Occupancy Estimation Using Sparse Sensor Coverage

Abstract

Estimation of occupant count in commercial and institutional buildings enables energy experts to make better decisions on which buildings to prioritize for green upgrading and retrofitting from a large building portfolio. A cheap easy-to-install solution will enable energy experts to obtain occupancy estimation by easily scaling to large building portfolios. This paper presents a method for estimating occupancy based on sparse coverage of low-cost IoT sensors. The method is tested on 2 datasets, one academic building in Denmark (DK) and one academic building in Canada (CAN). The datasets contain PIR, CO2 measurements, and electric energy data together with ground truth occupancy counts. We show that 20% sensor coverage is comparable to full sensor coverage (60%) with an NRMSE of 0.142 (DK) and 0.174 (CAN) for 20% sensor coverage and an NRMSE of 0.129 (DK) and 0.163 (CAN) for full sensor coverage. Results show that with less sensor coverage, sensor placement becomes more important and that even with 20% it is possible to get as good of an accuracy as full coverage. The occupant count is used for key performance indicators of the buildings’ energy usage which shows higher energy use per occupant at low occupancy.