Informatics Publishing Limited

Abstract

A few papers on studies of global solar radiation estimation for some locations in Nigeria using artificial neural network have been reported. This paper is the first report of gamma test (GT)-driven artificial neural network (ANN) approach adopted for nonlinear data model development and simulation of global solar radiation () within a coastal region and tropical location, Warri-Nigeria (N, E). Gamma test-run with seven(7) years(2003-2009) data consisting of 2201 unique data points contributed by maximum temperature (), minimum temperature  (), rainfall (), Relative humidity at 9hours()  and extraterrestrial radiation () is adopted for the ANN modelling and simulation of   . Using near neighbour number() of 10, the results of the GT scatter plots for all-input embedding dimension gives a gamma statistic() of 0.0005 and an average correlation coefficient of determination () of 0.9887. is simulated using data models developed from three different ANN algorithms (Two Layer back-propagation, Conjugate gradient and Bryoden Fletcher Golfab-Shanno) with feed-forward two layer network topology (5-7-7-1) which learnt at a rate of 0.25 The results obtained for various embedding dimensions and data length choices using both training and validation data sets shows the  ischolar_main mean square error (RMSE) and to range between 0.2373 to 13.1570 and 0.3545 to 0.9994 for the different algorithms. The significant contribution of  for all masks is revealed by the results of mask 11110 where critically low values are obtained when  is excluded and its also noted not to be adequate in predicting  for this location and other similar locations as exhibited by results for mask 00001

Keywords

Full Text