Implementation of machine learning techniques for the quasi real-time blind and electric lighting optimization in a controlled experimental facility

Abstract

Machine Learning techniques have been recently investigated as an alternative to the use of physical simulations, aiming to improve the response time of daylight and electric lighting performance-predictions. In this study, daylight and electric lighting predictor models are derived from daylighting RADIANCE simulations, aiming to provide visual comfort to office room occupants, with a reduced use of electric lighting. The aim is to integrate an intelligent control scheme, that, implemented on a small embedded 32-bit computer (Raspberry Pi), interfaces a KNX system for a quasi-real-time optimization of the building parameters. The present research constitutes a step towards the broader goal of achieving a unified approach, in which the daylight and electric lighting predictor models would be integrated in a Model Predictive Control. A verification of the ML performance is carried-out by comparing the model predictions to data obtained in monitoring sessions in autumn, winter and spring 2020-2021, resulting in an average MAPE of 19.3%.

Enhancing user acceptance in automated systems with human-centric lighting: the role of visual comfort, personality, and preference

“…Prior to participating in the study, participants were provided with detailed information on how to operate the system and what changes might be operated by the system. Participants were instructed to pursue with their routine work activities, during the whole experiment duration, in their respective offices and while the automated system was operating to optimize daylight provision (for more details about the system see [14]). Each day, when participants arrived at their offices, they activated the automated system by indicating their presence.…”