PREDICTION OF PRODUCTIVITY 50 WATT SOLAR PANEL USING NEURAL NETWORK BACKPROPAGATION

Abstract

This paper explain the prediction of 50 watts solar panel output using back propagation neural network based on the data of Semarang weather station climatology. Motivation of this research based on the problem that is difficult to estimate the average of solar panels power output daily.The output of solar panels influenced by the wether condition. Data of climatology used as input of artificial neural network are the intensity of solar radiation, humidity, wind speed, and temperature which are obtained from the Meteorology, Climatology and Geophysics Agency monitoring stations of Semarang. The output of neural network predicts power output of 50 watt solar panel. The neural network is trained by measurement data during nine months started from June 2012 until February 2013.Artificial neural network architectures are built using backpropagation method with four (4) inputs, one (1) unit hidden layer and one output. The inputs include the intensity of solar radiation, humidity, temperature and wind speed, and the output is the predicted energy output of 50 watt solar panels implemented in Semarang Indonesia. The amount of data used in this research was 273 days thatwere separated into two groups; 200 for training and 73 for testing. The results shows that the best configuration backpropagation neural network model with one hidden layer with eight (8) neurons, learning rate 0,01 and iteration 10000. This configuration carried out the coefficient of determination 98.5%,standard error 8,2 watt-minute, MSE 0,00369 and could predict 1026,4 watt-minute from 1071,6 watt-minute targets output 50 watt solar panel in Semarang.