Akademik Veri Yönetim Sistemi
Electrical Engineering, 2024 (SCI-Expanded)
The comparison of commonly used prediction models for solar irradiation and the determination of the most suitable prediction model regionally are crucial aspect. In this study, the results of solar irradiation prediction obtained from artificial neural network (ANN) models created using the same input parameters as frequently used calculation models for solar irradiation determination have been evaluated. Furthermore, by ensuring that the coefficients used in the calculation models are determined according to each study area, the calculation of solar irradiation at the optimum value has been achieved solely through empirical equations. The study encompasses Ağrı, Hakkari, Siirt, Şırnak and Van provinces in southeastern Turkey, which have high solar energy potential. The basic objective of this study is to present a comparative study among different solar irradiation calculation models and various ANN models enabling the prediction of solar irradiation using different input parameters in these models. In this context to investigate their applicability, and to determine successful models for predicting solar irradiation in the study areas. In this context, the accuracy and suitability of 5 sunshine-based, 4 temperature-based, and 4 hybrid calculation models containing different meteorological parameters, as well as 1 sunshine-based, 2 temperature-based, and 4 hybrid ANN-based models in which the parameters in each calculation model are used as input parameters, have been evaluated. The performance of all models has been evaluated using performance criteria such as mean bias error (MBE), root mean square error (RMSE), normalized root mean square error (nRMSE), and correlation coefficient (R). According to the results obtained, the performance of all ANN models for many provinces lies within acceptable prediction limits. However, according to the MBE, RMSE, nRMSE, and R values, the best performance results were obtained in Siirt for the sunshine-based “Exponential” calculation model with values of −1.2369, 3.3951, 20.9577 and 98.93, respectively, and for the hybrid-based ANN model “Meenal Model-I” with values of 0.0189, 2.1965, 12.6898 and 96.42, respectively. The results obtained from the temperature-based models did not perform well for the designated study areas. Accordingly, the worst solar irradiation prediction performance in Siirt was recorded with the temperature-based calculation model “Meenal” at 2.1452, 9.0254, 36.4125 and 54.46, and in Van with the “Bristow and Campbell” ANN model at −0.1313, 4.3909, 25.2921 and 83.63, respectively. Based on the prediction results obtained from all regions, it was determined that sunshine-based calculation models and hybrid-based ANN models are sufficient as forecasting models for successfully estimating the solar radiation required for Photovoltaic systems.