Office buildings are key energy consumers and thus require attention to achieve efficient operation. While individual office spaces are dynamically used, current building automation does not receive information on utilisation that could be used to adaptively adjust energy consumption. In this work, we propose an approach to estimate people count per office space using distributed strategically placed PIR sensors and algorithms that can process the distributed sensor information.We detail our sensing node and evaluate its performance in an office installation. A sensor model was subsequently used in a floor-wide simulation of realistic occupant behaviours to investigate two algorithms to estimate people count per office space. The occupant behaviour simulations confirmed that our estimation algorithms can accurately predict people count in different office use scenarios. The errors introduced by the PIR masking time after a detection can be partially compensated when using distributed sensor information. Our approach can be used for dynamic, occupancy-dependent lighting, climate, and appliances control of office spaces.

No abstract

In this work, we recognised office worker activities that are relevant for energy-related control of appliances and building systems using sensors that are commonly installed in new or refurbished office buildings. We considered desk-related activities and people count in office rooms, structured into desk-and room-cells. Recognition was performed using finite state machines (FSMs) and probabilistic layered hidden Markov models (LHMMs).We evaluated our approach in a real living-lab office, including three private and multi-person office rooms. As example devices, we used different ceiling-mounted PIR sensors based on the EnOcean platform and plug-in power meters. In at least five days of study data per office room, including reference sensor data and occupant annotations, we confirmed that activities can be recognised using these sensors. For computer and desk work, an overall recognition accuracy of 95% was achieved. People count was estimated at 87% and 78% for the best-performing two office rooms. We furthermore present building simulation results that compare different control strategies. Compared to modern BEMS, our results show that 21.9% and 19.5% of electrical energy can be saved for controls based on recognised desk activity and estimated people count, respectively. These results confirm the relevance of building energy management based on activity sensing.

No abstract