Advancements in Burst Localization Through Real-Time Hydraulic Gradient Analysis with Deep Neural Networks in Complex Water Transmission Systems

Abstract

In urban water management, the rapid detection and localization of bursts in water transmission lines (WTLs) is a critical step for efficient response, aiming to reduce service disruptions and minimize infrastructure damage. Transient methods primarily used for WTL burst detection lack practicality for application in real WTLs. Traditional pressure and flow-based data analysis methods have limitations in pinpointing the locations of bursts. To overcome these issues, this paper introduces an innovative method for real-time burst detection and localization in complex WTLs based on the analysis of hydraulic gradient (HG) variations. The methodology involves tracking discrepancies in real time between estimated and actual HG values across segmented WTLs, using deep learning. The developed models learn patterns and nonlinear relationships among various factors such as pump switching, valve statuses, and flow variations. This approach offers a clear advantage for burst localisation; as a burst in any segment causes actual HGs to be higher than the estimated ones at the upstream segments, while the opposite effect is observed at the downstream segments due to energy loss from the burst. This innovative method has been tested in two burst incidents in two real case studies and accurately detected a segment that had a burst in both case studies. In comparison, traditional pressure-based methods, while successful in detecting both bursts, misidentified the locations of these incidents. This underscores the proposed method's enhanced accuracy in pinpointing burst locations. The integration of this methodology with existing supervisory control and data acquisition (SCADA) systems highlights the method's practical applicability, significantly contributing to the development of robust and resilient urban water infrastructures.