Research On The Short-term Prediction Model Of Solar Proton Events

Solar proton events may bring hazard and danger to the satellites and spaceships that in the orbits near the earth, and may cause damage to the sensitive equipments on board the satellites. Thus precise prediction of the solar proton events especially short-term prediction is very important to assure the security of space task and coordinate the use of sensitive equipments. Based upon the existing study of physical mechanism, statistics character , correlative analysis and the prediction methods towards solar proton events, in this paper, we proposed to make use of the BP neutral network to create a new short-term proton events prediction model. We collected the proton events data between 1986 and 2006. Taking the traditional parameters into consideration, we proposed and studied new parameters. In addition, we studied the CME parameter and magnetic evolvement parameter, analysis the potential use of the parameters. Finally we chose these ten parameters as the inputs of our model, including sun spot number , area of active region , soft x ray burst level , solar longitude, radio burst type, 10.7cm radio flux, low energy particle flux, magnetic type, soft x ray bursts frequency.We designed a three-layer BP network base on the neutral network theory, and applied the supervised training method to assure the correct convergence of the network. We selected 50 samples as training samples, 20 samples as supervising samples, 72 samples as testing samples. 88% of the proton events samples are successfully predicted by the model, and 35% of the non-event samples are predicted as events. On whole, the model demonstrated a good performance to predict solar proton events and proved the usefulness of the new parameters we propose.Then, we modified the designed network and changed the inside parameters of the model, we chose the 0.5-4A and 1-8A band soft x ray data, abstracted the peak flux, peak integral flux and background flux as inputs, to simulate the peak flux of those high flux solar proton events. As to the 16 testing samples, 12 of them, the simulative error were controlled in one order, which was acceptable.