Potential of Machine Learning Based Support Vector Regression for Solar Radiation Prediction

Measurements of the solar radiation quantities profoundly affect on the energy output ratios. A decline in solar radiation measurements in many countries, which is due to reasons high cost, difficulty of measurement that necessitated developing different methods to estimate the proportion of solar radiation. Many empirical models have been developed using special variables and coefficients, such as Angstrom and Prescott models. The development of machine-learning algorithms makes these algorithms as a possible application instead of the empirical models to decrease the error rate and obtaining better results. In this paper, radial basis function is applied as the kernel function of support vector regression (SVR) method to calculate the amount of monthly average daily of the global solar radiation in four sites in Egypt. Five variables used as input (sunshine duration, air temperature, relative humidity, solar declination angle and extraterrestrial solar radiation). The experimental results have a good estimation in all locations according to root mean square error, however, this study proved that SVR models can be as an efficient machine-learning technique with a higher accuracy.