PREDICTION OF HOURLY SOLAR RADIATION USING AN ARTIFICIAL NEURAL NETWORK

The aim of the presented study is to apply artificial neural networks (ANNs) method for prediction hourly solar radiation of the selected six provinces of Turkey. Six neurons which receive input signals of latitude, longitude, altitude, day of the year, hour of the day and hourly mean atmospheric air temperature were used in the input layer of the network. One neuron producing corresponding output signal of the hourly solar radiation was utilized in the output layer of the network. The model for training and testing in the formed ANNs was analyzed. Neuron numbers in the hidden layer (from 6 to 30 neurons step by step) and epoch numbers for 100 epochs were tested for different values. The obtained results for this model was compared by using different neurons, mean squared error (MSE), coefficient of determination (R-2), mean absolute error (MAE). The best results for the training were obtained as 25 neurons in terms of minimum MSE value of 0.000607. The R-2 values of the ANN for training and testing data of the 25 neurons are determined as 0.9879 and 0.9891 while the MAE values of which are 18.33 W/m(2) and 18.94 W/m(2), respectively.