Research On Estimation And Prediction Of Urban Link Travel Time Based On Low-Frequency Floationg Car Data

Traffic Guidance System plays an important role in the Intelligent Transportation Systems (ITS), which is an effective way to ease traffic jams. Dealing with the traffic status information effectively and accurately is the foundation and guarantee of Traffic Guidance System implementation. The research on estimation and prediction of urban link travel time based on low-frequency floationg car data was conducted in this paper. The work was supported by the project"Research and Application of the Key Issues in Urban Intelligent Traffic Guidance System".An intelligent algorithm of map-matching and vehicle-tracing was proposed in this paper. Because of the particularity of map-matching of GPS floating car, the algorithm takes into account comprehensive factors which influence map-matching, including errors in GPS positioning data, geometrical characteristic of roads , the direction of vehicle and data continuity of vehicle. The algorithm improves the effectiveness and applicability of map-matching and vehicle-tracing magnificantly.An estimation model of urban link travel time based on low-frequency floating car data was advanced considering hybrid traffic in China. The model used map-matching and vehicle-tracing data. In order to improve correctness of source data of estimation of link travel time, an ameliorated time-distance method was put forward to estimate link travel time of single vehicle considering the influence of intersections. The weekly-similar historical data were classified into three speed patterns by Fuzzy C-means Clustering. Finally, link travel time was estimated using real-time data and center values of three speed patterns considering the special traffic behavior of taxi, which provids data for travel time prediction. In order to solve the contradiction that real-time performance and accuracy couldn't be obtained simultaneous in practical prediction, a forecasting model based on support vector machine and considering time-space properties of urban links was introduced in the paper. Regardless of space-time effects, the traditional support vector machine prediction models only analyse the time series themselves. Via transforming the training samples, accuracy and real-time performance have been improved simultaneously.Finally, A Traffic Guidance System with mixed B/S and C/S constructure was developed , which included three subsystems. The validity of the theories and methods was demonstrated by application of the system.