Abstract
In this study, spatial distribution, temporal variations, annual distribution, estimation and prediction of solar radiation in Nigeria was carried out using ANNs. Levenberg-Marquardt backpropagation algorithms was used for the training of the network using solar radiation data along the years (1979-2014). The data records were divided into three portions (training, testing and validation). The network processed the available data by dividing it into three portions randomly: 70% for the training, 15% for validation and the remaining 15% for testing. Input parameters were chosen as latitude, longitude, day of the year, year while observed solar radiation was chosen as targeted data (from a processed file). The output parameter was the estimated solar radiation. The network designs were tested with root mean square error and then the most successful network (taken to be best network) which is network with less error was used to carry out the study. The hyperbolic tangent sigmoid transfer function was also used between the input and the hidden layers as activation function, while the linear transfer function was used from hidden layers to the output layer as the activation function. The performance of ANNs was validated by; estimating the difference between the annual measured and estimated values were determined using coefficient of determination (R2). Results revealed that the R2 result was 0.82 (82%). The result of spatial variations indicated that both wet and dry seasons have their highest concentration in North-East of Nigeria. It is pertinent to also note that the lowest concentration occurred in North-West during wet season, while the lowest occurred at the South-South and South-West of Nigeria in dry season. In addition, the lowest in dry season is about 25W/m2, while that of wet season is about 15W/m2. The agreement between the temporal and annual variation of observed and estimated solar radiation reveals that the model exhibits good performance in studying solar radiation. The model was further used to predict two years ahead of the years of study.