Data-driven leak localization in WDN using pressure sensor and hydraulic information

Abstract

Maintaining a good quality of service under a wide range of operational management is challenging for water utilities. One of the significant challenges is the location of water leaks in the large-scale water distribution networks (WDN) due to limited data information throughout the system, generally having only flow sensors at the entrance of the system and some pressure sensors in some selected nodes. In addition, most systems do not have a hydraulic model of the network. Therefore, when using the hydraulic model, the presence of model errors such as nodal demand uncertainty and measurement noise can interfere with the performance of the leak location method. This work presents a fully data-driven technique to reduce the area of the leak localization in the WDN, using Graph theory to represent the network. To do so, we have developed a distance clustering with pre-defined centroids that are the sensor pressure information and some selected nodes. Furthermore, some extra pressure information of leaks events in the selected centroids is studied to develop a correlation between the pressure measurement and the event. Finally, the approach is evaluated in real-world water systems and discusses graphical results and key performance indicators.