基于xDeepFM的铁路货物运输时间预测

Prediction of Rail Freight Time Using xDeepFM

铁路货物运输时间会影响物流交付、列车调度等,准确的铁路货物运输时间预测是合理制定运输组织方案的关键。货物列车的运营受很多复杂因素的耦合影响,而既有研究普遍缺乏对各因素特征交互的深入探索,为了探索铁路货物运输时间预测新的特征融合机制、提高整体预测效果,本文创新性地将智能推荐算法领域的xDeepFM算法引入货运时间预测问题。基于该算法的因子分解机、深度学习等思想构建了货运时间预测模型,设计了数据预处理、特征映射及参数寻优模块,利用模型能自动高效学习复杂因素的显式和隐式高维特征交互关系来提升预测效果,为解决铁路货物运输时间预测问题提供了新思路。在案例研究中,本文选取2种经典机器学习模型(LSSVM、随机森林模型)和3种新颖深度学习模型(DNN、CNN、LSTM)作为对比模型。实验结果表明:本文所建的xDeepFM模型的预测误差MSE为0.4991,MAPE为3.473%,相较于对比模型,xDeepFM模型具有更高的预测准确度,适合运营环境复杂的货物运输预测问题,能够实现较好的预测效果。

Railway freight transport time influences logistics delivery,train scheduling,and many other aspects of railway transportation.The prediction of freight transport time is crucial for transportation organization schemes.Numerous complex factors influence the operation of freight trains,and most previous studies did not investigate the characteristic interaction of each factor in detail.In this study,a new model for predicting rail freight travel time based on the xDeepFM algorithm was developed to clarify a new feature fusion mechanism for freight transport time prediction and improve the overall prediction effect.The xDeepFM algorithm is an up-to-date intelligent recommendation algorithm innovatively applied to predict the rail freight travel time.The proposed model mainly includes the factorization machine and deep learning components to accomplish data preprocessing,feature mapping,and parameter optimization.Because the proposed model can automatically and efficiently learn freight train operation complications of explicit and implicit high-dimensional feature interaction,it improves the prediction of freight transport time and is a new approach for predicting freight transport travel time.In case studies,two classical machine learning models(the least-squares support vector machine and the random forest model)and three novel deep learning models(the deep neural network,convolutional neural network,and long short-term memory)were selected as comparison models.The experimental results showed that the mean square error and mean absolute percentage error of the proposed model were 0.4991 and 3.473%,respectively.The xDeepFM model exhibits a better prediction accuracy than the comparison models.Therefore,the proposed model is suitable for freight transport prediction in complex operating environments and improves the prediction effect.