Relativistic Electron Event Prediction Method Based On Support Vector Machine Geosynchronous Orbit Research

The relativistic electron enhancement event is one of the most devastating phenomena in space weather, with high occurrence rate in Earth's magnetosphere. Relativistic electrons, also known as"killer electrons", are major causes for spacecraft malfunction at geostationary orbit. Therefore, prediction of the relativistic electron enhancement event at geosynchronous orbit is crucial for the safety of the spacecraft. Furthermore, forecasting the relativistic electron enhancement event is an important part of the entire space environment prediction.Considerable studies have been done on relativistic electrons, which include statistical features from the observation, acceleration/ loss mechanisms, and various prediction models. In this paper, we utilize the Support Vector Machine (SVM) to predict the relativistic electron flux at geostationary orbit. We choose the input parameters of the model by their Mean Impact Values (MIV), including the electron flux, solar wind speed, solar wind density, Dst index on the previous day and AE index during the preceding two days. In addition, our SVM model can report both the magnitude of the electron flux (regression model) and the level of relativistic electron flux event (classification model) on the coming day, in order to improve current models which only predict the daily average of relativistic electron flux.By comparing with original data in 2002 and 2008, we show the predicting efficiency for the regression model is 0.67 and 0.71 respectively. This regression model can simulate the evolution of relativistic electron enhancement events more precisely than the neural network model, comparable to the result of radial diffusion model or REFM. The classification model can correctly categorize the level of energetic electron enhancement events at most of the time (83% and 82%), with better accuracy than the regression model, especially during active intervals. Our result demonstrates this forecasting technique based on SVM is viable and of practical values.