Based on the CMA T213 global ensemble forecast system, two different schemes of model stochastic perturbation are designed in this paper. The first scheme imposes the perturbation on the total tendency of physical process and dynamical process (STPS). The second scheme perturbates the parameterized physical process (SPPS). We separately introduce the two schemes into model to perform numerical tests. Then we evaluate the performance of ensemble forecast, analyze the characteristics of stochastic perturbation and understand the perturbation propagation mechanism. The result shows that both schemes of stochastic perturbation significantly influence the T213 model forecast variables. Perturbation values in vertical direction and horizontal direction increase along with the increase of integration time, especially in the region of middle and high latitudes. The difference between the two schemes is that the forecast variable of STPS scheme is more disturbed than SPPS scheme at the total stage of integration. The forecast variable of SPPS is more reasonable compared with STPS. The center of SPPS maximum perturbation value propagates from low latitudes to high latitudes with the increase of integration time. The total perturbation energy has the characteristics of propagating from small scale to large scale. Compared with STPS, SPPS ensemble forecast has advantages on the aspects of ensemble forecast dispersion and root mean square error of ensemble at the late stage of integration, improving the accuracy in forecasting the geopotential height field and wind field, and the techniques in forecasting precipitation. Finally， this paper proves that SPPS scheme can more reasonably reflect the uncertainty of the model physical process and it enhances the forecast ability of the ensemble forecast system at a certain degree.