Summary

Introduction

Some traditional models of queue dissipation (NETSIM, VIPAS) were mainly concerned with the calibration of models in terms of queue discharge headways, while ignoring other queue dissipation characteristics. These models produce outputs which are unacceptable for certain situations, as have been shown by recent studies. There are three basic requirements that should be satisfied by a queue dissipation model:

·          Model should provide a realistic representation of the probabilistic characteristics of queue discharge headways.

·          After the green offset, vehicles will occupy a specified space in the approach lane for different lengths of time, and the model should reflect and reproduce this real-life observation.

·          Model should be sufficiently calibrateable to different conditions.

Car Following Model

The authors develop a car-following model to calculate ‘actual’ acceleration/deceleration of a following vehicle based on speeds of the two vehicles, the distance between them at any point of time and a ‘possible’ acceleration/deceleration rate (modified at various places in the equation by a positive constant K to denote a ‘desired’ acceleration/deceleration rate – K is hence an indicator of the degree of risk that a following driver is willing to take). The value of the actual acceleration rate has to be within some constraints used in NETSIM (and also in this model).

Queue Dissipation Simulation Model

A queue dissipation simulation model is then developed based on the car-following model. It updates the positions, velocities and accelerations of all the vehicles in the system per second. The input variables to this model include the vehicle and driver characteristics, desired speeds, etc.; each of which is represented by a probability distribution. The most important part of model calibration involves the calibration of K. It is assumed to be the sum of a deterministic component of K as a function of speed of following vehicle, and a random component H due to behavioral differences between the drivers.

After applying driver characteristics (reaction times) and preferences (speeds, etc.) during the calibration process, the value of K was found to have values between 0.5 and 1.8. The value of H was calibrated to be in the range of -0.5 to 0.8.

Model Sensitivity

The model’s sensitivity was tested by varying the values of E, H and the velocity of the following vehicle. E and H, as mentioned before are random and probabilistic in nature. To explore the issue of whether input probability distributions could be replaced by their mean values without invalidating the simulation model outputs, seven cases based on different input configurations were considered, and the discharge headways and dwell times were compared across them.

The simulated discharge headways and dwell times do not differ by a large amount from their average values and distributions, implying that the calibrated K value can compensate for lack of variation in driver characteristics, vehicle spacings and car lengths. This part of the study also found that, lack of variation in desired speeds will lead to shorter dwell times and higher saturation flow rates. Finally, for the case in which all drivers were assigned the same K value (H=0), the headway and dwell-time distributions were highly unrealistic.

Conclusions