Estimating the number of occupants and activity intensity in large spaces with environmental sensors

Abstract

Recently, occupant-centered control models have been widely discussed, with intelligent control models looking for more refined and dynamic regulation based on occupant information. This article introduced a novel method for using environmental sensors and machine learning algorithms to identify the number of occupants and activity intensity in large spaces. A multi-functional space was monitored for approximately two months using PIR, CO2, sound decibel, temperature and humidity sensors. To address the challenge of identifying the number of occupants in large spaces, the study proposed a non-uniformly distributed interval of occupant numbers that offsets the small fluctuations in the number of people. The study also demonstrated that PIR and CO2 level measurements could be used to estimate the headcount interval with an accuracy rate of 84.5%. Furthermore, the study employed K-means clustering to identify low-, medium-, and high-level activities in the studied space, achieving an overall accuracy rate of 89.3%. A new metric of activity intensity was introduced to measure the activities carried out indoors, which incorporated CO2 and sound decibel levels, PIR readings, and the number of occupants. This proposed metric was found to be appropriate for quantifying the activity intensity in the studied space. Overall, the method presented in this study provided a promising approach for enabling occupant-based control strategies that leverage advanced sensor data to optimize building service systems in large spaces.