Research On Occupancy-Based Thermal Comfort Evaluation For Smart Buildings

In the context of the digital economy,the physical world of architecture is gradually merging with the digital world,giving rise to the concept of smart buildings.Comfort and energy efficiency are two major objectives of smart buildings,and the thermal comfort model is key to achieving both goals.It can perceive the thermal comfort status of indoor occupants and help buildings take effective actions.However,existing research typically only considers common factors such as temperature and humidity that affect human thermal comfort,which is not comprehensive enough to meet the modeling needs of high-accuracy thermal comfort models.In addition,thermal comfort modeling typically requires sufficient data from building sensing and management systems.However,in the model construction and training process,the problem of insufficient and imbalanced sample data is inevitable,which has become the main constraint on the accuracy of thermal comfort prediction models.To overcome this dilemma,this paper explores the optimization problem of human thermal comfort modeling in the context of smart buildings.Starting from the mechanism of human perception,a new thermal comfort modeling strategy is proposed,aiming to comprehensively improve the performance of the thermal comfort model and user experience.The main contributions of this paper are shown as follows:Firstly,in response to the problem that the thermal comfort factors are not comprehensive enough to meet the modeling needs of high-precision thermal comfort models,this paper studied a new thermal comfort factor-indoor occupancy.Based on Fanger’s theory and actual investigations,the impact of indoor occupancy on human thermal comfort was confirmed.Secondly,aiming at the detection of indoor personnel in smart building scenarios,indoor personnel counting method oriented to thermal comfort is designed,which can achieve accurate indoor personnel counting on the basis of protecting user privacy.Secondly,this paper proposes a thermal comfort evaluation method based on transfer learning under heterogeneous datasets to address the issue of limited and unbalanced sample data in the modeling process of the thermal comfort model,which restricts the performance of the thermal comfort prediction model.It can be further divided into three parts: sample balancing based on the CTGAN network,feature extraction based on transfer learning,and feature fusion based on improved XGBoost.The experiments show that compared with competing methods,the TCTL method proposed in this article has better reconstruction performance.Finally,in real life,we are more faced with a large amount of unmarked thermal comfort data.in order to mine the potential value of unmarked data,a pseudo labelbased domain adaptation thermal comfort evaluation is proposed.It can be further divided into three parts: pre-trained feature representation,feature alignment,and cyclic pseudo-label weighted learning based on multidimensional confidence.The designed models and algorithms have significant classification performance advantages on two real-world datasets.