Toward explicit learning frameworks for predicting the solubility of CO₂ - N₂ gas mixtures in brine: Implication for...

Carbon capture and storage (CCS) is a crucial technology for reducing industrial CO₂ emissions and mitigating climate change. However, its large-scale deployment faces significant financial challenges, with CO₂ capture and compression accounting for the great part of total costs. To reduce these expenses, the injection of impure CO₂ streams, particularly CO₂-N₂ mixtures, into saline aquifers is among the viable alternatives to minimizing the need for costly separation processes. The effectiveness of this approach depends on consistently estimating CO₂-N₂ solubility in brine under varying scenarios, as solubility directly influences storage efficiency, injection feasibility, and long-term formation stability. Experimental methods, while precise, are expensive and time-intensive, highlighting the need for efficient predictive models. In this study, robust and interpretable white-box correlations are introduced using the multigene genetic programming (MGGP) approach to estimate CO₂-N₂ gas mixtures solubility in brine. The correlations were trained and validated on a representative experimental database, ensuring reliable predictions across diverse pressure, temperature, and salinity conditions. The MGGP-based correlations achieved high overall R² of 0.9967 for CO₂ and 0.9914 for N₂, with small RMSE values of 0.000363 and 0.000052, respectively. These correlations not only ensure accuracy and physical consistency, confirmed through detailed trend analyses, but also significantly outperform previous approaches reported in the literature. Furthermore, SHAP technique was employed to enhance model interpretability and deepen understanding of input parameter contributions. These correlations offer a powerful, practical tool for real-world CCS applications, including techno-economic assessments and the design of injection strategies, especially when considering impure gas streams.