Solar Wind Speed Prediction With Two-Dimensional Attention Mechanism

As more and more high-technical systems are exposed to the space environment, extreme space weather becomes a great threat to human society. In the solar system, space weather is influenced by the solar wind, such that reliable prediction of solar wind conditions in the near-Earth environment effectively reduces the impact of space weather on human society. Solar wind speed prediction is improved by making full use of OMNI data measured at Lagrangian Point 1 (L1) by the National Aeronautics and Space Administration (NASA) and image data observed by the Solar Dynamics Observatory (SDO) satellite in this work. Specifically, we propose a model based on the "two-dimensional attention mechanism" (TDAM) to predict solar wind speed. In this study, we first analyze and preprocess data from 2011 to 2017. Second, considering the characteristics of time series data, we adopt the gated recurrent units (GRU) model which can deal with long-term dependence as the prediction part of our model. Third, we design a TDAM, which enables our prediction network to focus on important parts. Three performance indices are used: root-mean-square error (RMSE), mean absolute error (MAE), and correlation coefficient (CC). By comparing TDAM with other models, we find that the TDAM model achieves the best prediction results, with RMSE of 62.8 km/s, MAE of 47.8 km/s, and CC of 0.789 24 h in advance. The experimental results show that the proposed TDAM model can improve the prediction accuracy of solar wind speed.