Abstract:Numerical simulation of multi-phase fluid dynamics in porous media is critical to a variety of geoscience applications. Data-driven surrogate models using Convolutional Neural Networks (CNNs) have shown promise but are constrained to regular Cartesian grids and struggle with unstructured meshes necessary for accurately modeling complex geological features in subsurface simulations. To tackle this difficulty, we build surrogate models based on Graph Neural Networks (GNNs) to approximate space-time solutions of multi-phase flow and transport processes. Particularly, a novel Graph U-Net framework, referred to as AMG-GU, is developed to enable hierarchical graph learning for the parabolic pressure component of the coupled partial differential equation (PDE) system. Drawing inspiration from aggregation-type Algebraic Multigrid (AMG), we propose a graph coarsening strategy adapted to heterogeneous PDE coefficients, achieving an effective graph pooling operation. Results of three-dimensional heterogeneous test cases demonstrate that the multi-level surrogates predict pressure and saturation dynamics with high accuracy, significantly outperforming the single-level baseline. Our Graph U-Net model exhibits great generalization capability to unseen model configurations.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **How to construct an efficient surrogate model in the numerical simulation of multiphase fluid dynamics to accurately predict the pressure and saturation dynamics in complex geological features, especially when dealing with multiphase flow problems in heterogeneous porous media on unstructured grids**. Specifically, although traditional numerical methods (such as the finite - volume method) can accurately solve partial differential equations (PDEs), the computational cost is relatively high, especially in cases of high - dimensional and complex geometries. And data - driven surrogate models based on convolutional neural networks (CNNs) can accelerate prediction, but are limited to regular Cartesian grids and have difficulty handling unstructured grids. Therefore, this paper proposes a new surrogate model based on graph neural networks (GNNs) - AMG - GU (Algebraic Multigrid Graph U - Net) to overcome these limitations. The following are the core objectives and problem decomposition of the paper: 1. **Problem Background**: - Numerical simulation of multiphase fluid dynamics in porous media is crucial for many geoscience applications, such as hydrocarbon recovery, groundwater management, and geological carbon dioxide sequestration. - Due to the highly nonlinear and complex parabolic - hyperbolic characteristics of the multiphase flow PDE system, it is especially important to develop efficient and accurate simulation techniques. 2. **Limitations of Existing Methods**: - CNNs are suitable for regular grid data, but perform poorly when dealing with unstructured grids. - Sub - surface simulations usually require corner - point grids or unstructured grids to describe complex geological features, which further limits the application of CNNs. 3. **Solutions Proposed in the Paper**: - Construct a surrogate model based on GNNs to approximate the spatio - temporal solutions of multiphase flow and transport processes in porous media. - Propose a novel Graph U - Net framework (AMG - GU), which realizes effective hierarchical graph learning through an algebraic multigrid (AMG) - inspired graph coarsening strategy. - Design a specialized network architecture for the parabolic pressure component to capture multi - scale features and reduce computational complexity. 4. **Specific Objectives**: - Develop a graph - learning model that can handle heterogeneous PDE coefficients, which is an unresolved problem in previous work. - Verify the model performance in three - dimensional heterogeneous test cases, prove its high accuracy in predicting pressure and saturation dynamics, and significantly outperform single - level baseline models. - Demonstrate the good generalization ability of the model on unseen model configurations. In summary, the goal of this paper is to construct an efficient, accurate, and surrogate model suitable for complex geological features by combining graph neural networks and algebraic multigrid methods to solve the computational challenges in multiphase fluid dynamics simulations.

A Multigrid Graph U-Net Framework for Simulating Multiphase Flow in Heterogeneous Porous Media

Graph Convolutional Networks for Simulating Multi-phase Flow and Transport in Porous Media

Uncertainty Quantification of Two-Phase Flow in Porous Media Via the Coupled-TgNN Surrogate Model

Simulating multiphase flow in fractured media with graph neural networks

A Gradient-based Deep Neural Network Model for Simulating Multiphase Flow in Porous Media

Learning Pore-scale Multi-phase Flow from Experimental Data with Graph Neural Network

Solving the Discretised Multiphase Flow Equations with Interface Capturing on Structured Grids Using Machine Learning Libraries

AMGNET: multi-scale graph neural networks for flow field prediction

Mesh-based Super-Resolution of Fluid Flows with Multiscale Graph Neural Networks

Multi-Grid Graph Neural Networks with Self-Attention for Computational Mechanics

GNN-Surrogate: A Hierarchical and Adaptive Graph Neural Network for Parameter Space Exploration of Unstructured-Mesh Ocean Simulations

Computationally Efficient Multiscale Neural Networks Applied to Fluid Flow in Complex 3D Porous Media

Multi-scale rotation-equivariant graph neural networks for unsteady Eulerian fluid dynamics

A Graph Neural Network Framework for Grid-Based Simulation

Three-dimensional granular flow simulation using graph neural network-based learned simulator

Learning CO$_2$ plume migration in faulted reservoirs with Graph Neural Networks

Up-sampling-only and Adaptive Mesh-based GNN for Simulating Physical Systems

PowerFlowMultiNet: Multigraph Neural Networks for Unbalanced Three-Phase Distribution Systems

A graph neural network-based framework to identify flow phenomena on unstructured meshes

Graph Neural Network-based surrogate model for granular flows

Multipole Graph Neural Operator for Parametric Partial Differential Equations