Fault Detection and Condition Monitoring in District Heating Using Smart Meter Data

Abstract

Currently, large amounts of smart meter data are mainly used for billing purposes only, although they could also be valuable for decision support and business process optimization in customer service and maintenance. Therefore, this paper presents several relevant use cases for prognostics and health management based on a case study of a real meter data set of a medium-sized geothermal district heating network in southern Germany. First, we show the implementation of a machine learning algorithm for automatic fault detection. Our approach uses cluster analysis to categorize the different load cycles of a district heating substation and calculates a regression function between the demanded thermal energy and the outdoor temperature for each cluster. Thereby, the substation's control behaviour is learned and deviations due to a malfunction or failure can be detected in advance. In addition to the regression-based fault detection, we present two key performance indicators that can be computed relatively simple but resulting in very effective insights for condition monitoring and identifying substations with highly negative effects on the overall network. Our findings' correctness and usefulness were verified by the corresponding domain experts of the geothermal district heating company.
 Finally, we provide an outlook on smart meter data's role for the further development of intelligent district heating networks and the realization of highly complex approaches such as smart grids. To foster future research, we provide exemplary our RapidMiner processes as well as some sample data.