Prediction of Daily Flows and Disinfection Byproducts of the Union Water Plant using Artificial Neural Networks

Zoe J. Y. Zhu, W. Guo, B. MacKay and E. McBean


To provide for improvements in efficiency and ability to respond to changing external conditions, an ANN model is described to characterize reservoir/water tower contents sufficient to meet water demands on the Union Water Treatment plant. Maintaining water levels is a necessary dimension to reduce the precursors of disinfection byproducts, such as dissolved organic carbon, and to control the rate of disinfection byproduct formation. In this research, predictive models are developed to investigate the effect of maximum daily temperatures, incoming solar radiation and total daily precipitation (which influence water demands) and disinfection byproduct formation at the Union Water Treatment plant.

Artificial neural networks (ANNs) are essentially semi-parametric regression estimators that are well suited for predicting water demands and quality as they are able to approximate virtually any function, to varying degrees of accuracy. In this application, predictions of water demand and DBP are obtained using a simple back-propagation neural network. A comparative evaluation of the stepwise regression method and the ANN model has also been carried out. It shows that ANN obtains the better accuracy than the regression method. The model produced a R2 value of 0.785 instead of 0.705 obtained by the regression method.

Permanent link: