AN ARTIFICIAL NEURAL NETWORK BASED APPROACH FOR ESTIMATING DIRECT NORMAL, DIFFUSE HORIZONTAL AND GLOBAL HORIZONTAL IRRADIANCES USING SATELLITE IMAGES

摘要

This study proposes the use of an artificial neural network approach to estimate the direct normal irradiance (DNI), diffuse horizontal irradiance (DHI) and global horizontal irradiance (GHI) at temporal and spatial resolutions of 15min and 3km, respectively. Inputs to the models are six thermal channels of the SEVIRI instrument, onboard Meteosat Second Generation, along with solar zenith angle, latitude, longitude, solar time, day number and eccentricity correction. The study will show the generalization of the results when using an ensemble approach as opposed to a single network. For all sky conditions the testing dataset for DNI estimations have relative root mean square error (rRMSE) and relative mean bias error (rMBE) values of 17.8% and -3%, respectively. Results for DHI estimations are 13.4% and +1.6%, respectively, and finally GHI estimation results show error values of 7.3% and -1.7%, respectively.