Urban water demand forecasting for the Island of skiathos

Abstract

We present an analysis of historical water demand data from the utility of Skiathos, Greece and demonstrate suitable demand forecasting methodologies. We apply linear and nonlinear forecasting methods to a three-year time series water demand. The best fit for quarterly averaged data was observed for the Winters' additive method; for monthly-averaged data, ARIMA, Artificial Neural Network and a hybrid approach performed best. Given the intense seasonality of demand in Skiathos, monthly time series proved to be the best data set for forecasting, while the best forecasting method was the hybrid, which combines the advantages of ARIMA and Artificial Neural Networks. © 2014 The Authors.