基于强化学习的航班动态定价研究  

Research on Flight Dynamic Pricing Based on RResearch on Flight Dynamic Pricing Based on Reinforcement Learning einforcement Learning

作　　者：朱星辉[1] 赵谦 陈欣[2] 简露露 梁龙文 ZHU Xing-hui;ZHAO Qian;CHEN Xin;JIAN Lu-lu;LIANG Long-wen(Nanjing University of Aeronautics and Astronautics,Nanjing 211000,China;Nanjing University of Finance and Economics,Nanjing 210000,China)

机构地区：[1]南京航空航天大学,江苏南京211000 [2]南京财经大学,江苏南京210000

出　　处：《航空计算技术》2025年第2期1-5,共5页Aeronautical Computing Technique

基　　金：国家自然科学基金项目资助(52302391)。

摘　　要：旅客订购机票时具备票价提醒功能,为了减少由于旅客行为差异而产生的经济损失,在对航空公司进行航班动态定价售票研究中,增加了旅客行为的考虑。将旅客分为两类:耐心型和短视型。耐心旅客倾向于等待票价降至其心理预期,而短视旅客则根据当前价格迅速做出决策。在市场环境中模拟不同类型旅客购票行为,将航班动态定价建模为马尔可夫决策过程(MDP),应用DQN、ARS和PPO强化学习算法来解决复杂市场中的定价问题。结果显示,PPO算法在处理复杂环境时更加稳定且能获得较高收益。根据评估后的平均定价策略得出,航空公司的定价应随市场中存在的不同类型的旅客数量进行调整。With the integration of a price alert feature in airline ticket bookings,this study incorporates passenger behavior into the research on dynamic pricing for airline ticket sales to reduce economic losses caused by behavioral differences among passengers.Passengers are categorized into two types:patient and short-sighted.Patient passengers tend to wait for the price to drop to their anticipated level,while short-sighted passengers make quick decisions based on the current price.By simulating the purchasing behavior of different passenger types in a market environment,the dynamic pricing of flights is modeled as a Markov Decision Process(MDP).The DQN,ARS,and PPO reinforcement learning algorithms are applied to address pricing challenges in complex markets.The results show that the PPO algorithm is more stable and achieves higher returns when handling complex environments.Based on the evaluated average pricing strategy,it is concluded that airline pricing should be adjusted according to the proportion of different passenger types in the market.

关 键 词：强化学习算法 收益管理 旅客行为 动态定价 

分 类 号：F562[经济管理—产业经济] V35[航空宇航科学与技术—人机与环境工程]