Prediction of Poisson's ratio for a petroleum engineering application: Machine learning methods

Area of Science:

• Petroleum Geoscience
• Geomechanics
• Data-driven Modeling

Background:

• Static Poisson's ratio (νs) is critical for fracture pressure (FP) calculations in petroleum engineering.
• Laboratory determination of νs is time-consuming and costly, driving the need for alternative methods.
• Existing data-driven models for νs lack the accuracy and physical relationship insights required for critical applications.

Purpose of the Study:

• To develop a reliable and accurate data-driven model for predicting static Poisson's ratio (νs).
• To enhance model robustness by incorporating physical behavior alongside data-driven insights.
• To evaluate the impact of improved νs prediction on fracture pressure (FP) determination accuracy.

Main Methods:

• Developed and evaluated nineteen common machine learning methods using a large dataset (1691 samples).
• Selected and enhanced the best-performing model, Gaussian Process Regression (GPR), with trend analysis.
• Compared the enhanced GPR model against published methods for both νs prediction and subsequent FP calculations.

Main Results:

• The enhanced GPR model achieved a coefficient of determination (R2) of 0.95 and an average absolute percentage relative error (AAPRE) of 2.73% for νs prediction.
• GPR demonstrated accurate input-output relationships and superior precision across all practical ranges, confirmed by cross-plotting and error analyses.
• The GPR model significantly reduced the residual error in fracture pressure (FP) determination from 87% to 26%.

Conclusions:

• The proposed enhanced GPR model provides a highly accurate and robust method for predicting static Poisson's ratio (νs).
• This improved νs prediction capability leads to a substantial increase in the accuracy of fracture pressure (FP) calculations.
• The GPR model's ability to capture physical trends makes it a valuable tool for critical petroleum engineering applications.