Lane Changing Models for Arterial Traffic

Driving behavior models for lane-changing and acceleration form an integral component of microscopic traffic simulators and determine its value in evaluation of different traffic management strategies. The state-of-art model for lane changing adopts a two-level framework: the first level involves a latent or unobserved choice of a target lane; the second level models the acceptance of adjacent gaps in the direction of the target lane. While this modeling approach has several advantages over past works, it assumes drivers to execute lane change within the same time step in which gap was found to be acceptable. In other words, under time steps typically adopted in model applications, the lane change duration is assumed to be negligibly small. However, past works report average lane change durations to the order of 5-6 seconds. Besides this practical maneuvering requirement, the assumption fails further in moderate or low density traffic conditions with ample gap sizes or low speed conditions, where lane changing maneuver can take longer than average. The work outlined in this thesis proposes an extension to the twolevel framework for lane changing models through a third level that explicitly models the lane change duration. Traffic conditions in the driver's neighborhood that are likely to influence lane change duration are accounted for in the third level. The extended model is applied to data obtained from video observations on traffic on a stretch of an arterial corridor in California. Apart from possessing distinctive features including signalized intersections and multiple access locations that result in lower average speeds, the arterial dataset used in this study represents a relatively low density scenario in terms of gap availability, thereby presenting an ideal test-bed for the proposed model extension. Since arterial datasets have not received predominant attention in literature, this work uncovers some traffic aspects not encountered in past studies. The model is estimated using a sample of the overall dataset available in the form of disaggregate vehicle trajectories. The estimated model is implemented in a microscopic traffic simulator MITSIMLab, and model validation is done using aggregated traffic data. Estimation and validation results showcase the improved modeling capabilities achieved through the proposed extension. Thesis Supervisor: Moshe E. Ben-Akiva Title: Edmund K. Turner Professor of Civil and Environmental Engineering