Oh Seung-Taek, Ga Deog-Hyeon, Lim Jae-Hyun
Smart Natural Space Research Center, Kongju National University, Cheonan 31080, Republic of Korea.
Department of Computer Science & Engineering, Kongju National University, Cheonan 31080, Republic of Korea.
Sensors (Basel). 2023 Jan 12;23(2):883. doi: 10.3390/s23020883.
The light intensity and color temperature of natural light periodically change and promote the circadian entrainment of the human body. In addition, the color temperature cycle of natural light that is unique to each region is formed by its location and geographic and environmental factors, affecting the health of its residents. Research on lighting and construction to provide the color temperature of real-time natural light has continued to provide the beneficial effect of natural indoor lighting. However, lighting technology that provides the real-time color temperature of natural light could not be realized since it is challenging to select a color temperature cycle zone due to abrupt color temperature changes at sunrise and sunset. Such drastic shifts cause an irregular measurement of color temperature over time due to general weather or atmospheric conditions. In a previous study, a method of generating a color temperature cycle using deep learning was introduced, but the performance at the beginning and end of the color temperature cycle was unreliable. Therefore, this study proposes generating a real-time natural light color temperature cycle for the circadian lighting service. The characteristics of the daily color temperature cycle were analyzed based on the measured natural light characteristics database, and a data set for learning was established. To improve the color temperature cycle generation performance, a deep learning (TadGAN) model was implemented by searching for the lowest point of the color temperature at the start and end points of the color temperature cycle and applying the boot and ending datasets to these points. The color temperature cycle zone was accurately detected in real-time in the experiment, and the generation performance of the color temperature cycle was maintained at the beginning and end of the color temperature cycle. The mean absolute error decreased by about 67%, confirming the generation of a more accurate real-time color temperature cycle.
自然光的光强和色温会周期性变化，并促进人体的昼夜节律同步。此外，每个地区独有的自然光色温周期是由其地理位置以及地理和环境因素形成的，会影响当地居民的健康。关于照明和建筑以提供实时自然光色温的研究一直在持续发挥自然室内照明的有益作用。然而，由于日出和日落时色温的突然变化，选择色温周期区域具有挑战性，因此无法实现提供自然光实时色温的照明技术。由于一般天气或大气条件，这种剧烈变化会导致色温随时间的测量不规则。在先前的一项研究中，引入了一种使用深度学习生成色温周期的方法，但色温周期开始和结束时的性能并不可靠。因此，本研究提出为昼夜节律照明服务生成实时自然光色温周期。基于实测的自然光特性数据库分析了每日色温周期的特征，并建立了学习数据集。为了提高色温周期生成性能，通过在色温周期的起点和终点寻找色温的最低点，并将引导和结束数据集应用于这些点，实现了一个深度学习（TadGAN）模型。在实验中实时准确地检测到了色温周期区域，并且在色温周期的开始和结束时保持了色温周期的生成性能。平均绝对误差下降了约67%，证实生成了更准确的实时色温周期。
