Research On Railway Freight Volume Forecast Based On Deep Learning

With the continuous expansion of the scale of the railway network and the rapid development of information technology,the operation and management mode of railway freight transport has gradually transformed from an extensive type to a refined type.The refined railway freight operation management mode is an inevitable choice for railway enterprises to respond to the increasingly personalized transportation service market driven by the profit goal of railway enterprises,and it is also the primary choice to respond to the call for sustainable and green development of the current society driven by the social responsibility of railways.Refined railway freight operation management is not only reflected in the refinement and standardization of operational workflows and the digital integration of information systems,but also in the adjustment of the operation of the railway macro road network to the railway junctions and railway freight stations and the transportation capacity.optimization.Therefore,it is necessary to forecast the short-term freight volume of railways,one is to provide decision support for railway traffic organization,and the other is to provide information support for the allocation of logistics infrastructure and the development of railway logistics services.Considering that the railway freight volume sequence affected by the external environment has the characteristics of multi-dimensionality,nonlinearity and sequence dependence,the traditional model is difficult to predict.The deep learning prediction model has a strong learning ability,which can better extract relevant features and fit the nonlinear characteristics of freight volume.Due to the numerous factors affecting railway freight volume,firstly,grey relational analysis was used to solve the correlation between each influencing factor and freight volume,and then the number of employees in the mining industry,the added value of the secondary industry,the total population at the end of the year,the average daily freight rate,There are 7 main influencing factors including expressway mileage,GDP and PPI;secondly,based on the selected influencing factor data and historical freight data,a feature matrix is defined,and a multivariate LSTM model,a multivariate GRU model and a multivariate CNN-LSTM combined prediction model are established;Finally,the forecast effect of the model is verified by using the freight volume data of a certain region from 2015 to 2019,and compared with the forecast results of the univariate LSTM model,the univariate GRU model,and the univariate CNN-LSTM combined prediction model.The experimental results show that because the multivariate forecasting model predicts the freight volume from multiple dimensions,it not only analyzes the time series characteristics of the historical freight volume,but also fully taps the correlation characteristics between the influencing factors and the freight volume,so the prediction effect is better than that of the univariate model.Okay.In addition,among the multivariate models,the multivariate CNN-LSTM model has the best prediction effect,and its EVS,R^2,MAE,and RMSE are 0.96,0.96,120.75,and 157.42,respectively.