Alert-based wearable sensing system for individualized thermal preference prediction

Much evidence has shown that each individual has different needs, preferences, and expectations of the indoor thermal environment, which may cause potential excessive energy use. Incorporating an individualized comfort model is a new approach to characterizing individual thermal comfort and serves as a basis for a smart building paradigm. The major issue for such model development is related to the complexity of individual data collection and the ignorance of micro-environmental parameters. This paper intends to design, develop, and demonstrate an innovative alert-based occupant-responsive framework for capturing individualized thermal comfort-related data and predicting individualized thermal preference by leveraging wearable sensors and computing technologies. By applying a pre-defined alert algorithm to micro-environmental and individual physiological data, this sensing system can automatically capture pronounced data fluctuations. The data collection efficiency and effectiveness are enhanced with the alert algorithm compared with the scheduled surveys used in previous studies. The continuous participation of the researchers to disrupt the subjects for the surveys during the experiments is also avoided. The individual thermal preference prediction models achieved high overall accuracy, >94% with about 110 data points for subject A and 150 for B. Through the analysis of features' relative importance and their interactive effects, this work also shows the features’ characteristics and distinctions in predicting the thermal preferences of different individuals. This feature contribution analysis highlights the key influential parameters for an individual and may support the optimization potential of individualized comfort conditions and energy usage for building heating and cooling. The developed system also features the role of alerting functions in the automatic collection of subjective user data and facilitating smart building integration as well.