Abstract: The accuracy of the planetary boundary layer (PBL) parameterization scheme directly affects the forecast of near-surface meteorological variables and the simulation of thermodynamic and kinematic structures in the lower-troposphere, both of which are very important for the forecast of strong convective weathers such as thunderstorms. However, the inherent uncertainties in the PBL schemes obviously limit the model capability for deterministic forecast. In order to improve the forecast skill of the PBL scheme in convection-permitting numerical model, the stochastic parameter perturbation (SPP) method is applied to the Mellor-Yamada-Nakanishi-Niino (MYNN) PBL parameterization scheme that is incorporated in the Weather Research and Forecasting (WRF) model. Three important uncertain parameters are perturbed and the influence on the forecast of a thunderstorm process in Beijing is explored. Considering the characteristics of the convection-permitting ensemble prediction system, the decorrelation time scale, the decorrelation spatial length scale and the standard deviation in the grid point of the SPP method are adjusted to explore the optimal patterns to perturb parameters in the MYNN scheme. Results show that the method of stochastic perturbation of parameters in the MYNN PBL scheme (SPPM) effectively improves the spread of near-surface variables and low-level variables below 700 hPa. The position and intensity prediction of the short-term heavy precipitation is also improved. The test of the three parameters shows that the ensemble spread can be significantly improved when the decorrelation time scale is set to 12 h. The skill is also slightly improved when the standard deviation of the grid point is increased to 0.20. The decorrelation spatial length scale is recommended to maintain at a default value of 700 km. The skill is reduced when it is too small (150 km). In conclusion, the SPPM method can effectively address the uncertainty of the PBL parameterization scheme and improve the forecasting skill of the convection-permitting ensemble prediction system.