HB.se

Huvudmeny

Data Analytics for Fault Detection in District Heating (DAD)


The purpose of this project is to, together with industry partners, apply, improve and develop new methods and algorithms for predictive data analytics in order to increase energy efficiency in heating systems via fault monitoring, detection and prediction activities.

Start date

2018-01-01

End date

2020-12-31

Such efficiency improvements are necessary for Sweden, where the energy needed to deliver district heating continues to increase, to meet the EU energy directive targeting 30% energy efficiency by 2030, of which requirements include implementing energy efficient measures, improve efficiency of heating systems, and empowering better management of consumption via metering data.

Scientifically, the project targets several of the most dynamic areas in machine learning and data analytics – e.g., deep learning, anomaly detection, imbalanced learning, interpretability, prediction with confidence and concept drift, and loss function optimization – in order to address the problem of how to automatically monitor, detect and predict faults in a district heating system (DHS) using sensor data collected from DHS substations, e.g., data collected with Individual Metering and Debiting sensors.

Research indicates that many heating substations suffer from inefficiencies due to faulty components, erroneous configuration, suboptimal control strategies, delayed fault and error detection, etc., thus improvements can be made in terms of both efficiency and cost. The University of Borås, together with industry partners Borås Miljö och Energi, NODA Intelligent Systems and AB Bostäder, have identified the following industrial challenges to be addressed in the project:

  • On-line detection of abnormal behavior on a customer basis
  • Detection of sub-optimally tuned substations in a DHS
  • Classification of detected abnormal behavior

Overall, the project is organized into six Work Packages (WP):

  • WP1 - Project management and research environment
  • WP2 - Deep learning for time series anomaly detection, prediction and classification
  • WP3 - Fault detection and prediction with confidence in time series
  • WP4 - Integrating data analytics for fault prediction into an Intelligent Decision Support System framework
  • WP5 - Dissemination and knowledge transfer
  • WP6 - Replicability and potential for commercialization