Skip to main content

A Multiscale and Multimodal Modelling Approach for Green Urban Traffic Management

Periodic Reporting for period 3 - MAGnUM (A Multiscale and Multimodal Modelling Approach for Green Urban Traffic Management)

Reporting period: 2018-09-01 to 2020-02-29

The MAGnUM project aims to (i) create a consistent set of interrelated dynamic and multimodal traffic models able to capture driver behaviours at the different urban scales and (ii) apply this variety of models to design efficient and green traffic management strategies.
Traffic flow dynamics is well reproduced at a local urban scale by the kinematic wave model and its numerous extensions. Even if this model is parsimonious compared to other modelling approaches, it can hardly be applied at larger urban scales for traffic control applications. Very recently, a new modelling approach has been proposed to represent congestion dynamics at large scales. It relates the total travel production to the vehicle accumulation in a traffic network with for now a restrictive condition about network homogeneity. This approach is very promising for designing new traffic management systems but heterogeneous situations should be handled by properly connecting with the local scale to account for the effects of the local distributions and variations of the driver behaviour (demand) and the network structure (supply). Investigating these relationships and proposing a full set of consistent models representing traffic dynamics at several relevant scales (successive spatial and temporal integration) is very challenging with high potential gains for traffic control applications. This is the primary goal of MAGnUM and will be achieved by mixing analytical investigations on idealized but insightful test cases with explanatory approaches based on data gained from dynamic simulations or serious game sessions on more realistic and complex cases.
The second goal of the project concerns the design of innovative traffic management strategies at multiple urban scales. Breakthroughs will be achieved by considering multiple and competitive objectives when optimizing with a tight focus on environment issues and multi-modality.
From the first half the project the main results achieved so far are:
• Set-up of a dynamic and microscopic simulation environment corresponding to the city of Lyon with realistic demand patterns and trip purposes. It is used in our simulation game platform to study user behaviors simultaneously. Thanks to the ERC funding, this simulation game platform has been upgraded and integrated in a Cloud environment. We have already organized 6 simulation game sessions with about 30 to 50 players focusing on route choices. New sessions are going to be carried out during the second part of the project using the full Lyon city environment and focusing on different behaviors like mode choices and departure times.
• The trip-based reservoir approach has been successfully used to assess the impacts of parking search processes on the global network performance.
• Design of a simulation framework to systematically investigate the influence of path selections on network performance during dynamic network loading. We were able to show that when traffic conditions are critical (close to capacity) some route patterns may prevent the system from collapsing.
• Introduction of the concept of 3D congestion maps for traffic data analysis. It provides a unique picture of the daily pulse of congestion at a city scale. We were able to exhibit regularity in daily patterns that permits to define a small set of congestion maps as representatives for multiple days. A first application was to design a completely new travel time estimation method in real-time.
• Numerical and analytical investigations of trip-based reservoir model. In particular, we designed a very efficient numerical scheme (event-based) for the trip-based formulation of the reservoir traffic model that permits to easily simulate large urban areas but also to account for heterogeneous user preferences when considering the dynamic traffic assignment models (choice models).
• Further analysis of the trip-based approach in a multi-reservoir context permitted to define how flow exchanges should be tackled at reservoir perimeters to account for heterogeneous loadings and congestion spillbacks.
• development a new methodology to scale-up the real network into a simpler and aggregate network for calculating the regional (between reservoirs) routing strategies.
• design of a simplified but fully analytic framework to calculate user equilibrium in transportation network represented by a succession of reservoirs. A complete analysis of two parallel reservoirs mimicking a freeway and the side urban network has been finalized.
• Definition of a new multiclass and multimodal equilibrium framework with a probabilistic method to calculate the solution on large network. This last method appears to outperform existing methods based on fixe-point and successive averages.
• Investigation of innovative traffic management strategies to improve multimodal traffic conditions in urban area. For now, we have mainly focused our researches on strategies that are improving bus operations. The first one is about bus holding methods at bus stops. The second case study is the application of perimeter control along a bus instead of the usual set-up for a city center. We were able to highlight in a simulation study that perimeter control can almost achieve the same level of service for buses than a dedicated lane while being much more efficient for car traffic.
So far, the main contributions of the MAGnUM project in terms of new methodologies are:

• Network traffic modelling

Traffic flow dynamics is well reproduced at a local urban scale by the kinematic wave model and its numerous extensions. Even if this model is parsimonious compared to other modelling approaches, it can hardly be applied at larger urban scales for predicting traffic evolution. Earlier developments for large-scale networks have proposed to represent a city as a partition of multiple reservoirs where traffic dynamics can be more easily tracked by considering flow exchanges between reservoirs. However, this approach tends to aggregate too much user characteristics. In this ERC project, we have investigated a new approach for large-scale networks, which keeps the parsimony of traffic dynamics inside a reservoir while tracking all user individually inside the system. This trip-based approach appears to be much more accurate in free-flow and lightly saturated traffic. By a thorough analysis of the mathematical and physical properties of flow exchanges at the reservoir perimeter for the trip-based but also the more classical accumulation-based approaches, we were able to demonstrate that congestion spillbacks were not properly reproduced during saturation and over-saturation. We were able to designed a common framework to handle such cases. An important novelty here is the proper treatment of multiple trips with different lengths inside the same reservoir with respect to traffic dynamics in free-flow and congestion.

• Dynamic traffic assignment

With respect to dynamic traffic assignment, we have proposed the first global framework to handle route choices using macroscopic routes crossing multiple reservoirs. A challenge here was the definition of the aggregate network and to achieve a consistent scaling-up of the real network characteristics to the reservoir scale. This is true in particular for trip-lengths.

Another methodological contribution was the comparison of different choice models (utility theory, prospect theory, regret theory, bounded rationality…) at large-scale based using traffic simulation based on reservoirs. This permits to determine the effect at the network level of different rationales for users but also the important of properly considering correlations when handling uncertainties.

Finally, we have very promising preliminary results about computational time improvements that can be made when solving network equilibrium problems. Further studies are undergoing to confirm the trend.

• Data collection and analysis

To our best knowledge, the simulation game we developed is a unique data collection platform about user behaviors in a transportation network. It can obviously not replace experiments in real networks but it permits to simultaneously collect information for a large number of users, which is usually not achievable. Furthermore, as the simulation environment can be fully controlled and finely tuned, it permits to have a perfect vision of the local and global network scales. We have not yet finalized the fitting of choice models based on the first experiments (simulation game sessions) but we already noticed that because we can replicate much more diverse situations deeper analysis of the choice motivations.

A cutting-edge contribution in data analysis was the concept of 3D congestion maps. We benchmarked different clustering algorithms (k-means, DBSCAN, Ncut) to determine the best description of the daily evolution of link speeds into a few space-time domains with homogeneous speed. From daily congestion maps over the same city, we were able to cluster days with homogeneous patterns and apply consensual learning to determine a common shape per group. Our test case, i.e. the city of Amsterdam, shows that with only 4 consensual maps it was possible to describe the city traffic dynamics over 35 days.