Water Distribution Networks (WDNs) are critical infrastructures that are exposed to deliberate or accidental chemical, biological or radioactive contamination. A monitoring system capable of protecting a WDN against contamination events in real time is a big challenge needed to be accomplished. Powerful online sensor systems are currently developed and the prototypes are able to detect a small change in water quality. Consequently, the main objective of the project SMaRT-OnlineWDN is the development of an online security management toolkit for WDNs that is based on sensor measurements of water quality as well as water quantity. A new approach for the fast and reliable detection of abnormal events in the WDNs by an alarm generation module is presented in this paper. Although in the past several approaches have been investigated and implemented (e.g. CANARI of EPA), so far these alarm generation concepts haven't been widely applied in real WDNs. Two reasons for that are: (1) The parameterization of existing alarm generation software products is too complex and time consuming, (2) a lot of abnormalities in the data appear due to special operational actions (e.g. sensor calibrations, flushing of pipes, rapid changes of water quality due to mixing of different water resources). To cope with this difficulties, in our approach the alarm generation module is trained both by historical data and in online mod using OPC technologies. Multi-variate statistical methods which need only a few parameters (e.g. Principal Component) are used. A fingerprint database is built up by the water utility experts and it is used to label known events. Results based on real WDN data of Berlin, Strasbourg and Paris are presented.
Kühnert, Christian; Bernard, Thomas; Montalvo Arango, Idel; and Nitsche, Reik, "A New Alarm Generation Concept For Water Distribution Networks Based On Machine Learning Algorithms" (2014). CUNY Academic Works.