Application of various machine learning techniques in predicting coal wettability for CO2 sequestration purpose

Abstract

Carbon dioxide (CO2) sequestration in underground formations is one of the effective processes of decreasing carbon emissions. CO2 injection in coalbeds improves methane recovery from coal formations (ECBM) with storing CO2 for environmental purposes. The performance ECBM process and CO2 injection depend on the wettability behavior in the coal formation. The wettability can be measured using different experiments; however, these measurements are time-consuming, expensive, and highly inconsistent. Therefore, this paper aims to apply function networks (FN), support vector machine (SVM), and random forests (RF) to predict the contact angle in the coal formation. A dataset of 250 points was collected for different coal samples at different conditions. The different machine learning (ML) tools were used to predict contact angle (CA) as a function of coal properties, system pressure, and temperature. The training to testing data set ratio was used to be 70:30. A set of data hidden from the model was used for the validation purpose of the predictive models. The ML models showed a high capability to accurately predict the CA as a function of coal properties and the system conditions. The R-value (correlation coefficient) and the AAPE (average absolute percentage error) were used to evaluate the models' performance. R-values between actual and estimated CA from the FN model were calculated to be 0.97 and 0.97 comparing to 0.99, and 0.97 from the RF model in the case of training and testing datasets, while it was 0.95 and 0.96 for the SVM model. AAPE was less than 9% in the different ML models. This study delivers ML tools to accurately estimate the contact angle in the coal formation based on the coal properties, pressure and temperature, and water salinity without the need for experimental measurements of complicated calculations.