Artificial Neural Networks for Solar Radiation Prediction: Case Study, Al-Qadisiyah, Iraq

Abstract

For solar energy to develop a clean, renewable alternative to fossil fuels, it is important to be able to correctly predict surface longwave radiation. To improve cost-efficiency and accuracy in surface longwave radiation (SLR) predictions, forecast systems are increasingly utilizing artificial neural networks (ANNs). This study uses two different procedures for predicting solar radiation in great detail. The first model uses weather statistics from a station in Al-Qadisiyah, Iraq. The second model, on the other hand, uses daily statistics from 2022 from NASA's Prediction of Worldwide Energy dataset for the same site. The scaled conjugate gradient technique was used in both models. The goal is to find the best mix of meteorological factors that can be used with an ANN model to achieve accurate predictions. Based on the findings of this study, temperature, relative humidity, and Rainfall all seem to have a big effect on SLR. On the other hand, windy weather doesn't have much of an effect on SLR. ANN models also did very well when trained with data from NASA's Prediction of Worldwide Energy, with an R2-value of 0.823 and an RMSE value of 0.0106. The results show that this mix does better than other models in terms of performance score.