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

The significance of CO2 wetting behavior in shale formations has been emphasized in various CO2 sequestration applications. Traditional laboratory experimental techniques used to assess shale wettability are complex and time-consuming. To overcome these limitations, the study proposes the use of machine learning (ML); artificial neural networks (ANN), support vector machines (SVM), and adaptive neuro-fuzzy inference systems (ANFIS) tools to estimate the contact angle, a key indicator of shale wettability, providing a more efficient alternative to conventional laboratory methods. A dataset comprising various shale samples under different conditions was collected to predict shale-water-CO2 wettability by considering shale properties, operating pressure and temperature, and brine salinity. Pearson's correlation coefficient (R) was utilized to assess the linearity between the contact angle (CA) value and other input parameters. Initial data analysis showed that the elements affecting the shale wettability are primarily reliant on the pressure and temperature at which it operates, the total organic content (TOC), and the mineral composition of the rock. Between the different ML models, the artificial neural network (ANN) model performed the best, achieving a training R-2 of 0.99, testing R-2 of 0.98 and a validation R-2 of 0.96, with an RMSE below 5. The adaptive neuro-fuzzy inference system (ANFIS) model also accurately predicted the contact angle, obtaining a training R-2 of 0.99, testing R-2 of 0.97 and a validation R-2 of 0.95. Conversely, the support vector machine (SVM) model displayed signs of overfitting, as it achieved R-2 values of 0.99 in the training dataset, which decreased to 0.94 in the testing dataset, and 0.88 in the validation dataset. To avoid rerunning the ML models, an empirical correlation was developed based on the optimized weights and biases obtained from the ANN model to predict contact angle values using input parameters and the validation data set revealed R-2 of 0.96. The parametric study showed that, among the factors influencing shale wettability at a constant TOC, pressure had the most significant impact, and the dependency of the contact angle on pressure increased when TOC values were high.

Automated summary

This study proposes the use of machine learning techniques to estimate the contact angle of shale, providing a more efficient alternative to traditional laboratory methods. By considering shale properties, operating conditions, and brine salinity, a dataset was collected and used to train artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) models. The results show that the ANN model performed the best in predicting the contact angle.