Implementation of BernFlow-Net: a 3D convolutional neural network
- Download the desired data from my google drive (or create your own via your preferred simulation method)
- Create a new conda env
conda create --name bernie_env python=3.6 keras-gpu matplotlib spyder
conda activate bernie_env
pip install git+https://github.com/je-santos/livelossplot
pip install hdf5storage- Use the train.py script to train a model. We can play with the features that go in, to assess which are the most relevant.
This is how our network looks like:
The rest of the necessary packages should be available via pip
The full publication and all the training/testing data can be found here. An excel file is provided with the list of samples available.
The keras tunner could be used to optimize the number of filters on each encoding branch
We welcome collaborations
If you use our code for your own research, we would be grateful if you cite our publication AWR
@article{PFN2020,
title = "PoreFlow-Net: a 3D convolutional neural network to predict fluid flow through porous media",
journal = "Advances in Water Resources",
pages = "103539",
year = "2020",
issn = "0309-1708",
doi = "https://doi.org/10.1016/j.advwatres.2020.103539",
url = "http://www.sciencedirect.com/science/article/pii/S0309170819311145",
author = "Javier E. Santos and Duo Xu and Honggeun Jo and Christopher J. Landry and Maša Prodanović and Michael J. Pyrcz",
}