Establishment And Application Of Noncontact Personal Comfort Model Based On Machine Learning And Infrared Image Analysis

With the improvement of people’s living standards,the requirements for the comfort of the indoor environment are getting higher and higher.The traditional thermal comfort model cannot accurately describe the real-time status of indoor occupants,so the research of data-driven personal comfort model is getting more attention.The personal comfort model prioritizes cost-effective and easy-to-obtain data,and always contains parameters that represent the instantaneous state of the occupants,and can be continuously trained and updated.At present,most of these models use invasive sensors to collect subject data,which makes the process of data collection interference with occupants,which affects the accuracy of the model,so the actual application is limited.Secondly,this type of model is only developed for a certain type of population,and its performance to other types of population tends to be relatively poor.Until now,few scholars have established a thermal comfort sample database for the Chinese population and trained data to develop personal comfort model.In addition,the current research is mostly focused on the development of the model itself,but subsequent application of the model is less concerned.The integration of the personal comfort model into the control loop of the HVAC system is an important step for its practical application.In view of the shortage of existing research,this paper carries out theoretical and experimental research on the non-contact personal comfort model.First,in order to extract the skin temperature of the key facial area without contacting individuals,the USTC-NVIE database was used to develop an infrared facial landmark detection algorithm,and an improved high-resolution network(HRNet)was developed by deep learning on the database.We achieve the superior accuracy over the state-of-the-art algorithm.Subsequently,subjects were recruited to conduct personal thermal comfort experiments.The physiological parameters for the characteristics of the Chinese population,the environmental parameters in the experiment,and the thermal comfort feedback of the subjects were collected to establish a thermal comfort sample database suitable for the characteristics of the Chinese population.Then,based on experimental data,hypothesis testing and correlation analysis was performed.Personal comfort model was constructed and optimized using Support Vector Machines.Analyzing the performance of different input parameters,the result shows that the input parameters of all skin temperature + environmental parameters + individual parameters have the best effect,which can reach an accuracy of 85%.However,when the acquisition conditions are limited,the combination of the whole face skin temperature + environmental parameters can also achieve high accuracy.Finally,based on the developed personal comfort model,a control strategy for air temperature and wind speed is proposed,and an intelligent air-conditioning system is built.The control effect of the model on air-conditioning is studied and analyzed,and the experiment is compared with the self-control method of air-conditioning by occupants.The experimental results show that the control method using the personal comfort model can perceive changes in the indoor environment and the subject’s immediate state with little interference to the subject,and make indoor individual reach a comfortable state more quickly.Under the same environmental conditions,the average comfort level of subjects using personal comfort model to control air conditioning is higher,and the energy consumption of air conditioning is slightly better than the self-control method.