Ensemble Forecasting of Daily Water Demand

Daily weather conditions influence water demand and have a key role in short-term water demand forecasting. An ensemble technique is employed to integrate the uncertainty associated with weather variables in forecasting models. Detailed procedures of the ensemble modeling are described. A case study is presented in which the ensemble method is used to forecast water demand in the warm season with a three-day lead time for the Louisville Water Company water supply system, Louisville, Kentucky. A multivariate regression model is developed to address the relationship between water demand and selected variables, including temperature, dry-rain index, weekday and weekend day index. The ensemble prediction deals with temperature uncertainty by combining a perturbation generated from a uniform distribution to the original temperature time series, and deals with rainfall uncertainty by modifying the occurrence of rainfall in randomly selected days. The number of modified days is determined according to precipitation forecasting probability. The confidence band obtained from the ensemble model provides more reliable and useful information to water utilities which benefits the optimization of water supply system.