Novel dynamic thermal comfort prediction in buildings with a vision-based deep learning method by using thermographic imaging

Occupants’ thermal comfort is vital to individuals' well-being and energy consumption in buildings. The widely used thermal models like Predicted Mean Vote (PMV) and Standard Effective Temperature (SET) require numerous inputs that can be hard to obtain and often ignore individual variability. Occupants’ comfort levels can change over time due to the continuously changing human and environmental factors. Therefore, a real-time thermal comfort prediction method will be useful for building energy management systems and improving the occupants’ stratification. With the advancement of data science, machine learning methods are increasingly being introduced for predicting human thermal comfort. Compared to traditional data collection methods which usually involve a high level of cooperation and continuous engagement from individuals, the infrared or thermal cameras do not require direct interference with users and reduce the interference caused by wearable sensors. The infrared thermography technology can capture complete thermal images composed of the temperature values of each pixel. In this study, a novel real-time vision-based deep learning method is proposed to predict occupants’ comfort level and tested with 14 participants. This study is distinguished from existing methods as it does not require the measured skin temperature from the infrared cameras. Instead, it built a dataset based on thermographic images and thermal comfort survey to predict thermal comfort in real time. The proposed method provides a potential solution for improving occupancy thermal comfort while reducing energy consumption with low-cost and non-intrusive methods.