基于GRA-ARIMA-LSTM组合模型的中欧班列回程开行数量预测  

Prediction of Return Trips of China-Europe Railway Express Based on GRAARIMA-LSTM Combination Model

作　　者：赵海文 魏海蕊 ZHAO Haiwen;WEI Hairui(Business School,University of Shanghai for Science&Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学管理学院,上海200093

出　　处：《物流技术》2024年第10期136-148,共13页Logistics Technology

基　　金：教育部人文社科项目(22YJC630153)。

摘　　要：针对中欧班列回程开行数量的预测问题,提出一种基于灰色关联分析-差分整合移动平均自回归-长短期记忆网络(GRA-ARIMA-LSTM)组合模型开行数量预测方法。首先采用GRA方法选取相关性高的影响因素作为神经网络输入,并通过ARIMA模型处理回程开行数量时间序列的历史信息以获得线性预测值及其残差序列。随后采用LSTM模型对这些残差和其他相关因素进行深入研究,预测残差序列中的非线性因子。最终通过均方根误差(RMSE)、平均绝对百分比误差(MAPE)和决定系数(R2)三个评价指标评估组合模型与两个单一模型的预测结果。研究结果表明,GRA-ARIMA-LSTM联合模型的指标为R2=0.9876,MAPE=0.0124,RMSE=0.083。该组合模型预测精度最高,误差最小,更适合于中欧班列回程开行数量数据的预测分析,不仅为中欧班列运力资源的合理调度和降低回程货运提供了理论支持,而且对于提高预测准确性和决策效率具有重要意义。With the advancement of the"Belt and Road"initiative,the China-Europe Railway Express has not only strengthened the economic ties between China and the countries along the route but also promoted cultural and technological exchanges.By the end of 2023,the China-Europe Railway Express service has covered 217 cities in 25 European countries,having operated 17,523 runs and shipped 1.9 million TEUs.However,there is significant imbalance in the operation of outbound and return trains,which adversely affects the reduction of operating costs and the improvement of efficiency for the China-Europe Railway Express.Therefore,accurately predicting the number of return trains is crucial for assessing the future development of the railway transportation market and further adjusting and optimizing the balance of outbound and return trains.This paper addresses the prediction of the number of return trains for the China-Europe Railway Express by proposing a prediction method based on a Grey Relational Analysis-Autoregressive Integrated Moving Average-Long Short-Term Memory Network(GRA-ARIMA-LSTM)combination model.First,the GRA method is used to select highly correlated influencing factors as inputs for the neural network,and the ARIMA model processes the historical information of the time series data of the return train number to obtain linear predictions and their residual sequences.Subsequently,the LSTM model is employed to conduct in-depth study of these residuals and other related factors to predict the nonlinear factors in the residual sequence.Finally,using three metrics:Root Mean Square Error(RMSE),Mean Absolute Percentage Error(MAPE),and the Coefficient of Determination(R2),the combination model's prediction results are compared with those of two single models.The research results indicate that the GRA-ARIMA-LSTM combination model has metrics of R2=0.9876,MAPE=0.0124,and RMSE=0.083,showing that the combination model achieves the highest prediction accuracy and the smallest error,making it more suitable for the predicting

关 键 词：中欧班列 回程开行数量 长短期记忆网络 预测模型 平均绝对百分比误差 

分 类 号：F532.4[经济管理—产业经济] F224