justinbt1 / mystic-bit

Learning whilst drilling through real-time, near-bit prediction ahead of the drill-bit, using offset well log data.

Mystic Bit

Project from OGA Hackathon London 2018.

• Connor Tann
• Justin Boylan-Toomey
• Patrick Davies
• Lawrie Cowley
• Alessandro Cristofori
• Dan Austin
• Jeremy Fortun

For a summary of the project, see the PowerPoint presentation.

Vision

Real-time, near-bit prediction 1-40m ahead of the drill-bit, using offset well log data

Also predict the uncertainty range.

Delivering value through:

• Improved drilling safety
• Faster decision making
• Improved well targeting
• Leveraging existing field observations

Results

A set of 30 Gradient Boosting Decision Tree Regressors were successfully trained on the well data, enabling prediction ahead of the bit. Lagged OH features were created, and a quartile loss function was used to capture uncertainty. 30+ separate models trained!

A mysticbit Python module was created to deploy the ML framework

Web app created with Flask, Plotly and Dash.

Repository layout

• mysticbit: core python module containing ML models
• notebooks: Jupyter notebooks
• data: anonymized well log data data
• webapp/petex-hackathon: plotted/interactive charts

Conda environment

To create the python environment (windows), use:

conda create -n mysticbit python=3 anaconda
conda activate mysticbit
python -m ipykernel install --name mysticbit