Developing artificial neural networks and response surface methodology for evaluating CO2 absorption into K2CO3/piperazine solution

Abstract

In this research, artificial neural networks (ANNs) were used to predict the mass transfer flux of CO2 in K2CO3/Piperazine solutions (NCO2). ANN models including multilayer perceptron (MLP) and radial basis function (RBF) networks were used in the modelling. response surface methodology (RSM) was used to assess the impact of the process variables to achieve the optimal conditions. The optimal values of the input parameters, including temperature, loading means and predicted, coefficient of gas and liquid mass transfer, partial pressure of CO2, and equilibrium partial pressure of CO2 were obtained 56.94, 0.472, 0.787, 3.321, 0.843, 55786.409 and 32334.814 respectively. The maximum CO2 mass flux at optimum conditions was obtained 449.915 (kmol/m2.s. It has been observed that reducing CO2 loading and PCO2* and increasing PCO2-b have a greater effect on NCO2 increase. The experimental data on CO2 absorption in K2CO3/Piperazine solutions were used as a case study to learn, test, and evaluate the 0.8, 0.1, and 0.1 values, respectively. The optimal structure of a MLP includes 4 and 5 neurons in two hidden layers, and for RBF it is equal to 50. The comparison of R2 values for MLP, RBF, and RSM shows that they were 0.9953, 0.9944, and 0.9819, respectively. This indicates that MLP has high accuracy and compatibility, as evidenced by its high R2 value. In addition, the results of the RSM and the MLP were compared, and they demonstrated the desired accommodation.

Keywords

• CO2 absorption
• Artificial neural network (ANN)
• Response surface methodology (RSM)
• K2CO3/Piperazine solution