One-Hour Ahead Prediction of Solar Irradiance Using Support Vector Machines

Abstract

In this work, Support Vector Machines (SVM) are trained to perform one-hour ahead prediction of solar irradiance using K-step ahead prediction. Several SVMs regression model, such as Fine Gaussian, Medium Gaussian, Coarse Gaussian, Linear, Cubic, and Quadratic SVMs, are implemented and compared. Other than solar irradiance, five other meteorological features, namely average wind speed, air temperature, relative humidity, air pressure, and rain accumulation, are evaluated to determine the appropriate features for the prediction task. The result shows that the Root Mean Square Error (RMSE) of the prediction model can be reduced up to 9.09% by including either air temperature, relative humidity, air pressure, or rain accumulation as the additional feature to the regression learner. However, the RMSE is increased by 1.29% when the average wind speed is included as an additional feature. Overall, the Fine Gaussian SVM with all the meteorological features except average wind speed has the lowest RMSE of 67.08 Wm−2, which is 23.66% lower as compared to solar irradiance only Medium Gaussian SVM model.