Evaluation of Neuro-Fuzzy Inference Systems (ANFIS) performance in global solar radiation estimation

Solar radiation plays an important role in surface energy balance and agroclimatic studies. In order to estimate daily global solar irradiance, Adaptive Neuro-Fuzzy Inference System (ANFIS) was used and compared with extended Angstrom Prescott model using 11-years data (2007-2017) climatic data of 6 meteorological stations across Iran namely, Mashhad, Kerman, Shiraz, Yazd, Zanjan and Bandar Abbas. The daily weather data including sunshine duration(n), Relative humidity (RH%), Dew point temperature (Td), water vapor pressure, maximum/Minimum air temperature, mean air temperature, cloudiness and soil temperature were used as inputs of ANFIS model to estimate the daily solar radiation, as the output. Performance of two models was evaluated using t-student index. According to the results, ANFIS model did a better job in estimation of solar radiation comparing to Angstrom-Prescott model as it showed closer agreement with observed data. For evaluation of each model's skill in estimating solar radiation at daily and monthly time scales, the statistical indices of RMSE, MBE, r, t were used .Finally, the obtained values of t-test statistic of radiation estimation using ANFIS model, in Kerman, Mashhad, Shiraz, Zanjan, Yazd and Bandar Abbas station in daily time scale were 4.78, 7.52, 3.25, 2.20, 93.3, and 0.26, respectively. The corresponding monthly values were 0, 0.29, 0.59, 0.44, 0.21, and 0.20 respectively. Similarly, the t-index of radiation estimations by Angstrom model compared to observations for daily time scale were also 20.16, 36.31, 28.90, 2.39, 34.91 and 49.96, respectively. For monthly series, t values were 6.74, 9.53, 4.68, 0.59, 8.38 and 9.71, respectively. According to the findings of this study, the nonlinear modeling process of ANFIS model improves the estimation of solar radiation compared to the Angstrom-Prescott method.