Machine Learning for Water Supply Supervision

Thomas Schranz, Gerald Schweiger, Siegfried Pabst, Franz Wotawa

Abstract: In an industrial setting water supply systems can be complex. Constructing physical models for fault diagnosis or prediction requires extensive knowledge about the system’s components and characteristics. Through advances in embedded computing, consumption meter data is often readily available. This data can be used to construct black box models that describe system behavior and highlight irregularities such as leakages. In this paper we discuss the application of artificial intelligence to the task of identifying irregular consumption patterns. We describe and evaluate data models based on neural networks and decision trees that were used for consumption prediction in buildings at the Graz University of Technology.