At present, a prototype of Convection Permitting Ensemble Prediction System of North China is developed by Beijing Institute of Urban Meteorology, China Meteorological Administration. In order to construct initial condition perturbation for this ensemble prediction system, an ensemble data assimilation (EDA) initial condition perturbation method is developed based on observation perturbation and background state from global ensemble using 3-dimension variation data assimilation (3D-Var). The batch experiments of EDA method in the Convection Permitting Ensemble Prediction System of North China are carried out and compared with another initial condition perturbation method of dynamical downscaling. The results show that the construction of observation perturbation is scientific, and can produce observation perturbation of normal distribution with the magnitude in accordance with observation error. For the initial field, the members of dynamic downscaling are sufficiently dispersive, and the EDA with observation will constrain the dispersion of members, while the EDA with perturbed observation can maintain the ensemble dispersion, and the generated perturbation can represent the uncertainty of both the background field and the observation. From the batch test results, the EDA can significantly reduce the forecast error of the short range compared with the dynamic downscaling, with the ensemble spread slightly decreased. The probability score also shows some improvements from EDA. The results of precipitation forecast show that the EDA method can significantly improve the precipitation forecast that it can provide more accurate magnitude and time period of local precipitation. The statistical score of precipitation forecast also show that EDA method has better probability forecast skill for light rain, moderate rain and heavy rain. unperturbed