Occupant Behavior Revealed from Sensor-Fusion-Based Clustering Analysis: Case of a University Building Office

Occupant behavior is among the main drivers in building energy consumption. However, due to its stochastic nature, it isn’t easy to model and is often oversimplified in building simulations. Moreover, obtaining explicit occupancy data for modeling purposes is a tedious endeavor. This has led to the implementation of indirect occupancy sensing systems, and when coupled with machine learning algorithms, have proven effective for occupancy estimation. This work applies k-means clustering to experimental data recorded for 25 days to reveal occupancy profiles in a university office. Four clusters were obtained from electric current, light luminosity, and motion sensor data. These clusters are interpreted as occupancy profiles, with varying degrees of occupancy level and arrival and departure times. While electric current and motion sensor data clusters could bring forth relevant occupancy information, light luminosity was not as informative in our case study. This approach to occupancy profiling has the advantage of not requiring ground truth occupancy data, thus facilitating the modeling process. The information obtained from clustering analysis is useful for building systems’ control strategy design, supervised learning algorithm tuning, and building energy simulation.