This study to develop lighting is advanced for reproducing natural light color temperature beneficial to humans. Methods were introduced to provide daily color temperature cycles through formulas based on the measured natural light characteristics or real-time reproduction of natural light color temperature linking sensors. Analysis results for the measured natural light showed that irregular color temperature cycles were observed for more than 90% of the year due to the influence of regional weather and atmospheric conditions. Regular color temperature cycles were observed only on some clear days. The color temperature cycle dramatically affects the health of the occupants. However, since irregular color temperatures are difficult to predict and cannot easily generate cycles, only the color temperatures of some clear days are currently used, and the actual color temperature of natural light cannot be reproduced. There is little research on deriving real-time periodic characteristics and lighting services targeting irregular color temperatures of natural light. Therefore, this paper proposes a TadGAN (Time Series Anomaly Detection Using Generative Adversarial Networks)-based daily color temperature cycle generation method that responds to irregular changes in the natural light color temperature. A TadGAN model for generating the natural light color temperature cycle was built, and learning was performed based on the dataset extracted through the measured natural light characteristic Database. After that, the generator of TadGAN was repeatedly applied to generate a color temperature cycle close to the change of natural light. In the performance test of the proposed method, it was possible to generate periodic characteristics of the irregular natural light color temperature distribution.
Keywords: CCT; TadGAN; daily color temperature; natural light; recursive TadGAN.