Urban medium voltage distribution network planning is a large-scale combinatorial optimization issue. This paper analyses the cable single-ring network mode which is commonly used in current urban medium voltage distribution network, and proposes a new method of planning. Use the global convergence properties of genetic algorithms to determine the load grouping, use the traveling salesman problem path optimization method to plan the specific path, and then obtain closed-loop network. On this basis, select the contact line by minimizing power loss. Practical example proves that the proposed method optimizes coding strategy, improves the rate of convergence. It can effectively solve the issue of urban medium voltage distribution network planning, and it is more instructive to actual urban medium voltage distribution network planning.