Data-Driven Modeling Approach for Pore Pressure Gradient Prediction while Drilling from Drilling Parameters.

Area of Science:

• Petroleum Engineering
• Machine Learning Applications in Geosciences

Background:

• Real-time formation pressure gradient prediction is vital for optimizing drilling operations and economics.
• Existing methods often rely on costly or unavailable well-logging data.
• Current models struggle with data limitations, necessitating alternative prediction approaches.

Purpose of the Study:

• To develop and validate machine learning models for real-time pore pressure gradient prediction.
• To utilize readily available mechanical and hydraulic drilling parameters for prediction.
• To compare the performance of Support Vector Machines, Functional Networks, and Random Forest algorithms.

Main Methods:

• Development of three predictive models: Support Vector Machines, Functional Networks, and Random Forest (RF).
• Utilized field data including mud flow rate (Q), standpipe pressure, rate of penetration, and rotary speed (RS).
• Trained models on a dataset of 3239 field data points and validated on an unseen dataset.

Main Results:

• All developed models demonstrated high accuracy, with correlation coefficients (R) up to 0.99.
• Root-mean-squared error (RMSE) was low, ranging from 0.008 to 0.021 psi/ft.
• The Random Forest model exhibited superior performance, achieving R=0.99 and minimal RMSE, with validation confirming high accuracy.

Conclusions:

• The developed machine learning models reliably predict pore pressure gradients using drilling parameters.
• These models offer a cost-effective and accessible alternative to traditional methods.
• The study highlights the potential of RF for accurate real-time drilling hazard prediction.