Research On Prediction Technology Of Occupancy Level Of Building Interior Space Based On Environmental Data

Accurate and reliable prediction of occupancy levels plays a key role in building energy efficiency and occupant-centric building operations.Existing HVAC energy saving systems place more emphasis on occupancy detection,estimation,and localization to trade off energy consumption and thermal comfort.Different from methods such as cameras and radio frequency identification technology,the estimation method that uses indoor environment variables as input has attracted the attention of researchers at home and abroad because of its advantages of low cost,high efficiency,and privacy protection.However,existing research still faces some challenges.On the one hand,the generalization of the indoor occupancy level prediction model is difficult to guarantee due to the lack of resources and low quality of the data set used for model construction and training.On the other hand,current single-view-based models have limited predictive capabilities,resulting in low predictive accuracy.In this work,a non-intrusive indoor occupancy prediction method jointly optimized by transfer learning and multi-view learning is proposed to solve the above problems.Specifically,the main work and innovations are as follows:First of all,in order to improve the generalization ability of the model and solve the situation of a large amount of unlabeled data,a migration learning architecture based on environmental feature adaptation is proposed,which uses multi-sensor data sets in the source domain for initial model training,and the learned knowledge Shared to the local,so that the model can better adapt to different scenarios.In addition,with the help of the powerful computing power support and large storage database in the cloud,knowledge and features can be obtained from the trained model,reducing the training time of the local model and the usage of computing resources,and realizing real-time staff occupancy level estimation.Then,a method for estimating indoor occupancy level based on multi-view collaboration,that is,T-MCCI algorithm,is designed.First,to solve the sparsity problem of multi-sensor data,selfsupervised learning based on multiple environmental parameters is performed on the cloud to learn knowledge,and the knowledge is transferred to the local;then,the local model combines cloud knowledge and local data for collaborative learning,and The weight-based label combination algorithm mines the consistency and non-consistency information between the classifier labels;finally,through the measurement of the diversity of the classifiers,the collaborative training process is tracked and controlled,and the staff occupancy level is estimated.Experimental results on the collected data sets show that compared with existing supervised learning algorithms and semisupervised learning algorithms,the performance of the occupancy level algorithm proposed in this paper is higher.Finally,on the basis of the indoor occupancy level estimation algorithm based on multi-view collaboration,this paper proposes an optimization scheme that is more suitable for actual smart building scenarios,that is,the T-MACI method.On the one hand,the metric loss based on the transfer learning network constrains the view features of sensor data,making the learned features more discriminative and generalizable.On the other hand,starting from the two dimensions of global features and local detail features,the correlation between sensor data and occupancy level is studied,and an attention classifier based on classification quality is constructed.Finally,the feature extraction network is updated based on the collaborative iterative network,which enables the occupancy level model to have better occupancy estimation performance.Experimental results on real datasets show that the proposed algorithm outperforms the comparison algorithms.