Study On The Estimation And Prediction Methods Of Link Average Travel Time Of Traffic Flow Based On On-board GPS Data

At present, traffic congestion has become a social problem concerned by nationalurban, which hinders the urban economic development and affects people’s dailytravel activities, and this situation is increasingly serious.Link average travel time of traffic flow is the most intuitive indicator reflectingthe road traffic flow status，getting real-time and accurate link average travel time oftraffic flow is the effective way to improve the traffic congestion status. The wideapplications of “GPS+GIS+Wireless communication” technology in vehiclesupervision provide the low-cost basic data for the estimation and prediction of linkaverage travel time of traffic flow based on on-board GPS data. This paper is based onon-board GPS data and studies on the related methods deeply, for the purpose offurther improving the effect of the estimation and prediction of link average traveltime of traffic flow data. The research works and progress of this paper is mainlymanifested in the following aspects.1) Study on on-board GPS data pretreatment methodThis paper excludes the on-board GPS drift data by threshold technology,identifies the on-board GPS redundant data using vehicle parking judgment method,process the on-board GPS location deviation data by map matching method andimproves the quality of the follow-up technical module of the input data.2) Study on the estimation method of link travel time of single car based onon-board GPS dataContrary to the shortcomings of location-time interpolation algorithm andspeed-time integral algorithm, this paper designs the corresponding improvedalgorithm using Newton Interpolation and Chebyshev Polynomial Fitting Modelrespectively, and designs a merge estimation method based on BP neural. Theempirical analysis shows that the proposed improved location-time interpolationalgorithm, improved speed-time integral algorithm and the merging method canimprove the estimation effect of link travel time of single car based on on-board GPSdata. 3) Study on the estimation method of link travel time of traffic flow based onon-board GPS dataFor the situation that the estimation methods of the link average travel time oftraffic flow are seldom and the effect is to be further improved, this paper proposesthe estimation method of link travel time of traffic flow based on on-board GPS databy SVM method on the basis of analyzing the influence of different sample carspecies and traffic state on the estimation of link average travel time of traffic flow.The empirical analysis shows that the proposed method can lower the estimation errorfurther compared with the literatures.4) Study on the prediction method of link average travel time of traffic flowbased on on-board GPS dataConsidering the advantage of combination forecasting in the aspect of loweringerror, according to the frequency of use and the effect of prediction, this paperchooses improved Kalman filter prediction method and genetic neural networkmethod as the basic methods, then designs a combination forecasting method based onSVM method. The empirical analysis result shows that compared with the chosenbasic methods and the existing combination forecasting methods, the proposedmethod can lower the prediction error further.The research content, research methods and research conclusions arecomplement and discovery of estimation and prediction methods of link averagetravel time of traffic flow. The results can provide information infrastructure for thedynamic decision of transportation managers, transportation travelers andtransportation researchers, and it is of important academic significance and practicalvalue for alleviating traffic congestion.