TOWARDS NON-INTRUSIVE, DATA-DRIVEN DETECTION OF DISTRICT HEATING SYSTEM CONSUMER BEHAVIOR

Abstract

This paper takes initial steps in proposing non-intrusive, data-driven methodology for detecting consumer behavior patterns in District Heating Systems (DHS), focusing specifically on occupant presence and window opening events. Recognizing the limitations of legacy supply-driven control logic in DHS, paper aims to show the feasibility of a novel detection approach without ground truth data, utilizing environmental sensor data collected from residential apartments, including CO₂ concentration, indoor temperature, relative humidity, and outdoor temperature. Occupancy detection is performed by using analysis of CO₂ concentration signal. For window opening detection, a robust multivariate outlier detection based on the Mahalanobis distance computed over first-order differences in sensor signals was applied. Events are inferred by identifying anomalous combinations of rapid CO₂, temperature, and humidity changes during periods of thermal imbalance between indoor and outdoor environments. The results show consistent and interpretable behavioral patterns that demonstrate the feasibility of detecting consumer behavior using passive sensor data. This enables the integration of behavior-aware control strategies in DHS operations, with the potential to improve thermal comfort, reduce energy waste, and lower emissions.