Descriptive and predictive modeling

Water distribution networks (WDNs) are hydraulic infrastructures that provide a continuous supply of pressurized safe drinking water to all consumers, playing an important role in public health. Leakages cause service interruptions, waste resources, and compromise water quality. Although we can find many methods that support the monitoring and control of WDNs, they exhibit limited ability to detect anomalies and are not yet consistently applied to Portuguese WDNs. We show that it is possible to (1) describe the dynamics of a WDN through spatiotemporal correlation analysis of pressure and volumetric flowrate sensors, and (2) analyze disruptions on the expected correlation to detect burst leakage dynamics using standard classifiers. Our approach is promising in a synthetic setting and offers initial support towards leakage detection in real WDNs despite the presence of highly irregular consumption patterns, a limited number of recorded leakages, and highly heterogeneous leakage profiles. We discovered that the disruption caused by leakages is higher shortly after the burst. Furthermore, a comprehensive pairing of heterogeneous sensors and data balancing in the real setting is also promising. Our results suggest that it is important to access data fromWDNs with good sensor coverage and complete information about leakages. Accordingly, we believe that our WDN would benefit a lot from sensor expansion and relocation. Lastly, given the simplicity, novelty, and accuracy of the proposed correlation-based principles for anomaly detection in heterogeneous and georeferenced time series, we anticipate our work to contribute to the study and development of automated leakage detection in portuguese WDNs.