Inferring occupant ties: automated inference of occupant network structure in commercial buildings

Abstract

To design and manage office buildings that are both energy-efficient and productive work environments, we need a better understanding of the relationship between building and occupant systems. Past data-driven building research has focused on energy efficiency and occupant comfort, but little work has used building sensor data to understand occupant organizational behavior and dynamics in buildings. In this initial work, we present a methodology for using distributed plug load energy consumption sensors to infer the social/organizational network of occupants (i.e., the relationships among occupants in a building). We demonstrate how plug load data can be used to model activities, and we introduce how statistical methods---in particular, the graphical lasso and the influence model---can be used to learn network structure from time-series activity data. We apply our method to a seven-person office environment in Northern California, and we compare the inferred networks to ground truth spatial, social, and organizational networks obtained through validated survey questions. In the end, a better understanding of how occupants organize and utilize spaces could enable more contextual control and co-optimization of building-human systems.