The simulation of the vehicle flow in large road networks is a
complex problem. Indeed, the traffic flow can be represented using
macroscopic models, where the traffic flow is assimilated to a fluid
going through a pipe. These models run efficiently on computers, but
they lack the information required to model satisfyingly urban
environments. The microscopic models can manage such environments,
since the vehicles and their interactions are explicitely represented.
Yet, they are not as efficient as macroscopic models, and they cannot
handle large road networks. In this talk, two approaches increasing
the running efficiency of the simulations while preserving qualitative
informations about the flow are introduced : 1) using General-Purpose
processing on Graphics Processing Unit (GPGPU) to increase the
efficiency of parallel computations using the Graphics Processing Unit
(GPU), and 2) using dynamic hybrid models to run simulations using
interacting macroscopic and microscopic representations.