| Ticket pricing refers to airlines formulating ticket price strategies and selling tickets of different ticket levels to different consumers in a timely manner to maximize flight seat revenue.It is the main way to increase ticket revenue.Existing air ticket pricing research has two shortcomings:(1)The pricing strategy is based on the known demand curve,where the curve is generated by historical data fitting or distribution assumptions,which will reduce performance due to deviations in demand forecasts.(2)The pricing strategy is static.It only triggers the adjustment rules of the ticket price based on factors such as different times and the number of remaining seats on the flight.The price cannot be adjusted in real time based on the market,which affects the overall revenue.In response to the above problems,the dynamic pricing of air tickets based on reinforcement learning is carried out,aiming to use reinforcement learning to build a real-time dynamic pricing strategy with the largest cumulative expectation.It is worth emphasizing that the dynamic pricing of air tickets is not based on big data and set prices by person.All passengers have the same air ticket price at the same time,but the price will dynamically change with factors such as sales time,competition and cooperation,and the number of remaining seats on the flight.At this time,how to construct a ticketing strategy that adapts to the actual demand environment and has the greatest profit has become the main challenge for the dynamic pricing of air tickets.In order to solve the above problems,an approach for dynamic flight pricing based on environmental model simulation is proposed.The core of which is to model the dynamic pricing of civil aviation air tickets as a Markov decision process,use the flight pricing time and the current remaining number of seats to describe the state and the fare class as action.First,the demand of each fare class is counted to build an environmental model,based on this model to produce the simulation experience of each action,and then take action and backtracking simulation experience to update the action value function.After repeatedly iterate,the learning system will obtain the dynamic pricing strategy with the highest expected return.After 104 groups and 92 groups of experiments,the rate of profit improvement on the two flights is empirically 22.16% and 14.10% over the actual flight revenue.The effect of flight pricing times is analyzed.In order to solve the problem that cannot fully consider the randomness of the environment in the modeling process,and at the same time cannot avoid the sampling error in sample collection,an approach for dynamic flight pricing based on strategy learning is proposed.Its core is no longer directly predict the demand of the fare level,but to model the dynamic pricing of tickets as an offline reinforcement learning process,only use the number of remaining seats on the flight to describing the status information.The strategy is learned by iteratively policy evaluation and policy improvement.The approach resorts to reinforcement learning to output pricing strategy with the highest expected return from historical ticket purchase data.After 717 groups and 610 groups of simulated pricing experiments,the rate of profit improvement on the two flights is empirically 30.94% and 39.96% over the existing pricing strategy,while that rate is 6.04% and 3.36% over the demand forecasting algorithm. |
