Artificial Neural Network for Prediction of SO2 Removal and Volumetric Mass Transfer Coefficient in Spray Tower

Abstract Sulfur dioxide has serious effects on the environment and on humans, and can cause bad odors and severe respiratory diseases. One of the most common methods for controlling SO2 is the process of desulfurization using a spray tower. Despite this process being able to eliminate much of the SO2, two of the most important parameters that need to be evaluated are the removal efficiency and the volumetric coefficient of mass transfer of the gas phase (kga). Due to the large number of parameters to be evaluated and the process complexity, there are difficulties in the proposal of mathematical models to predict those parameters over a wide range of the operation conditions of a spray tower. To overcome these drawbacks artificial intelligence methods can be an alternative. The aim of this study was to obtain an artificial neural network (ANN) to predict the removal efficiency and the kga of the SO2 removal in a spray tower. The input parameters considered were the spray nozzle orifice diameter, number of nozzles, liquid flow rate, gas flow rate and SO2 inlet concentration. Four different training algorithms, different combinations of transfer functions, and several networks structures were tested in search of the best ANN. The results showed that the choosing of the best model from the training and validation steps did not generate reliable results. The best structure was defined by analyzing the results of a simulation step, which used independent data. The best model was obtained with the structure 5-9-2, trained using the Levenberg-Marquardt algorithm with Bayesian Regularization and having the softmax and linear transfer functions in the hidden and output layers, respectively. This ANN presented accurate results and predicted the behavior of experimental data as expected. The best network presented an average error of 8.44% for the outlet SO2 concentration and 4.53% for the kga, which was lower than those obtained with the usual empirical correlations. This work showed that the use of neural networks is promising in the prediction of important variables in the processes of removal of air pollutants in spray towers.