Paper Conference

Proceedings of BSO Conference 2020: Fifth Conference of IBPSA-England

     

Detection of Window Opening Using a Deep Learning Approach for Effective Management of Building Ventilation Heat Losses

Paige Wenbin Tien, Shuangyu Wei, John Kaiser Calautit

Abstract: Occupant behaviour within buildings has a significant impact on building energy consumption and represents one of the main sources of excessive energy use. This paper introduces a data-driven deep learning framework that enables the detection and recognition of opening and closing of windows. This approach is based on a control strategy which can detect and recognise the period and state of the window opening in real-time and simultaneously adjust the heating, ventilation and airconditioning (HVAC) system to minimise energy wastage and maintain indoor environment quality and thermal comfort. The framework is based on a trained deep learning algorithm deployed to an artificial intelligence (AI)-powered camera. To assess the capabilities of the proposed deep learning framework, building energy simulation (BES) was used with various operation profiles of the opening of windows; fixed profile, actual observation profile and deep learning influenced profile (DLIP). The DLIP is the profile generated via the framework which uses the data obtained from the realtime window detection. A university lecture room with a south-facing window was selected for the modelling and testing of the method. The initial results using the deep learning model showed that it can recognize the state of the window openings with an accuracy of up to 77.8%. Further developments include framework enhancement to improve detection accuracy for multiple window opening types and sizes and to provide automated setpoint adjustment for HVAC systems.
Pages: 194 - 199
Paper:
bso2020_Tien