Snowdon, James, Waterson, Ben and Fangohr, Hans
The evolution of driver route switching behavior in stochastically perturbed networks
At 4th International Symposium on Dynamic Traffic Assignment (DTA2012), United States.
04 - 06 Jun 2012.
Full text not available from this repository.
Traffic assignment, the process of ‘assigning’ expected vehicle origin-destination flows to discrete routes traversing a road network, has traditionally been approached as an optimization problem yielding one fixed set of ‘equilibrium’ road link flows. This approach has been criticized for oversimplifying both traveler behavior representation and the effects of stochastic variations in network conditions. Here the connection is explored between day-to-day driver learning models and correlated perturbations occurring spatially around the network on simulated ‘days’. As elsewhere in the literature, a perturbation is defined as any incident which increases travel time along the perturbed stretch of road for the same level of demand. Equipped with a plausible correlation learning mechanism, drivers are modeled inferring downstream road conditions without the need for signage provision such that strategic en-route path switching can occur with the goal of potentially avoiding delays. It is found that when drivers can undertake strategic behavior the negative impact of perturbations is lessened due to some drivers diverting upstream to avoid them, meaning that potentially delayed routes are seen as relatively more attractive when compared against the outcome when no driver is modeled as being able to switch as is captured by existing assignment techniques. Thus even on days where no perturbations occur, initial driver route choices differ when compared against traditional algorithmic assignment results.
Conference or Workshop Item
|Venue - Dates:
||4th International Symposium on Dynamic Traffic Assignment (DTA2012), United States, 2012-06-04 - 2012-06-06
||02 Nov 2012 16:39
||17 Apr 2017 16:26
|Further Information:||Google Scholar|
Actions (login required)