Investigating Smart Thermostat Data And Its Potential Relationship To Occupant Thermal Comfort

Crowdsourcing and advanced hardware have provided new ways of collecting data about energy use and occupant behaviour in residential buildings. This research uses machine learning algorithms to predict an occupant’s thermal comfort using general building properties by using data analytics to understand thermal comfort patterns from smart thermostat data. The analysis utilized classification algorithms and temperature data collected from smart thermostats. The dataset consists of 7500 homes in major cities in North America. Over 100 models were trained using a variety of clustering algorithms and combinations of features. The highest-performing models had classification accuracies ranging between 7-9% higher than random. The study did not observe a relationship between thermostat adjustments and general building properties. Instead, the study found positive correlations between the number of thermostat adjustments and the age of the home. The contribution of this work is to showcase a detailed approach for conducting such analysis.M.A.S