Virtual Occupancy Sensors (VOS) model for Building Cooling demand Prediction and Cooling System Control
DescriptionThe energy consumption in buildings is significantly affected by the behaviours of the occupants. Studieshave revealed that, with an occupancy-driven control, the energy consumption of a Heating, Ventilation andAir-Conditioning (HVAC) system can be reduced by as much as 56%. Occupancy level can also be used forbuilding energy consumption prediction. However, precise and reliable estimation of occupancy remains achallenge. The reasons are that current detection technologies suffer from sensor drift, privacy concern, poorquality, intrusiveness, change of use, and insufficient commissioning. Recently sensor fusion techniques areintroduced which takes advantages of salient features of various sensors and keeps errors within control,thus enabling better occupancy estimation possibilities. They can be applied to supplement existing sensorsfor building system controls and monitoring, and to enable occupancy based energy consumption predictions.Redundant sensors as well as combinations of ambient sensors have been experimented for multiple sensorfusion algorithms. Non-ambient sensors such as wearable sensors and ICT- based sensors have also beenutilized for the study. A robust and reliable multi-sensor fusion algorithm for occupancy detection inbuildings may enable a more efficient building systems control and operation, and provide an alternativeapproach to building energy consumption prediction.The aim of this study is to develop a Virtual Occupancy Sensors (VOS) based on data fusion algorithm foroccupancy estimation in buildings. The effectiveness of the VOS model in predicting energy usage and inimproving the thermal comfort and energy efficiency of a CAV system with PI control will also beevaluated. Occupancy space electrical power demand, visual information from video camera, and Wi-Ficonnection profile are selected for constructing the VOS model. They are readily available in buildings, easyto install and replace, comparatively inexpensive, not susceptible to long term drift and/or any drifting caneasily be noticed, and not easily affected by changes in indoor ambience. The VOS data can be processedeasily so that the VOS model can be adapted to long term sensor drifts. This research will adopt existingVOS sensors, the practical implementation of which will not have adverse effect on the present privacystatus quo. A research staff office with an area of 135 m2will be used as the test office for the VOS model.The occupancy space electrical power information is recorded by the building power monitoring system.Visual information will be captured by surveillance cameras already installed in the test office. Wi-Ficonnection profile will be extracted from the existing Wi-Fi routers. A genetic based CFS classifier will beadopted to estimate the best combination of features for representing the occupancy. Ground truthoccupancy information will be retrieved from the staff card access control system of the test office for theconstruction of the CFS classifier. Various combinations of the features will be studied so as to examinewhich feature(s) are more appropriate for certain conditions. Moreover, the effectiveness of the VOS modelin cooling demand prediction will be evaluated for a 25-seat classroom. Artificial neural network (ANN)based simulations will be conducted to examine the cooling load prediction accuracy when variouscombinations of input parameters are adopted. The performance of the proposed cooling load predictionmodel will be compared with the performance of other published cooling load prediction models.Furthermore, the effectiveness of the VOS model based cooling system control on improving thermalcomfort and energy efficiency will be evaluated. The VOS model will be incorporated into a direct adaptiveANN model, that will adjust the gain of the PI control of the CAV system in the 25-seat classroom. Thedirect adaptive ANN algorithm exhibits a learning-then-control feature, the feedback control can update theANN weights by the difference between the desired output and the actual output. This relaxation of apriorimodel of the system renders the controller the flexibility in design and implementation. In particular,detailed studies will be carried out to determine the safe range of the estimation gain to ensure the stabilityand robustness of the control system. An appropriate updating interval of the control would also beexamined to ensure fast and stable response of the control system.
|Effective start/end date||1/01/18 → 9/12/20|