Predicting solubility of CO2 in brine by advanced machine learning systems: Application to carbon capture and sequestration

Abstract

Carbon dioxide (CO2) capture and sequestration in saline aquifers have turned into a key focus as it becomes an effective way to reduce CO2 in the atmosphere. The solubility of CO2 in brine is of vital role in monitoring CO2 sequestration. In this study, based on molality of NaCl, pressure and temperature, modeling of CO2 solubility in brine has been carried out utilizing multilayer perceptron (MLP) and radial basis function neural network (RBFNN). Levenberg-Marquardt (LM) algorithm was implemented to optimize the MLP model, while genetic algorithm (GA), particle swarm optimization (PSO) and artificial bee colony (ABC), were applied to optimize the RBFNN model. To this end, a widespread experimental databank including 570 data sets gathered from literature was considered to implement the proposed models. Graphical and statistical assessment criteria were considered to investigate the performances of these models. The obtained results revealed that all the proposed techniques are in excellent correspondence with experimental data. In addition, the performance analyses showed that RBFNN-ABC model exhibits the higher accuracy in the prediction of CO2 solubility in brine compared with the other proposed smart approaches and the existing well-known models. The RBFNN-ABC model yields a root mean square error (RMSE) value of 0.0289 and an R2 of 0.9967. Finally, the RBFNN-ABC model validity was confirmed and a small number of probable doubtful data was detected.