Evaluation of support vector machine for estimation of solar radiation from measured meteorological variables

Abstract

Solar radiation is an essential and important variable to many models. However, it is measured at a very limited number of meteorological stations in the world. Developing method for accurate estimation of solar radiation from measured meteorological variables has been a focus and challenging task. This paper presents the method of solar radiation estimation using support vector machine (SVM). The main objective of this work is to examine the feasibility of SVM and explore its potential in solar radiation estimation. A total of 20 SVM models using different combinations of sunshine ratio, maximum and minimum air temperature, relative humidity, and atmospheric water vapor pressure as input attributes are explored using meteorological data at 15 stations in China. These models significantly outperform the empirical models with an average 14 % higher accuracy. When sunshine duration data are available, model SVM2 using sunshine ratio and air temperature range is proposed. It significantly outperforms the empirical models with an average 26 % higher accuracy. When sunshine duration data are not available, model SVM19 using maximum temperature, minimum temperature and atmospheric water vapor pressure is proposed. It significantly outperforms the temperature-based empirical models with an average of 18 % higher accuracy. The remarkable improvement indicates that the SVM method would be a promising alternative over traditional approaches for estimation of solar radiation at any locations.