Deep learning-based parameterization of heterogeneous geological parameter fields

Li Feng, Shaoxing Mo, Alexander Y. Sun, ..., Jichun Wu, Xiaoqing Shi

This is a PyTorch implementation of deep learning (DL)-based parameterization model for effectively representing the heterogeneous geological parameter fields using low-dimensional latent vectors. The DL-based parameterization method generates a low-dimensional representation of the non-Gaussian permeability fields with multi- and intra-facies heterogeneity in a geological carbon storage (GCS) problem. Once trained, the DL model's decoder is able to produce a non-Gaussian permeability field given an arbitrary low-dimensional latent vector as input. This DL-based parameterization strategy can then be integrated with inverse algorithms to estimate the heterogeneous non-Gaussian permeability field, thereby enhancing the characterization of $CO_2$ plume migration in the aquifer.

Dependencies

• python 3
• PyTorch
• h5py
• matplotlib
• scipy

DL model architecture

Three original log-permeability field realizations (a-c), histograms of DL model-encoded latent variables (d-f), the DL model reconstructed log-permeability fields (g-i) and three random log-permeability field realizations generated by the DL model’s decoder given inputs drawn from the standard normal distribution (j-l)

Network Training

python3 train_DL.py

Citation

See Feng et al. (2024) for more information. If you find this repo useful for your research, please consider to cite:

@article{FENG2024104833,
	author = {Li Feng and Shaoxing Mo and Alexander Y. Sun and Dexi Wang and Zhengmao Yang and Yuhan Chen and Haiou Wang and Jichun Wu and Xiaoqing Shi},
	title = {Deep learning-based geological parameterization for history matching {CO}$_2$ plume migration in complex aquifers},
	journal = {Advances in Water Resources},
	pages = {104833},
	year = {2024},
	issn = {0309-1708},
	doi = {https://doi.org/10.1016/j.advwatres.2024.104833}
}

or:

Feng, L., Mo, S., Sun, A. Y., Wang, D., Yang, Z., Chen, Y., Wang, H., Wu, J., & Shi, X. (2024). Deep learning-based geological parameterization for history matching CO$_2$ plume migration in complex aquifers. Advances in Water Resources, 104833. https://doi.org/10.1016/j.advwatres.2024.104833

Related article: [Mo, S., Zabaras, N., Shi, X., Wu, J., 2020. Integration of adversarial autoencoders with residual dense convolutional networks for estimation of non-Gaussian hydraulic conductivities. Water Resources Research 56, e2019WR026082. doi:https://doi.org/10.1029/2019WR026082.]

Questions

Contact Li Feng ([email protected]) or Shaoxing Mo ([email protected]) with questions or comments.