Estimation of CO2 solubility in aqueous solutions of commonly used blended amines: Application to optimised greenhouse gas capture

One of the key concerns in the 21st century, alongside the growing population, is the increase in energy consumption and the resulting global warming. The impact of CO2, a prominent greenhouse gas, has garnered significant attention in the realm of CO2 capture and gas purification. CO2 absorption can be enhanced by introducing some additives into the aqueous solution. In this study, the accuracies of some of the most up-to-date computational approaches are investigated. The employed machine learning methods are hybrid-adaptive neurofuzzy inference system (Hybrid-ANFIS), particle swarm optimization-adaptive neuro-fuzzy inference system (PSO-ANFIS), least-squares support vector machines (LSSVM) and genetic algorithm-radial basis function (GARBF). The developed models were used in estimating the solubility of CO2 in binary and ternary amines aqueous solutions. i.e. blends of monoethanolamine (MEA), triethanolamine (TEA), aminomethyl propanol (AMP), and methyldiethanolamine (MDEA). This modeling study was undertaken over relatively significant ranges of CO2 loading (mole of CO2/mole of solution) as a function of input parameters, which are 0.4-2908 kPa for pressure, 303-393.15 K for temperature, 36.22-68.89 g/mol for apparent molecular weight, and 30-55 wt % for total concentration. In this work, the validity of approaches based on different statistical graphs was investigated, and it was observed that the developed methods, especially the GA-RBF model, are highly accurate in estimating the data of interest. The obtained AARD% values for the developed models are 18.63, 8.25, 12.22, and 7.54 for Hybrid-ANFIS, PSO-ANFIS, LSSVM, and GA-RBF, respectively.