Predicting Solar Wind Speed at L1-point based on Convolutional Neural Network and PFSS Magnetogram

An accurate model for the solar wind speed at the L1-point is important for space weather predictions, catastrophic event warnings, and other issues concerning the solar wind magnetosphere interaction. In this work, we construct a model under the Convolutional Neural Network (CNN) framework, considering a source surface of r_SS = 2.5*R_sun for the globally open field distribution, aiming to predict solar wind speed at the L1-point. The input of our model consists of four Potential Field Source Surface (PFSS) Magnetograms at r_SS, which are 7, 6, 5, and 4 days before the target epoch. The model provides predictions with an averaged correlation coefficient (CC) of 0.50, and a root-mean-squared error (RMSE) of 94km/s. These two indicators are better than that of the Wang-Sheeley-Arge (WSA) model. Furthermore, this model is independent of input other than the magnetogram (e.g., the solar wind speed one or several 27 days before) so that it has the potential to generate a solar wind speed map that covers an extensive range of solar latitude (from ~45N° to ~45S°) and the whole solar longitude.