Data-driven automation of HVAC systems: An experimental study in a university study room

Abstract

This study proposes a framework for controlling HVAC systems for university study rooms that includes a data-driven model capable of identifying the probability of user interaction with air conditioning and ventilation systems, depending on the thermal sensation vote (TSV), perceived air quality (PAQ) and microclimate parameters (air and operative temperature, air velocity, relative humidity and CO2). The experimental setup allowed the participants to carry out their usual study/work activities without the need to be supervised. This allowed the occupants not to be psychologically conditioned and made their experience and interaction with the environment/systems as realistic as possible. The analysis of the experimental data showed that the operative temperature mainly influences the thermal sensation of the occupant inside the room, while the perceived air quality depends not only on the CO2 concentration but also on thermal perception and air velocity. Furthermore, three predictive models (heating, cooling and IAQ) were obtained from the experimental data, indicating the probability of user interaction with the system (R2 between 0.85 and 0.94). The heating phase model was also verified by automatizing the heating system through the developed framework and comparing the user's sensations before and after control (users unaware of the changes). The use of the models increased users' thermal comfort from 39 % to 82 %, confirming the effectiveness of the system.