ESTIMATING DAILY GLOBAL SOLAR RADIATION OVER AKURE, NIGERIA USING ARTIFICIAL NEURAL NETWORK

Abstract

This research work aims at developing an artificial neural network (ANN) model for estimating global solar radiation over Akure (7°15’0”N, 5°11’42”E). The goal was to estimate solar radiation using a combination of meteorological parameters and to identify the combination that gives better estimates when solar radiation data is not readily available. Several artificial neural networks were created with three different architectural designs. A feed forward back-propagation algorithm with three layers (input, hidden and output layer) was developed, trained and tested on ten years (1981-1991) daily data of sunshine hours, temperature ratio and temperature mean over Akure, Nigeria. The data was divided into two sets, the first set from 1981-1987 was used for training and the second set from 1988-1991 was used for testing the artificial neural network. After a successful training of the networks, it was tested on new inputs (1988-1991). The performance of the three networks were compared .The one input (sunshine hours) network gave a root mean square error (RMSE) value of 2.328 MJ/m2, mean biased error (MBE) value -0.075 MJ/m2, and correlation coefficient (R) of 0.762. The two input (temperature mean and temperature ratio) network gave a RMSE value of 2.692 MJ/m2, MBE value -0.926 MJ/m2, and correlation coefficient (R) of 0.69. The three input (sunshine hours, temperature mean and temperature ratio) network gave a RMSE value of 1.942 MJ/m2, MBE value -0.553 MJ/m2, and correlation coefficient (R) of 0.849. Although the results of the three input model using sunshine hours, temperature mean and temperature ratio is reliable ,the model using sunshine hours as the only input showed more accuracy in estimating daily global solar radiation in Akure with a mean bias error (MBE) of - 0.075MJ/m2 and root mean square error (RMSE) of 2.328MJ/m2. The study therefore concluded that synthetic daily global solar radiation can be generated using artificial neural network with Sunshine hours as the only input to a reasonable level of accuracy for stations where ground measurements of global solar radiation are not readily available.