Evaluation of the Capability of Intelligent Models in Estimating Monthly Global Solar Radiation

The concern of the present research was to do a comparative study between the GEP, ANN and ANFIS models to estimate monthly global solar radiation. For this purpose, long-term (24-years) monthly data of global solar radiation (RS, MJ m−2), sunshine hours and air temperature (°C), from Tabriz synoptic station were used. To perform the artificial intelligence models, a new combination of inputs including monthly mean clearness index (KT), monthly temperature range (ΔT), relative sunshine hours (n/N) and extraterrestrial global solar radiation (Ra) were employed. Since the lowest values of MBE and RMSE (0. 13 and 1.97 MJ m−2 respectively) and the highest value of R2 (0.92) were obtained for ANN model, and therefore, the ANN model was selected as the best model to estimate the monthly global solar radiation. Using quarter-quarter (Q-Q) plots revealed that although the ANN model generally presents the best fit for monthly global solar radiation data, this model is found to be not successful in estimating the higher values of monthly global solar radiation data. Therefore, the application of ANN model is recommended for regions with lower solar radiation values. The performance of the ANFIS model was better than other models in covering the highest and lowest values (the first and fourth quarter). Therefore, it can be concluded that the ANFIS model gives more accurate results in the areas with the higher values of solar radiation. The findings also show that unlike previous researches which were carried out in daily scale, the performance of GEP technique for modeling monthly global solar radiation is satisfactory especially in the ranges of 250 to 800 MJ m−2. Thus, it can be inferred that GEP can be more powerful in modeling the phenomena which have low fluctuations or a limited range.