Identifying the most significant input parameters for predicting global solar radiation using an ANFIS selection procedure

There are several variables that influence the global solar radiation (GSR) prediction; thus, determining the most significant parameters is an important task to achieve accurate predictions. In this paper, adaptive neuro-fuzzy inference system (ANFIS) is employed to identify the most relevant parameters for prediction of daily GSR. Three cities of Isfahan, Kerman and Tabass distributed in central and south central parts of Iran are considered as case studies. The ANFIS process for variable selection includes evaluating several combinations of input parameters for three cases with 1, 2 and 3 inputs to recognize the most relevant sets. To achieve this, nine parameters of sunshine duration (n), maximum possible sunshine duration (N), minimum, maximum and average air temperatures (Tmin, Tmax and Tavg), relative humidity (Rh), water vapor pressure (VP), sea level pressure (P) and extraterrestrial radiation (Ho) are considered. The results reveal that an optimum sets of inputs are not identical for all cities due to difference in climate conditions and solar radiation characteristics. According to the results, considering the most relevant combinations of 2 input parameters is the more appropriate option for all cities to achieve more accuracy and less complexity in predictions. The survey results emphasize the importance of appropriate selection of input parameters to predict daily GSR. Such suitable, simple and accurate prediction is profitable to properly design and evaluate the performance of solar energy systems, which subsequently leads to technical and economic benefits.