1. Physics‐Informed Convolutional Decoder (PICD): A Novel Approach for Direct Inversion of Heterogeneous Subsurface Flow.
- Author
-
Wang, Nanzhe, Kong, Xiang‐Zhao, and Zhang, Dongxiao
- Subjects
- *
HYDRAULIC conductivity , *GROUNDWATER flow , *PHYSICAL laws , *DEEP learning - Abstract
We propose a physics‐informed convolutional decoder (PICD) framework for inverse modeling of heterogenous groundwater flow. PICD stands out as a direct inversion method, eliminating the need for repeated forward model simulations. The framework combines data‐driven and physics‐driven approaches by integrating monitoring data and domain knowledge into the inversion process. PICD utilizes a convolutional decoder to effectively approximate the spatial distribution of hydraulic heads, while Karhunen–Loève expansion (KLE) is employed to parameterize hydraulic conductivities. During the training process, the stochastic vector in KLE and the parameters of the convolutional decoder are adjusted simultaneously to minimize the data‐mismatch and the physical violation. The final optimized stochastic vectors correspond to the estimation of hydraulic conductivities, and the trained convolutional decoder can predict the evolution and distribution of hydraulic heads. Various scenarios of groundwater flow are examined and results demonstrate the framework's capability to accurately estimate heterogeneous hydraulic conductivities and to deliver satisfactory predictions of hydraulic heads, even with sparse measurements. Plain Language Summary: Inverse modeling refers to estimate the unknown model parameters with measurements of model responses. In groundwater flow problems, the information about subsurface formation parameters is very limited, so inverse modeling is required to inference the uncertain formation parameters with sparse measurements. Many conventional inversion methods necessitate repeated forward calculations to compare the predictions with measurements and evaluate the likelihood of different estimations, resulting in a substantial computational burden. In this work, we propose a novel physics‐informed convolutional decoder (PICD) framework, which, as a direct inversion method, can circumvent the need for multiple forward calculations during the inversion process. In addition to measurements, physical laws are leveraged to provide extra information for inversion, alleviating the dependence on data, and enforcing the predictions align with measurements as well as domain‐specific knowledge. Several groundwater flow problems are considered to validate the effectiveness of the proposed PICD framework, and satisfactory performance can be obtained. The proposed PICD framework emerges as a promising tool for efficient and informed groundwater flow inverse modeling. Key Points: A physics‐informed deep learning framework is proposed for inversion of groundwater flowInversion can be performed directly without iterative forward modelingSatisfactory inversion performance can be achieved even with sparse measurements [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF