Characteristic Analysis And Forecast For Highway Traffic Volume Based On Toll Data

With the sustainable development of the social economy and urbanization,the demand for public travel and freight transportation is increasing expeditiously.Due to highway capacity is not sufficient to satisfy travel demand,traffic congestion and traffic safety or the other problems will gradually emerge.Finding out the spatial-temporal feature of highway traffic flow and accurately forecasting future OD traffic volume has a great crucial to reasonably scheme establishing for traffic control and improve the operating environment of highway network.Based on toll data,this paper has analyzed the spatial-temporal feature of highway traffic flow from different time and space scale.According to the similarity of traffic flow changing trend,the scenarios of vehicle travel time are divided into four categories:working days,the day before the holiday,weekends and holidays.Based to correlation analysis of historical traffic volume,the OD traffic forecasting model is proposed in different travel time scenarios.Moreover,aim to further improve the accuracy of OD traffic forecasting in weekends,this paper proposed a OD traffic forecasting model based on vehicle travel archives.Through deeply dig into indexes that describe the vehicle travel behavior characteristics and spatial-temporal selection preference in the toll data,highway vehicle travel archives is established.Afterwards,the aggregate analysis of all the indexes in the vehicle travel archives,constructs the characteristic variable set of the OD traffic forecasting model,then the method of maximum mutual information coefficient is used to extract the optimal feature subset.The training sample set of the random forest forecast model is established to realize OD traffic forecasting of highway traffic flow.The main research results were summarized as follows:(1)This paper by studying the spatial-temporal feature of highway traffic flow found out that traffic flow change has obvious periodicity in the time scale.In the spatial scale,traffic flow has the features of unbalanced distribution.Furthermore,according to the results of traffic volume correlation analysis,the historical traffic volume has correlation with the historical traffic volume of previous n interval under different travel time scenarios.(2)According to the results of correlation analysis,the OD traffic forecasting model based on historical traffic flow is established under different travel time scenarios by using the time series data and cycle series data of historical traffic volume.From the forecast results,the model prediction error using time series data is 9.19 percent in the weekday scenario,which is 2.43 percent lower than the prediction error of the model with periodic cycle data.In the scenario the day before the holiday,the model prediction error is only 11.30%.However the prediction error of the model using time series data and cycle series data is about 20 percent in the weekend.Through a specific example analysis shows that the use of OD traffic forecasting model based on historical traffic volume has high precision under the scenario of working day and the day before the holiday,while the prediction accuracy of the model is relatively limited in the weekend.(3)Using toll data,the OD traffic forecasting model based on vehicle travel archives is established.In the weekend scenario,the RMSE,MAE and MAPE evaluation indexes were optimized.Compared to the OD traffic forecasting model based on historical traffic,the MAPE decreased by about 7%.The results of the example analysis show that it is feasible to adopt the OD traffic forecasting model based on vehicle travel archives in the weekend.