PH neutralization plant optimization using artificial neural network

Abstract

This study deals with optimization techniques for pH neutralization process in order to predict the pH value. The process is Single Input Single Output (SISO) system, where the input is alkaline dosing pump percentage and the output is pH value. The experiment is in the open-loop test. The data was analyzed by three algorithms of neural networks, i.e. Bayesian Regularization Neural Network (BRNN), Levenberg Marquardt Neural Network (LMNN) and Scaled Conjugate Gradient Neural Network (SCGNN). Among the three algorithms of artificial neural networks (ANN), BRNN gave the most accurate predictions for pH value. Based on the correlation coefficient, R-value, BRNN, and LMNN are equally efficient. However, in terms of the mean square error, MSE value, BRNN is performed better compare with LMNN. Results indicated that the ANN with ten hidden neurons achieved the best prediction accuracy based on R-value and MSE value. The identified ANN model architecture will be used to apply at the pH neutralization process plant to evaluate the actual performance.