Occupancy prediction using different levels of sensor data

Abstract

With growing concern over climate change and environmental issues, energy saving on building has attracted wide attention in recent years. Occupancy information, such as the presence state and the number of occupants, allows a robust control of the indoor environment to improve the performance of building energy. However, the ability to determine the actual number of occupants in a room is beyond the reach of most current sensing modalities. Where sophisticated methods are required for the prediction of the room or building occupancy. Many different approaches have also been proposed in literature. To address this issue, an experiment is setup in an office space location in the National University of Singapore to examine the feasibility of the different approaches where three prediction model is selected (physical based, support vector machine, artificial neural networks) to predict occupancy based on different levels of existing infrastructure sensors data such as CO2, temperature, relative humidity, plug load and more. Where the performance of the models is evaluated based on four difference performance metrics (error measures). The results suggest a significant correlation between the measured environmental conditions and occupancy. The selection process of the input variables also have critical impact on the performance of the algorithms. With reference to the research findings, the result show a Mean Absolute Error (MAE) of 15, 2 and 2 on the occupancy prediction was achieved by physical based model, ANN and SVM model respectively during testing periods.