Speaker：Jie Chen（XJTLU）

Time：2022-04-20, 10:30

Location：Tencent Meeting ID： 680829103（No Password）

Abstract:

In multiscale modeling of subsurface fluid flow in heterogeneous porous media, standard polynomial basis functions are replaced by multiscale basis functions to acquire multiscale properties. To produce such functions, a number of Partial Differential Equations (PDEs) must be solved, so it makes sense to replace PDEs solvers with data-driven methods, given their great capabilities and general acceptance in recent decades. A mixed Generalized Multiscale Finite Element Method (mixed GMsFEM) has been recently proposed for subsurface flow problems, which approximates the pressure and velocity in the multiscale and fine grid space, respectively. The main purpose of this talk is to develop four distinct Convolutional Neural Network (CNN) models to predict four different multiscale basis functions for the mixed GMsFEM. These models have been applied to the 249,375 samples generated by the MatLab software, with the permeability field as the only input. The statistical results indicate that the four developed models yield satisfactory outputs with a coefficient of determination (R2) of 0.8328 - 0.9049 and Mean Squared Error (MSE) of 0.0109 - 0.0261. Graphically, all models follow the observed trend in each coarse block. Looking at this work as an image (matrix)-to-image (matrix) regression problem, the constructed deep learning-based models may have applications beyond reservoir engineering, such as hydrogeology and rock mechanics.