yohanesnuwara / volve-machine-learning

Exploration of machine learning in geoscience use cases

MLGeo

This repository is my exploration on bringing machine learning to geoscience use cases. One of the datasets is Volve oil field open dataset disclosed by Equinor since 2018. For more information about this dataset and how to access the original database, see here.

P-sonic (DT) log prediction in well 15/9-F-11B and 15/9-F-1C (Notebook)

I was glad to be invited by NEUSTRA community in collaboration w/ SPE Alexandria University in Egypt in October 29, 2020, where I shared this work and presented a demo. You can see the notebook here as a shortened version from the original notebook above.

Background: Two wells (well 15/9-F-11B and 15/9-F-1C) don't have P-sonic log.

Workflow:

• Well 15/9-F-11A, 15/9-F-11A, and 15/9-F-1B are used as training data.
• The training data are normalized using Yeo-Johnson method of power transformation, and outliers have been removed using One-Class SVM method.
• Regressors in Scikit-Learn used are: Linear Regression, Random Forest, Support Vector Machine, Decision Tree, Gradient Boosting, and K-Nearest Neighbors.
• Best performance achieved by Gradient Boosting.
• Hyperparameter tuning showed the best hyperparameter of GB as follows; n_estimators=1,000 and max_depth=100
• Average R² and RMSE score achieved are .95 and .12

Result:

The predicted results are written in CSV files; well 15/9-F-11B and 15/9-F-1

Lithology prediction from drilling data

volve-machine-learning's Issues

drilling data

Can you shared your drilling data for me ? thank you.