The effect of climatic parameters in the modeling of solar radiation using artificial intelligence (Case study: Yazd synoptic station)

Solar radiation is one of the important parameters that affects on many soil and water processes such as evaporation, snow melting and plant growth. Considering the importance of the amount of radiation in the the application of solar energy and the many problems in recording this parameter and the success of intelligent models in predicting complex parameters, it is reasonable to use ANFIS and ANN models to predict the radiation parameter.
In this study, using a large database on a wide period which contained a set of meteorological and geographical data such as latitude, longitude, months of the year, the average temperature, the sunshine duration, relative humidity and the average of the monthly global solar irradiation, the performance of two techniques, artificial neural network and Active Neuro-Fuzzy Inference System, was investigated for the next 12 months in the Yazd station. Sensitivity analysis of different climate parameters such as maximum temperature, average temperature, sunshine hours, relative humidity, solar radiation and evaporation, showed that they were important factors in predicting of solar radiation. Then, the two models were analyzed with different combinations of data. After ensuring the performance of the two models in the testing phase and achieving the best results with the highest efficiency and lowest error rate in the prediction of solar radiation, only by entering the most effective climatic parameters of 2005, the solar radiation value of 2006 was forecasted, and the predicted values were compared with actual values. The results of this study showed that both methods have the ability to simulate the amount of solar radiation. High values ​of the correlation coefficient and low error, confirm the reliability of the results. According to the results, although the highest correlation coefficient was obtained using artificial neural network, the results of both models were satisfactory in two stages of testing and evaluation and are estimated to be close to each other. In total, artificial neural networks with a correlation coefficient of 0.91 and RMSE and MAE rates of 0.19 and 0.08, respectively, produced less error in predictions in comparison with fuzzy-neural networks. Also, the BIAS value is -0.30, which shows a small negative overstimation in the data. Finally, the composition of sunny hours, average temperature, maximum temperature as the optimal combination was identified. In addition, it was determined that sunny hours and average temperatures are the most effective parameters in prediction of solar radiation, while relative humidity has the least effect on it.