不确定需求下铁路集装箱动态定价研究  

Research on Dynamic Pricing of Railway Containers under Uncertain Demand

作　　者：唐慧敏 谭雪 张小强[1] 梁越 TANG Hui-min;TAN Xue;ZHANG Xiao-qiang;LIANG Yue(School of Transportation and Logistics,Southwest Jiaotong University,Chengdu 611756,China;Transport and Urban Planning and Design Research Institute,China Railway Eryuan Engineering Group Co.,Ltd.,Chengdu 610031,China)

机构地区：[1]西南交通大学,交通运输与物流学院,成都611756 [2]交通与城市规划设计研究院,中铁二院工程集团有限责任公司,成都610031

出　　处：《交通运输工程与信息学报》2021年第3期133-142,共10页Journal of Transportation Engineering and Information

基　　金：四川省科技厅软课题(2020JDR0127);成都市科技局软科学课题(2020-RK00-00102-ZF)。

摘　　要：针对铁路集装箱运输需求的不确定性,本文将传统线性需求模型、多元logit需求模型与基于LSTM神经网络的深度学习需求模型进行对比,以铁路承运企业利润最大为目标函数建立集装箱动态定价模型。首先,提取与集装箱运输需求相关的特征向量,采用梯度下降算法标定线性模型、多元logit模型参数和监督式学习方法训练LSTM神经网络。需求模型实验显示,与线性、多元logit模型相比,LSTM深度学习模型对集装箱运输需求的拟合精度更高。在此基础上,设计精英选择策略遗传算法结合LSTM网络的动态定价反馈机制求解集装箱动态定价模型。实例结果表明,基于LSTM网络的铁路集装箱动态定价模型能有效提升铁路货运企业收益。In this study,the traditional linear demand function and the multi-logit demand model with the deep learning demand model,based on the long short-term memory(LSTM)neural network,are compared,considering the uncertainty of the railway container transportation demand model.This study maximizes the profit of the railway carrier by establishing a dynamic container pricing model.We extract the feature vectors related to the container freight demand,use the gradient descent method to solve the linear function and logit model parameters,and train the LSTM neural network through the supervised learning method.The results show that LSTM is more accurate relative to demand than linear and the logit function.We design an e-GA combined with the dynamic pricing feedback mechanism of LSTM to solve the pricing model.Finally,calculations based on actual cases show that the dynamic pricing model of railway containers based on LSTM can effectively increase the revenue of railway freight enterprises.

关 键 词：铁路运输 动态定价 需求模型 收益管理 长短期记忆网络 

分 类 号：U294.3[交通运输工程—交通运输规划与管理] F532.5[交通运输工程—道路与铁道工程]