Lost circulation intensity characterization in drilling operations: Leveraging machine learning and well log data.

Area of Science:

• Geosciences
• Petroleum Engineering
• Data Science

Background:

• Lost circulation poses significant financial and operational risks during drilling.
• Geological parameters, particularly in challenging formations, are primary drivers of lost circulation incidents.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting lost circulation intensity.
• To leverage well-log data for enhanced forecasting of lost circulation events.

Main Methods:

• Applied seven machine learning algorithms: Random Forest, Extra Trees, Decision Tree, XGBoost, k-Nearest Neighbors, Support Vector Machine, and Hard Voting.
• Utilized nine well logs from a gas field in northern Iran, categorizing lost circulation into six intensity classes.
• Conducted rigorous exploratory data analysis and preprocessing.

Main Results:

• Random Forest, Extra Trees, and Hard Voting demonstrated superior performance in predicting lost circulation intensity.
• Extra Trees and Hard Voting models exhibited very high predictive accuracy.
• Ensemble methods proved effective in handling the multivariate nature of the problem.

Conclusions:

• Machine learning integration with well-log data offers a robust foundation for real-time drilling decision-making.
• The developed models have the potential to significantly lower operational risks, enhance drilling safety, and minimize nonproductive time.
• Ensemble techniques, like Hard Voting, are highly effective for complex prediction tasks in petroleum operations.