Estimation of Hourly Solar Radiation in Australia Using Machine Learning Models

The prediction of global hourly solar radiation is of great significance for the development of solar energy resources in energy conversion countries. Based on this view, this study used RF and SVM models to simulate solar radiation data distributed at three weather stations in Australia based on different input combinations. In the training of the model, the meteorological data included in the parameter combination selected in this study include average temperature, relative humidity, extraterrestrial radiation and precipitation. The results showed that the simulation accuracy of SVM model was better than that of RF model, and the average RMSE was 0.61 and 0.68 MJ m−2 h−1, respectively. The model with Z2 input was significantly better than the model with Z1 input. The RMSE was reduced by 25% and the R2 was increased by 10%. Temperature had an important influence on solar radiation.