Application of Artificial Neural Network in Estimating Hourly Global Solar Radiation from Satellite Images
Abstract
The assessment of Earth-reaching solar radiation continues to be relevant in the utilization of this form of energy. In this work, we have used satellite images of the Earth in the visible spectrum obtained from EUMETSAT and Artificial Neural Network to estimate the solar radiation at our location, Akoka (lat. 6.510N; lon 3.400E), for the months of January, February and March, 2010. The estimates were obtained for the hours between 10:00hours UT and 13:00hours UT. The reflectance of the Earth was determined for each image and from this, cloud index were estimated. The cloud index obtained for each hour for half the total number of days in the months, were used to train a Feed-forward back propagation Artificial Neural Network. Clear-sky index for the hours of the days used for the training served as target for the network. For the three months considered in their usual order, the relative values of RMSE obtained for the hours ending at 11:00, 12:00 and 13:00 were 0.0349, 0.0276, and 0.0356 while the corresponding relative MBE values were -0.120, -0.040 and 0.013.