In Ref.[1] a simple Solar Proton Events (SPE) short-term forecast model is built with three solar photospheric magnetic physical properties (the maximum horizontal gradient of longitudinal magnetic field |▽
hB
z |
m, the length of neutral line with strong gradients L, and the number of singular points η), which suggested magnetic physical properties are effective in forecasting SPE. The traditional SPE forecasting models, which have not used magnetic physical properties as input parameters, often have low Probably of Detections (POD) or high False Alarm Rates (FAR) for SPE. This paper built a more effective SPE short-term forecasting model with the traditional SPE forecasting parameters and magnetic physical properties by BP neural network. Model A uses only the traditional SPE forecasting parameters and Model B uses the traditional SPE forecasting parameters as well as those three magnetic physical parameters. In testing 973 samples during 2002--2003, Model A and B have the same POD while Model B has a lower FAR than Model A. It further testified that magnetic physical properties are effective for forecasting SPE. And the most important thing is that it will largely improve our practical ability in forecasting SPE.