A multistep short-term solar radiation forecasting model using fully convolutional neural networks and chaotic aquila optimization combining WRF-Solar model results

Solar energy is one of the most promising new energy sources, and making full use of it is the main way to reduce carbon emissions. The prediction of short-term solar radiation is of great significance to the stable operation of grid-connected photovoltaic power stations and the efficient conversion of solar energy. In this paper, a multistep short-term solar radiation prediction method based on the WRF-Solar model, deep fully convolution networks and a chaotic aquila optimization algorithm is proposed. First, the WRF-Solar model is used to predict solar radiation, and the results are spliced with historical satellite observations. Then, the spliced data are fed into five fully convolution networks for separate prediction, and each network has multilayer convolution networks to extract spatial features of different scales. Finally, the final solar radiation prediction is obtained using a chaotic aquila optimization algorithm and combining the results of the five networks. Experiments in Northwest China show that although the prediction performance varies from month to month, on the whole, the proposed method is better than other models, making it easier for the optimizer to jump out of the local optimal solution. The accuracy and robustness of the proposed model can better guide power grid dispatching.