The traditional methods used for distribution network planning are challenged because of the access of the stochastic distributed generation. A new distribution network planning model is proposed by converting the distribution network planning with distributed generation into solving the Steiner tree problem. Moreover, in order to solve the problem that the time of calculation multiplied along with the increase in the amount of Steiner nodes, an improved particle swarm optimization algorithm is proposed to solve the Steiner tree Non-deterministic Polynomial problem. In this algorithm the problem of premature convergence is solved by applying a small disturbance to the global optimum and the searching results are more reliable and accurate by improve the search ability for global optimum. Typical results of the proposed algorithm and the traditional algorithm are further illustrated and compared through an actual 54 nodes system to prove the validity of the methods in this paper.