Abstract:Data assimilation for parameter and state estimation in subsurface transport problems remains a significant challenge due to the sparsity of measurements, the heterogeneity of porous media, and the high computational cost of forward numerical models. We present a physics-informed deep neural networks (DNNs) machine learning method for estimating space-dependent hydraulic conductivity, hydraulic head, and concentration fields from sparse measurements. In this approach, we employ individual DNNs to approximate the unknown parameters (e.g., hydraulic conductivity) and states (e.g., hydraulic head and concentration) of a physical system, and jointly train these DNNs by minimizing the loss function that consists of the governing equations residuals in addition to the error with respect to measurement data. We apply this approach to assimilate conductivity, hydraulic head, and concentration measurements for joint inversion of the conductivity, hydraulic head, and concentration fields in a steady-state advection--dispersion problem. We study the accuracy of the physics-informed DNN approach with respect to data size, number of variables (conductivity and head versus conductivity, head, and concentration), DNNs size, and DNN initialization during training. We demonstrate that the physics-informed DNNs are significantly more accurate than standard data-driven DNNs when the training set consists of sparse data. We also show that the accuracy of parameter estimation increases as additional variables are inverted jointly.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is parameter and state estimation in underground transportation problems. Specifically, due to the sparsity of measurement data, the heterogeneity of porous media, and the high computational cost of forward numerical models, data assimilation in underground transportation problems (i.e., estimating parameters and states from observed data) faces significant challenges. To address these issues, the authors propose a Physics - Informed Neural Networks (PINNs) method for estimating spatially - dependent hydraulic conductivity, water head, and concentration fields from sparse measurement data. This method approximates unknown parameters (such as hydraulic conductivity) and states (such as water head and concentration) by jointly training multiple DNNs and achieving it by minimizing a loss function, which includes not only the error with respect to the measurement data but also the residuals of the governing equations. The following are the specific problems that this paper attempts to solve: 1. **Data Sparsity**: In underground transportation problems, measurement data is usually very limited, making it difficult to accurately estimate parameters and states. 2. **Heterogeneity of Porous Media**: The complexity and heterogeneity of underground media increase the difficulty of parameter estimation. 3. **High Computational Cost**: Traditional numerical models are computationally expensive, which limits their application range. 4. **Non - linear and Non - Gaussian Characteristics**: The inverse problems of underground fluid and transport equations are highly non - linear, and the states and parameters are non - Gaussian distributed, which poses challenges to both direct solution methods and Bayesian parameter estimation methods. By introducing PINNs, the authors aim to use machine - learning techniques to overcome the above challenges and improve the accuracy of parameter and state estimation. Specifically, PINNs can significantly improve the estimation accuracy in the case of sparse data by combining physical laws (such as Darcy's law and the advection - dispersion equation) and measurement data. ### Key Formulas 1. **DNN Approximation Formula**: \[ u(x) \approx \hat{u}(x; \theta) = y_{n_l+1}(y_{n_l}(...(y_2(x)))) \] where, \[ y_2(x) = \sigma(W_1 x + b_1) \] \[ y_3(y_2) = \sigma(W_2 y_2 + b_2) \] \[ ... \] \[ y_{n_l}(y_{n_l-1}) = \sigma(W_{n_l-1} y_{n_l-1} + b_{n_l-1}) \] \[ y_{n_l+1}(y_{n_l}) = \sigma(W_{n_l} y_{n_l} + b_{n_l}) \] 2. **Loss Function**: \[ J(\theta, \gamma) = J_d(\theta, \gamma) + \omega_f J_f(\theta, \gamma) + \omega_b J_b(\theta, \gamma) \] where, \[ J_d(\theta, \gamma) = \frac{1}{|T_u|} \sum_{x \in T_u} (\hat{u}(x; \theta) - u^*(x))^2 + \frac{1}{|T_p|} \sum_{x \in T_p} (\hat{p}(x; \theta) - p^*(x))^2 \] \[ J_f(\theta, \gamma) = \frac{1}{|T_f|} \sum_{x \in T_f} (L(\hat{u}(x; \theta); \hat{p}(x, \gamma)))^2 \] \[ J_b(\theta, \gamma) = \frac{1}{|T_b|} \sum_{x \in T_b} (B(\hat{u}(x; \theta)))^2

Physics-Informed Neural Networks for Multiphysics Data Assimilation with Application to Subsurface Transport

Physics‐Informed Deep Neural Networks for Learning Parameters and Constitutive Relationships in Subsurface Flow Problems

Physics-Based Deep Learning for Flow Problems

Physics Informed Deep Learning for Flow and Transport in Porous Media

A Neural-Network Based MPAS—Shallow Water Model and Its 4D-Var Data Assimilation System

Physics-informed neural networks for solving nonlinear diffusivity and Biot's equations

Physics-Informed Machine Learning Method for Large-Scale Data Assimilation Problems

Physics-informed learning of governing equations from scarce data

Groundwater contamination source identification and high-dimensional parameter inversion using residual dense convolutional neural network

A High Resolution Physics-informed Neural Networks for High-Dimensional Convection-Diffusion-reaction Equations.

Applying physics informed neural network for flow data assimilation

Surrogate Modeling for Porous Flow Using Deep Neural Networks

A Novel Sequential-Based Hybrid Approach Incorporating Physical Modeling and Deep Learning for Multiphase Subsurface Flow Simulation

Investigating Steady Unconfined Groundwater Flow using Physics Informed Neural Networks

1-D Coupled Surface Flow and Transport Equations Revisited Via the Physics-Informed Neural Network Approach

Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations

Physics Constrained Learning for Data-driven Inverse Modeling from Sparse Observations

Physics-informed neural network to augment experimental data: an application to stratified flows

Physics-Informed Neural Network for the Transient Diffusivity Equation in Reservoir Engineering

A physics-constrained deep learning model for simulating multiphase flow in 3D heterogeneous porous media

Physics informed deep learning for computational elastodynamics without labeled data