基于EMD与Attention-LSTM的铁路货运站短期装车量预测研究  

Short-term Prediction of Car Loading Quantities for Railway Freight Station Based on EMD and Attention-LSTM

作　　者：汪岗 马亮 陈奕霖[2,4] WANG Gang;MA Liang;CHEN Yilin(Dispatching Command Center,Baoshen Railway Group,China Energy,Baotou 014000,Inner Mongolia,China;The School of Information Science and Technology,Southwest Jiaotong University,Chengdu 611756,Sichuan,China;National Research Center of Railway Intelligence Transportation System Engineering Technology,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China;Sichuan Engineering Research Center of Train Operation Control Technology,Southwest Jiaotong University,Chengdu 611756,Sichuan,China)

机构地区：[1]国能包神铁路集团有限责任公司调度指挥中心,内蒙古包头014000 [2]西南交通大学信息科学与技术学院,四川成都611756 [3]中国铁道科学研究院集团有限公司国家铁路智能运输系统工程技术研究中心,北京100081 [4]西南交通大学四川省列车运行控制技术工程研究中心,四川成都611756

出　　处：《铁道货运》2023年第12期60-69,共10页Railway Freight Transport

基　　金：四川省自然科学基金面上项目(2022NSFSC0466);中国国家铁路集团有限公司科技研究开发计划课题(L2021X001)。

摘　　要：在铁路运输企业的日常货运工作中,提前掌握货运站未来短期装车量和变化趋势有助于调配空车和编制日常作业计划。铁路货运站装车量不仅受客户需求影响,还与车站计划、机车调度和装卸机具运用等影响因素密切相关,具有一定的复杂性和随机性。通过提出一种基于经验模态分解(EMD)和引入注意力机制(AttentionMechanism)的长短期记忆(LSTM)网络的铁路货运站短期装车量组合预测模型(EMD-Attention-LSTM),以某铁路运输企业某区域重点货运站的545d历史装车量为研究对象,选取平均绝对百分比误差、平均绝对误差和均方根误差作为评价指标,实验结果表明:EMD-Attention-LSTM模型与ARIMA、SVM和Attention-LSTM相比预测精度最高,为准确预测铁路货运站短期装车量提供了一种新途径。In the daily freight transportation operations of railway transportation enterprises,advance knowledge of future short-term loading quantities and trends at freight stations is conducive to the allocation of empty cars and the preparation of daily operational plans.The loading quantities of railway freight stations are not only influenced by customer demand but also closely related to factors such as station planning,locomotive scheduling,and the use of loading and unloading equipment,with a certain degree of complexity and randomness.This paper proposed an EMD-Attention-LSTM model that combines empirical mode decomposition(EMD)and attention mechanism with long short-term memory(LSTM)networks for predicting the combined short-term loading quantities of railway freight stations.Using the historical loading data of a key freight station in a certain region of a railway transportation enterprise for 545 days as the research object,the paper selected the mean absolute percentage error,mean absolute error,and root-meansquare error as evaluation indicators.Experimental results show that the EMD-Attention-LSTM model has the highest prediction accuracy compared to ARIMA,SVM,and Attention-LSTM,providing a new way to accurately predict the short-term loading quantities of railway freight stations.

关 键 词：铁路货运 装车量预测 经验模态分解 注意力机制 长短期记忆网络 

分 类 号：U294.13[交通运输工程—交通运输规划与管理]