Real-time burst detection in Water Distribution Systems using a Bayesian demand forecasting methodology

Abstract

The negative consequences of non-revenue water losses from Water Distribution Systems (WDS) can be reduced through the successful and prompt identification of bursts and abnormal conditions. Here we present a preliminary investigation into the application of a probabilistic demand forecasting approach to identify pipe bursts. The method produces a probabilistic forecast of future demand under normal conditions. This, in turn, quantifies the probability that a future observation is abnormal. The method, when tested using synthetic bursts applied to a demand time-series for a UK WDS, performed well in detecting bursts, particularly those >5% of mean daily flow at night time.