Deep learning analysis of Ar, Xe, Kr, and O2 adsorption on Activated Carbon and Zeolites using ANN approach

Abstract

In the present study, the artificial neural network (ANN) has been used to predict the adsorbed amount of Ar, Xe, Kr, and O2 on zeolites and activated carbon. The experimental (data set 1400) on adsorbent type, gas type, and pressure of this adsorption, defined by the 303°K isotherm, have been applied as input datasets for ANN development. The adsorbed amount was used as an ANN's output dataset. The Bayesian Regularization (BR) algorithm was applied as the two-layer network training from a Multi-layer perceptron (MLP) technique with 25 neurons. Furthermore, it was compared to the Radial based functions (RBF) algorithm, which uses a single 60-neuron hidden layer. The MLP and RBF networks had the best Mean Square Error (MSE) validating the efficiency of 0.00004 and 0.00071, respectively, after 100 epochs. The square of the coefficient of correlations (R2) for the MLP and RBF models were 0.9998 and 0.9978, respectively. The generated network weight matrix can predict the adsorption behavior of different adsorbents under different process conditions with high efficiency (accuracy over 99%).