Highlights Propose a data-driven framework to reroute some users to improve system performance. Define trips that contribute the most to system congestion based on trajectory features. Unravel system-optimum path by applying machine learning to trajectory feature data. Propose a robust simulation-based regression method to compute path marginal costs. Validate the proposed methods through case studies on a medium size real-world network.

    Abstract This work investigates network-related trajectory features to unravel trips that contribute most to system under-performance. When such trips are identified, feature analysis also permits determining the best alternatives in terms of routes to bring the system to its optimum. First, we define a combination of network-related trajectory features that helps us unravel the critical trips which contribute the most to the network under-performance, based on the literature review and a factor selection process. Second, based on supervised learning methods, we propose a two-step data-driven methodological framework to reroute a part of the users and make the system close to its optimum. The learning models are trained with trajectory features to identify which users should be selected, and which alternative routes should be assigned, thanks to the data and features obtained from two reference dynamic traffic assignment (DTA) simulations, under User-Equilibrium (UE) and System-Optimum (SO). We only focus on trajectory features that are accessible in real time, such as network features and regular travel time information, so that the methods proposed can be implemented without requiring cumbersome network monitoring and prediction. Finally, we evaluate the efficiency of the methods proposed through microscopic DTA simulations. The results show that by targeting 20% of the users according to our selection model and moving them onto paths predicted as optimal alternative paths based on our rerouting model, the total travel time (TTT) of the system is reduced by 5.9% in comparison to a UE DTA simulation. This represents 62.5% of the potential TTT reduction from UE to SO, when all the users choose their path under the SO condition.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Unravelling System Optimums by trajectory data analysis and machine learning


    Contributors:


    Publication date :

    2021-07-15




    Type of media :

    Article (Journal)


    Type of material :

    Electronic Resource


    Language :

    English