Prediction of solar radiation resources in China using the LS-SVM algorithms

Solar radiation knowledge is important for the solar energy conversion and utilization. In this work, least squares-support vector machine (LS-SVM) algorithms were applied to estimate the yearly and monthly average daily global solar radiation in China using the ordinary meteorological data and geographic parameters. The monthly climatic data from 101 radiation measurement stations were divided into one testing data sets, and two validation data sets. An efficient optimization algorithm known as the grid search are applied to tune parameters in LS-SVM model. The results indicated the superior performance and satisfactory prediction of LS-SVM model (R²=0.9832, RMSE=0.7278 MJ·m^-2·d^-1 for training data, and R²>0.948, RMSE < 1.2 MJ·m^-2·d^-1 for validation data). The work finally took the LS-SVM model to map 10-minute grid of yearly and monthly average daily global solar radiation in China using climatic data of CRU-LC2.0. The spatial and temporal distributions of the atlases are generally similar with other researches, but show more advantages on spatial resolution and continuity.