Research On Personal Thermal Comfort Model Based On Machine Learning

With the continuous development of thermal environment control technology,people put forward higher and more personalized requirements for the thermal comfort of indoor environment.In order to better solve people’s increasing personalized thermal comfort needs,the main topic of this study is to establish a personal thermal comfort model with accuracy,stability,applicability and expansibility.In this paper,the office building crowd in hot summer and warm winter area is selected as the main research object.After in-depth research on the adaptability of machine learning and thermal comfort,the integrated learning model based on stacking strategy is designed,and the iteration from general model to personal thermal comfort model is completed.Finally,the online prediction module design and development of personal thermal comfort model are completed.In this paper,we have conducted in-depth research from four aspects and obtained the following results:One is to study the model performance of machine learning.Aiming at the specific data form of thermal comfort field data set,based on the field test data of office in hot summer and warm winter area,this paper models and analyzes different types of machine learning algorithms in machine learning.The purpose is to explore the most suitable machine learning model for the research of personal thermal comfort model,and lay a solid theoretical foundation for the subsequent establishment of personal thermal comfort model.The second is to build a general thermal comfort model based on stacking strategy.Using stacking integration strategy and using the existing high-quality ASHRAE global thermal comfort data set Ⅱ,this paper integrates multiple optimal machine learning models and establishes a general thermal comfort model.On the one hand,it can effectively solve the problem of model training difficulty caused by small amount of individual data in previous machine learning research,on the other hand,it can effectively improve the prediction performance of general thermal comfort model.Third,further individual adaptation based on general thermal comfort model.Based on the training of the original general thermal comfort model,the personal data set is added for retraining to make the model more in line with the individual needs.Through comparative analysis,it is verified that the personal thermal comfort model proposed in this paper has engineering value.Fourthly,based on the personal thermal comfort model,the online prediction module and visual web interface are designed and developed,which can predict the input parameters in real time and monitor them in real time.As a completely independent module,the application can access to different scenarios,which greatly reduces the limitation of application scenarios.The work of this paper effectively solves the problems of poor model performance and not meeting the personalized needs in the traditional application of machine learning technology.It provides an effective method for the establishment of personal thermal comfort model,and also provides a reliable scheme for practical engineering application.