Comparative analysis of estimated solar radiation with different learning methods and empirical models

Abstract

Solar radiation, which is used in hydrological and agricultural modeling, agricultural, solar energy systems, and climatological studies, is the most important element of the energy reaching the earth. The present study compared the performance of two empirical equations -Angstrom and Hargreaves-Samani equations- and three machine learning models -Artificial Neural Networks (ANN), Support Vector Machine (SVM), and Long Short-Term Memory (LSTM)-. Various learning models were developed for the variables used in each empirical equation. In the present study, monthly data of six stations in Turkey, three stations receiving the most solar radiation and three stations receiving the lowest solar radiation, were used. In terms of the mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and determination coefficient (R²) values of each model, the LSTM was the most successful model, followed by ANN and SVM. The MAE value was 2.65 with the Hargreaves-Samani equation and decreased to 0.987 with the LSTM model, while MAE was 1.24 in the Angstrom equation and decreased to 0.747 with the LSTM model. The study revealed that the deep learning model is more appropriate to use than the empirical equations, even in cases with limited data.