Occupant counting model development for urban building energy modeling using commercial off-the-shelf Wi-Fi sensing technology

Abstract

Urban Building Energy Models (UBEMs) are vital for estimating building energy use and related greenhouse gas emissions. However, their reliability needs boosting by using more real-world data, especially regarding occupancy behavior. Presently, UBEMs often use standard occupancy patterns, which may not reflect the real building use, especially for commercial and institutional buildings.Wi-Fi sensing is a reliable approach that can improve UBEMs due to its wide availability and cost-effectiveness. In this study, we leverage detailed signal data derived from Wi-Fi sensing technology to create a realistic, scalable and cost-effective occupancy model. A framework has been developed to derive the number of people in a building, which will influence energy usage patterns and enhance UBEMs.The developed model employs various machine learning techniques and achieves a test accuracy of 77 %. Limited availability and diversity of the initial dataset necessitated the use of data augmentation techniques, enabling the model to learn varied representations and thus achieve better test performance of 91 % post-augmentation.To evaluate the effectiveness of the developed model, it has been applied to two institutional buildings of a specific inner-city district in Montreal, Canada, to compute their heating and cooling demands. The outcomes are then compared with those obtained using standard schedules, revealing a considerable discrepancy in annual peak and total annual cooling demand of about %5 and 20 % respectively.