Neural Network Modeling of The Sulfur Dioxide Removal by Activated Carbon Sorbent

Abstract

An artificial neural network (ANN) model of three-layers was advanced to predict the efficiency of the sulfur dioxide (SO2) removal from the flue gas stream (SO2+air) in a fixed bed reactor using granulated activated carbon sorbent. The experimental data were collected from varying six process variables, namely, initial SO2 concentration, reaction temperature, flue gas flow rate, sorbent particle size, bed height and reaction time. The data were used to create input-base information to train and test the NN strategy. Back propagation algorithm with two hidden layers was used for training and tests the NN. The neural network predictions of SO2 removal efficiency agree with experimental data with the minimum mean squared error (MSE) for training and testing with values of 0.112*10-4 and 0.817*10-3, respectively.