Short-term prediction of solar proton events by neural network method

Based on a large number of statistical results, we study the prediction of solar proton events in depth. By summarizing the relations of the proton events with the area, position, McIntosh structure, magnetic structure of the active region, and the number of solar flare bursts which happened in the active region two days before, a short-term prediction model of solar proton events is built on the lines of artificial neural network. Tests on 12 samples later than 2000 indicate a prediction accuracy of about 83%. Further test predictions are made on the proton events which occurred in Jan.–Apr. 2002. It is found that all the 6 events that occurred in this period are correctly predicted. Among them, 3 are predicted 3 days ahead, 2 events — 2 days ahead, and 1 event — 1 day ahead.