In the current article, the thermostat keypress actions with concurrent occupancy, temperature, and relative humidity data in private office spaces were analyzed. The authors observed that the occupants interact with their thermostats infrequently (on average once every 56 h of occupancy); and when they did so, occupants changed the temperature set-point, on average, by 1°C. The authors observed that about one-third of the thermostat overrides were either to decrease the set-points during the heating season, or to increase the set-points during the cooling season. The temperatures leading to the thermostat overrides changed by ∼3°C seasonally. It was noted that the frequency of thermostat interactions can be approximated as a univariate logistic regression model that inputs the indoor temperature as the predictor variable. A recursive algorithm was formulated to develop univariate thermostat use models inside building controllers. It was implemented inside four commercial controllers serving seven private office spaces to choose operating set-points. One year after the implementation, the iterative learning process resulted in a 2°C–3°C reduction in the set-points during the heating season and a 2°C–3°C increase in the set-points during the cooling season with respect to the default 22°C set-point in both seasons.

Additional Metadata
Persistent URL dx.doi.org/10.1080/23744731.2017.1328956
Journal Science and Technology for the Built Environment
Citation
Gunay, H.B, O'Brien, W, Beausoleil-Morrison, I, & Bursill, J. (Jayson). (2018). Development and implementation of a thermostat learning algorithm. Science and Technology for the Built Environment, 24(1), 43–56. doi:10.1080/23744731.2017.1328956