Author(s): Sanghoon Jun; Donghwi Jung
Linked Author(s): Donghwi Jung, SANGHOON JUN
Keywords: Dvanced Metering Infrastructure Deep Learning Leakage Smart Meter Water Distribution Network
Abstract: Advanced metering infrastructure (AMI) is the representative smart meter and associated communication system of water distribution systems (WDSs) that has high spatial and temporal resolutions. AMIs are installed on individual service lines and measure water consumption data on much shorter time intervals (e. g., 5 min). WDS within AMIs is not a future upcoming technology, rather, it is already being widely implemented. For example, the city of San Diego, California, authorized up to $25 million to install smart meters at all households while the Madison Water Utility, Wisconsin, installed more than 64,300 AMIs. In Europe, the city of Gandia in Valencia, Spain, manages smart meters on their over 700,000 customer services (Jun et al. 2024). However, the benefits of using AMIs have been focused on the downstream of the meters. Less efforts have been devoted to utilizing AMI data within the distribution network for WDS operation and management such as leakage detection. Leakages are one of the major failure events in WDSs. They can be classified into three categories: background leaks (tiny weeps and seeps at joints that are not detectable even with field inspection), unreported leaks (small unseen water losses), and reported leaks (visible losses) (AWWA 2008). The large, reported failures, such as pipe bursts, are visible over the ground. Thus, these large leaks do not require a detection model to be identified as they can be reported directly through the phone calls from pedestrians. On the other hand, small leaks (unreported leaks) may remain unobserved for a long time and decrease water utility service level such as low pressures at customer taps and low water quality. A good leak detection model is needed to identify these small failures. However, the main drawback of current leak detection models is that they are based on collecting and employing data from supervisory control and data acquisition (SCADA) systems. These systems have sparse flow and pressure meters, often limited to monitoring tank water levels, pump flow rates and heads, and a few pressures that may be inadequate to identify small leaks that have not reached the ground surface. Thus, a current data collection networks are insufficient and a clear next step is to utilize AMI smart meter data. While generally focusing downstream of the meters, pressure meters have been added to customer meters and can provide a fuller picture of the WDS (Jun and Lansey 2023). If pressure-supplemented AMI (PS-AMI) systems are deployed, in which all end-user pressures and demands are measured, leak detection performance can significantly improve including detecting small leaks. These systems can capture leak signals that are often masked by other noises, such as measurement errors, by looking at network’s full pressure surface generated from the AMI data. Further, high density of AMI meters provides more information to locate leaks. However, the full potential value of using PS-AMI systems for leak detection and localization has not been fully explored. This presentation explores the benefits of leveraging smart WDS AMI data for leak detection and localization. Several leak identification methods utilizing AMI demand and pressure data are introduced, including system mass balance method, sensitivity matrix and optimization-based method, convolutional neural network (CNN) deep learning method. Each approach addresses different types of data, such as the summation of end-user demands and the difference between estimated and measured AMI pressures. By examining these methods’ outstanding leak detection performance across several WDSs, the full potential value of smart WDS for leak detection is highlighted. Additionally, the smallest detectable leak sizes of AMI-based methods are determined for varying WDSs, demonstrating the advantages of using AMI data. Finally, future challenges in AMI-based leak detection studies are discussed, such as handling the impact of WDS uncertainties.
Year: 2025