A Study On Real Time Building Occupancy Prediction Based On Environmental Parameters

Building occupancy prediction is of great importance for promoting energy efficiency control of buildings.In this paper,a comprehensive review on occupancy measurement methods and prediction methods was first presented,and the advantages and disadvantages of the methods were also summarized and compared.Existing researches mainly applied machine learning methods to predict occupancy based on indoor environmental parameters.They can automatically mine the correlation between environmental parameters and occupancy.However,such methods have difficulties in extracting the informative and effective features from the environmental parameters and tent to easily ignore the dynamic characteristics of occupancy and thus generate bias.In this view,this paper proposed two novel models to predict the presence/absence and the number of occupants separately.First,this paper developed an inhomogeneous hidden markov model based on state transitions（TIHMM）using environmental parameters and applied it to predict the presence/absence in an office and also compared its results with the HMM and the IHMM.Second,this paper developed a deep learning model（i.e.gcForest）integrated with wavelet denoising method to predict the number of occupants based on CO₂ concentration.To evaluate the effectiveness of the proposed model,a validation experiment in an office was conducted and its results was compared with SVM,CART and IHMM models.The main conclusions are as follows:In terms of the proposed model for estimating the binary occupancy,the accuracy can be improved to about 98.1%by TIHMM compared with the 94.3%by HMM and95.4%by IHMM.Moreover,compared with the HMM and the IHMM,the TIHMM can more effectively capture the changes of the presence/absence,in particular the first arrival time and the last departure time.In terms of the proposed model for estimating the number of occupants,the experimental results show that the wavelet denoising method could filter the noise and preserve the data features of CO₂ concentration.The estimation accuracy after wavelet denoising can be increased by 3.5%.Moreover,the proposed model could achieve higher estimation accuracy up to 82.5%,whereas the accuracy is 75.7%,74.9%and73.9%for the CART,the SVM and the IHMM respectively.Additionally,this model can capture both the first arrival time and the last departure time.As the maximum depths of classifiers impact the gcForest’s performance,the results also show that a proper selection of the maximum depth combination could lead to a significant improvement of the model estimation accuracy.In the future,it would be interesting to developed models to predict more comprehensive information of building occupancy,such as the location of the occupants.Also,it is promising and significant to investigate the applicability of the proposed models,which can be coupled to building control systems and building energy simulation software.