Core Concepts
Developing a novel strategy using deep learning to predict online multiscale basis functions for the Richards equation.
Abstract
The content introduces a new approach combining online GMsFEM with deep learning to predict local online multiscale basis functions for the Richards equation. It discusses the importance of soil moisture, challenges in modeling unsaturated flow, and the use of GMsFEM. The process involves creating offline and online multiscale basis functions, utilizing neural networks to predict these functions efficiently, and solving the nonlinear Richards equation. The study aims to improve accuracy and reduce computational complexity through machine learning methods.
Stats
Multiple numerical experiments show good performance in predicting online multiscale basis functions.
The study uses stochastic permeability realizations and neural networks to develop a nonlinear map between permeability fields and online multiscale basis functions.
Deep neural networks are employed to speed up computing local online multiscale basis functions.
The research focuses on solving the nonlinear single-continuum Richards equation using the online GMsFEM method.
Quotes
"We employ training set of stochastic permeability realizations and computed relating online multiscale basis functions to train neural networks."
"Deep neural networks have demonstrated efficacy in solving pattern recognition tasks."
"The main motivation is to speed up computing local online multiscale basis functions."