Improved Water Demand Forecasting to Promote Sustainable Water Management

Abstract

The Region of Durham in Ontario is a fast growing urban area east of Toronto and has a population of 650,000, covering an area of 2500 km2. It has a single tiered water supply system with the regional agency acting as a retailer to provide water to households, businesses, institutions and farms. In 2014 its output was 63,555 Mega Litres. The Region of Durham’s water agency faces many challenges including growing demands, ageing infrastructure, water quality concerns and rising costs operations. Forecasting water demands on a daily basis is remarkably difficult. Variables such as weather conditions, operational changes, watermain breaks, business cycles, human behaviour, economic and social factors effect water demand forecasting, but it is difficult to quantify those factors and thus difficult to make an accurate prediction. The water industry has responded to this challenge by developing sophisticated procedures for forecasting. The approaches used include Artificial Neural Networks (ANN) and time series statistical modeling, which takes into consideration all possible factors as input variables to build forecasting model. The Region of Durham has thus far relied upon ANN with mixed results. Through several years of observation, overall the ANN forecasting model can predict a relatively accurate water demand for next 24 hour period (R2 >0.7) in some pressure zones. Winter forecasting is more accurate than summer because outdoor water use is extremely variable.