Date & Time: February 13 at 14am
These defence of MOLLIER Stéphane
Subject of the thesis: Two-dimensional models for large-scale traffic networks
Doctoral School : EEATS
Localisation of the Thesis Defense: GIPSA-lab – room Mont-Blanc
Thesis directed by: CANUDAS-DE-WIT Carlos and DELLE MONACHE Maria-Laura
Resume of the thesis:
Congestion in traffic networks is a common issue in big cities and has considerable economic and environmental impacts. Traffic policies and real-time network management can reduce congestion using prediction of dynamical modeling. Initially, researchers studied traffic flow on a single road and then, they extended it to a network of roads. However, large-scale networks present challenges in terms of computation time and parameters’ calibration. This led the researchers to focus on aggregated models and to look for a good balance between accuracy and practicality. One of the approaches describes traffic evolution with a continuous partial differential equation on a 2D-plane. Vehicles are represented by a two-dimensional density and their propagation is described by the flow direction. The thesis aims to develop these models and devises methods for their calibration and their validation. The contributions follow three extensions of the model. First, a simple model in two-dimensional space to describe a homogeneous network with a preferred direction of flow propagation is considered. A homogeneous network has the same speed limits and a similar concentration of roads everywhere. A method for validation using GPS probes from microsimulation is provided. Then, a space-dependent extension to describe a heterogeneous network with a preferred direction of flow propagation is presented. A heterogeneous network has different speed limits and a variable concentration of roads. Such networks are of interest because they can show how bottleneck affects traffic dynamics. Finally, the case of multiple directions of flow is considered using multiple layers of density, each layer representing a different flow direction. Due to the interaction between layers, these models are not always hyperbolic which can impact their stability