Research On Modeling Of The Global Evolution Of Energetic Electron Precipitation Based On Deep Learning Technology

The Earth’s magnetosphere is one of the most important regions in solar-terrestrial physics.The magnetosphere,formed by the deflection of the solar wind around the Earth’s magnetic field,creates a comet-like region known as the magnetosphere.It serves as a shield that blocks solar wind and other particles from entering the Earth’s atmosphere,protecting the planet from harm caused by solar wind.Within the Earth’s inner magnetosphere,there is a region filled with high-density cold plasma known as the plasmasphere.The plasmasphere includes energetic particle populations that are important components of the radiation belts and ring current.As a high-density cold plasma region within the Earth’s inner magnetosphere,the plasmasphere is primarily composed of electrons and protons diffusing and migrating from the lower ionosphere and partially overlaps with the radiation belt region.Therefore,variations in the background conditions of the plasmasphere play a crucial role in influencing the dynamics of the Earth’s radiation belts.It is one of the factors that determines the efficiency of wave-particle interactions,which in turn affect particle precipitation into the atmosphere.Therefore,global modeling of the plasmasphere and energetic electron precipitation can provide valuable insights into the relationship between the plasmasphere and energetic electron precipitation.The work presented in this paper is of great significance for studying the dynamics of the magnetosphere,ionosphere,and plasmasphere.The paper focuses on the following two main objectives:First,in this paper,a global model of energetic electron precipitation flux is constructed using deep learning techniques,utilizing energetic electron precipitation flux data from POES satellites and geomagnetic parameter data from the omni database.During model training,the five-fold cross-validation between different satellite data is employed to ensure model stability and generalization ability.Among the different models compared,a deep neural network-based precipitation flux model is ultimately selected.After validating the model output against satellite measurements,the model is applied to a magnetic storm period for reconstructing and analyzing the global dynamic evolution of >30 ke V and >100 ke V electron precipitation fluxes during July15-20,2017.The "magnetosphere-ionosphere-atmosphere" system is complex and dynamically changing,and the physical processes of energetic electron precipitation play an important role.Our reconstruction results indicate that the deep neural network model for energetic electron precipitation flux(EPFN)can accurately quantify the dynamics of global energetic electron precipitation.Second,in this paper,global modeling of electron density is conducted using deep learning techniques with electron density data from RBSP satellites and geomagnetic parameter data from the omni database.After comparing models based on onedimensional neural networks and long short-term memory networks,a global electron density model based on long short-term memory networks and attention mechanisms is developed.To validate the effectiveness of the model in practical applications,it is applied to the magnetic storm on July 15,2017,to study the global dynamic changes in electron density.Through analysis of the global dynamic evolution of electron density during the storm period,the model successfully reproduces plasma erosion and the formation and evolution of plumes.This result demonstrates that the model can be applied to the study of plasmasphere dynamics,not only providing new ideas and methods for the study of the plasmasphere but also having important implications for predicting and responding to space weather impacts.