Research On Prediction Methods Of Freeway Traffic Flow Driven By Spatial-Temporal Data

With the development of economy,the increasing car ownership and regional travel demand has brought great pressure to expressway traffic.In order to solve the contradiction between supply and demand of expressway traffic,while speeding up the construction of expressway infrastructure,it is of great importance and necessity to use intelligent transportation system(ITS)to comprehensively monitor and warn the traffic status of expressway,so as to improve the management and operation level and service level of expressway.Traffic volume and travel time are important parameters of ITS.In the actual application scenarios,ITS must be able to make real-time response to the traffic state changes of road network.Therefore,accurate and real-time prediction of traffic volume and travel time is the key technical problem to be solved in expressway intelligent transportation system.Expressway toll collection system can automatically collect and upload the path data of vehicles in the road network,and a large amount of historical data has been accumulated in the system,which provides data support for the analysis of vehicle traffic characteristics and traffic flow parameter prediction of the road network.The expressway network is a systematic whole,and the vehicles in the network interact with each other.Therefore,in order to improve the accuracy and reliability of the prediction,it is necessary to consider the spatio-temporal correlation characteristics of each node of the road network.Based on the massive toll data,this paper analyzes the temporal and spatial distribution of traffic flow in expressway network.On this basis,the prediction method of expressway traffic flow and travel time with spatio-temporal correlation characteristics is deeply studied.Different models are comprehensively divided from the perspectives of average error,peak prediction ability and prediction stability under different time step The analysis and evaluation provide the basis for the selection of forecasting methods for different traffic flow forecasting demands.The main research contents and innovations are as follows:1.In view of the two different types of toll data,i.e.entrance data and portal data,this paper presents the statistical logic and method of transforming the original data into the sample data of traffic volume and travel time.In order to provide reliable sample data for expressway traffic flow prediction,this paper makes statistical analysis on the original data of two different types of toll data.Through analysis,it is found that there are data exceptions in both types of data.Different recognition and processing methods are given for different abnormal data: for data redundancy,the duplicate records are deleted according to the VEHICLEPLATE,ENTIME,ENSTATIONID,EXTIME,EXSTATIONID and other related fields;For the case of data missing,the sample data is queried for null value,and interpolation method is used to fill in according to the actual demand;For the case of data outliers,the upper and lower quartile method is used to screen and eliminate outliers.2.Based on the data of high-speed entrance and exit tolls and gantry tolls in Guangdong Province,the temporal and spatial correlation characteristics of high-speed traffic flow in Guangdong Province are analyzed.Through the spatial correlation analysis of the traffic flow of macro region and micro node,it is found that the entrance flow of toll station and the average arrival mileage of entrance vehicles present the opposite distribution law.Through the analysis of the temporal correlation characteristics of traffic flow,it is found that the factors influencing the change trend of traffic flow from the time dimension include the historical distribution trend of traffic flow itself and the historical distribution law of the factors related to traffic flow space.These spatiotemporal correlation analyses provide some ideas for the input structure of the model.3.In order to improve the accuracy of the prediction,this paper selects Shuiguan Expressway of Shenzhen City,Guangdong Province as the research object.The input feature space of the model is constructed,and five commonly used models,LSTM,random forests,XGBoost,KNN and SVR,are applied to the prediction.Finally,the Bayesian regression method is used to comprehensively optimize the prediction effect of each model,and a fusion prediction model based on Bayesian regression is proposed.