Partitioning the city to tackle congestion

© istock

© istock

In his thesis, Mohammedreza Saeedmanesh proposes a model to replicate urban congestion propagation and utilize it for traffic management purposes.

To control traffic in a city optimally, the traffic lights at each crossroads should be modulated in real time, which is computationally challenging. "Efficiency is based on selectivity, focusing on defined control points," summarizes Mohammedreza Saeedmanesh of the Urban Transport Systems Laboratory (LUTS). He has just finished his thesis on the subject in which he developed a dynamic urban traffic network model in order to offer a more efficient traffic management.

Road congestion affects almost all big cities, with different topology, land-use and population distribution. It takes different forms and spreads across the network in different directions from day to day. One way of improving traffic conditions is to develop microscopic models capable of replicating these phenomena on a road scale. 

However, these microscopic models must take into account the drivers’ behaviour, sometimes unpredictable, and, ideally, allow real-time regulation. The alternative modeling approach is topartition the urban heterogeneous network into more homogeneous regions, between which flows can be regulated. In other words, to partition the city into homogeneous sub-regions according to traffic congestion level.

A dynamic model

To know the traffic situation, cities have different types of data sources: loop detectors, sentinel vehicles, cameras, GPS.... Once the homgenious subregions have been defined, the macroscopic tool for modeling congestion is the concept of Macroscopic Fundamental Diagram (MFD). Developed in particular by LUTS director Nicolas Geroliminis, it relates total vehiclefluidity, i.e. the number of vehicles kilometer travelled per hour, to density, i.e. the number of vehicles. The curve roughly forms a polynomial at the top of which the fluidity is maximum and then declines.

When a sub-region is congested, the idea is therefore to regulate traffic at the borders of this zone, typically by encouraging vehicles to leave (green lights) and limiting their entry (red lights). The interest of Mohammedreza Saeedmanesh's work is first to propose a new static method to partition a network into homogeneous and compact sub-regions, based on an original defined spatial similarity metric. The proposed method is not sensitive either to the network structure or to the parameters calibration.

The doctoral student then refined this model so as to make it dynamic, making it possible to reconfigure the clusters obtained (by merging or division) according to the evolution of traffic. Finally, he adapted the MFD model to make it operational in real life and computationally efficient on the basis of loop detector real-time measurements only.

Closer to reality

Then, this MFD-based microscopic modeling makes it possible either to regulate traffic using data-driven approaches; or to utilize model predictive and thus anticipate congestion evolution over space and time. "With the proposed modeling and control framework, we are even closer to reality," concludes the researcher. The model was successfully tested on a traffic simulator (AIMSUN). LUTS scientists are now testing it in the field.

Dynamic clustering and propagation of congestion in heterogeneously congested urban traffic networks, Saeedmanesh, Mohammadreza, Geroliminis, Nikolaos.