Note: This content is accessible to all versions of every browser. However, this browser does not seem to support current Web standards, preventing the display of our site's design details.


Learning Freeway Traffic Dynamics


M. Keller

Semester Thesis, FS16 (10513)

In this semester project we investigate how Gaussian Processes (GPs) can be used to represent and learn the traffic dynamics of a freeway. The proposed approach is data-driven and builds on the popular macroscopic Cell Transmission Model where we also model process noise and uncertainty. There are two main challenges: First, process noise in traffic dynamics seems to be poorly representable by stationary kernels and second, the abundance of data makes standard GP learning intractable. We present two approaches for the representation and estimation of non-stationary process noise in GPs: learning parametric functions by maximizing the data likelihood with a gradient-descent method, and learning process noise directly from data with an approximative EM algorithm. To keep computations tractable, we use a combination of the Hierarchical Mixture of Experts model (HGP) and reduce the data set size based on data density. We further introduce a simple approach for the interpolation of traffic data between sensors. Our work is the first step towards a robust control system for ramp metering based on reinforcement learning with Gaussian Processes.

Supervisors: Chithrupa Ramesh, Marius Schmitt, John Lygeros


Type of Publication:

(13)Semester/Bachelor Thesis

File Download:

Request a copy of this publication.
(Uses JavaScript)
% Autogenerated BibTeX entry
@PhdThesis { Xxx:2016:IFA_5445
Permanent link