Artificial neural networks and genetic algorithms: An efficient modelling and optimization methodology for active chlorine production using the electrolysis process

Abstract

This study evaluates the effectiveness of a modelling and optimization methodology based on artificial neural networks and genetic algorithms in the prediction of the behaviour of an electrolysis process of active chlorine production from a synthetic saline effluent. Multilayer perceptrons feedforward neural networks were developed for the active chlorine production and energy consumption based on the following inputs: electrolysis time, current intensity, hydrochloric acid concentration, and chloride ion concentration. In order to diagnose and prevent the over-fitting problem during the learning process, learning curves and the regularization factor were utilized. The trained ANN models were able to successfully predict the active chlorine production and energy consumption of the process (R 2 =0.979 and MSE=3.826 for active chlorine production and R 2 =0.985 and MSE=6.952 for energy consumption). Multi-objective optimization for maximizing active chlorine production and minimizing energy consumption was carried out by a genetic algorithm using the best derived ANN models. The Pareto front obtained led to multiple non-dominated optimal points, which result in insights regarding the optimal operating conditions for the process.