A Model-Driven Learning Approach for Predicting the Personalized Dynamic Thermal Comfort in Ordinary Office Environment

Abstract

Occupants’ thermal comfort plays a critical role in the optimization of building operation, which has thus attracted more and more attention in recent years. However, diversity and uncertainties in the thermal comfort, which is caused by not only the physical environment, but also the psychology and physiology, provide challenges in the modeling of the thermal comfort. In this paper, based on cyber-physical system framework, we develop a thermal comfort model by a model-driven learning approach to dynamically predict the personalized thermal comfort through online learning and computation. This model consists of a physical part and a data-driven part. The physical part is developed based on the traditional heat balance equation. Since in the physical part there are some parameters (such as skin temperature) are difficult to be measured in practice, a data-driven part is thus developed based on the regression model to estimate the uncertain parameters with the feedback of occupants. By integrating the data-driven part into the physical part, the developed model could take both advantages of the model-driven and data-driven methods. The effectiveness and performance of the developed thermal comfort model are demonstrated using field experiments.