Artificial Neural Networks based Prediction of Insolation on Horizontal Surfaces for Bangladesh

In this work, Artificial Neural Network (ANN) based model for predicting the solar radiation in Bangladesh has been developed. Standard multilayer, feed-forward, back-propagation neural networks with different architecture have been designed using MATLAB's Neural Network tool. The training and testing data of 64 different locations spread all over Bangladesh were obtained from the NASA surface meteorology and solar energy database. The input parameters for the network are: latitude, longitude, elevation, month, average daylight hours, mean earth temperature and relative humidity while the solar insolation on horizontal surfaces are the target parameters. The overall Mean Square Errors (MSE) during training 0.0029, regression value of 0.99707, small percentage of error (0.16% to 1.71%) in response to unknown input vectors indicate that the developed model can be used reliably for predicting insolation of locations where there is no direct irradiance measuring instruments.