The application of artificial neural networks to water demand modelling

The cost of electricity for the pumping of water in water distribution systems, accounts for a large portion of the operating budgets of many water utilities. In North America there is currently a move towards the deregulation of the power industry that will change the rate structure for water utilities. It is therefore necessary for water utilities to better understand their power usage and pumping requirements to optimize their power usage to take advantage of the rate structure. An important component of this project is the accurate prediction of water demands. An artificial neural network model is presented which has been developed to predict the daily and 12-day water demands for the City of Edmonton. The developed daily model has an average error of 2.3%, while the 2 to 12 day model has an average error of 3.1%. An hourly prediction method that was developed has a 3.4% average error.