Research On Railway Freight Rate Forecast Using The Deep Learning

As the main adjustment lever of the transportation industry,freight rates affect the flow of freight volume between various modes of transportation.To improve the market competitiveness and operating efficiency of railway transportation,the railway implements a freight rate floating mechanism based on a reasonable price comparison between highways and railways.Therefore,analyzing the influencing factors of railway freight rate fluctuations and accurately predicting freight rate trends are extremely important for railway transportation companies and cargo owners.Transportation costs;on the other hand,clarifying the influencing factors of railway freight rates can help railway companies adjust freight rates reasonably and improve the competitiveness of the freight market.Considering the combined action of multiple factors,the railway freight rate fluctuation has the characteristics of high dimensionality,nonlinearity,non-stationarity.The deep learning prediction model based on LSTM has powerful functions of memory,learning and feature capture,and can well capture data features.And remember the development trend of freight rates to achieve accurate forecasting.Under this background,this paper starts with the formation mechanism and the adjustment mechanism of railway freight rates,gives the calculation formula of railway freight rate fluctuation rate,and analyzes the influencing factors of railway freight rate fluctuations.The degree of correlation of freight rate fluctuations and screening of 9 keys:total amount of shipments,warehouse price index,PMI,GDP,retail commodity price index,road freight index,road freight volume,iron ore output,and iron ore price index Then,based on the selected influencing factors,ARIMA,BP neural network,LSTM univariate prediction model and LSTM multivariate prediction model were established;finally,the iron ore vehicle transportation order data of C Railway Bureau was brought into the weekly forecast of railway freight rates.Comparative analysis with monthly forecasts.The experimental results show that the LSTM multivariate prediction model has the best prediction effect in terms of the floating prediction of railway weekly freight rates,with an accuracy rate of 99.5%,The model can not only memorize the time series characteristics of historical freight rate changes,but also fully mine the correlation characteristics between influencing factors and freight rate fluctuations,extract useful information from multiple dimensions,and perform best in the evaluation indicators MSE,RMSE,and MAE.For the floating forecast of railway monthly freight rates with a small amount of data and a larger time granularity,the ARIMA model with historical data as a single input has the best forecasting effect,with a forecasting accuracy of 95.2%.