Skip to content

1.13 Leakage detection in water distribution networks

– using machine-learning strategies


In this work, we consider a set of real water pressure measurements from the water and wastewatercompany of Stockholm, Sweden (SVOA, Stockholm Vatten och Avfall). Specifically, we analyze the observed water pressure data collected from January 2018 to March 2019 in four selected pumping stations. These stations are located in a district-metered area (DMA) of the water distribution network (WDN). The DMA corresponds to a residential area that has a total population of 70,250 people.


We proposed a non-hydraulic modeling method for detecting leakage in WDNs through
low-complexity learning strategies. Our proposed methodology uses learning strategies fromunsupervised learning (K-means and cluster validation techniques), and supervised learning (learning vector quantization algorithms).

Impact on society

  1. Our algorithm can be applied to leakage detection scenarios where we have access to water pressure measurements at different points of the WDN.
  2. Our proposed methodology is suitable to real-world scenarios where measurements are available but without a detailed description of the architecture of the WDN.