For medium and long term planning of distribution network,a multi-stage planning method is proposed to describe the dynamic changes of load during plan years and the large-scale distributed generation(DG)integrated into distribution network．The method can complete the sizing and siting of DG and the network expansion．The model of distribution network multi-stage planning based on the minimum life cycle cost(LCC)is presented．To solve the complex model,an improved genetic membrane algorithm(IGMA)is proposed．IGMA includes exchange recombination rule,arc cotangent-exponential self-adaptive cross transposition rule and arctangent-exponential self-adaptive mutation rewrite rule．An encoding method based on the adjacency matrix of digraph is used,which does not need to decode in the optimization process．Proposing a new method to quickly generate feasible initial solutions and revise all infeasible conditions(the isolated node,chain and the closed loop)by changing the in-degree of node．Finally,the numerical simulation on IEEE 54-bus system shows that IGMA has the fast convergence speed and the strong capability of global search as well as stability．