A comprehensive study on CO2 solubility in brine: Thermodynamic-based and neural network modeling

Phase equilibrium data are required to estimate the capacity of a geological formation to sequester CO2. In this paper, a comprehensive study, including both thermodynamic and neural network modeling, is performed on CO2 solubility in brine. Brine is approximated by a NaCl solution. The Redlich–Kwong equation of state and Pitzer expansion are used to develop the thermodynamic model. The equation of state constants are adjusted by genetic algorithm optimization. A novel approach based on a neural network model is utilized as well. The temperature range in which the presented model is valid is 283–383 K, and for pressure is 0–600 bar, covering the temperature and pressure conditions for geological sequestration. A two-layer network consisting 5 neurons in its hidden layer, was chosen as the optimum topology. The regression coefficient for the neural network model was calculated R2 = 0.975. In addition, the neural network model showed lower mean absolute percentage error (3.41%) compared to the thermodynamic model (3.55%).