Estimating oil RF at exploration

This repository is the companion to the manuscript Estimating oil recovery factor at exploration stage using XGBoost classification. It contains all the code used to generate the figures and tables used in that manuscript. The software is meant specifically for training and evaluating machine learning models for predicting oil recovery factors, but much of it is sufficiently general-purpose that it could be adapted for any ordinal categorical regression task with multiple databases.

Project setup

Run

python -m venv venv
source venv/bin/activate
pip install -U pip
pip install -r requirements.txt
pip install .[dev]

You might have to install "numba==0.60.0", then ".[dev]"

The code used for the 2024 submission is in src/recovery_factor. The notebooks that generate the plots are in notebooks.

The models were trained at the command line with the bash one-liner

echo "TCA TC CA TA" | tr ' ' '\n' | parallel -j 2 rf-train -n 200 --training-data {}

The databases

TORIS: https://edx.netl.doe.gov/dataset/toris-an-integrated-decision-support-system-for-petroleum-e-p-policy-evaluation

TORIS comes from the Department of Energy.

Atlas: https://www.data.boem.gov/Main/GandG.aspx

This is an atlas produced by the Bureau of Ocean Energy Management.

Commercial This data is not publicly available.