solution-seeker-as/manywells

ManyWells: simulation of multiphase flow in thousands of wells

The ManyWells datasets contain simulations of multiphase (gas, oil, water) flow in thousands of wells. The datasets were created and shared by Solution Seeker AS to support research on data-driven methodologies and industrial applications of machine learning and AI.

Details

• —Curated and shared by: Solution Seeker AS
• —License: Creative Commons BY-NC 4.0
• —Code repository: ManyWells GitHub repository
• —Paper: ManyWells paper (preprint)

Description of datasets

There are currently three datasets for three different operating conditions:

• —manywells-sol-1: simulations with stationary boundary conditions and open-loop control (no feedback)
• —manywells-nsol-1: simulations with non-stationary boundary conditions and open-loop control (no feedback)
• —manywells-nscl-1: simulations with non-stationary boundary conditions and closed-loop control (with feedback)

Each dataset has one million data points from 2000 vertical wells (500 data points per well). Each data point corresponds to one simulation performed by the simulator code in the ManyWells GitHub repository.

The datasets include a feature called "ID", which uniquely identifies a wells. Please note that ID are unique only within each dataset, meaning that well 1 in dataset A is different from well 1 in dataset B.

The parameters used in the simulation can be found in the config files, named [dataset-name]_config.zip. The configuration is stored as a dictionary with the keys being the well IDs.

Dataset features

The following relations apply:

• —WLIQ = WOIL + WWAT
• —WTOT = WLIQ + WGAS + WGL
• —QLIQ = QOIL + QWAT
• —QTOT = QLIQ + QGAS + QGL

The flow regime features, FRBH and FRWH, may take on one of the following string values: 'bubbly', 'slug-churn', 'annular'.

The WEEKS feature is only present in non-stationary datasets (manywells-nsol-1 and manywells-nscl-1).

Personal and sensitive information

The datasets do not contain any personal, sensitive, or private data.

How to use the data?

The following snippet shows how to use the HuggingFace datasets package to download the dataset in Python.

import datasets

# Download 'manywells-sol-1'
data = datasets.load_dataset("solution-seeker-as/manywells", name='manywells-sol-1')

# Cast dataset to a Pandas DataFrame
df = data['train'].to_pandas()

# Print the data
print(df)

Citation

We kindly ask that you cite our paper if you use the ManyWells datasets or code in an academic work. You can cite it as shown in the BibTex entry below.

@article{Grimstad2026,
	title = {{ManyWells: Simulation of multiphase flow in thousands of wells}},
	author = {Bjarne Grimstad and Erlend Lundby and Henrik Andersson},
	journal = {Geoenergy Science and Engineering},
	volume = {257},
	pages = {214226},
	year = {2026},
	issn = {2949-8910},
	doi = {https://doi.org/10.1016/j.geoen.2025.214226},
}

Dataset contact

Queries about the dataset can be directed to Bjarne Grimstad (e-mail: bjarne.grimstad@solutionseeker.no).