Applications of Different Classification Machine Learning Techniques to Predict Formation Tops and Lithology While...

Area of Science:

• Geosciences and Petroleum Engineering
• Artificial Intelligence and Machine Learning

Background:

• Traditional methods for formation top and lithology identification are costly and time-consuming.
• Limitations include high costs, lower accuracy, and time lags, hindering real-time estimation in hydrocarbon operations.

Purpose of the Study:

• To leverage machine learning models for predicting formation tops and lithologies using accessible drilling parameters.
• To overcome the limitations of conventional techniques by developing a cost-effective and accurate real-time estimation method.

Main Methods:

• Collected real-field drilling data, including rate of penetration (ROP), weight on bit (WOB), and torque, from two wells in the Middle East.
• Trained and tested Gaussian naive Bayes (GNB), logistic regression (LR), and linear discriminant analysis (LDA) models on well data.
• Validated model performance on unseen data from a separate well to assess generalization capabilities.

Main Results:

• Gaussian naive Bayes (GNB) demonstrated superior performance, achieving high accuracy, precision, recall, and F1 scores for lithology prediction.
• Weight on bit (WOB) and rate of penetration (ROP) were identified as key parameters influencing lithology identification.
• Models achieved validation accuracies of approximately 0.96 for GNB, 0.95 for LR, and 0.92 for LDA.

Conclusions:

• Machine learning models effectively predict formation lithology and tops in real time using readily available drilling parameters.
• The developed approach offers a highly accurate and cost-effective solution for optimizing drilling processes and mitigating risks in hydrocarbon exploration.