To deal with the volatility of photovoltaic, wind resources and the randomness of load in the active distribution system, the multi-state models for photovoltaic, wind generation and load are established respectively, and turn the uncertainties into the desired scene set. The model of bi-level programming of active distribution system considering grid structure and distributed generation is proposed. The lowest annual cost is taken as the objective function of the upper layer planning and the largest generating capacitySexpectation of distributed generation is taken as the objective function of the lower layer planning. Network connectivity, radial, power balance, reactive power compensation switching and other constraints are considered. The upper and lower layer of the planning model are solved by improved genetic algorithm and prime-dual interior point method. The validity and effectiveness of proposed model and algorithm are verified by the case study of a 54-bus distribution network.