Modeling and predicting occupancy profile in office space with a Wi-Fi probe-based Dynamic Markov Time-Window Inference approach

Demand-based HVAC control methods in buildings show great energy saving potential when accurate occupancy information is available. Appropriate service based on actual occupant demand could prevent unnecessary energy waste caused by system overcooling or overheating. Therefore, various occupancy detection approaches had attracted increasing attentions in recent years. Among them, Wi-Fi based detection approaches have been thoroughly discussed since Wi-Fi access points (APs) and wireless devices are ubiquitously used in modern buildings. Compared with traditional request and response based occupancy assessment, the newly developed Wi-Fi probe technology can actively scan Wi-Fi enabled devices even if they are not connected to the network. However, Wi-Fi probe detection still subjects to significant errors due to unstable signal and unpredictable occupant behavior. This study stresses the time-series and stochastic characteristics of detected signals and proposes a novel Dynamic Markov Time-Window Inference (DMTWI) model to predict reliable occupancy. The conventional Auto-Regressive Moving Average (ARMA) model and Support Vector Regression (SVR) model are also examined and compared with the proposed approach. Also, an on-site experiment was conducted to validate the proposed model, and the results reveal that the prediction accuracy is over 80% when x-accuracy tolerance is less than 4 for weekdays, 3 for holidays, and 2 for weekend days.