<p>Dynamic pricing, as a core component of Revenue Management (RM), has created significant revenue growth for the airline industry through a two-stage decision-making paradigm of demand forecasting and price optimization within the framework of discrete price points. The full implementation of the New Distribution Capability (NDC) provides policy support for the continuous dynamic pricing mechanism. However, limited by the theoretical framework of the traditional pricing paradigm, the current airline passenger market still generally adopts pricing strategies based on discrete price points. Reinforcement Learning (RL), with its model-free learning ability and online optimization features, offers new ideas for solving continuous pricing problems in complex dynamic environments. This study introduces reinforcement learning algorithms into the scenario of continuous dynamic pricing in the airline industry and explores a continuous pricing model for multiple-class cabins. Furthermore, based on discussions of the decision-making characteristics of the agent during the decision-making process and the inventory relationships among different-class tickets during the ticket sales process, a reward shaping method for balancing inventory is proposed. Experimental results using the Multinomial Logit as the choice model show that the proposed reward shaping method can significantly improve the performance of the original algorithm.</p>