Prediction of solar irradiance with machine learning methods using satellite data
Uğur Ercan, Abdülkadir Koçer
Abstract
Solar energy is used in many domestic and industrial areas. To make maximum use of solar energy, it is important to know the irradiance value per unit area. Irradiance measurements are carried out at many stations. In cases where the irradiance value is not measured, empirical models are generally used. With the widespread use of machine learning methods in prediction problems in many fields, it has gained increasing popularity recently. This study aims to predict the solar irradiance of all cities in the Mediterranean region in Turkiye using statistical method and popular machine learning methods. The performances of these models are compared. Ensemble Learning methods (Gradient Boosting, Extreme Gradient Boosting) which are among the popular machine learning methods, and Artificial Neural Networks were used. The data used belong to 8 cities in the Mediterranean Region, Türkiye. Declination angle, wind speed, ambient temperature, relative humidity, and cloudiness index were chosen as input variables. When the results of the models established for each city are examined, it is seen that the models established with machine learning methods are more successful than statistical methods. The best results were obtained from models established with the Extreme Gradient Boosting method (R2 = 0.9993, MAPE = 0.0119).