Research Of Link Speed Estimation Based On Traffic GPS Data Quality Analysis

Link speed is one of the most important parameters intransportation field, it is the index to estimate the trafficsituation and the key factor which influence people trip. However,probe vehicle data is the foundation of link speed estimation, butthe existent traffic data has a wide range of sources, the varioustypes, the various forms of expression and huge amount of information,result in many GPS data cannot be treated as reliable basis of datamining in ITS. There are several main reasons cause data qualityissue: GPS data mainly comes from all kinds of sensors, videomonitoring, and vehicle GPS system. Data collection and transmissionare susceptible to signal propagation, vehicle terminal receivingequipment, weather and other factors, result in some quality problemsuch as the data collected is invalid, lost, have redundant, mistake,time points drift. In there has been of the research, link speed atsome point is got through calculate arithmetic mean value of manyvehicles speed, this bring about a high hardness requirements to thequantity of vehicle. When the quantity of vehicles fails to meetspecimen requirement, the data will have big error, finally cause theincorrect estimation of link speed.To solve the vehicle data quality problem, the paper studies theestimation of link speed based on data quality. Extract some roadsegments which have more GPS data in Guangdong Provincial Expresswaysystem, filter and fuse these GPS data, and put forward three index system about the estimation:1, integrality: the vehicle data at somepoint should be as much as possible.2, stability: the quantity ofvehicle data should be steady.3, consistency: the vehicle data shouldmatch with traffic capability of link. We firstly create a dataquality appraisal system through fusing the three indexes， andcalculate the comprehensive quality value of every probe vehicle,then verify the appraisal system, obtain the mass distribution. Thenuse the comprehensive quality value as weight of link speedestimation, finally estimate the link speed based on exponentialsmoothing. Numerical experiments and comparing with other algorithmshow the high accuracy of link speed estimation.