Back to Projects
C
Data Science / Analysis
Choke Performance Prediction
ML Engineer (Research)
Scikit-learnRandom ForestGridSearchCVPython
About the Project
A research project comparing machine learning approaches to classical empirical correlations for predicting choke performance in Niger Delta oil wells. The Random Forest regression model was benchmarked against the Gilbert empirical correlation on ~1,000 observations across 7 wells, demonstrating superior prediction accuracy with data-driven approaches over traditional petroleum engineering formulas.
Key Highlights
- Random Forest model outperformed classical Gilbert empirical correlation
- Trained on ~1,000 observations across 7 Niger Delta wells
- GridSearchCV hyperparameter tuning for optimal model configuration
- Feature importance analysis revealing key production variables
- Residual error analysis comparing ML vs. classical approaches
Technical Challenges
Convincing domain experts that ML can outperform established petroleum engineering formulas required rigorous benchmarking. The residual analysis showed systematic errors in Gilbert's correlation that the Random Forest captured, but the model needed to be transparent enough for engineering teams to trust.