A data-driven approach for the control of a daylight–artificial light integrated scheme

This paper presents a data analytics model-based predictive control approach for a daylight–artificial light integrated scheme. The essential data are collected from an automated test room with dimmable LED luminaires and motorized Venetian blinds. This study considered machine learning techniques to develop a novel control strategy for all the four-window orientations for maintaining comfort and energy conservation. The shades are operated one at a time, and the annual data collected were used to develop the predictive models. The irradiance, altitude, temperature and daylight on the window are the predictors, and the blind position is the response variable to establish the models for the windows on all four sides of the test room. The standard support vector regression, Bayesian support vector regression and Gaussian process regression models are analysed in comparison with the baseline model. The luminaire dimming control signals generated using the predicted optimum blind position and exterior illuminance based on a building information illuminance model is commissioned for a given room. This approach mainly concentrated on the implementation of an industrial-level product by reducing the computational complexity of the rule-based blind positioning system. At present, the models are in a reconfigurable embedded WiFi-enabled operating system.