Skip to content

Skoltech-CHR/DeepField

Repository files navigation

Python Pytorch

DeepField

Machine learning framework for reservoir simulation.

img

Features

  • reservoir representation with Grid, Rock, States, Wells, Aquifer and PVT-tables components
  • interactive 3D visualization with some advanced options
  • common reservoir preprocessing tools
  • working with arbitrary large datasets of field simulations
  • constructor of neural network models
  • generative models for field augmentation
  • various model training scenarios for arbitrary long simulation periods
  • detailed documentation and step-by-step tutorials
  • complete pipelines of the reservoir simulation steps

img

Installation

Clone the repository:

git clone https://github.com/Skoltech-CHR/DeepField.git

Working with a remote server, it is recommended to install VNC for remote rendering of 3D graphics (follow this instruction)

Another option is to build the docker image with DeepField inside. Instructions and dockerfile are provided in the docker directory.

Note: the project is in developement. We welcome contributions and collaborations.

Quick start

Load a reservoir model from .DATA file:

  from deepfield import Field

  model = Field('model.data').load()

See the tutorials to explore the framework step-by-step and the documentation for more details.

Model formats

Initial reservoir model can be given in a mixture of ECLIPSE, MORE, PETREL, tNavigator formats. However, there is no guarantee that any mixture will be understood. Main file should be in .DATA file. Dependencies can be text and binary files including common formats:

  • .GRDECL
  • .INC
  • .RSM
  • .UNRST
  • .RSSPEC
  • .UNSMRY
  • .SMSPEC
  • .EGRID
  • .INIT

Within the DeepField framework it is recommended to use the HDF5 format to speed up data load and dump in Python-friendly data formats. In this case all data are contained in a single .HDF5 file. At any point the model can be exported back into .DATA text and binary files to ensure a compatibility with conventional software.

Citing

Plain text

E. Illarionov, P. Temirchev, D. Voloskov, R. Kostoev, M. Simonov, D. Pissarenko, D. Orlov, D. Koroteev, 2022. End-to-end neural network approach to 3D reservoir simulation and adaptation. J. Pet. Sci. Eng. 208, 109332. https://doi.org/10.1016/j.petrol.2021.109332

BibTex

@article{ILLARIONOV2022109332,
author = {E. Illarionov and P. Temirchev and D. Voloskov and R. Kostoev and M. Simonov and D. Pissarenko and D. Orlov and D. Koroteev},
title = {End-to-end neural network approach to 3D reservoir simulation and adaptation},
journal = {Journal of Petroleum Science and Engineering},
volume = {208},
pages = {109332},
year = {2022},
issn = {0920-4105},
doi = {https://doi.org/10.1016/j.petrol.2021.109332},
url = {https://www.sciencedirect.com/science/article/pii/S0920410521009827}
}