Graph networks for porous media flow prediction

Paper Title:

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

Published on:

10 July 2023

Primary Category:

Computational Physics

Paper Authors:

Jiamin Jiang,

Bo Guo

Bullets

Key Details

•

Proposes graph networks for porous flow prediction on unstructured grids

•

Combines graph and edge convolution to capture pressure and saturation

•

Models tested on 2D heterogeneous cases with different meshes

•

Predictions match reference solutions closely despite unseen meshes

•

Around 200x speedup compared to numerical simulation

Explore the topics in this paper

generalization

graph convolutional networks

multiphase flow

porous media

surrogate models

AI generated summary

Graph networks for porous media flow prediction

This paper proposes graph convolutional networks to learn surrogate models that can rapidly predict multiphase flow dynamics in porous media, for problems with complex geometries. The models are trained on simulation data, then tested on unseen cases. Results show the models accurately predict pressure and saturation evolutions, and generalize to new meshes.