Abstract
Abstract
Executive Summary
Subsurface carbon dioxide is viewed as a critical technology to combat and mitigate climate change in the upcoming decades. Storing carbon dioxide in deep saline formations is regarded as a promising option for geological carbon storage (GCS), but to fully comprehend the CO2 trapping processes in these formations, reliable methods must be established to evaluate the efficiency of CO2 trapping. In this project, 4 machine learning techniques (Decision Tree, Random Forest, Gradient Boosting and Stacked Generalization) were applied to develop a robust model to assess CO2 trapping efficiency. Additionally, Artificial Neural Network (ANN), a deep learning tool was also developed and tested. The developed ML models demonstrated promising results for predicting trapping indices in saline aquifers, with correlation factor R2 > 0.9. Moreover, partial dependance plots (PDP) were utilized as a sensitivity analysis tool and it showed that some input variables are affecting the results (trapping indices) more than others (i.e. Salinity, temperature & time). The developed ML models RF, GB, and SG specifically exhibited a propitious outcomes.