From the perspective of high-speed rail revenue management, in the case of multi-train and multi-OD, this paper considers the passengers’ travel preferences, takes the “days” in the pre-sale period of the railway as the stage division, describes the ticket sales process as a Markov decision process, and constructs a dynamic pricing model. Considering that the passenger demand is unknown in real life, it is difficult to construct an accurate prediction model, and this paper uses the Q-learning method in reinforcement learning to select the optimal pricing strategy. This approach eliminates the need to build an accurate environmental forecasting model, and takes into account past sales data, resulting in a more realistic pricing strategy. Furthermore, this paper calculates the performance of the Q-learning algorithm under different occupancy conditions, and proves the feasibility of applying the Q-learning algorithm to the field of dynamic pricing of high-speed rail tickets, and enriches the theoretical method of high-speed rail revenue management.