A systematic approach to occupancy modeling in ambient sensor-rich buildings

Abstract

With ever-rising energy demand and diminishing sources of inexpensive energy resources, energy conservation has become an increasingly important topic. Building heating, ventilation, and air conditioning (HVAC) systems are considered to be a prime target for energy conservation due to their significant contribution to commercial buildings’ energy consumption in the US. Knowing a building’s occupancy plays a crucial role in implementing demand-response HVAC controls, with a corresponding potential for reduction of HVAC energy consumption, especially in office buildings. This paper evaluates occupancy modeling (both binary detection and multi-class estimation) using twelve ambient sensor variables. Performance of six machine-learning techniques is evaluated in both single-occupancy and multi-occupancy offices. Of the six, the decision-tree technique yielded the best overall accuracy (i.e. 96.0% to 98.2%) and root mean square error (RMSE) (i.e. 0.109 to 0.156). The contribution of each individual ambient sensor variable is evaluated via information gain. It is found that CO2, door status, and light variables have important contributions to the final modeling results. It is observed that the overall accuracy generally increases as the number of sensors increases. This paper also examines the possibility of building a global occupancy model, and explores the reasons for low performance of global occupancy estimation. Lastly, the occupancy model is used to estimate and visualize the accumulative room and thermal zone usage in an office test-bed building for three months. The results reveal that the effective vacancy accounts for a substantial portion of the operational hours, varying from 19.8% to 29.8% with an average of 23.3%, which bears significant potential for energy savings. Furthermore, the authors simulated HVAC energy consumption of the test-bed building for three months in DesignBuilder and EnergyPlus, and compared energy consumption of occupancy-based demand-response HVAC controls using the authors’ occupancy-modeling results to the conventional HVAC controls currently implemented in the test-bed building. The results demonstrate that 20% of gas and 18% of electricity could be effectively saved if occupancy-based demand-response HVAC control is implemented.