Data analytics and modelling for indoor occupied states

Abstract

This research focused on data analytics and modelling for indoor occupied states in smart buildings, with the aim of improving energy efficiency and environmental sustainability by accurately predicting and controlling indoor occupancy states. The study was based on five months of real smart building data provided by Keppel Ltd., mainly sensor data, including carbon dioxide concentration, humidity and temperature. In the initial stage of the project, unsupervised learning methods were used to perform cluster analysis and feature importance assessment, and successfully identified the influence of indoor occupancy patterns and environmental indicators on the prediction model. Subsequently, supervised learning models were used to predict the indoor occupancy states, with an accuracy of up to 99%. Leveraging these predictive models and analyzing historical unoccupied periods during working hours, suggested that energy savings of about 12% could theoretically be achieved through intelligent regulation. This significant finding not only provided a strong support for the energy efficiency management of smart buildings, but also indicated that energy use efficiency, resource utilization efficiency and indoor environmental quality could be significantly improved through fine prediction and control of indoor occupancy states.