A Comprehensive Review of Characterizing CO2-Brine Interfacial Tension in Saline Aquifers using Machine Learning

Abstract

The alarming increase in global warming, primarily driven by the rising CO₂ concentration in the atmosphere, has spurred the need for technological solutions to reduce CO₂ concentrations. One widely successful approach is geological sequestration, which involves pressurizing and injecting CO₂ into underground rock formations. Saline aquifers, containing saltwater, are often used for this purpose due to their large storage capacity and broad availability. However, to optimize CO₂ storage and reduce the risk of gas leakage, it is essential to account for capillary forces and the interfacial tension (IFT) between CO₂ and brine within the formation. Traditional methods for characterizing CO₂-brine IFT in saline aquifers, both experimental and theoretical, are well-documented in the literature. Experimental methods, though accurate, are labor-intensive, time-consuming, and require expensive equipment, while theoretical approaches rely on idealized models and computationally demanding simulations. Recently, machine learning (ML) techniques have emerged as a promising alternative for IFT characterization. These techniques allow models of CO₂-brine IFT to be automatically "learned'' from data using optimization algorithms. The literature suggests that ML can achieve superior accuracy compared to traditional theoretical methods. However, in its current state, the literature lacks a comprehensive review of these emerging methods. This work addresses that gap by offering an in-depth survey of existing machine learning techniques for IFT characterization in saline aquifers, while also introducing novel, unexplored approaches to inspire future advancements. Our comparative analysis shows that simpler ML models, such as ensemble tree-based models and small multi-layer perceptrons, may be the most accurate and practical for estimating CO₂-brine IFT in saline aquifers.