72‐Hour Time Series Forecasting of Hourly Relativistic Electron Fluxes at Geostationary Orbit by Deep Learning

Abstract

AbstractIn this study, we forecast hourly relativistic (>2 MeV) electron fluxes at geostationary orbit for the next 72 hr using a deep learning model based on multilayer perceptron. The input data of the model are solar wind parameters (temperature, density and speed), interplanetary magnetic field (|B| and Bz), geomagnetic indices (Kp and Dst), and electron fluxes themselves. All input data are hourly averaged ones for the preceding 72 consecutive hours. We use electron flux data from Geostationary Operational Environmental Satellite (GOES)‐15 and ‐16, and perform a mapping for matching these two data. Total period of the data is from 2011 January to 2021 March (GOES‐15 data for 2011–2017 and GOES‐16 data for 2018–2021). We divide the data into training set (January–August), validation set (September), and test set (October–December) to consider the solar cycle effect. Our main results are as follows. First, our model successfully predicts hourly electron fluxes for the next 72 hr. Second, root‐mean‐square error of our model is from 0.18 (for 1 hr prediction) to 0.68 (for 72 hr prediction), and prediction efficiency is from 0.97 to 0.53, which are much better than those of the previous studies. Third, our model well predicts both diurnal variation and sudden increases of electron fluxes associated with fast solar winds and interplanetary magnetic fields. Our study implies that the deep learning model can be applied to forecasting long‐term sequential space weather events.