Short-time Prediction Of Solar Proton Events Based On Ensemble Learning

Solar proton event is a kind of harmful space weather phenomenon caused by solar activity,which will spread a large number of high-energy particles to near earth space,cause spacecraft failure,and cause serious harm to the health of astronauts.Therefore,the short-term prediction of solar proton events is of great significance for the disaster prevention of space activities.In recent years,great amounts of methods have been developed by many researchers for the prediction of solar proton events in the next 24 hours,but they all have a high false alarm rate.Based on these research,this paper proposes an ensmble prediction model based on multiple machine learning methods.Firstly,we built the data set with solar flare,CME and sunspot related features used as the prediction factors,and the occurrence of solar proton events used as the label.Then,the model is established based on eight widely used machine learning methods,and an ensmble prediction model is obtained by optimizing the model parameters and selecting the prediction factors.This model can predict whether solar proton events would occur in the next 24 hours,which achieves a high probability of detection of80.95%,while the false rate is only 19.05%,and the F1 value reaches 0.8095.Compared with the existing proton event prediction methods,it has made significant progress.After completing the prediction of solar proton events in 24 hours,this paper explores the prediction of solar proton events in a longer time range,and obtains the ensemble model to predict the occurrence of solar proton events in 48 hours and 72 hours.Compared with the 24-hour model,the performance of the two models has a certain gap,but they still have a higher accuracy rate and a lower false reporting rate.These models could make the prevention time of solar proton events advance one to two days.The research of this paper provides technology accumulation for building short-term prediction model of solar proton events with higher accuracy and better stability by using machine learning method.This method can be extended to a variety of space weather and phenomenon prediction fields,and also provides application ideas for future research on observation data of smile satellite and other space probes.