Cross-Space Building Occupancy Modeling by Contextual Information Based Learning
Yang Zheng, Burçin Becerik-Gerber · 2015
Research on building occupancy modeling is becoming increasingly important and prevalent, as occupancy is closely related to building energy efficiency, lighting control, security monitoring, emergency evacuation, and rescue operations. Building occupancy is time-sequenced occupancy changes that represent how occupants occupy a building, including both occupant presence and occupant number. Extensive research has been carried out to model building occupancy based on ambient sensing. However, occupancy models reported in previous work are mostly used to detect and estimate the occupancy in the same spaces, where the models are trained, and the generalizability of existing occupancy models to different spaces has not been explored. In general, collecting precise and continuous occupancy data for each space for training can be time consuming and intrusive. In most cases, it might not be realistic to have access to actual occupancy for all of the spaces in a building, thus it is difficult to establish an individual model for each space and improve large-scale occupancy awareness at the building level. We present a systematic framework for cross-space occupancy modeling through non-intrusive ambient sensing and relationship learning, by which the model was trained in one space and then applied to other geometrically similar spaces. Six rooms in two buildings were selected to validate the effectiveness of the framework. It was found that daily F-measure ranged from 0.71 to 0.94 with the mean of 0.80, and daily RMSE ranged from 0.42 to 1.72 with the mean of 1.23 for all classes of occupancy. Daily occupied/unoccupied detection accuracy ranged from 0.79 to 0.99 with the mean of 0.91; and the daily number estimation accuracy ranged from 0.63 to 0.78 with the mean of 0.71, when only the occupied periods were considered. The results showed that the F-measure and daily RMSE had an approximately negative linear relationship while the daily occupied/unoccupied detection accuracy had an approximately positive logarithmic relationship with the daily number estimation accuracy. Three potential influential factors on cross-modeling performance were also investigated. It was found that the level of real-time occupancy variation, the degree of long-term occupancy difference between spaces, and the operation of HVAC systems had negatively logarithmic, linear, and logistic relations with the daily cross-space occupancy modeling accuracies, respectively.