Author(s): Li Gu; Brent Burton

Linked Author(s):

Keywords: Urban water demand forecast; Artificial neural networks; Multiple linear regression; Annual average demand; Land-use; Serviced population; Climatic conditions

Abstract: The accurate forecasting of water demand is essential to the planning, design, operation and maintenance of drinking water supply infrastructure. Traditionally, this forecasting has been predominantly completed using parametric approaches, such as the multiple linear regression method. Such techniques require the development of an equation, or set of equations, to describe the relationship between water demand and the explanatory variables, such as population, land-use, and climate condition. The true form of the functional relationship between demand and the explanatory variables is often unknown and is, at least to some extent, subjective in its practical application. This paper explores the potential use of artificial neural networks (ANNs) to predict long-term urban water usage. An ANN is essentially a nonlinear mathematical structure capable of the representation of complex nonlinear processes that relate the inputs to the outputs of any system. Based upon data from seventeen municipalities in the Greater Vancouver region located in the west coast of Canada, an ANN has been developed and evaluated. Both the variable selection and the sensitivity of the prediction accuracy to network structure were investigated. It is shown that the ANN provides more accurate forecasts than the traditional multiple linear regression (MLR) approach.

DOI:

Year: 2005