In the SCCER “Future Energy Efficient Buildings & Districts” (FEEBD) at the iHomeLab, methods of machine learning are being developed in order to understand user behavior and thus increase the acceptance of building control.
HSLU iHomeLab is leading two tasks within SCCER FEEB&D Phase II. In the first Task, methods are explored for the recognition of user activities in order to enable user-centric control strategies. First, insight is gathered about the interaction between the building control system and its occupants. Then, the most suitable solution (single- or multisensory) is determined for the reliable recognition of predetermined activities and behaviors. Finally, concepts are developed based on machine learning and sensor fusion in order to recognize the selected activities reliably. It is expected that up to five TWh/a could potentially be saved in Switzerland alone through user-centric building control systems.
The second task takes place in collaboration with HSLU IGE. It is explored how data-driven building energy models can be automatically created and consecutively used for building characterization (“fingerprint”) and energy usage prediction. Data from various buildings of different types, both residential and functional, will thereby be used as input for the modelling task. The goal is to define energy fingerprints (i.e. typical behavior) of buildings automatically, and deviations from them. This in turn allows one to identify energy optimization opportunities and to predict electrical energy consumption of buildings that is driven by user behavior. The self-generation of building models will reduce the initial engineering cost for predictive control systems and therefore help to accelerate their market penetration.