Modeling driver behavior during merge maneuvers.

The major objective of this study is to develop empirical methodologies for modelling ramp driver acceleration-deceleration and gap acceptance behaviour during freeway merge manoeuvres. A large quantity of freeway merge data were collected from several entrance ramps including both parallel and taper type acceleration lanes capturing a wide traffic flow range to suit different analysis purposes. Comprehensive freeway merge traffic analyses were conducted using the collected data. Both graphical presentations and independence tests in contingency tables indicated that ramp vehicle merge behaviour is insignificantly related to any single traffic parameter, such as ramp vehicle approach speeds, freeway flow levels, and speed differentials as well as time or distance gaps between ramp vehicles and surrounding freeway and ramp vehicles. Combination forms of these traffic parameters were found to be better indicators for modelling freeway merge driver behaviour. Initially, ramp vehicle acceleration-deceleration behaviour models were conceptually formulated as extended forms of conventional nonlinear car-following models incorporating joint freeway and ramp vehicle effects. These sophisticated nonlinear specifications, although theoretically attractive, have been proven to be infeasible to predict dynamic ramp vehicle acceleration-deceleration rates. A multinomial probit model, using speed differentials, distance separations of ramp vehicles to corresponding freeway and ramp vehicles, distance to the acceleration lane terminus, and Markov indexes as attributes, predicted ramp driver acceleration, deceleration, or constant speed choice behaviour. The resulting acceleration or deceleration rate magnitudes were predicted by a family of exponential curves using ramp vehicle speed as an explanatory variable. Calibration results of a binary logit gap acceptance function indicated that perceived ramp driver angular velocity to a corresponding freeway lag vehicle and remaining distance to the acceleration lane end are the best gap acceptance decision criteria. (Author/publisher)