| Exploring new thermal comfort evaluation methods is essential for increased building efficiency,comfort and productivity.With the spread of the Internet of Things in the construction industry,non-invasive thermal adaptive behavior recognition provides a new method to evaluate user thermal comfort in real time.However,high quality large-scale dataset and high-precision multi-view thermal adaptive behavior recognition models are still lacking in the thermal comfort evaluation of office users.Besides,the relationship between thermal adaptive behaviors and thermal comfort,as well as the interaction between thermal adaptive behaviors and environmental factors need to be further explored.Based on these,this paper studied the thermal adaptive behavior recognition and thermal comfort evaluation of indoor users in office buildings,and further developed the non-invasive thermal comfort evaluation method.Questionnaire surveys on thermal adaptive behaviors were firstly carried out for office building users in five typical climate zones of China.From 654 valid questionnaires,12 thermal adaptive behaviors related to thermal discomfort were clustered,including putting on a coat,crossing arms around the chest,warming hands with breath,crossing legs,stamping feet,and shoulder shaking in a cold environment;and taking off of a coat,rolling up sleeves,fanning with hand,fanning with papers,wiping sweat,and shaking a coat in a hot environment.Secondly,62 subjects were recruited for the thermal adaptive behavior recognition experiment,and 14 kinds of thermal adaptive behavior video data were collected from sitting and standing at 8 views(front,back,left,right,left-front,right-front,left-rear,and right-rear)and 2 heights(1.5m and 2.0m).A large-scale dataset for thermal adaptive behavior recognition with 27,504 video samples was developed.Then,two thermal adaptive behavior recognition models were constructed based on the Two-Stream Inflated 3D Conv Net(I3D)and Slow Fast neural network models.The established Two-Stream I3 D and Slow Fast recognition models adopted an end-to-end approach for recognition,which overcame dependence on human key point recognition.The average prediction accuracy of both models reached 95% or above,even when complex background and human key points were partially obscured.Additionally,the recognition time of the two models is millisecond level,which could support real-time recognition.Finally,temperature control experiments were conducted in an office in Guangzhou.Logistic regression method was used to study the relationship between thermal adaptive behaviors and thermal comfort.The interaction between thermal adaptive behaviors and environmental factors was also analyzed.The results showed that the indoor air temperature and globe temperature were typical representatives used to predict the probability of thermal adaptive behaviors(P<0.005).The overall regression accuracy was higher than 70%.The characteristic coefficients of the probabilistic prediction models of users’ thermal adaptive behaviors in cold and hot environments were approximately 26°C and 28°C(indoor air temperature),which were the critical temperatures of indoor thermal satisfaction and dissatisfaction,respectively.The results of this study provide data and model basis for indoor thermal comfort evaluation of office users based on thermal adaptive behaviors.It is also helpful to better understand the thermal comfort of office users and provide reference for non-invasive thermal comfort evaluation. |
