- Notifications
You must be signed in to change notification settings - Fork1
Learning whilst drilling through real-time, near-bit prediction ahead of the drill-bit, using offset well log data.
License
justinbt1/Mystic-Bit
Folders and files
| Name | Name | Last commit message | Last commit date | |
|---|---|---|---|---|
Repository files navigation
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 thePowerPoint presentation.
Real-time, near-bit prediction 1-40m ahead of the drill-bit, using offsetwell log data
Also predict the uncertainty range.
Delivering value through:
- Improved drilling safety
- Faster decision making
- Improved well targeting
- Leveraging existing field observations
A set of 30 Gradient Boosting Decision Tree Regressors were successfullytrained on the well data, enabling prediction ahead of the bit.Lagged OH features were created, and a quartile loss functionwas 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.
- mysticbit: core python module containing ML models
- notebooks: Jupyter notebooks
- data: anonymized well log data data
- webapp/petex-hackathon: plotted/interactive charts
To create the python environment (windows), use:
conda create -n mysticbit python=3 anacondaconda activate mysticbitpython -m ipykernel install --name mysticbit