CO2 storage capacity estimation in underground geological sites using innovative data-driven techniques

Underground CO 2 storage, a core component of Carbon Capture and Storage (CCS) technologies, plays a pivotal role in mitigating greenhouse gas emissions and addressing global climate change. The importance of CO 2 storage capacity lies in its direct correlation with the potential for long-term, secure sequestration of CO 2 , thereby allowing for the continued use of fossil fuels while reducing greenhouse gas concentrations in the atmosphere. In this study, three machine learning methods (Multilayer Perceptron (MLP), Least Squares Support Vector Machine (LSSVM), and Adaptive Neuro-Fuzzy Inference System (ANFIS)) along with their hybrid combinations (hybrid MLP-LSSVM, hybrid MLP-ANFIS, and hybrid LSSVM-ANFIS) were employed to predict the storage capacity in various underground CO 2 storage sites. These sites include salt caverns, saline aquifers, depleted oil and gas reservoirs, coal seams, and basalt formations. Eight technical parameters influencing the storage capacity of these sites were utilized, comprising a total of 4545 data points. The results indicate that the three hybrid methods employed were highly effective in predicting CO 2 storage capacity, achieving a determination coefficient (R 2 ) value of 0.9999. Additionally, a sensitivity analysis performed using feature importance method revealed that the depth parameter had the greatest impact on the results, while the permeability parameter had the least influence. This study presents a comprehensive machine learning framework utilizing various types of underground CO 2 storage sites, incorporating a new set of technical parameters and innovative machine learning methods to predict the storage capacity of these sites. This approach significantly enhances both the comprehensiveness and accuracy of the findings. The findings of this study are significant for technical and economic evaluations of underground CO 2 storage sites, aiding in macro-decision-making for future projects in this sector while helping to minimize costs and risks. • Prediction of CO 2 storage capacity was done using a new set of 8 technical parameters. • A wide range of various geological CO 2 storage sites were investigated. • Storage capacity prediction was done using innovative hybrid machine learning methods. • The novel hybrid models achieved the highest accuracy. • The most significant feature for making predictions was the depth of CO 2 storage sites.