Hourly Solar Radiation Estimation Using DM and GRNN Models
Abstract
Accurate information on solar radiation intensity is essential for the development of solar energy projects. The real-time solar radiation measurement on smaller time intervals is preferred for some precise data. However, since the values are greatly affected by the time and location and the calibration and maintenance work is costly, only professional institutes will perform this type of measurement for short periods. As such, usually properly recorded radiation data are hardly available in most developing countries like Ethiopia. This leads to the requirement of using estimation models established by climatological and geographical parameters of locations. Among various models, hourly solar radiation estimation models are important for much more accurate prediction because of the detailed changes that can be recorded in a day. In this study, Data Mining (DM) and Generalized Regression Neural Network (GRNN) methods were proposed. The correlation coefficients for the models were determined using calculated sun-earth parameters and measured irradiance in Mekelle, Ethiopia. The methods of statistical analysis were used to evaluate and verify the performance of the models. The study showed that the calculated variable coefficients of the DM model and GGRNN = 0.62 + 0.57 (Gmes) can predict the nature of hourly solar radiation in the study area. The GRNN method showed better estimation compared to the DM technique. The DM technique also showed a better estimation for clear sky days. The limitations for accurate prediction of the models could be mainly due to the short-term measured average solar radiation values and the outlier input features of the training data space. Hence, further study is recommended for effective predictions, especially for cloudy days.
Keywords
Solar radiation, Estimation models, Dm, Grnn, Mekelle