Role of artificial intelligence in improving the operation of district heating systems in Finland
Bahronov, Abbos (2026)
Kandidaatintyö
2026
School of Energy Systems, Energiatekniikka
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2026052352429
https://urn.fi/URN:NBN:fi-fe2026052352429
Tiivistelmä
District heating (DH) systems play an important role in providing efficient heat supply for residential and industrial buildings, particularly in Finland because of its cold climate zone. With increasing digitalization of energy systems, operational data from DH networks enables the development of data-driven monitoring methods for improving system reliability and anomaly detection.
This thesis investigates the use of Artificial Intelligence (AI)-based predictive models for anomaly detection in DH systems. Particularly, linear regression and decision tree models are applied to predict system behavior using operational variables. Anomalies are detected using Shewhart Statistical Process Control (SPC) charts applied to residual signals between measured and predicted values. The proposed approach is also compared with a traditional threshold-based anomaly detection method.
The case study is conducted using operational time-series data from the PreDist dataset obtained from a German district heating operator. The results show that both machine learning models were capable of detecting anomalies in DH operation, with the decision tree model detecting a larger number of anomaly samples and the linear regression model producing fewer false alarms compared to the traditional model. Overall, the study demonstrates the potential of AI-based methods for supporting operational monitoring in DH systems.
This thesis investigates the use of Artificial Intelligence (AI)-based predictive models for anomaly detection in DH systems. Particularly, linear regression and decision tree models are applied to predict system behavior using operational variables. Anomalies are detected using Shewhart Statistical Process Control (SPC) charts applied to residual signals between measured and predicted values. The proposed approach is also compared with a traditional threshold-based anomaly detection method.
The case study is conducted using operational time-series data from the PreDist dataset obtained from a German district heating operator. The results show that both machine learning models were capable of detecting anomalies in DH operation, with the decision tree model detecting a larger number of anomaly samples and the linear regression model producing fewer false alarms compared to the traditional model. Overall, the study demonstrates the potential of AI-based methods for supporting operational monitoring in DH systems.