A Machine Learning Approach to Forecasting SEP Events with Solar Activities
Torres, Jesse Scott
MetadataShow full item record
Solar energetic particles (SEPs) are fast-moving events which can cause severe damage to astronauts and their equipment, and can disrupt communications on Earth. There are no clear patterns in solar activities which indicate whether an SEP is about to occur, making physics-based methods inaccurate for SEP forecasting. Therefore, in order to provide an advance warning so that astronauts are able to get to safety, we apply neural networks to the problems of forecasting SEP occurrence and intensity. Our algorithm for predicting SEP occurrence uses a combination of observed CME properties and derived features and achieves a TSS of 0.846 and an F1 score of 0.277. To forecast SEP intensity, we use electron and proton flux time series data. The data is separated into intensity ranges, which are used to train separate models. The model can be selected either through manual thresholds or another model to predict the intensity range. Doing so, we are able to accurately forecast proton flux near the onset of each event with a small lag.