With the advancement of high-performance computing and sensor technology, it has become possible to directly identify physical laws from data. Although data-driven methods for identifying partial differential equations have demonstrated potential in various fields, their applicability and effectiveness in seepage problems still require systematic evaluation. For this, the PDE(partial differential equations)-LEARN framework based on sparse regression was employed to systematically identify the governing equations for one-dimensional linear, two-dimensional linear, and one-dimensional nonlinear seepage flows. The impacts of data volume and noise level on identification accuracy were investigated. The results indicate that the framework can accurately identify the equation structure even with limited data and under noisy conditions, while maintaining low errors in equation coefficients. Moreover, moderate noise is found to enhance equation identification performance through the stochastic resonance effect. Furthermore, by constructing test cases containing high-order derivative terms, the predictive performance of the identified equations is compared with that of fully connected neural networks, revealing that the former exhibits significant advantages in extrapolation predictions. Finally, the method's applicability to real physical processes is validated through variable-diameter sand-filled tube seepage experiments. The results show that the model predictions are highly consistent with experimental data, demonstrating the potential of this approach in practical seepage modeling.
ZHANG Kai, DANG Da-feng, ZHANG Li-ming, LIU Pi-yang, YAO Jun, YAN Xia.
Data-driven Sparse Identification of Seepage Control Equations[J].
Science Technology and Engineering, 2026, 26(11): 4486-4497 DOI:10.12404/j.issn.1671-1815.2506681
CaoL, YuanH, PanZ, et al. Dynamic scaling prediction model and application in near-wellbore formation of ultradeep natural gas reservoirs[J]. SPE Journal, 2024, 29(5): 2476-2493.
ZhangShunkang, AdilaiKuerban. Study on microscopic factors of early reasonable productivity of low permeability tight reservoir[J]. Science Technology and Engineering, 2024, 24(31): 13351-13358.
ZhuangChunhui, LiYajun, ZhangHuan, et al. The law of oil-water two-phase seepage and the calculation method of relative permeability in nanoscale pores[J]. Science Technology and Engineering, 2025, 25(11): 4505-4514.
[6]
YangG, XuR, TianY, et al. Data-driven methods for flow and transport in porous media: a review[J]. International Journal of Heat and Mass Transfer, 2024, 235: 126149.
[7]
SimonovO A, ErinaY Y, PonomarevA A. Review of modern models of porous media for numerical simulation of fluid flows[J]. Heliyon, 2023, 9(12): e22292.
[8]
AlhubailA, FahsM, LehmannF, et al. Modeling fluid flow in heterogeneous porous media with physics-informed neural networks: weighting strategies for the mixed pressure head-velocity formulation[J]. Advances in Water Resources, 2024, 193: 104797.
[9]
LvS, LiD, ZhaW, et al. Solving seepage equation using physics-informed residual network without labeled data[J]. Computer Methods in Applied Mechanics and Engineering, 2024, 418: 116563.
[10]
AliA S A, JazaeiF, ClementT P, et al. Physics-informed neural networks in groundwater flow modeling: advantages and future directions[J]. Groundwater for Sustainable Development, 2024, 25: 101172.
[11]
BruntonS L, ProctorJ L, KutzJ N. Discovering governing equations from data by sparse identification of nonlinear dynamical systems[J]. Proceedings of the National Academy of Sciences, 2016, 113(15): 3932-3937.
[12]
RudyS H, BruntonS L, ProctorJ L, et al. Data-driven discovery of partial differential equations[J]. Science Advances, 2017, 3(4): e1602614.
[13]
StephanyR, EarlsC. PDE-LEARN: using deep learning to discover partial differential equations from noisy, limited data[J]. Neural Networks, 2024, 174: 106242.
[14]
BruntonS L, KutzJ N. Promising directions of machine learning for partial differential equations[J]. Nature Computational Science, 2024, 4(7): 483-494.
[15]
MinorS, MessengerD A, DukicV, et al. Learning weather models from data with WSINDy[J]. arXiv Preprint, arXiv: 2501. 00738, 2025.
[16]
SongW, JiangS, Camps-VallsG, et al. Towards data-driven discovery of governing equations in geosciences[J]. Communications Earth & Environment, 2024, 5(1): 589.
JinG, XingH, ZhangR, et al. Data-driven discovery of governing equations for transient heat transfer analysis[J]. Computational Geosciences, 2022, 26(3): 613-631.
JiangHao, WangBofu, LuZhiming. Data-driven sparse identification of governing equations for fluid dynamics[J]. Chinese Journal of Theoretical and Applied Mechanics, 2021, 53(6): 1543-1551.
[21]
ZengJ, XuH, ChenY, et al. Deep learning discovery of macroscopic governing equations for viscous gravity currents from microscopic simulation data[J]. Computational Geosciences, 2023, 27(6): 987-1000.
[22]
SongW, ShiL, WangL, et al. Data-driven discovery of soil moisture flow governing equation: a sparse regression framework[J]. Water Resources Research, 2022, 58(8): e2022WR031926.
[23]
XuH, ZhangD, ZengJ. Deep-learning of parametric partial differential equations from sparse and noisy data[J]. Physics of Fluids, 2021, 33(3): 037132.
[24]
AbdolazimiR, JinS, VarshneyP K, et al. Harnessing the power of noise: a survey of techniques and applications[J]. arXiv Preprint, arXiv: 2410. 06348, 2024.