Feature Importance Analysis For A Personalized Thermal Comfort Model Using Meta-learning

Abstract

The continual demand for energy-efficient devices has a far-reaching impact, and the HVAC system is no exception. To account for this energy demand, the HVAC system could potentially focus exclusively on energy efficiency while sacrificing user satisfaction. However, the thermal comfort of occupants is also a vital factor when designing HVAC control systems. In this paper, we present a personalized thermal preference model using meta-learning and feature importance analysis (FIA) in order to predict the thermal sensation of an individual with a minimum number of individual surveys and sensor measurements. The use of meta-learning reduces the number of required personalized thermal sensation votes. The FIA algorithm attempts to identify the most important features of the thermal preference model under the framework of meta-learning. With the ASHRAE database II, we find that the use of the FIA algorithm with meta-learning provides a clear distinction between the most important features in the model as compared to when only supervised learning is used.