Prediction of Contaminants in Water Distribution Systems Using Artificial Neural Networks

Abstract

Transport phenomena of contaminants in water distribution systems are examined using Artificial Neural Networks (ANNs). First, a contaminant is introduced in the reservoir, and its concentration throughout the system is predicted as a function of water demands at regional levels. For this purpose, a back-propagation ANN is trained, tested and validated with data obtained from both experiments in an actual water system and EPANET. Next, the most likely location of the chemical’s intrusion point is tracked based on readings collected from several sensors placed in the water network. In order to minimize intrinsic errors, several parameters of the architecture and functions of these ANNs were thoroughly tested: a number of processing units in hidden layers, transfer functions, and learning rules, to name a few. The present study provides relevant information for alertness and preparedness in potential intentional and accidental contamination events.