Advanced Machine Learning Models for CO<sub>2</sub> and H<sub>2</sub>S Solubility in Water and NaCl Brine: Implications for Geoenergy Extraction and Carbon Storage
Wei Wei, Peng Lü, Chen Zhu, Pan Luo, Rabah Mesdour
Abstract
Accurate determination of CO 2 and H 2 S pure gas and mixture solubility of CO 2 and H 2 S in water and brine is important to predict the chemical reactions, phase behavior, and solubility trapping of the sour gases in petroleum reservoir engineering and underground carbon storage. In this study, three machine learning (ML) models, backpropagation neural networks (BPNN), generalized regression neural network (GRNN), and eXtreme Gradient Boosting (XGBoost) models, were implemented to predict gas solubility. In addition, a fourth model called fusion model has been developed for higher accuracy by stacking the three aforementioned models. The models were trained and validated with a database of experimental data with a wide range of temperature, pressure, NaCl salinity, and initial H 2 S concentration (2784 data points). The results from the four models are highly consistent and agree well with the experimental data. Among these models, the fusion model shows the best performance in estimating the gas solubility with a root-mean-square error (RMSE) and an adjusted R 2 of 0.0271 and 0.9995, respectively, which are better than previous ML and correlation methods in the literature. In addition, a software was developed in this study for visually analyzing the gas solubility trends with different variables. Sensitivity analyses show that pressure and temperature are the most sensitive parameters. The gas solubility in water and brine increases monotonically with temperature, but logarithmically with increasing pressure. The results predict that the CO 2 + H 2 S mixture is more soluble than the pure CO 2 . Salinity has an inhibitory effect on the solubility. At high salinity, the solubility increases with increasing pressure or temperature compared to that at low salinity. The CO 2 and H 2 S solubility modeling is essential for several applications such as reservoir souring, CO 2 sequestration, CO 2 -enhanced oil recovery (CO 2 -EOR), thermal recovery/steam-assisted gravity drainage via aquathermolysis, and corrosion-related issues in sour gas fields. A library of charts and tables of solubility data were generated for the relevant gas solubility in a wide range of conditions, which provides a ready reference for geoscientists and chemical and petroleum engineers to acquire CO 2 and H 2 S solubilities at desired conditions.