Estimation of CO2-Brine interfacial tension using Machine Learning: Implications for CO2 geo-storage

Created by W.Langdon from gp-bibliography.bib Revision:1.8010

@Article{MOUALLEM:2024:molliq,

• author = "Johny Mouallem and Arshad Raza and Guenther Glatz and Mohamed Mahmoud and Muhammad Arif",
• title = "Estimation of {CO2-Brine} interfacial tension using Machine Learning: Implications for {CO2} geo-storage",
• journal = "Journal of Molecular Liquids",
• volume = "393",
• pages = "123672",
• year = "2024",
• ISSN = "0167-7322",
• DOI = "doi:10.1016/j.molliq.2023.123672",
• URL = "https://www.sciencedirect.com/science/article/pii/S0167732223024790",
• keywords = "genetic algorithms, genetic programming, CO geo-storage, Interfacial tension, Artificial intelligence, IFT correlation, Relevance factor analysis, Optimal storage depth, ANN",
• abstract = "Carbon capture and storage (CCS) is a promising technique to reduce anthropogenic gases causing climate change. This efficient strategy contributes toward reaching net-zero emissions and consists of capturing CO2, transporting it, and sequestering it deep down into selected geological formations. Subsurface storage of carbon dioxide (CO2) depends on several factors like injectivity, formation characteristics, seal integrity, and the associated rock-fluid and fluid-fluid interactions, etc. One critical parameter, in this context, is the interfacial tension (IFT) of the fluid-fluid system in question i.e., CO2-brine IFT for CO2 geo-storage. While experimental data for IFT of CO2-brine systems have been rigorously reported, and a few studies generated robust correlations to forecast the IFT as a function of its influencing factors, still the correlations lack in terms of accuracy and consideration of the most up-to-date data inventory. This paper thus presents a robust and accurate artificial intelligence (AI) based model to estimate the IFT of CO2-brine systems based on the largest data set (2896 points) used so far. A range of intelligent models such as Gradient Boosting (GB), Neural Network (NN), and Genetic Programming (GP) are used here to predict CO2-brine IFT. Furthermore, the most influencing factors are evaluated by using the relevance factor analysis method that helps in determining the weight of the contribution of each parameter on IFT. Our results suggest that: a) Gradient Boosting (GB) model with all its derivatives demonstrates the best accuracy for IFT prediction with a high coefficient of determination (R2) equal to 0.964, b) lowest performance is attributed to GP, and c) the impact of different factors is found to be in the order pressure > temperature > salinity > impurities. Moreover, an improved IFT correlation as a function of thermophysical and chemical properties i.e., temperature, pressure, and salinity is presented to quantify IFT with high precision (R2 = 0.886 and MRAE = 0.295) and significant time saving. This correlation is further validated and results show that it can capture the several chemical and physical processes leading to the various behavior trends of IFT stated in the literature. As a direct application in CO2 geo-storage projects, our proposed correlation is used to determine the optimal storage depth of a real carbonate saline aquifer located onshore of UAE. This study thus provides a robust model to estimate CO2-brine IFT which is important for storage capacity estimations and helps to better understand the factors influencing IFT. The model proposed here captures the dependence of CO2-brine IFT on six independent variables including pressure, temperature, brine ionic strength, cation type, and presence of impurities (CH4 and N2)",

}

Genetic Programming entries for Johny Mouallem Arshad Raza Guenther Glatz Mohamed Mahmoud Muhammad Arif Syed Hamid

Citations