Investigation and Forecasting of Economic and Environmental Performance of CO2 Sequestration in Shale Formations Using Data Analytics and Machine Learning

Due to its radiation absorption capacity in the atmosphere, carbon dioxide (CO2) has been recognized as the most critical greenhouse gas that is targeted for emission-reduction activities. Studies show that sequestration and storage of CO2 in geological formations is one of the potential approaches for reducing the impact of carbon dioxide on the climate. More recently, with the exploration and exploitation of unconventional resources such as fractured shales, these resources have been identified as ideal candidates for the sequestration and storage processes due to their deep nature, large areal extent and volume, existing infrastructure for injection, and potentially induced fracture network. While considered a potential solution, the uncertainties related to its long-term operational, financial, and sustainability aspects are still being investigated through modelling studies. In this study, by using produced methane, the net present value (NPV) of different scenarios was estimated by incorporating several economic constant parameters and applying an interest rate to take into account the time value of money. Also, by using injected CO2 volumes, carbon footprint due to the entire gas production operation was estimated for each scenario based on several assumptions. Then by considering reservoir, hydraulic fracture, and operational parameters as predictors, and estimated Net Present Value and carbon footprint as response variables, the applicability of different machine learning algorithms including Multiple Regression, Random Forest, and Neural Network was investigated to develop a predictive model. Performance comparison of these predictive models indicated that the deep neural network model predicted Net Present Value with the highest accuracy. However, none of the machine learning algorithms were successful in precisely predicting carbon footprint due to the absence of sufficient data related to the sources of carbon emissions. Once different models were developed, then the impact of predictors on the response variables was investigated to see which parameters were more important. According to the result of this investigation, Stimulated Reservoir Volume (SRV) fracture permeability was the most important variable in both cases.