U-FNO - an enhanced Fourier neural operator-based deep-learning model for multiphase flow

In this work, we introduce a model architecture, [U-FNO] (https://www.sciencedirect.com/science/article/pii/S0309170822000562), for solving a dynamic CO₂-water multiphase flow problem in the context of carbon capture and storage (CCS). The figure below shows that schematic of U-FNO, where we enhances the experssiveness of Fourier Neural Operator (FNO) by appending a mini U-Net path to the Fourier layer.

Data sets

The data set is available at: https://drive.google.com/drive/folders/1fZQfMn_vsjKUXAfRV0q_gswtl8JEkVGo?usp=sharing

Train set (n = 4,500):

• input: sg_train_a.pt, output: sg_train_u.pt
• input: dP_train_a.pt, output: dP_train_u.pt

Validation set (n = 500):

• input: sg_val_a.pt, output: sg_val_u.pt
• input: dP_train_a.pt, output: dP_train_u.pt

Test set (n = 500):

• input: sg_test_a.pt, output: sg_test_u.pt
• input: dP_test_a.pt, output: dP_test_u.pt

Pre-trained models

The pre-trained models is available at: https://drive.google.com/drive/folders/1eHTGITZUM55NokoWqaPSzLRoJMIoJQoD?usp=sharing

Requirements

• PyTorch 1.8.0

Citation

@article{wen2022u,
  title={U-FNO--An enhanced Fourier neural operator-based deep-learning model for multiphase flow},
  author={Wen, Gege and Li, Zongyi and Azizzadenesheli, Kamyar and Anandkumar, Anima and Benson, Sally M},
  journal={Advances in Water Resources},
  pages={104180},
  year={2022},
  publisher={Elsevier}
}