Applications of Artificial Intelligence for Static Poisson's Ratio Prediction While Drilling

Area of Science:

• Geosciences and Petroleum Engineering
• Artificial Intelligence in Earth Sciences

Background:

• Predicting static Poisson's ratio is crucial for reservoir characterization but is experimentally expensive and limited by current empirical methods.
• Existing methods for static Poisson's ratio determination suffer from measurement discontinuities and limited applicability.
• Artificial intelligence offers a promising alternative for real-time prediction in the oil and gas industry.

Purpose of the Study:

• To develop and compare artificial intelligence models for real-time static Poisson's ratio prediction during drilling in complex lithologies.
• To utilize mechanical drilling parameters as input features for AI-driven predictions.
• To assess the accuracy and reliability of functional networks (FN) and random forest (RF) models.

Main Methods:

• Employed functional networks (FN) and random forest (RF) algorithms.
• Utilized a dataset of 1775 records from a vertical well with complex lithology (shale, sand, carbonate).
• Validated model performance using a separate dataset from another well.

Main Results:

• Both FN and RF models demonstrated significant accuracy in predicting static Poisson's ratio.
• The RF model outperformed the FN model, achieving higher correlation coefficients (R=0.99 training, 0.94 testing) and lower average absolute percentage errors (AAPE=1.89% training, 5.19% testing).
• Validated models showed robustness, with R values of 0.94 (FN) and 0.86 (RF) and low AAPE.

Conclusions:

• Developed AI models provide an accurate and cost-effective solution for real-time static Poisson's ratio prediction.
• The random forest model is particularly effective for this application.
• These models can enhance drilling operations and reservoir analysis in complex geological formations.