Predicting trihalomethane formation in chlorinated waters using multivariate regression and neural networks

Recently, there has been increased interest in modelling disinfection by-products (DBP) in order to better understand and manage the presence of these compounds in drinking water. In this paper, the use of artificial neural networks (ANN) to predict trihalomethane (THM) formation resulting from chlorination bench-scale experiments is investigated and compared with the use of classical multivariate linear regression (MLR). ANN and MLR were developed from three databases which were generated through bench-scale chlorination essays carried out in the US and Canada. A detailed analysis of modelling results shows that for all three databases, ANNs have in general a greater ability than MLRs to predict THM formation for most water quality and chlorination conditions, with the exception of instantaneous THMs (formation immediately following chlorine addition).