A Time Varying Minimum Volume Ellipsoid (MVE) Method for Water Distribution Systems Event Detection

Water distribution systems are particularly vulnerable infrastructure systems as they comprise numerous exposed elements which can be deliberately infiltrated or may malfunction. Faults in pump operation, errors in sensors, power outages, cyber-attacks, contamination intrusions, and other abnormal complications harm water distribution systems operation in various ways. Therefore, implementing reliable events detection mechanisms in drinking water systems is vital for ensuring societal welfare. As such, the ability to rapidly detect such occurrences (and their origin) is a foremost objective. In this study a model for detecting abnormal events in water distribution systems using a minimum volume ellipsoid (MVE), adapted through time, is developed and demonstrated. A preliminary step to the ellipsoid finding is the clustering of different observation groups. It may be inaccurate to apply all observations in the same ellipsoid, as there are routine changes in water parameters. Applying some seasonal adjustments for example, is essential. Pattern recognition can help characterize typical behaviors of a time series. For example, water temperature rises in 2–3 degrees around 10 am, or increases in chloramine concentration don't last more than 10 minutes. Such characteristics may help define more appropriate ellipsoids for each time step thus generate a more sensitive model. The proposed method repeatedly constructs an MVE and modifies its structure according to new measurements and recorded events. The method is demonstrated on an example application through base runs and sensitivity analyses using hourly consumptions, flows, and pressures.