Abstract:

The amount of electrical energy to be obtained from solar energy systems varies greatly depending on the solar radiation value. The design and planning of a solar energy system are possible by knowing the radiation value. Since solar radiation intensity has a highly variable structure throughout the day, it is very difficult to capture these changes using a single prediction model. In this context, in recent years, different hybrid models and approaches have been proposed by researchers to overcome the limitations of single models and increase predictive accuracy. In this study, the applicability and performance of the method were investigated by using the Decomposition-Ensemble learning approach, which is a hybrid approach to the estimation of solar radiation intensity data. In addition, it is aimed to increase the time resolution of forwarding solar radiation forecasts. In this context, 15-day solar radiation value was forecasted hourly, using an annual solar radiation data measured hourly with a pyranometer located within Afyon Kocatepe University, Solar and Wind Energy Application and Research Center. In the learning approach, Empirical Mode Decomposition (EMD) method is used for decomposition and Least Squares Support Vector Regression (LS-SVR) method is used for individual predictions. The most appropriate parameter values of LS-SVR models were determined by using a grid search algorithm and 5 fold cross-validation methods. The results obtained from the study showed that the Decomposition-Ensemble learning approach is successful in estimating solar radiation data.

Keywords: