Evaluation of ANN, ANFIS and fuzzy systems in estimation of solar radiation in Iran

Solar radiation is one of the most salient factors in determining the optimal locations of solar farms. It is the main input of geological, ecological, meteorological and hydrological models. In Iran, there are 63 stations which measures solar radiation; compared to the extent of the country, solar radiation monitoring network has very low densities. In the present study, in order to increase the network congestion and continuous mapping of solar radiation, synoptic meteorological stations’ data were used. Considering the high correlation between solar radiation and meteorological data (sunshine duration, maximum temperature and negatively high correlated sea pressure), such data was used to calculate solar radiation in synoptic stations by using Fuzzy Inference System (FIS), Artificial Neural Network (ANN) and Adaptive Neuro Fuzzy Inference Systems (ANFIS). The evaluation of the results was performed by RMSE, MAE and MBE to rank the methods. Our results revealed that Sugeno method accompanied by Fuzzy C-mean clustering has RMSE=28.07 w/m2 that lays the least errors amongst the others. With respect to ANN, Cub-clustering and Grid partition ANFIS, Sugeno method showed 18, 39% and 42% improvement. MAE and MBE also implied the ability of the Sugeno fuzzy method. Such a method is more flexible for modeling complex and nonlinear systems. The implementation of the methods in prediction of solar radiation revealed that Sugeno is easier and faster to executable. Estimated Solar radiation for 333 synoptic stations was interpolated by Ordinary Kriging to generate a continuous surface for the country. The generated solar radiation atlas is suitable to identify solar throw areas of our country as well as for engineering applications and energy planning. Radiation atlas showed that 32 percent of the country has solar radiation above 500w/m2 that is the amount of radiation required for solar farms.