基于强化学习的平行航班动态定价

Dynamic Pricing of Parallel Flights Based on the Reinforcement Learning

由于平行航班之间的竞争越来越激烈,为提高航空公司收益,对机票销售系统中的航班和旅客分别建模。将航班的动态定价问题建模成马尔可夫博弈过程,对混合类型旅客建立Logit选择模型。利用多Agent的强化学习算法对实例进行求解,结果表明WoLF-PHC算法收敛所需迭代的次数大于Nash-Q算法,但在计算速度上WoLF-PHC算法优势明显,且具有较强的适应能力。此外,航空机票的定价策略与其他易逝品有所不同,整体呈现上升趋势。而旅客环境参数的变化,也会影响定价策略。基于WoLF-PHC算法得到的定价策略对于收益提升具有积极作用。

The competition between parallel flights is becoming increasingly fierce.In this study,to improve the airline’s revenue,the flights and the passengers were separately modeled in the ticket sale system.The problem of dynamic pricing of flights was modeled as Markov game,and the Logit choice model was used to model for the mixed-type passengers.The multi-agent reinforcement learning was adopted to solve the problem in reality.The results indicated that the number of convergence for WoLF-PHC algorithm was more than that of the Nash-Q,but the WoLF-PHC algorithm had higher convergence frequency with strong adaptability.In addition,the pricing strategy of flight ticket sale process was different from that of other perishable products,which generally reflected an upward trend.The pricing strategy would also be adjusted with the modification of environment parameters of passengers.The pricing policy obtained by WoLF-PHC algorithm has positive effects on improving revenue.