Data-driven models to predict shale wettability for CO₂ sequestration applications.

Area of Science:

• Geosciences and Petroleum Engineering
• Computational Science and Data Analytics

Background:

• Understanding CO2 wetting behavior in shale is crucial for CO2 sequestration.
• Traditional experimental methods for assessing shale wettability are complex and time-consuming.

Purpose of the Study:

• To develop efficient machine learning (ML) models for predicting shale wettability via contact angle estimation.
• To compare the performance of artificial neural networks (ANN), support vector machines (SVM), and adaptive neuro-fuzzy inference systems (ANFIS).

Main Methods:

• Collected a dataset of shale samples under various conditions (shale properties, pressure, temperature, brine salinity).
• Trained and validated ANN, SVM, and ANFIS models to predict contact angle (CA).
• Developed an empirical correlation based on the optimized ANN model.

Main Results:

• ANN model achieved the highest accuracy (R² of 0.98 on testing data, 0.96 on validation data).
• ANFIS also showed strong predictive performance (R² of 0.97 on testing data, 0.95 on validation data).
• SVM model exhibited overfitting; an empirical correlation yielded an R² of 0.96.

Conclusions:

• ML models, particularly ANN, provide an efficient alternative for predicting shale wettability.
• Pressure and total organic content (TOC) are key factors influencing shale wettability.
• The developed empirical correlation offers a practical tool for contact angle prediction.