Ereignisdetektion in Trinkwassernetzen mittels PCA und DPCA

Drinking water distribution networks are currently equipped with a large amount of multi-parameter probes. One important tasks of these probes is to detect contaminations in the network. At the moment, operators perform the analysis of the measurements manually to determine the current state of the network. Data-driven methods have not yet been proved suitable in practice since those algorithms detect a large amount of false positive alarms, which occur e.g. due to operational changes. Hence, in this paper a new approach based on the DPCA is proposed, which reduces the amount of false alarms, while still reacting sensitive on real contaminations. Experiments for the evaluation of the developed approach are conducted on a laboratory plant.