Personal Thermal Comfort Model for Cyber-Physical Human Centric Systems using Incomplete Supervised Learning Method

Abstract

A personal thermal comfort (PTC) model is a novel approach to predict the thermal sensation of an individual rather than a group of people. The relationship between the environmental and human factors of this model is not well investigated in the smart home domain. Moreover, the difficulties on collecting feedback of personal thermal sensation, especially neutral comfort, and the requirements of data size can lead to the failure of thermal comfort system to fulfill the individual’s comfort preference. In this paper, we present an incomplete supervised learning method for an enhanced PTC model to be more personalized to predict real-time personal thermal sensation in Cyber-Physical Human Centric System. In particular, the psychological parameters are always available for the continuous satisfactory control of a heating, ventilating, and air conditioning (HVAC) system in a timely manner. In the enhanced PTC model, we propose a personalized predictive classifier (PPC), which uses a learning algorithm from incomplete supervision to predict the personal thermal sensation. We also explore the PPC that uses two cascaded Random Forest classifiers can result a better performance and a faster learning speed in the implementation of energy efficient thermal comfort control (EETCC) system to improve individual’s thermal comfort level while optimizing the HVAC energy consumption at the same time. Through the experiment and evaluation studies, we conclude that the enhanced PTC model with PPC is able to offer psychological parameters inference to a continuous satisfactory control system in smart homes.