Feature Analysis And Index Prediction Of Highway Freight Transportation Driven By Big Data

Highway freight transportation,an important part of comprehensive transportation system in China,bearing the important responsibility of short-distance freight transport and "door to door" transport,plays a decisive role in economic and social development.With the continuous improvement of transportation infrastructure and the development of highway freight transportation,transportation departments at all levels in China attach great importance to the statistical analysis of highway freight transportation.However,in the practice,statistical analysis of highway freight transportation is still using traditional method,informationalized and intelligentized statistical analysis remains to be further development.In recent years,with the gradually improvement of traffic informatization system and the continuous accumulation of traffic data,which make it possible to make statistical analysis and index prediction of highway freight transportation based on big data,intelligent statistical analysis system of highway freight transportation driven by big data is gradually established and improved.This paper mainly improved the method or carried on innovative research on data processing and statistical methods,highway freight transportation feature analysis,truck flows and freight transport volume prediction methods,driven by traffic big data,based on theoretical basis like Applied Statistics,Traffic and Transportation Engineering,Big Data and Neural Network et al.The research of this paper,which provides the basic support for the construction and improvement of the intelligent statistical analysis system of highway freight transportation,is as follows:First,for the purpose of calculating statistical indicators of highway freight transportation driven by big data,this paper designed the data mining method of freight vehicle self-weight based on expressway toll data,and proposed a statistical method of expressway freight transportation indicators based on the classification of truck type,axle number of freight vehicles and abnormal data.This paper also calculated freight transport index of ordinary highway by using method based on the statistical results of expressway freight index,the proportion of truck flows between expressway and ordinary highway and average transport distance,used to calculate the freight traffic index of ordinary highway.Secondly,in the view of analyzing problem of the highway freight transport feature,this paper,based on the statistical results of the highway freight transport in Guangdong Province,first analyzed the characteristics of the highway freight transport production and attraction in Guangdong Province,and then analyzed the highway freight transport feature in spatial and temporal dimensions.In spatial dimension,the distribution regularities of origin-destination distribution,spatial origin and spatial direction of highway freight transport is analyzed from three levels of city,county,specific area.And in temporal dimension,freight transport feature is analyzed from two levels of truck flows and freight transportation.At the same time,the analysis of highway freight transport feature in spatial dimension and temporal dimension provides modeling ideas for the construction of truck flows and freight transportation prediction model.Thirdly,aiming at predicting the truck flows,this paper proposed a prediction model of expressway’s trucks arrival volume considering the spatial-temporal correlation based on the spatial correlation and temporal correlation between the expressway entrance toll station and the target area.Based on the feedforward neural network,spatial correlation mechanism and time correlation mechanism are designed to learn the spatial and temporal correlation between input and output,and bias items are set to correct the output of each time step.Considering different quantity of entrance toll stations’ truck flows sequence as input to the prediction result,this paper designed the experiments of different number of the input sequence and analyzed the prediction results,then chose the optimal input sequence to further verify the proposed model,the result showed that the prediction effect of the proposed model is superior to existing benchmark model and frontier model.Fourthly,aiming to predict the freight transportation volume,considering the spatial correlation of the predicted area and the temporal correlation of the transport volume series,this paper embed the graph convolution unit into the long-short term memory unit,set up the spatial-temporal attention mechanism to improve the ability of the model to extract spatialtemporal features,and build a deep learning network,Long-short term graph deep learning network,which is based on LSTM and GCN,to forecast highway freight transportation.The validity and practicability of the model are verified by the experiments of the districts and counties in the Pearl River Delta.