October 25, 2006, 11:00, Room salle LITEP C2 413 (b‚timent GC) (click here for the map)
We consider the problem of simulation-based traffic state estimation. Our traffic model is comprised of two major components: a mixed micro/macro traffic flow simulator and a behavioral model of combined route and activity location choice. The physical model moves individual agents based on a macroscopic representation of traffic flow dynamics. The behavioral model is simulated by a combination of a time variant best path algorithm and dynamic programming, yielding a behavioral pattern that minimizes a travellerís perceived cost. The problem of traffic state estimation is considered in a Bayesian setting. The behavioral model comprises the a priori information, which is combined with anonymous traffic measurements e.g. of flows or velocities. As a solution algorithm, we use an iterative procedure that repeatedly linearizes the available measurements' likelihood with respect to individual turning decisions in the traffic flow model. This results in a convenient solution update scheme which consists in an ordinary simulation run based on additively modified travel cost. An illustrative example is given and potentials for further enhancements are noted.