Modeling and prediction of CO2 partial pressure in methanol solution using artificial neural networks

Abstract

CO₂ capture techniques are being developed faster by developing models that predict the solubility of CO₂ in various solvents. Artificial neural network (ANN) model is developed in the current study to predict the solubility of CO₂ in CH₃OH + H₂O system. Correlations can predict CO₂ solubility in liquids (in different mole fractions) for the temperatures of 258–390.0 K and pressure of 0–10 MPa, respectively. In this study, prediction data for the pressure essential to dissolve CO₂ in methanol solution are reported for temperature of 258–395.0 K. Multi-layer perceptron (MLP) and radial basis functions (RBF) were applied in this study. The predictions of solubility of carbon dioxide in mixtures of water and methanol are more accurate with MLP-ANN (artificial neural network) than RBF-ANN. The proposed models and reports of experimental data on CO₂ partial pressure are found to be in good agreement. It has been found that the ANN technique provides high accuracy and good prediction. As a result, the correlation coefficient R² = 0.99 was highly accurate and the mean square error (MSE) was less than 0.1. Levenberg-Marquardt (trainlm) with the lowest MSE measured at 0.00072863 with the strongest regression coefficient (R²). The best MSE validation performance of MLP and RBF networks was 0.0066566 and 0.2166952 at 30 epochs and 50 epochs, respectively. This study showed that the MLP and RBF model explained in this study are suitable to predicting CO₂ solubility in methanol solution.