Exploring artificial neural network approach and RSM modeling in the prediction of CO2 capture using carbon molecular sieves

Abstract

In this work, adsorption and reduction of CO2 by carbon molecular sieves (CMS) was modeled using response surface methodology (RSM) and artificial neuron networks (ANNs). The CO2 adsorption experiments were carried out at temperature in range of 20–80 °C, pressure in range of 2–10 bar, and time in range of 0–1800 sec. 311 experimental data points of CO2 adsorption on the carbon molecular sieves are applied in the development of an ANN model. A Bayesian regularization algorithm was used as the best algorithm among of Scaled conjugate gradient back propagation and Levenberg-Marquardt back propagation for training the Multilayer perceptron (MLP) network. The best architecture is obtained after 100 epochs and it has 7 neurons in the first hidden layer, 8 neurons in the second hidden layer. The mean square error (MSE) value of 0.0006246 was obtained at 100 epochs for the best-developed MLP. Then, performance was compared with radial basis functions (RBF) algorithm. MLP, RBF and RSM models were efficient in the modelling of the CO2 adsorption with correlation coefficients of 0.99784, 0.9497 and 0.8958, respectively. The MLP showing good agreement between the experimental data and predicted data of CO2 adsorption via CMS (R2> 0.99). It is proved that ANN algorithm is a promising model for the prediction of CO2 adsorption by carbon molecular sieves.