An innovative shading controller for blinds in an open-plan office using machine learning.

Achieving visual and seasonal thermal comfort is an intractable issue for automated shading controllers over multiple blinds serving large side-lit, open-plan offices, especially when the occupied positions are spatially and temporally transient. In the current literature, few proposals have accounted for this. This study describes a novel model-based shading controller to fulfill this gap by optimizing the vertical eye illuminance conditionally at any occupied place. The controller was generated through real-time daylight simulations and two surrogate model techniques (online and offline) based on the radial basis function neural network. The offline surrogate model can predict timely vertical illuminance at any occupied position in the worst scenarios, and the online surrogate model yields the best combined shading results in a timely manner by an accelerated optimization procedure. The accuracy of the customized prediction models embedded in the controller was verified. Comparative simulations were performed for an open-plan office in Harbin, China. The performance regarding visual comfort, daylighting, electrical energy savings, and seasonal solar heat gains were explored and evaluated, demonstrating the advantages of our proposed control approach. • A novel shading controller to maneuver multiple blinds for visual comfort and energy savings for open-plan offices. • Machine learning-based surrogate modelling was adopted into controller for acceleration in adjustment. • Vertical illuminance was used as optimization criteria to alleviate the glare caused by the intense diffuse daylight. [ABSTRACT FROM AUTHOR]