Research On Railway Freight Demand Forecasting Based On Feature Selection And Algorithm Fusion

Demand forecasting is the foundation of railway freight organization and planning.It is the prerequisites for scientifically formulating railway transportation plans to predict the demand for railway freight accurately.Meanwhile,exploring the different impact levels of tnose factors on railway freight demand can make great reference value for the future investment direction of railway construction of the country.The realization of the demand for railway freight transportation requires a certain transportation capacity as a guarantee,which is the result of a balance between market demand and its own supply.Railway freight turnover is an important statistical index that can comprehensively reflect the demand of the macroscopic environment for freight transportation and the total amount of freight work provided by the transportation department.Therefore,this paper adopts the freight turnover as an alternative index to reflect the demand of railway freight.This paper first analyzes the effects of different influencing factors on railway freight turnover by combining the literature review and data analysis from four perspectives: macroeconomics,logistics environment,policy impact,and railway supply,a multistage hybrid feature selection method was built combined with the fuzzy decision theory and machine learning methods.Secondly,a stochastic forest regression algorithm is used to build a railway freight turnover forecast model.Considering that empirical parameters are usually used in standard random forest regression algorithms,which is difficult to ensure the best predictive performance of the model.Therefore,an optimized RF model by gray wolf algorithm was proposed in this paper.Then,conductd an empirical analysis based on the Nanchang Railway Bureau’s 2008-2017 influencing factors data set,Using the feature selection algorithm constructed in this paper,14 key factors affecting railway freight turnover were screened out.The optimal feature subset was input into a random forest model with different parameters to predict the railway freight turnover.By comparing and analyzing the experimental results,it is found that the GWO-RF model has the best prediction accuracy and calculation efficiency,which verifies the feasibility and effectiveness of the model and can be used for the forecasting of railway freight demand.There are two main points of innovation in this paper: firstly,aiming at the shortcomings of the existing feature selection methods,a new multi-stage hybrid feature selection method is proposed by combining fuzzy decision theory and machine learning,and the key influencing factors of railway freight demand are obtained,so as to provide policy Suggestions for the regulation of railway freight demand.Secondly,based on the key influencing factors,a random forest prediction model optimized by grey wolf algorithm is established to accurately predict the demand of railway freight,so as to provide a basis for railway departments to make transport plans.