Data-Driven Acid Fracture Conductivity Correlations Honoring Different Mineralogy and Etching Patterns.

Area of Science:

• Petroleum Engineering
• Geoscience
• Artificial Intelligence

Background:

• Acid fracturing is key for enhancing conductivity in tight carbonate formations.
• Accurate fracture conductivity estimation is vital for predicting well productivity.
• Existing models for acid fracture conductivity often lack specificity and introduce significant errors.

Purpose of the Study:

• To develop accurate conductivity correlations for acid-fractured carbonate rocks using machine learning.
• To create models that consider specific rock types and etching patterns.
• To improve the prediction of fractured well productivity.

Main Methods:

• Applied machine learning to 560 experimental acid fracture data points.
• Utilized data preprocessing, regularization, and feature selection.
• Developed a machine learning classifier for etching pattern prediction.
• Generated multivariate linear regression, ridge regression, and artificial neural network models.

Main Results:

• Achieved 93% accuracy in predicting etching patterns.
• Developed conductivity correlations with high correlation coefficients.
• Demonstrated superior performance compared to universal correlation models.
• Validated model performance using metrics like precision, accuracy, MSE, recall, and cross-validation.

Conclusions:

• Machine learning models effectively predict acid fracture conductivity in diverse carbonate rocks.
• Accounting for rock type and etching patterns enhances prediction accuracy.
• The developed models offer a more reliable approach to estimating fractured well productivity.