Real-Time Prediction of Petrophysical Properties Using Machine Learning Based on Drilling Parameters

Area of Science:

• Geoscience
• Petroleum Engineering
• Machine Learning

Background:

• Accurate prediction of rock porosity and permeability is vital for reservoir assessment.
• Traditional methods are time-consuming and expensive, hindering comprehensive reservoir evaluation.

Purpose of the Study:

• To develop and evaluate a novel, cost-effective approach for real-time prediction of rock porosity and permeability.
• To leverage readily available drilling parameters for efficient reservoir property estimation.

Main Methods:

• Collected drilling parameters (ROP, GPM, RPM, SPP, torque, WOB) and core analysis data (porosity, permeability) from two Middle Eastern wells.
• Employed and evaluated three machine learning models: Decision Trees (DTs), Random Forest (RFs), and Support Vector Machines (SVMs).

Main Results:

• All models achieved high correlation coefficients (R > 0.91) for porosity prediction.
• The Random Forest model demonstrated excellent permeability prediction (R > 0.92).
• Decision Trees showed slightly lower performance (R=0.88), while SVMs experienced overfitting (R=0.83 on test data).

Conclusions:

• Machine learning models successfully predict reservoir properties in real-time using drilling parameters.
• This approach provides a practical and efficient alternative to traditional methods for reservoir evaluation.
• The study enhances decision-making and exploration/production activities in the oil and gas industry.