A Prediction Model of Relativistic Electrons at Geostationary Orbit Using the EMD‐LSTM Network and Geomagnetic Indices.

In this study, the Empirical Mode Decomposition algorithm (EMD) and the Long Short Term Memory neural network (LSTM) are combined into an EMD‐LSTM model, to predict the variation of the >2 MeV electron fluxes 1 day ahead. Input parameters include the Pc5 power, AP, AE, Kp, >0.6 MeV, and historical electron flux values, are used for predictions. All the time resolution of parameters are daily integral values. As compared the prediction results of the EMD‐LSTM model with other classical prediction models, the results show that the 1 day ahead prediction efficiency of the >2 MeV electron fluxes possesses a prediction efficiency of 0.80, and the highest prediction efficiency can reach 0.93. These results are superior to the prediction accuracy of more previous models. Using two high‐energy electron flux storm events for validation, the results indicate that the performance of the EMD‐LSTM model in the period of the high‐energy electron flux storm is also relatively good, especially for the prediction of high‐energy electron fluxes at extreme points, and the predictions are closer to actual observations. Plain Language Summary: During the main phase of a high‐energy storm, the relativistic electron fluxes level at MeV energy from the outer radiation belt will be enhanced at geosynchronous orbit. In particular, the >2 MeV electrons could penetrate the surface of satellites and accumulate on their insides. After a long period, the effect of these electrons could result in satellites being unable to operate or being damaged beyond recovery. To mitigate against this damage by accurate forecast to take protective measures, we combine the empirical mode decomposition (EMD) and long short‐term memory (LSTM) algorithms to predict >2 MeV electrons flux values. The EMD‐LSTM model results show that the model can accurately predict the rapid changes in data series and extreme data points with little time offset. Key Points: Propose a prediction model of relativistic electrons using a deep learning algorithm, the Empirical Mode Decomposition algorithm‐Long Short Term Memory neural network model, to predict the >2 MeV electron fluxesUse the ultralow frequency Pc5 power and related geomagnetic indices as input parameters to predict the >2 MeV electron fluxesThe forecast is shown to be highly accurate during case studies of storm times due to small time offset between observation and forecast values [ABSTRACT FROM AUTHOR]