Modeling personalized occupancy profiles for representing long term patterns by using ambient context

• Presents a framework to model personalized occupancy presence profiles.
• The model represents occupant's long-term presence patterns.
• We compare four different modeling techniques to real occupancy.
• We integrate ambient context for distinguishing the regular and irregular presences.
• We evaluate modeled occupancy presence profile performances in energy simulation.

Occupancy is a crucial factor in deciding the load for a building's HVAC system. To incorporate and account for occupancy, most simulation tools use fixed design profiles that are based on statistical methods, which use large-scale occupant surveys and/or observations from a number of similar buildings. Such survey and observational data are labor- and time- intensive to gather, and do not accurately represent the actual occupancy patterns. Consequently, fixed design profiles may deviate from actual occupancies of a building. Data acquired by wireless sensor networks could provide high-resolution and accurate information for describing indoor ambient variations caused by occupancy status changes and screening irregular presence. This paper proposes a framework to model personalized occupancy profiles for representing occupants' long-term presence patterns. A personalized occupancy profile is described as typical weekday/weekend occupancy probability as a function of time for a specific occupant. Regression modeling, time-series modeling, pattern recognition modeling and stochastic process modeling are tested to model the expected occupancy status and their performances are compared in terms of the degree of statistical approximation to actual occupancy. The paper evaluates the impact of implementing personalized occupancy profiles on energy simulation results by simulating energy consumption of four thermal zones in a building for four months using OpenStudio. The results show that the personalized occupancy profiles acquired through time-series modeling, pattern recognition modeling and stochastic process modeling outperform the fixed design profiles and observation-based profiles currently used in building energy simulations.