Abstract:
This aim of this work was to use Artificial Neural Network, ANN to estimate the daily average
global solar radiation in Owerri and Port-Harcourt. The specific objectives were to develop and
train ANN for estimating global solar radiation, validate daily global solar radiation data using (the
trained) neural networks and select the best among the trained ANN model for estimating daily
average global solar radiation on a horizontal surface by meteorological data in the study areas.
The database used consisted of 5 years, (2009 - 2013), daily data over Port-Harcourt and Owerri
(altitude 19m, latitude, 4.09oN, longitude 7.01o E and altitude 90m, latitude 5.05oE and longitude
7.04oE respectively), both in Nigeria. Daily average sunshine hours, wind speed, maximum and
minimum temperature, and rainfall were used as inputs, and the daily average global solar radiation
was the targeted output of the estimation. Three different neural networks were created for each
area using one input (sunshine), three inputs (sunshine, minimum and maximum temperature), and
five inputs (sunshine, minimum and maximum temperature, rainfall and wind speed) in other to
select the network that gives the best estimation. The method involved two steps; training and
testing. The training section involved the use of part of the dataset (2009-2012) for each of the
networks and also varying the number of neurons in the hidden layer to minimize error. In the
testing section the remaining one year dataset (2013) was used to validate the trained networks.
The performance of ANN models was tested with the well known statistical tests of performance
namely; Mean Bias Error (MBE), Root Mean Square Error (RMSE), and correlation coefficient,
(R). The results showed that the ANN model performed very well in estimating daily global solar
radiation on a horizontal surface from other meteorological parameters over the station. The
network with the best performance for Owerri and Port-Harcourt is three input (sunshine,
minimum and maximum temperature) network (R= 0.75, MBE= -1.124MJ/m2, RMSE=
3.269MJ/m2) and five input (sunshine hours, minimum and maximum temperature, rainfall and
wind speed) network (R, 0.74, MBE, 0.670MJ/m2, RMSE, 3.2180MJ/m2 ) respectively. The
network with the least performance for both Owerri and Port-Harcourt is the one input (sunshine)
network (R, 0.675, RMSE, 1.9486MJ/m2, MBE, -0.0158 MJ/m2,) and (R, 0.54, MBE, 0.9158
MJ/m2, RMSE, 4.1593 MJ/m2) respectively.
This study concluded that ANN method can be used to accurately estimate daily global solar
radiation on a horizontal surface from other meteorological parameters over the study areas.
