Accurate prediction of global solar radiation (GSR) is very important for all solar energy applications such as design of solar energy system, management of solar power plants and so on. But GSR data is not easily available in all the locations of India due to cost and other technical issues in measurement techniques. Hence it is essential to predict the solar radiation by employing solar radiation prediction models with available meteorological parameters as inputs. The input parameters include sunshine duration, temperature, wind speed, atmospheric pressure, and relative humidity and so on. This work is focused on the review of Artificial Neural Network (ANN) based solar radiation prediction. The ANN based model is applicable wherever the radiation data is unavailable. From the review results, it is concluded that ANN model gives good prediction accuracy than conventional regression models with correlation coefficient ranges from 0.97 to 0.99.
26 December 2016 in Indian Journal of Science and Technology
Due to cost and difficulty in measurement, solar radiation data are not available in some of remote locations where there is a potential solar application. In this paper, linear Angstrom and non-linear polynomial relations were used to estimate global solar radiation(GSR) on horizontal surface using sunshine duration, maximum temperature and minimum temperature during a period of 10 years from 2003 to 2012 in Chennai city. The data for the analysis was obtained from Indian Meteorological Department (IMD), Pune. The performances of both sunshine and temperature dependent models were evaluated for the estimation of GSR and a comparison assessment of the models were carried out using measured GSR in MJ/m2/day in Chennai. The result of the correlations shows that sunshine duration based model gives the best result with a coefficient of determination of R2 =0.9784. The results revealed that models using sunshine hours had the smallest values of MBE, MAPE, MABE and RMSE of 0.0516, 0.6449, 0.5855 0.6953 respectively. Hence the regression equations using sunshine duration can be used to predict the GSR of the representative regions in the absence of experimental data. The daily GSR values produced by this approach can be used in the design and estimation of the performance of solar applications.