Abstract

The expansion of cities puts a high demand on water utilities to provide fresh water amidst global water stress. One of the most challenging issues that continuously menaces water sector sustainability is dealing with anomalous consumption as commercial water losses. To this end, this paper proposes a novel and elaborate unsupervised machine learning method with an effortless computation to detect end-user anomalies in water distribution systems. The method relies on density-based spatial clustering of applications with noise (DBSCAN) as the core and Lempel-Ziv Complexity as the key features to overcome the curse of dimensionality and low data resolution, while only users' billing records are needed. Building upon the method's outputs, six latent abnormal patterns were perceived in a large unlabeled real-life dataset after validation by a smaller but labeled (fraudulent or normal consumption) real-life dataset. The results exposed that the method meets reliable and efficacious outputs for anomaly detection, as the derived clustering could capture high accuracy of 98% and a hit rate of 92%. Also, this study provides valuable information that can be used for action planning like on-site inspections and improving water loss management practices, exceptionally when water utilities/authorities' budgets are constrained, and revenue declines occur.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call