Research On Traffic State Identification Of Expressway Stations Based On Network Toll Data

With the increasing demand for expressway travel,the contradiction between expressway service capacity and demand is gradually intensifying,and traffic congestion and safety problems are constantly emerging.The highway toll data contains rich traffic flow operation characteristics information,which can well reflect the changing characteristics of the traffic state.Therefore,it is of great significance to improve the level of road management and control by mining online toll data and extracting valuable traffic flow state characterization parameters,so as to grasp the laws of expressway traffic state changes.In this paper,starting from the mining of highway toll data,the inter-station traffic volume and travel time are used to characterize the traffic flow,and on this basis,the inter-station traffic status recognition and prediction are studied.The main research contents are as follows:(1)In order to make up for the lack of cross-section data caused by the lack of crosssection detection facilities,combined with the cross-section traffic volume of highway sections,a cross-section traffic volume estimation method was established based on this,which provided a data foundation for inter-station travel time prediction and traffic status recognition;based on LSTM The neural network constructs a cross-section traffic volume prediction model,which is verified by multi-segment instance analysis.Compared with the ARIMA prediction model,the MAE is increased by 12.58 and 74.07 at 15 min and 1h,and the MAPE is increased by 4.54% and 8.3% respectively,MRE improved by 70.92 and 593.57 respectively,the model has higher accuracy.(2)For the missing part of the travel time data between individual stations,a corresponding supplementary method is proposed to improve the integrity of the travel time data.Through the analysis of the influencing factors of the travel time between stations,the main influencing factors are traffic volume,time period and adjacent time period.Average travel time;in order to further improve the accuracy of the travel time prediction model,the average travel time of adjacent time periods,the traffic volume of the small car cross section,the traffic volume of the large car cross section and the day of the week are used as the prediction model considerations.Time prediction model;after multi-segment instance analysis and verification,the results show that the multi-feature random forest travel time prediction model is superior to the time series model in the three evaluation indicators of MAE,MSE and MAPE,and the prediction effect is better than the classic BP neural network.Has good adaptability.(3)In order to accurately describe the traffic state of the expressway,based on the principle of selecting the traffic state identification index,the traffic volume and the average travel speed between the stations are determined as the traffic state characterization parameters;The C-means model builds a recognition model for traffic conditions between stations;in order to solve the uneven distribution of samples in the actual situation,the cluster center range is limited to improve the recognition model in combination with the service level of the highway;through multi-segment instance analysis,the stations are realized Recognition and prediction of traffic status,the results show that the model has high accuracy and adaptability,and can provide some theoretical support for highway traffic control.