Data-driven occupancy schedules for commercial buildings

Abstract

Standard schedules of occupancy are one of the backbones of building energy simulations. Some schedules that are in use today were published 40 years ago and have not been modified ever since. In this work, we aim at reviewing the representativeness of such standard schedules by comparison to a large data set. We extracted popular times data for commercial buildings, which has the same data structure as occupancy profiles, from the Google maps platform for 13 representative US cities in different climate zones. We use the mean absolute error and the earth mover’s distance as measures of difference in profile scale and shape, respectively. Additionally, we define energy impact metrics, such as the peak value and the time of the peak, to quantify differences that potentially have significant impacts on simulation results. We compared data of restaurant and retail buildings to the respective standard schedules. We found significant differences between standards and data, especially in energy impact metrics. Observed mean peak values were 10 - 40% (occupant capacity) different in the city with the overall best agreement to standards. Moreover, our results indicate that the categorization into weekdays, Saturday and Sunday day types should be reconsidered. In a second step, we compared data among the different cities and found relatively smaller differences, which might be rooted in climatic or socioeconomic influences on peoples’ behavior. This leads us to believe that location-specific data should be considered to more precisely capture occupant behavior.