Using Correlation between Data from Multiple Monitoring Sensors to Detect Bursts in Water Distribution Systems

Abstract

Many burst detection methods, including a prediction stage, have been developed in order to identify bursts in a timely manner. These methods require vast historical data to produce accurate predictions. The clustering-based method proposed in this paper only requires one day of time series data. In clustering analysis, cosine distance is used to evaluate dissimilarity between flow data. Incorporating cosine distance enables this method to fully use the temporal varying correlation between the data from multiple flow sensors. By doing this, data variations caused by sudden weather changes, festivals, and periodic changes in water demand are correctly classified as normal conditions in pipe networks. This method was applied in a real multi-inlet and multioutlet district metering area (DMA). The results show that it can achieve a low false positive rate and few false alarms and be sensitive to relatively large bursts. This method has the potential to be used in different types of DMA.