Revised Correlation for Accurate Estimation of CO2-Brine Interfacial Tension at Reservoir Conditions

Abstract

A recently reported correlation of CO2–brine interfacial tension (IFT) for the full range of reservoir pressure, temperature, and formation water salinity necessary to evaluate CO2 sequestration in hydrocarbon reservoirs has been revised and simplified. The new CO2–brine correlation predicts CO2–brine IFT normalized by water surface tension from temperature, CO2-water density difference normalized by the density difference between water and it's vapor(function of pressure and temperature), and the valence-weighted molality of the cations. The correlation was derived from a database that covers 1298 CO2–brine IFT data including 484 for pure water. The revised CO2–brine IFT correlation accurately predicts available data for pure CO2–pure water IFT data with an average absolute relative error (AARE) of 2.4% and standard deviation (SD) of 3.1%. Similarly, for CO2–brine IFT correlation achieves an AARE of 2.4% and a SD of 3.5%. We also examine simple methods for predicting the interfacial tension between water and mixtures of CO2 and other gases. We find that while under ideal conditions the mole-fraction weighted average is a viable approximation to the mixture interfacial tension, it typically overstates IFT in many situations. We propose a minor generalization of Kashefi's (2012) method which has better qualitative behavior and improves predictions with an AARE of approximately 5%.

Thermodynamic Properties of a Gas–Liquid–Solid System during the CO2 Geological Storage and Utilization Process: A Review

“…However, the SAFT EoS is too complicated and requires too many factors and parameters. Most recently, Jerauld et al [58] proposed a revised correlation method for the accurate estimation of the CO 2 -brine IFT based on over 1600 data points from the literature and concluded that the Kashefi [59] method has better qualitative behavior and improves predictions with an AARE of approximately 5%. Neural network-and machine learning-based models have high computing speeds, strong adaptability, and a fault-tolerant ability; however, the normalization method lacks thermodynamic meaning.…”