Jaap Storm (TU/e)

Introduction and Research Overview

Kinematic wave models have been used since the 1960’s to describe traffic flows on networks as a fluid. This approach has been fruitful, but does not take into account one important aspect, namely, that traffic dynamics are stochastic. The intuition for this is that traffic flows are influenced by both physical laws, and uncontrollable variables, e.g., human behavior, the weather, and differences between vehicles. In the talk, I will present (on a high level) this background, and how we generalized this into a stochastic framework. This stochastic model can be analyzed using scaling methods, leading to a Gaussian process approximation as a powerful predictive instrument. In short, the vehicle density distribution (in time and space) can be accurately approximated by a multivariate normal distribution. We obtain similar results on the travel-time distributions of individual vehicles. The framework is very general, can handle multiple vehicle types, and networks of arbitrary topology.