Variable Selection Technique for Solar Global Irradiance Prediction

Sun’s energy is variable and intermittent by nature and has straight influence on output yielded by photo voltaic (PV) systems. Solar global irradiance prediction is crucial for conducting various research projects in emerging field of renewable energy sources (RESs). Selecting the most significant input variables that influence the solar irradiance prediction is important as it can reduce computational time and burden, enhance the prediction accuracy, increase the convergence speed and simplify the structure of the model. This work has been aimed at finding input variables that affect the prediction of solar global irradiance the most based on backward elimination technique integrated with Pearson correlation coefficient approach. 14 input variables—3 geographical variables, 9 meteorological and 2 calendar variables have been considered for the analysis. Feed forward neural network (FFD) technique has been trained on 25 different climatic regions of India and has been used to predict irradiance for Chandigarh, India. The proposed approach has yielded that surface pressure (SP), wind speed (WS), year (Y) are the most insignificant variables, whereas Clearness Index (CI), maximum temperature (MaT) are most influential ones.KeywordsErrorMeteorological parametersPredictionSolar irradiance