A Real-Time Method to Detect the Leakage Location in Urban Water Distribution Networks

Abstract

Water distribution networks (WDNs) are prone to infrastructure failures that lead to water loss. This paper presents a practical and efficient leakage detection and localization method based on the supervisory control and data acquisition (SCADA) system combined with state estimation. In WDNs, normal node demand patterns, seasonal variations, and leakage can cause pressure sensor values to fluctuate; accordingly, the pressure sensor values in a SCADA system can be regarded as exhibiting a mixture of normal and abnormal fluctuations. To obtain the fluctuation signal arising from leakage, the node demands must first be calibrated. For leakage detection, the proposed method involves comparing the errors between historical monitored values and simulated values to assess whether a change in these errors is due to normal system changes or the impact of leakage. At the leakage location, an emitter coefficient is used to simulate unknown leakage flow. Then, similarity indicators of the leakage locations are constructed using the Euclidian distance, Pearson correlation coefficient, and Jaccard similarity. The leakage location can be determined by comparing these similarity values. The performance of this method was tested on the L-Town WDN provided in the BattLeDIM competition. Our model’s ability to quickly and accurately detect and locate leakages was validated on the 2018 data set of leakage events. The model then was applied to predict the times and locations of leakages in 2019, and eight leakages were detected, six of which were true positives. This paper presents a practical and efficient leakage detection and localization method based on the SCADA system combined with state estimation. At leakage detection, the main steps include real-time demand calibration, difference analysis and threshold setting between simulated value and historical monitoring value, and leakage detection judgment. At the leakage location, the main steps include calculating the change of pressure value after adding a leakage, selecting the pressure sensor with the largest change, applying clustering algorithm to narrow the location scope, leakage simulation, similarity analysis, and leakage location. The performance of the method was tested on the L-Town WDN with a population of about 10,000 and about 800 pipes. Based on the SCADA data for 2019, we identified 8 pipe leakage events in total—6 true positives and 2 false positives—and 17 false negatives. This method has the potential to be applied in other cities. The data required for this method include SCADA data, historical leakage reports, and the water distribution network with distributed node demand. The method assumes that there is only one leakage at a time in the network. When multiple leakages occur simultaneously, the reliability of detection results is affected.