Microscopic traffic simulation has proven to be a useful tool for the design and assessment of Intelligent Transport Systems by its ability to capture the full dynamics of the time dependent phenomena. Depending on the level of detail, the transport analyst requires a local or a global view of such dynamics. Dynamic traffic assignment is one of the tools when the global level is necessary. Unfortunately the analytical tools are not yet mature to provide the answers for networks of sensible size. This seminar will discuss a heuristic procedure based on an origin-destination microscopic simulation using time sliced origin-destination matrices to account for time variations in traffic demand. Paths from origins to destinations are timely updated depending on changes on link travel times according to changes in traffic conditions. Vehicles follow those time changing paths to which they are assigned according to route choice models. A percentage of vehicles defined by the analyst can be allowed to change dynamically the route en route. The analyst also sets up the number of alternative routes to be used between each origin-destination pair. The seminar will also address two of the modeling aspects involved in this approach:
Discrete route choice models, as for example logit based models, have been traditionally used to model the way in which travels (vehicles) are assigned to the various available routes. Further that the typical problems of logit models, that can be partially overcome using modified logit models, experiments show that the driveršs sensitivity to route changing conditions shoild also take into account other factors that will be discussed in the seminar.
There are an increasing number of traffic systems based on intelligent transport applications that require real-time and short time predictions of the traffic demand patterns described by local O/D matrices. Assuming that an historical time sliced O/D matrix is available, and that real-time flow measurement are provided from a set of traffic detectors, then a neural network based procedure can be used for short-term forecasting the O/D matrix from the current time interval into the next.