Distribution network planning is a multi-goal multi-phase and multi-restriction problem that involves the geographical condition of power line route， construction and maintenance expenses， power line losses， power flow and security constraint. A model is proposed to minimize the work of planner while simplifying the network planning procedure. The geographical feature is carefully considered to minimize the optimal search scope and improve the efficiency. An improved differential evolution（DE） is used also in optimal calculation. The DE is far more efficient and robust compared to PSO and GA. The DE is improved to prevent the results from becoming locally optimized by adopting the probability distribution feature and regenerating a diverse population of individuals. The application of the method in a provincial capital power network shows that the optimal speed and robustness can be improved by adopting the improved DE and geographical feature pretreatment.