In view of the huge workload and long cycle of manual line selection at present, as well as the deviation among the horizontal, vertical and horizontal way of line estimation and the actual optimal path. A method for the agent to complete line selection through trial and error and learning manual route selection experience is proposed. In this method, Q-learning algorithm is applied to neural network, and deep Q-learning is used to learn circuit experience, so that the neural network can quickly reach convergence, and finally generate an optimal path with environmental adaptability. Compared with the example, it is proved that the route is practical and effective.