一种基于深度强化学习的酒店收益管理模型与方法  

Hotel Revenue Management Method Based on Deep Reinforcement Learning

作　　者：刘显峰 于忠清 LIU Xianfeng;YU Zhongqing(College of Computer Science&Technology,Qingdao University,Qingdao 266101,China)

机构地区：[1]青岛大学计算机科学技术学院,山东青岛266101

出　　处：《青岛大学学报（工程技术版）》2022年第2期47-54,共8页Journal of Qingdao University(Engineering & Technology Edition)

基　　金：山东省重点研发计划项目(2019JZZY020101)。

摘　　要：针对深度强化学习方法在酒店收益管理上的应用问题,本文基于深度强化学习,构建了收益管理决策过程的模型与方法,通过对收益管理问题的马尔可夫性质进行了界定,描述了其模型和参数的统计学性质。同时,编写程序,实现基于深度强化学习的收益管理方法,并通过实验,将本文方法与某供应商采用的传统方法进行对比分析。分析结果表明,强化学习方法与人工收益管理方法相比,总收益提升了约15%,与传统收益管理系统相比,总收益提升了约5%,说明传统的收益管理方法成本较高,监督学习模型过于强调全局泛化性,而增大了对最优结果的估计方差,且计算量过大,而本文提出的方法能够更快地梯度下降到最优位置。该研究为企业在数据驱动下的精准定价和营销决策提供了理论基础。Aiming at the application of deep reinforcement learning in hotel revenue management,this paper builds a model and method of revenue management decision-making process based on deep reinforcement learning.By defining the Markov nature of the revenue management problem,the statistical nature of its model and parameters is described.At the same time,a program is written to realize a revenue management method based on deep reinforcement learning,and through experiments,the method in this article is compared with the traditional method adopted by a certain supplier.The analysis results show that the total revenue of the reinforcement learning method is increased by about 15%compared with the manual revenue management method,and the total revenue is increased by about 5%compared with the traditional revenue management system.The supervised learning model puts too much emphasis on global generalization,which increases the variance of the estimation of the optimal result,and the amount of calculation is too large.The method proposed in this paper can descend to the optimal position faster.This research provides a theoretical basis for companies to provide accurate pricing and marketing decisions driven by data.

关 键 词：深度学习 强化学习 酒店管理 收益管理 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] F719.2[自动化与计算机技术—控制科学与工程]