Skip to content

Latest commit

 

History

History
59 lines (39 loc) · 1.32 KB

README.md

File metadata and controls

59 lines (39 loc) · 1.32 KB
Raw

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