On-Line Prediction of Consumption for Water Supply Network Control

Abstract

This paper describes two methods used for on-line Forecasting of consumer water demand on a water distribution system and compares their accuracy and computational feasibility for the development of a computer-based optimal control scheme on an actual network in, Worcestershire, England. The first method uses Kaiman filtering techniques with a temperature or rainfall variable controlling the propagation of the residual component, while the second method uses the Karhunen-Loeve spectral expansion using past load data only. Comparison indicates that the former is best for prediction up to 48 hours ahead, while the latter is best for prediction 1-2 hours ahead.