Research On The Thermal Comfort Prediction Of The Elderly Based On Machine Learning In Summer

Modern society is facing serious aging challenges.The number of elderly people over 60 in my country will exceed 300 million during the "14th Five-Year Plan" period.Many elderly people are more susceptible to extreme weather such as temperature fluctuations in summer or winter.Because they are often weak in physique,accompanied by various health problems,and most of them do not actively adjust indoor airconditioning and other thermal environment equipment.However,the elderly spend approximately 90% of their time indoors.The existing PMV model is used as a thermal comfort prediction model for the design standard of the indoor thermal environment of buildings.It does not consider the real-time thermal sensation of personnel.At the same time,the model is obtained based on the experimental data of young people in a steadystate thermal environment.There is a lack of correction methods for the elderly and dynamic thermal environment,and there are shortcomings such as inaccurate predictions in actual buildings.Therefore,in order to create a thermal environment that can meet the thermal comfort needs of the elderly,this study is based on human experimental data and uses machine learning methods to establish a thermal comfort prediction model for the elderly,which can be better adapted to the indoor thermal environment real-time adjustment control system.First,the objective physiological parameters and subjective thermal comfort of the elderly and young people are studied.In this paper,healthy elderly people are recruited to carry out human thermal comfort experiments under summer conditions of 18-34℃and 34-18℃.In order to better explore the differences of the elderly,young people are introduced as a comparison group.The results show that: when the ambient temperature is about 34℃ and 20℃,the overall thermal sensation of the elderly is significantly higher than that of the young under the temperature drop.Performing a linear regression of thermal sensation and ambient temperature,it can be concluded that the thermal sensitivity of the elderly is lower than that of the young,and the thermal expectation results indicate that the elderly are less willing to actively adjust the thermal environment.The neutral temperature ranges for young people and old people are 23～27℃ and 22～27℃respectively.In the temperature drop conditions of this study,the average skin temperature of the two groups did not differ significantly.The temperature of the forehead and the back of the hand detected by infrared,regardless of temperature rise or temperature drop,the forehead temperature has a significant difference when the ambient temperature is warm(P<0.05),but no difference is found in the skin temperature of the back of the hand.In the linear fit between local skin temperature and overall thermal sensation,forehead skin temperature can best characterize the overall thermal sensation of subjects.On this basis,the difference between the thermal comfort response of the elderly and the young in the temperature change is studied,that is,the impact of short-term thermal experience and the asymmetry of cold and heat on the thermal evaluation of personnel.At the same ambient temperature,subjects who experienced a warmer ambient temperature in the previous stage had lower thermal sensation votes than subjects who experienced a colder ambient temperature in the previous stage.Compared with the elderly,young people were more affected by short-term heat experiences.Regardless of whether it is the elderly or the young,there are asymmetries in the hot and cold conditions.The difference between the young people is significant,and the temperature drop has a more severe impact on the subject’s thermal sensation.Finally,a thermal comfort prediction model based on machine learning algorithms is established,which is suitable for the elderly.This research proposes two machine learning data-driven models to predict the thermal perception of people in the built environment.The first model considers the influencing factors of thermal sensation in detail.The input features are: age,gender,BMI,thermal experience temperature,temperature change direction,and local skin temperature of the head,chest,upper arm,lower arm,hand,and thigh.The second model is based on practical applications,using the forehead temperature detected by the infrared array sensor,the skin temperature on the back of the hand,and age and BMI as features to derive a simplified model.The overall prediction accuracy of the first model reached 82%.The machine learning algorithm of this model is Subspace KNN.The prediction accuracy of the second lessfeatured thermal sensation prediction model reached 74.4%.At this time,the optimal machine learning algorithm is random forest.The thermal comfort prediction model proposed in this research can effectively predict the thermal comfort of the elderly based on the objective parameters of the personnel and the real-time skin temperature detected by the infrared sensor.Combined with the automatic adjustment of the HVAC system,it provides technical measures for the construction of the indoor thermal environment for the elderly.