Predicting room occupancy based on carbon dioxide concentration levels : And potential energy-saving use cases in HVAC (heating, ventilation and air conditioning) control systems.

Sammanfattning: This thesis aims to explore the possibilities of recording room occupancy based on measured CO2 (carbon dioxide) concentration levels, and potential energy-saving uses in HVAC (heating, ventilation and air conditioning) - control systems. Furthermore, this thesis specifically aims to use unsupervised machine learning methods, which removes the need for generating correct occupancy labels and training the models before deploying them. The findings demonstrate that CO2 concentration levels can serve as a reliable predictor of occupancy in enclosed spaces, though the accuracy of this method varies depending on room configurations. Notably, a model that is based on the observed rates of change of CO2, and which requires no specific training for each room configuration, demonstrates comparable accuracy to traditional supervised machine learning models. The study further investigates if any other common air particles might serve as occupancy indicators. However, the findings predominantly suggest that CO2 demonstrates the most promise for this application. Additionally, by applying our model to predict office space usage, we derived potential HVAC optimisations, using set-point and set-back temperatures and then calculating energy and cost savings. The practical implications of these findings suggest that the model can be used to save energy and money.