In recent years, with the application of advanced intelligent technology, the level of automation in station train operations has greatly improved. However, due to the complexity and high uncertainty of shunting operations, the challenge of the implementation of automation in shunting operations is attracting more and more attention. Moreover, the establishment of shunting routes lacks the support of advanced technology to enhance efficiency and accuracy. In order to improve the efficiency of shunting route search, which has a significant impact on route establishment, a shunting route search algorithm based on reinforcement learning, and more specifically, Q-learning is proposed. By analyzing the characteristics of shunting operations, two more specific scenarios that should be considered in shunting route search are proposed: time-conflict scenario and spatial-disturbance scenario. Experiments under ideal scenario and spatial-disturbance scenario are conducted to evaluate the performance of the Q-Iearning approach. The simulation results show the efficiency and effectiveness of the Q-learning method for shunting route search under both scenarios.