Real-time short-term traffic state prediction is essential for traffic management and is often conducted, based on historical data, at different traffic detection locations. However, the spatial and temporal dependencies of different detectors and different traffic state indicators need to be considered simultaneously, which makes short-term traffic prediction challenging. In this paper, we propose a deep learning based approach to simultaneously predict multi-step ahead traffic flow and speed for multiple detectors on a freeway. Based on the data collected by multiple loop detectors on the freeway, this paper transforms the time-space tensor into an image. The convolutional neural network (CNN) is employed to handle the multi-input multi-output problem. Compared to the support vector regression (SVR) and historical average (HA) models, the CNN-based model performs higher prediction accuracy and has a better performance in multistep traffic prediction. The proposed method can predict traffic speed and flow of several future time intervals within one prediction step, which saves a lot of prediction time.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Short-Term Forecasting of Traffic Flow and Speed: A Deep Learning Approach


    Beteiligte:

    Kongress:

    18th COTA International Conference of Transportation Professionals ; 2018 ; Beijing, China


    Erschienen in:

    CICTP 2018 ; 2186-2196


    Erscheinungsdatum :

    02.07.2018




    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch