Modelling of Adsorption of Methane, Nitrogen, Carbon Dioxide, Their Binary Mixtures, and Their Ternary Mixture on Activated Carbons Using Artificial Neural Network

This work examines the use of neural networks in modelling the adsorption process of gas mixtures (CO2, CH4, and N-2) on different activated carbons. Seven feed-forward neural network models, characterized by different structures, were constructed with the aim of predicting the adsorption of gas mixtures. A set of 417, 625, 143, 87, 64, 64, and 40 data points for NN1 to NN7, respectively, were used to test the neural networks. Of the total data, 60 %, 20 %, and 20 % were used, respectively, for training, validation, and testing of the seven models. Results show a good fit between the predicted and experimental values for each model; good correlations were found (R = 0.99656 for NN1, R = 0.99284 for NN2, R = 0.99388 for NN3, R = 0.99639 for Q(1) for NN4, R = 0.99472 for Q(2) for NN4, R = 0.99716 for Q(1) for NN5, R = 0.99752 for Q(3) for NN5, R = 0.99746 for Q(2) for NN6, R = 0.99783 for Q(3) for NN6, R = 0.9946 for Q(1) for NN7, R = 0.99089 for Q(2) for NN7, and R = 0.9947 for Q(3) for NN7). Moreover, the comparison between the predicted results and the classical models (Gibbs model, Generalized dual-site Langmuir model, and Ideal Adsorption Solution Theory) shows that the neural network models gave far better results.