摘要
为了解决需求函数未知情况下的高铁客票动态定价问题。以最大化单列车期望收益为目标构建Markov多阶段决策模型并设计DQN(Deep Q Net)强化学习框架寻找动态定价最优策略。算法以当日收益为奖励,通过神经网络来逼近所有状态-动作组合的期望最优收益。为验证算法性能,基于市场动态和旅客行为,开发高铁客运需求模拟系统并进行仿真实验。实验结果表明,智能体动态定价策略可以在不同需求水平下灵活调整价格,其性能接近理论上界并且显著优于对比策略。
This paper aims to solve the dynamic pricing problem of high-speed rail tickets under the unknown demand function.To maximize the expected return of a single train,we constructed a Markov multi-stage decision model and designed a DQN(Deep Q Net)reinforcement learning framework to find the optimal strategy for dynamic pricing.The algorithm used the day's income as the reward,and approximated the expected optimal return of all state-action combinations using a neural network.A high-speed rail passenger transport demand simulator was developed based on the market dynamics and passenger behavior for verifying the performance of the algorithm.The experimental results show that the agent dynamic pricing strategy can adjust the price flexibly under different demand levels,and its performance is close to the theoretical upper bound and better than the comparison strategy significantly.
作者
毕文杰
陈功
Bi Wenjie;Chen Gong(School of Business,Central South University,Changsha 410083,Hunan,China)
出处
《计算机应用与软件》
北大核心
2024年第4期228-235,261,共9页
Computer Applications and Software
基金
国家自然科学基金项目(71871231,91646115)。
关键词
收益管理
高铁客票定价
动态定价
动态规划
强化学习
环境模拟算法
Revenue management
High-speed rail tickets pricing
Dynamic pricing
Dynamic planning
Reinforcement learning
Environment simulation algorithm