A data-driven method to minimize sensor use for dynamic shading control

Abstract

Dynamic facade systems are a promising strategy to balance the benefits and drawbacks of daylight penetration through the window in a built environment. A well-operated shading system should allow for sufficient daylight access while preventing glare in response to changing weather conditions. The dynamic shading system can also help with making use of daylight to reduce the lighting energy use when integrated with electric lighting control. This paper presented a data-driven approach to control the automated shading system. The control method allows the system to be responsive to outdoor weather conditions. With data from offline daylight simulation using typical meteorological year weather, three machine learning classification algorithms were applied to train predictive models for online shading control. The proposed method was validated using a climate-based simulation. It was found that all models had a satisfying prediction performance, with an accuracy of 0.85 to 0.92. The shading control could maintain the work plane illuminance for 97% to 99% of the time during the year. It was also associated with a potential of reducing electric lighting energy use by 40% to 43% compared to that of on/off lighting control.