@phdthesis{,
    author = {Rempe, Felix},
    title = {Traffic Speed Estimation and Prediction Using Floating Car Data},
    editor = {},
    booktitle = {},
    series = {},
    journal = {},
    address = {},
    publisher = {},
    edition = {},
    year = {2019},
    isbn = {},
    volume = {},
    number =  {},
    pages = {},
    url = {},
    doi = {},
    keywords = {Traffic State Estimation, Traffic State Prediction, Floating Car Data, Traffic Congestion Analysis, Traffic Jam Warning},
    abstract = {This thesis proposes novel methods to use Floating Car Data (FCD) for applications in traffic speed estimation and prediction. Three approaches are developed and evaluated using real FCD collected by a large fleet of vehicles. The first method targets traffic speed estimation on freeways. It describes how to process raw and sparse trajectory data using empirical traffic features described in the Three- Phase theory to compute a continuous traffic speed estimate in space-time. Therefore, first the three traffic phases are reconstructed, and second, traffic velocities inside each phase domain are estimated. In an evaluation with 101 congestion patterns the method achieves higher accuracies than comparable state-of-the-art approaches. An efficient implementation using the Fourier transform as well as a high degree of flexibility and robustness contribute to its practical utilization. The second method seeks to provide short-term congestion front forecasts. Continuous speed information is analyzed for current hazardous congestion fronts. Flow data and speed information in the proximity of the fronts are fused and processed with an ana- lytical front propagation model. The results of a comparison of several variants of the method and a naive model show that one of the proposed model variants forecasts more accurately in a 10-minute horizon than all others. The third method focuses congestion in urban road networks. Using one year of FCD, a small number of subnets showing regular congestion are extracted. A statistical analysis of the congestion level inside these subnets reveals patterns of spatio-temporal conges- tion. Based on these patterns, a network-wide congestion forecast method is developed and applied. Its higher accuracy compared to typical time series forecasts indicate that these subnets serve as valuable features for prediction models to reflect the network-wide status of congestion.},
    note = {},
    
    school = {Universität der Bundeswehr München},
    
}