IDENTIFYING LEVELS OF OCCUPANCY IN BUILDINGS USING AUTOMATED MACHINE LEARNING

Abstract

Buildings are the key contributors to energy consumption. Knowledge on the occupancy of indoor spaces plays an important role in the estimation of building's energy use. The indoor occupancy status and levels can be estimated based on information acquired from cameras, energy meters, person trackers and environmental sensors. Some of these methods such as image processing and tracking are causing concerns related to privacy, others require installation of smart devices, such as smart meters. As environmental sensors are embedded in many home appliances, it is easy and cheap to reach this information while ensuring privacy. This study focuses on exploring the accuracy of semi-automated and automated machine learning methods in identifying the levels of occupancy in indoor spaces based on data acquired from environmental sensors. The study involved a data acquisition stage and three stages of machine learning experiments. The results indicate that the automated predictions of occupancy status and levels can be completed with high accuracy using Automated Machine Learning (AutoML) methods.