Water quality supervision of distribution networks based on machine learning algorithms and operator feedback

Water Distribution Networks contain lots of quality sensors placed in the network. Generally, the analysis of this sensor data, e.g. to check for contaminations, is performed manually by the operators and not by data-driven methods. This has several reasons: First, the parameterization of these methods is time consuming; second, many false positive alarms are generated due to special operational actions. This paper addresses both problems: An alarm detection method is presented needing only a few parameters for configuration and the amount of false alarms is reduced, by using known events for training. The approach is tested on a laboratory plant.