Three Structures of a Multilayer Artificial Neural Network for Predicting the Solar Radiation of Baghdad City- Iraq

摘要

In this study, a multilayer neural network with three structures (4-4-4-1), (4-8-8-1) and (4-9-9-1) is investigated to predict the average daily solar radiation in the capital city of Iraq. A MATLAB algorithm is used to implement these structures with totally dataset of 2604 hourly points for each January and August in three years 2014, 2015 and 2016. The database was recorded in the Energy and Environment Research Center Station of Al-Jadriyah-Baghdad as 14 hour in day and then the average value were taken for one day as maximum daily ambient temperature, sunshine duration, relative humidity and wind speed as input parameters and average daily solar radiation as output parameter. The algorithm is trained through the back propagation technique with traingdm, learngdm, MSE, and tansig as the training, learning, performance, and transfer functions, respectively. The results show that the optimum testing structure for solar radiation in January was the 4-9-9-1 structure with coefficient of determination (R~2 = 0.925) and Mean square error (MSE=6.5), while in August the structure 4-8-8-1 was the best with coefficient of determination (R~2 = 0.934) and Mean square error (MSE=5.4).