Center for Research in Energy Systems Transformation (CREST)

Existing HVAC systems involve little feedback from indoor occupants, resulting in unnecessary cooling/heating waste and high percentage of discomfort. In addition, large thermal preference variance amongst people requires the development of personal thermal comfort models, rather than group-based methodologies such as predicted mean vote (PMV). This study focuses on assessing wearable solutions with the aim to predict personal thermal preference. We collected physiological signals (e.g., skin temperature, heart rate) of 14 subjects (6 female and 8 male adults) and environmental parameters (e.g., air temperature, wind speed, solar radiation, precipitation) for two weeks (at least 20 hr/d) to infer personal real-time thermal preference. The subjects reported their real-time thermal sensation and preference using cell-phones approximately every hour. We trained a Random Forest algorithm using data collected from individuals to develop a personal comfort model with the objective to predict thermal preference. The results show that subjects expressed needs for "warmer" or "cooler" conditions at about 30% (from 21% to 88%) of their daily time on average, implying the strong demand for a personalized indoor thermal comfort. In addition, the personal comfort model using Random Forest can infer individual thermal preference with a mean accuracy of 75% (53-93%) using physiological and environmental parameters, demonstrating the strengths of the proposed data-driven method.