Abstract
Identification of unknown physical processes and parameters of groundwater contaminant problems is a challenging task due to their ill-posed and non-unique nature. Numerous works have focused on determining nonlinear physical processes through model selection methods. However, identifying corresponding nonlinear systems for different physical phenomena using numerical methods can be computationally prohibitive. With the advent of machine learning (ML) algorithms, more efficient surrogate models based on neural networks (NNs) have been developed in various disciplines. In this work, a theory-guided U-net (TgU-net) framework is proposed for surrogate modeling of three-dimensional (3D) groundwater contaminant problems in order to efficiently elucidate their involved processes and unknown parameters. In TgU-net, the underlying governing equations are embedded into the loss function of U-net as soft constraints. Herein, sorption is considered to be a potential process of an uncertain type, and three equilibrium sorption isotherm types (i.e., linear, Freundlich, and Langmuir) are considered. Different from traditional approaches in which one model corresponds to one equation (Schoeniger et al. in Water Resour Res 50(12):9484–9513, 2014; Cao et al. in Hydrogeol J 27(8):2907–2918, 2019), these three sorption types are modeled through only one TgU-net surrogate. Accurate predictions illustrate the satisfactory generalizability and extrapolability of the constructed TgU-net. Furthermore, based on the constructed TgU-net surrogate, a data assimilation method is employed to identify the physical process and parameters simultaneously. The convergence of indicators demonstrates the validity of the proposed method. The influence of sparsity-promoting techniques, data noise, and quantity of observation information is also explored. Results demonstrate the feasibility of neural network learning a cluster of equations that have similar behaviors. This work shows the possibility of governing equation discovery of physical problems that contain multiple and even uncertain processes by using deep learning and data assimilation methods.
Similar content being viewed by others
References
Brunton SL, Proctor JL, Kutz JN, Bialek W (2016) Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proc Natl Acad Sci PNAS 113(15):3932–3937. https://doi.org/10.1073/pnas.1517384113
Cao T, Zeng X, Wu J, Wang D, Sun Y, Zhu X, Long Y et al (2019) Groundwater contaminant source identification via Bayesian model selection and uncertainty quantification. Hydrogeol J 27(8):2907–2918. https://doi.org/10.1007/s10040-019-02055-3
Chang H, Zhang D (2019) Identification of physical processes via combined data-driven and data-assimilation methods. J Comput Phys 393:337–350. https://doi.org/10.1016/j.jcp.2019.05.008
Chang H, Liao Q, Zhang D (2017) Surrogate model based iterative ensemble smoother for subsurface flow data assimilation. Adv Water Resour 100:96–108. https://doi.org/10.1016/j.advwatres.2016.12.001
Chen Y, Oliver DS (2013) Levenberg–Marquardt forms of the iterative ensemble smoother for efficient history matching and uncertainty quantification. Comput Geosci 17(4):689–703. https://doi.org/10.1007/s10596-013-9351-5
Chen J, Viquerat J, Hachem EJACP (2020) U-net architectures for fast prediction of incompressible laminar flows. arXiv preprint arXiv:1910.13532
Chen Y, Luo Y, Liu Q, Xu H, Zhang D (2022) Symbolic genetic algorithm for discovering open-form partial differential equations (SGA-PDE). Phys Rev Res. https://doi.org/10.1103/PhysRevResearch.4.023174
Chun-Yu G, Yi-Wei F, Yang H, Peng X, Yun-Fei K (2021) Deep-learning-based liquid extraction algorithm for particle image velocimetry in two-phase flow experiments of an object entering water. Appl Ocean Res 108:102526. https://doi.org/10.1016/j.apor.2021.102526
Dolz J, Ben Ayed I, Desrosiers C (2019) Dense multi-path U-net for ischemic stroke lesion segmentation in multiple image modalities. Brainlesion: glioma, multiple sclerosis, stroke and traumatic brain Injuries. Springer, Cham, pp 271–282. https://doi.org/10.1007/978-3-030-11723-8_27
Fetter CW (1999) Contaminant hydrogeology, 2nd edn. Prentice Hall, Englewood Cliffs
Ghanem RG, Spanos PD (1991) Stochastic finite elements: a spectral approach. Stochastic finite elements. Springer, New York
Glorot X, Bengio Y (2010) Understanding the difficulty of training deep feedforward neural networks. J Mach Learn Res 9:249–256
He T, Wang N, Zhang D (2021) Theory-guided full convolutional neural network: an efficient surrogate model for inverse problems in subsurface contaminant transport. Adv Water Resour 157:104051. https://doi.org/10.1016/j.advwatres.2021.104051
Imambi S, Prakash KB, Kanagachidambaresan GR (2021) Pytorch. Programming with TensorFlow. Springer, Cham, pp 87–104. https://doi.org/10.1007/978-3-030-57077-4_10
Jiang Z, Tahmasebi P, Mao Z (2021) Deep residual U-net convolution neural networks with autoregressive strategy for fluid flow predictions in large-scale geosystems. Adv Water Resour 150:103878. https://doi.org/10.1016/j.advwatres.2021.103878
Kingma DP, Ba JL (2015) Adam: a method for stochastic optimization. Paper presented at the international conference on learning representations.
Kuha J (2004) AIC and BIC: comparisons of assumptions and performance. Sociol Methods Res 33(2):188–229. https://doi.org/10.1177/0049124103262065
Lakshmi MVS, Saisreeja PL, Chandana L, Mounika P, U P (2021) A LeakyReLU based effective brain MRI segmentation using U-NET. Paper presented at the 1251–1256. https://doi.org/10.1109/ICOEI51242.2021.9453079
Le QT, Ooi C (2021) Surrogate modeling of fluid dynamics with a multigrid inspired neural network architecture. Mach Learn Appl. https://doi.org/10.1016/j.mlwa.2021.100176
Lee J-Y, Park J (2021) Deep regression network-assisted efficient streamline generation method. IEEE Access 9:111704–111717. https://doi.org/10.1109/ACCESS.2021.3100127
Loshchilov I, Hutter F (2017) Fixing weight decay regularization in Adam. arXiv preprint arXiv:1711.05101
Mangan NM, Kutz JN, Brunton SL, Proctor JL (2017) Model selection for dynamical systems via sparse regression and information criteria. Proc R Soc A Math Phys Eng Sci 473(2204):20170009. https://doi.org/10.1098/rspa.2017.0009
Mo S, Zabaras N, Shi X, Wu J (2019a) Deep autoregressive neural networks for high-dimensional inverse problems in groundwater contaminant source identification. Water Resour Res 55(5):3856–3881
Mo S, Zhu Y, Zabaras N, Shi X, Wu J (2019b) Deep convolutional encoder–decoder networks for uncertainty quantification of dynamic multiphase flow in heterogeneous media. Water Resour Res 55:703–728. https://doi.org/10.1029/2018WR023528
Newell A, Yang K, Deng J (2016) Stacked hourglass networks for human pose estimation. In: Computer vision—ECCV 2016. Springer, Cham, pp 483–499. https://doi.org/10.1007/978-3-319-46484-8_29
Oliver DS, Reynolds AC, Liu N (2008) Inverse theory for petroleum reservoir characterization and history matching. Cambridge University Press, Cambridge. https://doi.org/10.1017/CBO9780511535642
Raissi M, Perdikaris P, Karniadakis GE (2019) Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. J Comput Phys 378:686–707. https://doi.org/10.1016/j.jcp.2018.10.045
Rasmussen CE (2004) Gaussian processes in machine learning. Advanced lectures on machine learning. Springer, Berlin, pp 63–71. https://doi.org/10.1007/978-3-540-28650-9_4
Ronneberger O, Fischer P, Brox T (2015) U-net: convolutional networks for biomedical image segmentation. Springer, Cham, pp 234–241. https://doi.org/10.1007/978-3-319-24574-4_28
Schaeffer H (2017) Learning partial differential equations via data discovery and sparse optimization. Proc R Soc A Math Phys Eng Sci 473(2197):20160446. https://doi.org/10.1098/rspa.2016.0446
Schoeniger A, Woehling T, Samaniego L, Nowak W (2014) Model selection on solid ground: Rigorous comparison of nine ways to evaluate Bayesian model evidence. Water Resour Res 50(12):9484–9513. https://doi.org/10.1002/2014WR016062
Shelhamer E, Long J, Darrell T (2017) Fully convolutional networks for semantic segmentation. IEEE Trans Pattern Anal Mach Intell 39(4):640–651. https://doi.org/10.1109/TPAMI.2016.2572683
Srivastava D, Singh RM (2015) Groundwater system modeling for simultaneous identification of pollution sources and parameters with uncertainty characterization. Water Resour Manag 29(13):4607–4627. https://doi.org/10.1007/s11269-015-1078-8
Tang Z, Peng X, Geng S, Zhu Y, Metaxas DN (2018) CU-Net: coupled U-Nets. Paper presented at the BMVC
Tang M, Liu Y, Durlofsky LJ (2020) A deep-learning-based surrogate model for data assimilation in dynamic subsurface flow problems. J Comput Phys 413:109456. https://doi.org/10.1016/j.jcp.2020.109456
Tatang MA, Pan W, Prinn RG, McRae GJ (1997) An efficient method for parametric uncertainty analysis of numerical geophysical models. J Geophys Res Atmos 102(D18):21925–21932. https://doi.org/10.1029/97JD01654
Tibshirani R (1996) Regression shrinkage and selection via the lasso. J R Stat Soc Ser B Methodol 58(1):267–288. https://doi.org/10.1111/j.2517-6161.1996.tb02080.x
Troldborg M, Nowak W, Tuxen N, Bjerg PL, Helmig R, Binning PJ (2010) Uncertainty evaluation of mass discharge estimates from a contaminated site using a fully bayesian framework. Water Resour Res. https://doi.org/10.1029/2010WR009227
Wang N, Zhang D, Chang H, Li H (2020a) Deep learning of subsurface flow via theory-guided neural network. J Hydrol (amsterdam) 584:124700. https://doi.org/10.1016/j.jhydrol.2020.124700
Wang YD, Chung T, Armstrong RT, Mostaghimi P (2020b) Ml-lbm: machine learning aided flow simulation in porous media. arXiv preprint arXiv:2004.11675
Wang N, Chang H, Zhang D (2021a) Efficient uncertainty quantification for dynamic subsurface flow with surrogate by theory-guided neural network. Comput Methods Appl Mech Eng. https://doi.org/10.1016/j.cma.2020.113492
Wang N, Chang H, Zhang D (2021b) Efficient uncertainty quantification and data assimilation via theory-guided convolutional neural network. Paper presented at the SPE Reservoir Simulation Conference, Galveston, Texas, USA. Society of Petroleum Engineers
Wu H, Fang W, Kang Q, Tao W, Qiao R, Los Alamos National Lab. (LANL), Los Alamos, NM (United States) (2019) Predicting effective diffusivity of porous media from images by deep learning. Sci Rep 9(1):20387
Xu T, Gómez-Hernández JJ (2018) Simultaneous identification of a contaminant source and hydraulic conductivity via the restart normal-score ensemble Kalman filter. Adv Water Resour 112:106–123. https://doi.org/10.1016/j.advwatres.2017.12.011
Xu R, Wang N, Zhang D (2021) Solution of diffusivity equations with local sources/sinks and surrogate modeling using weak form theory-guided neural network. Adv Water Resour 153:103941. https://doi.org/10.1016/j.advwatres.2021.103941
Yang L, Zhang D, Karniadakis GEM, Brown Univ., Providence, RI (United States) (2020) Physics-informed generative adversarial networks for stochastic differential equations. SIAM J Sci Comput 42(1):A292–A317.https://doi.org/10.1137/18M1225409
Ye M, Meyer PD, Neuman SP, Pacific Northwest National Lab. (PNNL), Richland, WA (United States) (2008) On model selection criteria in multimodel analysis. Water Resour Res 44(3):W03428. https://doi.org/10.1029/2008WR006803
Ying S, Zhang J, Zeng L, Shi J, Wu L (2017) Bayesian inference for kinetic models of biotransformation using a generalized rate equation. Sci Total Environ 590–591:287–296. https://doi.org/10.1016/j.scitotenv.2017.03.003
Zhang D, Lu Z (2004) An efficient, high-order perturbation approach for flow in random porous media via Karhunen–Loève and polynomial expansions. J Comput Phys 194(2):773–794. https://doi.org/10.1016/j.jcp.2003.09.015
Zhang J, Zeng L, Chen C, Chen D, Wu L (2015) Efficient Bayesian experimental design for contaminant source identification. Water Resour Res 51(1):576–598. https://doi.org/10.1002/2014WR015740
Zheng C, Wang PP (1999) Mt3dms: A modular three-dimensional multispecies transport model for simulation of advection, dispersion, and chemical reactions of contaminants in groundwater systems; documentation and user’s guide. AJR Am J Roentgenol 169(4):1196–1197
Zhou Z, Tartakovsky DM (2021) Markov chain Monte Carlo with neural network surrogates: application to contaminant source identification. Stoch Env Res Risk Assess 35(3):639–651. https://doi.org/10.1007/s00477-020-01888-9
Acknowledgements
This work is partially funded by the National Natural Science Foundation of China (Grant No. 52288101), the Shenzhen Key Laboratory of Natural Gas Hydrates (Grant No. ZDSYS20200421111201738), and the SUSTech—Qingdao New Energy Technology Research Institute.
Author information
Authors and Affiliations
Contributions
TH: conceptualization, investigation, methodology, data curation, software, visualization, writing - original draft. HC & DZ: conceptualization, methodology, supervision, writing - review and editing, funding acquisition.
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no competing interests.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix
Appendix A
The fitting ability, generalizability, and extrapolability of the constructed TgU-net-based surrogate models are tested with sparse data points. In addition, the corresponding traditional U-nets are trained in the same configurations as comparisons. Predicted h fields of U-net-h and TgU-net-h are presented in Figs. 22 and 23 in Appendix A, respectively. RMSEs of these points predicted by U-net and TgU-net are given in Table 12 in Appendix A.
Appendix B
In this section, the trained TgU-net-based surrogate models are employed to process forward predictions, and are combined with the IES method for inverse modeling. Firstly, observation data of hydraulic head fields are used to estimate K fields. Then, the velocity fields from the estimated K fields are inputted into TgU-net-C to estimate unknown source parameters and physical processes according to observation data of contaminant concentration. Figure 24 shows prediction and error of the hydraulic head field of the estimated lnK field in the IES method. Comparison of convergence of initialized and estimated parameters of linear and Freundlich sorption is shown in Figs. 25 and 26 in Appendix B, respectively. Convergence of contaminant parameters of linear and Freundlich sorption are presented in Tables 13 and 14 in Appendix B, respectively. Tables 15 compares the convergence results of parameters by utilizing different sparsity-promoting methods under 10% noise (Tables 16, 17 and 18).
Appendix C
In this section, the influences of strong nonlinearity and information volume on data assimilation results are discussed. Estimated source parameters are displayed in Table 19.
Appendix D
The performances of U-net-h, TgU-net-h, U-net-C, and TgU-net-C in the forward problems are summarized in Table 20. The performances of TgU-net-h and TgU-net-C in the inverse problems are summarized in Tables 21 and 22, respectively. Noise is not considered. The number of test realizations is 100. The correlation length of lnK along each side is 0.2 times the corresponding geometrical length. \(\stackrel{-}{lnK(\mathbf{x})}\) is set to be 1, and the variance of lnK is 0.5. The sorption type is random. With the assistance of theory-guidance, the prediction accuracy can be significantly improved. And the estimated parameters converge well near the references, further demonstrating the validity of the proposed method.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
He, T., Chang, H. & Zhang, D. Identification of physical processes and unknown parameters of 3D groundwater contaminant problems via theory-guided U-net. Stoch Environ Res Risk Assess 38, 869–900 (2024). https://doi.org/10.1007/s00477-023-02604-z
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00477-023-02604-z