Quantitative Prediction of High-Energy Electron Integral Flux at Geostationary Orbit Based on Deep Learning

Abstract

The deep learning method of long short-term memory (LSTM) is applied to develop a model to predict the daily >2-MeV electron integral flux 1 day ahead at geostationary orbit. The inputs to the model include geomagnetic and solar wind parameters such as Kp, Ap, Dst, solar wind speed, magnetopause subsolar distance, and the value of >2-MeV electron integral flux itself over the previous five consecutive days. The model is trained on the data from the periods 1999–2007 and 2011–2016, and the efficiency of the model is tested on the 2008–2010 period. We experiment with different input combinations and find that when the model takes daily >2-MeV electron integral flux, daily averaged magnetopause subsolar distance, and daily summed Kp index as inputs, the prediction efficiencies for 2008, 2009, and 2010 are 0.833, 0.896, and 0.911, respectively. This value reaches 0.900 for 2008, when hourly >2-MeV electron integral flux, hourly magnetopause subsolar distance, and daily summed Kp index are taken as inputs, with training on the remaining data from 19 June 2003 to 13 April 2010. The prediction efficiencies of the persistence model and the 27-order autoregressive model for the same tested time period are 0.679 and 0.743, respectively. Therefore, the model developed based on the LSTM method can improve the prediction efficiency significantly for daily >2-MeV electron integral flux 1 day ahead at geostationary orbit.