|Title||Monitoring Support for Water Distribution Systems based on Pressure Sensor Data|
|Author(s)||Geelen, Caspar V.C.; Yntema, Doekle R.; Molenaar, Jaap; Keesman, Karel J.|
|Source||Water Resources Management 33 (2019)10. - ISSN 0920-4741 - p. 3339 - 3353.|
Mathematical and Statistical Methods - Biometris
Biobased Chemistry and Technology
|Publication type||Refereed Article in a scientific journal|
|Keyword(s)||Early warning system - Proactive leakage control - Real-time learning - Unsupervised learning - Water distribution network (WDS)|
The increasing age and deterioration of drinking water mains is causing an increasing frequency of pipe bursts. Not only are pipe repairs costly, bursts might also lead to contamination of the Dutch non-chlorinated drinking water, as well as damage to other above- and underground infrastructure. Detection and localization of pipe bursts have long been priorities for water distribution companies. Here we present a method for proactive leakage control, referred to as Monitoring Support. Contrary to most leak prevention methods, our method is based on real-time pressure sensor measurements and focuses on detection of recurring pressure anomalies, which are assumed to be indicative of misuse or malfunctioning of the water distribution network. The method visualizes and warns for both recurring and one-time anomalous events and offers monitoring experts an unsupervised decision support tool that requires no training data or manual labeling. Additionally, our method supports any time series data source and can be applied to other types of distribution networks, such as those for gas, electricity and oil. The performance of our method, including both instance-based and feature-based clustering, was validated on two pressure sensor data sets. Results indicate that feature-based clustering is the best method for detection of recurring pressure anomalies, with accuracy F1-scores of 92% and 94% for a 2013 and 2017 data set, respectively.