Machine Learning-Based Propped Fracture Conductivity Correlations of Several Shale Formations.

Area of Science:

• Petroleum Engineering
• Geomechanics
• Data Science

Background:

• Hydraulic fracturing relies on proppants to maintain fracture conductivity for reservoir fluid flow.
• Accurate prediction of propped fracture conductivity is crucial for optimizing fracture design.
• Existing models for fracture conductivity are often complex or lack accuracy due to data limitations.

Purpose of the Study:

• To develop simple and accurate machine learning-based correlations for propped fracture conductivity in unconventional gas shale.
• To introduce a novel application of machine learning for predicting fracture conductivity in specific shale formations.

Main Methods:

• Collected approximately 350 data points from experiments on Marcellus, Barnett, Fayetteville, and Eagle Ford shale formations.
• Employed machine learning models including artificial neural network (ANN), fuzzy logic, and functional network.
• Utilized input parameters such as fracture orientation, closure stress, proppant characteristics, and rock mechanical properties.

Main Results:

• The artificial neural network (ANN) model achieved the highest accuracy, with R-squared values of 0.89 for training and 0.93 for testing data.
• Machine learning models demonstrated superior performance compared to traditional methods.
• Individual correlations for each shale formation yielded higher accuracy.

Conclusions:

• Machine learning, particularly ANN, provides a robust and accurate method for estimating propped fracture conductivity.
• Closure stress was identified as the most significant factor influencing fracture conductivity.
• The developed correlations offer a practical tool for optimizing hydraulic fracturing design in shale gas reservoirs.