Rufno

README.md

This repository contains the code and figures associated with the paper:

Chu, A.K.; Benson, S.M.; Wen, G. Deep-Learning-Based Flow Prediction for CO2 Storage in Shale–Sandstone Formations. Energies 2023, 16, 246. https://doi.org/10.3390/en16010246

---

Model

Run_FNORUNet3_dP_4layer.py: trains dP model. Can pass arguments specifying parameters such as the training/validation data set size, error type, learning rate, modes, etc.

Run_FNORUNet3_SG_5layer.py: trains SG model.

Run_FNORUNet4_dP_4layer_0rerr.py: trains dP model, with a loss function excluding the r-error.

Run_FNORUNet4_SG_5layer_0rerr.py: trains SG model, with a loss function excluding the r-error.

FNORUNet_4layer_model.py: model architecture for RU-FNO with 4 ResNet layers.

FNORUNet_5layer_model.py: model architecture for RU-FNO with 5 ResNet layers.

---

Analysis

analysis.ipynb: plots for analysis of shale case studies.

calculateErr.ipynb: calculate R2 scores and mean errors of models.

dataExample.ipynb: plots for examples from training data.

dataGenerationExample.ipynb: plots illustrating data generation methodology.

plotResults.ipynb: plots for model prediction results.

R2plots.ipynb: R2 histograms and scatter plots (Fig 2)

R2training.ipynb: plots of R2 score over training process (Fig 2)

sleipnerSim.ipynb: model prediction for Sleipner-like reservoir (Fig 11)

speedup.ipynb: calculation of model speedup (Table 3)

The .npy data and PyTorch model files referenced in the code are available here.

---

Figures

.png files for figures are located in the Figures directory.