engineering,2010,84(3):253-304.
- [6] SHU C W. High-Order Finite Difference and Finite Volume WENO Schemes and Discontinuous Galerkin Methods for CFD [J].International Journal of Computational Fluid Dynamics,2003,17(2):107-118.
- [7] WEINAN E,HAN J,JENTZEN A,et al. Deep Learning-Based Numerical Methods for High-Dimensional Parabolic Partial Differential Equations and Backward Stochastic Differential Equations [J].Communications in Mathematics and Statistics,2017,5(4):349-380.
- [8] 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.
- [9] Tian Shifang, Li Biao Solving complex nonlinear problems using gradient optimization based physical information neural networks [J]. Acta Physica Sinica, 2023, 72 (10): 15-25
- [10] 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.
- [11] JIN X,CAI S,LI H,et al. NSFnets (Navier-Stokes Flow Nets): Physics-Informed Neural Networks for the Incompressible Navier-Stokes Equations [J/OL].Journal of Computational Physics,2021,426:109951(2021-01-05).https://doi.org/10.1016/j.jcp.2020.109951.
- [12] SARABIAN M,BABAEE H,LAKSARI K. Physics-Informed Neural Networks for Brain Hemodynamic Predictions Using Medical Imaging [J].IEEE Transactions on Medical Imaging,2022,41(9):2285-2303.
- [13] HAN J,XUE L,WEI Y,et al. Physics-Informed Neural Network-Based Petroleum Reservoir Simulation with Sparse Data Using Domain Decomposition [J].Petroleum Science,2023,20(6):3450-3460.
- [14] Xu Baochang, Zhang Xuezhi, Wang Yaxin, etc Adaptive physical information neural network model for predicting annular pressure in two-phase flow [J]. Acta Petroleum Sinica, 2023, 44 (3): 545-555
- [15] DU J,ZHENG J,LIANG Y,et al. Deeppipe:A Two Stage Physics Informed Neural Network for Predicting Mixed Oil Concentration Distribution [J/OL].Energy,2023,276:127452(2023-04-24).https://doi.org/10.1016/j.energy.2023.127452.
- [16] RAO C,REN P,WANG Q,et al. Encoding Physics to Learn Reaction–Diffusion Processes [J].Nature Machine Intelligence,2023,5(7):765-779.
- [17] SHI X,CHEN Z,WANG H,et al. Convolutional LSTM Network:A Machine Learning Approach for Precipitation Nowcasting [C]//NIPS'15: Proceedings of the 29th International Conference on Neural Information Processing Systems-Volume 1.Cambridge:MIT Press,2015:802-810.
- [18] KINGMA D P,BA J. Adam: A Method for Stochastic Optimization [J/OL].arXiv:1412.6980 [cs.LG].(2024-12-22).https://arxiv.org/abs/1412.6980.
- [19] FENG X,PROHL A. Numerical Analysis of the Allen-Cahn Equation and Approximation for Mean Curvature Flows [J].Numerische Mathematik,2003,94:33-65.
- [20] MCCLENNY L,BRAGA-NETO U. Self-Adaptive Physics-Informed Neural Networks Using a Soft Attention Mechanism [J/OL].arXiv:2009.04544 [cs.LG].(2024-06-18).https://arxiv.org/abs/2009.04544.
- [21] BONKILE M P,AWASTHI A,LAKSHMI C,et al. A Systematic Literature Review of Burgers’ Equation with Recent Advances [J].Pramana,2018,90:1-21.
- [22] KRUPA M,SANDSTEDE B,SZMOLYAN P. Fast and Slow Waves in the FitzHugh-Nagumo Equation [J].Journal of Differential Equations,1997,133(1):49-97.
- [23] REN P,RAO C,LIU Y,et al. PhyCRNet: Physics Informed Convolutional Recurrent Network for Solving Spatiotemporal PDEs [J/OL].Computer Methods in Applied Mechanics and Engineering,2022,389:114399(2021-12-18).https://doi.org/10.1016/j.cma.2021.114399.
About the authors
Fu Qian (1998.04-), female, Han ethnicity, from Guyuan, Ningxia, assistant engineer, bachelor's degree, research direction: intelligent solution methods for partial differential equations and intelligent oilfield construction.