以深度學習為基礎之智慧照明光譜生成技術

This proposal aims to develop a real-time arbitrary spectrum generator for smartlighting control. In this era of human centric technologies, how to effectivelymanage and control the lighting systems have become important research topics.Smart lighting technology facilitates sensor-based remote control on theluminosity, color temperature and usage occasions to offer lighting environmentswith energy efficiency and users’ well-being.This project anticipates to use illuminance, correlated color temperature, thedistance from the Planckian locus, color rendering index, and circadian stimulusas the target parameters in the spectrum generation. A customized geneticalgorithm will first be utilized to generate the spectrum database of a multichannelspectrum variable luminaire. A deep-learning based spectrum generatorwill then be trained to establish the model between the channel weights of theluminaire and the lighting parameters.After the model optimization is completed, input a specific lighting scenario orlighting parameters specified by the user, the spectrum generator will quicklyoutput the weights required for each channel of the luminaire and generate thecorresponding spectrum. This arbitrary spectrum generation technique will bewidely applicable to the smart lighting control of various multi-channel spectrumvariable luminaires.