Prediction of Daylight Illuminance on Classroom Desk Surface Using Machine Learning Model and Illuminance Sensor

The purpose of this study is to predict the desk illumination of the classroom using the results obtained through sensors that can measure the natural lighting environment, and to save energy by dimming the artificial lighting of the classroom. In addition, a theoretical equation for predicting desk illumination through sensor values was produced. However, the desk illumination measured through the theoretical formula had a disadvantage of 70% accuracy and could not consider the blinds in the classroom. Machine learning was conducted to solve these shortcomings. Since constructing a machine learning dataset through actual measurement has time and space limitations, a simulation was produced to construct a dataset. As a result of comparing the produced simulation and the measured value, there was an error within 5%, which was used as machine learning data. There are a total of 192 sets and 1,728 data of natural light sensor values and classroom desk illumination values measured by changing direction, blind use, season, and time through simulation. As a result of learning this through the XGBoost regression model, the simulation value showed an accuracy of 95.4%, and the actual measurement value showed an accuracy of 95.2%. In addition, as a result of predicting desk illumination by natural lighting in each season and calculating the lighting energy saving effect, it was found to have an energy saving effect of 13.0% in winter, 29.6% in spring and fall, and 40.7% in summer.