Enhanced Prediction of CO2 Solubility under Geological Conditions for CCUS via Improved Pitzer Parameters and Physics-Informed Machine Learning

TL;DR

This paper enhances CO2 solubility prediction in geological formations by improving Pitzer parameters and developing a physics-informed machine learning model, leading to more accurate and reliable CCS operations.

Contribution

It introduces an improved set of temperature-dependent Pitzer parameters and a physics-informed machine learning model for better CO2 solubility prediction.

Findings

Up to 76% reduction in average deviation with new Pitzer parameters.

14% reduction in prediction error with the machine learning model.

Up to 40% improvement at high salinities.

Abstract

The solubility of CO2 in formation brines plays a critical role in the efficiency of carbon capture and storage (CCS) operations. It is strongly influenced by pressure, temperature, and brine composition. Various experimental studies and modeling approaches have been developed to estimate CO2 solubility under wide ranges of pressure, temperature, and salinities. This work makes three key contributions. First, we present an extensive literature review of experimental, theoretical, and simulation-based approaches for measuring and predicting CO2 solubility across a wide range of conditions and also a discussion of how the different parameters affect solubility. Second, we introduce an improved set of temperature-dependent Pitzer interaction parameters, yielding up to a 76% reduction in average absolute deviation compared to conventional values in the geochemical simulation software PHREEQC. Third, we develop a physics-informed machine learning model that integrates thermodynamic intuition with data-driven learning, achieving a 14% reduction in prediction error over the state-of-the-art and up to 40% improvement at high salinities. Together, these advances provide a robust and accurate framework for predicting CO2 solubility, supporting more reliable CCS design and deployment.