Research On Real-time Occupancy Detection And Comfort Adjustment System For Indoor Environment

The building occupancy detection has become an active research field in recent years.Seeking suitable methods to obtain the occupancy information of indoor environment plays a vital role in reducing building energy consumption.In addition,under the background of Novel Coronavirus pneumonia pandemic,occupied situation detection can also help to control the density of people in room to reduce the spread of disease.In this study,we established an indoor environment real-time occupancy detection and comfort adjustment system based on cloud platform and sensor array.The occupied situation of indoor environment was divided into three categories represented as unoccupied,normally and abnormally occupied,respectively.“Unoccupied”represents that there are no occupants（empty state）in the indoor environment.“Normally occupied”represents that there are occupiers in the indoor environment,and the number of occupants is within an appropriate range.“Abnormally occupied”refers to the situation that the number of occupants is large and exceeds the setting limit.Through controlling the number and location of sensor nodes,single sensor node was used to detect the occupation of small-area indoor environment,and multi-sensor nodes were used for the occupancy detection of large-area indoor environment in subregion.Meanwhile,to provide a comfortable living and office environment,the system was equipped with a feedback control and adjustment module.According to the actual situation of the indoor environment,the system could adjust the comfort level by lighting and ventilation regulation.Firstly,nine algorithms were trained on the occupancy dataset provided by Candanedo et al.to determine the appropriate occupancy detection algorithm.Regsubsets function in R language is used for full subset regression of the characteristic variables collected by sensors to evaluate the contribution of each characteristic variable and determine the best feature combination to optimize the size of the sensor array.The simulation results showed that the voting based weighted extreme learning machine（WV-ELM）model achieved the highest prediction accuracy.Additionally,the combination of light and CO₂ sensors could achieve a satisfying classification result,which worked as the sensor array in the proposed system.Cloud platform for data visualization and storage,infrared photoelectric sensor correction module,small embedded system raspberry pie and lighting and ventilation module together constituted the final system.WV-ELM model was combined with the proposed system to predict the real-time occupancy situation.To verify efficiency of the system,two parts of experiments were designed.The first part uses a single sensor node in a 12 m² laboratory to verify the feasibility of the system for occupancy detection in a small indoor environment.The results manifested that the system realized a detection accuracy of 97.32%with running time less than 30 s.The second part was conducted in the laboratory in the first part and a large office（54 m²）.To further evaluate the performance of the system,multi-sensor nodes were arranged in different regions in the small and large indoor environment for occupancy detection.The results demonstrated that the system using multi-sensor nodes also achieved satisfactory occupancy detection accuracy（higher than 97.12%）.Simultaneously,the system also had good performance in real-time indoor environment comfort adjustment.