ESTIMATION OF DAILY IRRADIATION EXPOSURE OF GLOBAL RADIATION USING ELMAN NEURAL NETWORK

Using the data of three M eteorological Stations at Fushan, Jiʼnan and Juxian, 2000-2003, the Elman neural network model was established to estimate the daily solar irradiance. The modeling results showed that a the three M eteorological Stations, the M ean Percentage Error(M PE) ranged from 17.3% to 21.3%, the Root M ean Square Error(RM SE) from 1.7 to 2.02 M J·m~ (-2). The difference between the estimated and observed daily solar irradiance was smallest at Fushan Stations among the three M eteorological Stations, ranging from-2 to 4 M J m~ (-2). The estimated daily solar irradiances were greatly affected by the weather conditions, which were more accurate under the clear weather than those in other weather. Compared with the generalized regression neural network modeling results, the M PE decreased by 9%-16% and the RM SE decreased by 0.506 M J·m~ (-2) on average at the three M eteorological Stations.