Error coming from treatment of physics processes is one of major reasons of numerical weather prediction errors, ensemble forecast is an effective way to diminish the errors. In this paper, 10 members' ensemble forecast scheme consists of five kinds of resolvable scale precipitation processes, two kinds of implicit cumulus parameterization schemes, two kinds of atmospheric radiation parameterization and one kind of terrain resolution, which was run to predict 48 hours' weather of a typical northeast cold vortex. The results show the results of ensemble forecast are better than that of each member's, the simple average of ensemble members' prediction results can improve forecast accuracy; at the same time, prediction accuracy of little probability case may be improved by analyzing moving route chart of weather system, characterized isoline〖KG*2〗chart and precipitation probability forecast chart.