References and Literature: Queue Discharge

NGSIM Task E.1-1: Core Algorithms Assessment
Queue discharge models

Overview

• Van Zuylen, H. J., and H. Taale, 2002, Dynamic and Stochastic Aspects of Queues at Signal Controlled Intersections , Proceedings of the 13th Mini-EURO Conference – Handling Uncertainty in the Analysis of Traffic and Transportation Systems, Bari, Italy.
• Smilowitz, K., C. F. Daganzo, M. J. Cassidy, and R. L. Bertini, 1999, Some Observations of Highway Traffic in Long Queues , Transportation Research Record 1679.

Start up lost time

• FHWA, 2002, CORSIM 5.1 User Manual, U.S. DOT, McLean, Virginia.
  Cohen, S., 2001, Application of Car-Following Systems to the Queue Discharge Problem at Signalized Intersections , Proceedings of the Transportation Research Board Annual Meeting, Washington, D.C.
• Li, H., and P. Prevedouros, 2001, Detailed Observations of Saturation Headways and Start-Up Lost Times , Proceedings of the 81 st Annual meeting of the Transportation Research Board.
• Bonneson, J., 1992, Study of Headway and Lost Time at Single-Point Urban Interchanges , Transportation Research Record, 1365, pp. 30-39. Bonneson, J., 1992, Modeling Queued Driver Behavior at Signalized Junctions , Transportation Research Record 1365, pp. 99-107.
• Axhausen, K. W., M. Fellendorf, and D. Hook, 1989, Zur Abhängigkeit der Zeitbedarfswerte von der Knotenpunktgeometrie (Saturation Flows as a Function of Intersection Design) , Straßenverkehrstechnik , 33 (5), pp. 170-174.
• Fellendorf, 2003
• Rioux, T., 1977, The Development of the Texas Traffic and Intersection Simulation Package , Ph.D. Dissertation, University of Texas, Austin, Texas. ;
• Lee, C. E., 1977, The Texas Model for Intersection Traffic: Development, University of Texas at Austin, Center for Highway Research, Research Report No. 184-1.
• Lin, F., and D. Cooke, 1986, Modeling of Queue Dissipation for Signal Control , Journal of Transportation Engineering, Volume 112, Number 6, pp. 593-608.

Car following regimes

• Van Aerde, et al., 1996
• FHWA, 2002
• Sha’aban, et al., 2003
• Benekohal, R., and J. Treiterer, 1988, Carsim: Car-Following Model for Simulation of Traffic in Normal and Stop-and-Go Conditions , Transportation Research Record 1194, pp. 99-111.

Other references

• Kimber, R., M. McDonald, and N. Hounsell, 1986, The Prediction of Saturation Flows for Road Junctions Controlled by Traffic Signals , TRRL Research Report RR67, Transport and Road Research Laboratory, Department of Transport, Berkshire, England.
• PTV (Planung Transport Verkehr), 2003, VISSIM 3.70 User's Manual , Germany.
• Quadstone, 2003, Paramics 4.0 User Guide , Internal Report, www.paramics-online.com/techsupport/docum.htm .
• Rioux, T., 2003, Personal communication.
• Sha'Aban, J., A. Tomlinson, B. Heydecker, and L. Bull, 2003, Adaptive Traffic Control Using Evolutionary Algorithms , web document, www.iasi.rm.cnr.it/ewgt/13conference/68_shaaban.pdf .
• Tarko, A., and M. Tracz, 2000, Uncertainty in Saturation Flow Predictions , Fourth International Symposium on Highway Capacity, Transportation Research Board, Hawaii, June 27 to July 1.
• Transportation Research Board, 2000, Highway Capacity Manual , Transportation Research Board Special Report.

Extracted text:

2.3 Queue Discharge Models

2.3.1 Overview

The concept of queue discharge is an important element in the behavior of drivers at both signalized intersections and in queues on freeways and other roads during congested periods. As microsimulation systems are frequently used to predict capacity and performance of alternative signal schemes or facilities, one key element in this prediction is the modeling of how queues evolve and dissipate (van Zuylen and Taale, 2002). Queue dissipation and growth is of primary interest typically due to oversaturated conditions (on one or more approaches to an intersection) when volume of traffic exceeds the capacity of the signal system to handle the demand. As van Zuylen and Taale observe, the standard deviation of the queue length is often twice as large as the average value. This indicates that, in practice, a simple model and a more complex model may not be distinguishable from one another if average value analysis is used alone for validation. In this research, van Zuylen is referring to approximation models rather than representations of individual driver behaviors such as used in microsimulation; however, the conclusion remains the same; accurate modeling of the stochastic nature of queue discharge is a primary component of microsimulation. For example, Smilowitz, et al. (2001) detail observations of interesting properties of queue dynamics.

In a similar way, when conditions become congested on freeways stop and go phenomena are observed in the field as queues grow and dissipate over space and time. Such features are represented in microsimulation systems as emergent features of the traffic stream (system state) rather than as specific behaviors.

2.3.2 Model Assessment

Start-Up Lost Time

Typically, queue discharge at signals begins with a start-up lost time , which is a stochastic delay time before the vehicles begin to move representing the driver’s response time to recognize that the signal has changed from red to green (FHWA, 2002; TRB, 2000). For example, CORSIM allows the specification of the start-up lost time by driver class, although this does not allow the start-up lost time of any particular driver to vary over time (FHWA, 2002). Start-up lost times (reaction times of drivers) of drivers further back in the queue are typically dealt with by the normal sluggishness of the car-following equations rather than explicitly applying an additional delay factor (Cohen, 2001). Recent studies have shown that typical values of saturation flow rates recorded in the HCM may be lower than observed in the field and left-turn movements may have different means and probability distributions than through movements (Li and Prevedouros, 2001). Recent research also indicates that variability in saturation flow rates can significantly affect LOS estimates of signal capacity (Tarko and Tracz, 2001).

Bonneson (1992) found that traffic pressure (defined as lane volume and queue length per cycle) has a negative effect on the saturation headway of the first 12 vehicles in a queue. This implies that the saturation headway of the vehicles at the front of the queue may be lower due to the knowledge of those drivers that there are many vehicles waiting behind them. A companion study also indicates that the minimum headway may occur later in the queue (ninth to 12 th vehicle) than suggested in the HCM (fifth vehicle) (Bonneson, 1992). A similar study in Germany identified the saturation flow rate process as a u-shape where both the beginning and end of the queue exhibit larger headways than the middle vehicles, due to anticipation (Axhausen, et al., 1989). This result has been verified in the field in unpublished validation study of VISSIM (Fellendorf, 2003). Several other simulations have indicated that headway distributions have been compared favorably to real world data collected at intersections (Rioux, 1977; Lee, 1977).

This finding was also verified by Li and Prevedouros (2001). This study includes an additional finding that indicates that the last vehicle in the queue tends to have a much smaller average headway due to anticipation of the possibility of being cut-off by the signal indication and, thus, being the only vehicle not to get through the signal cycle. Li and Prevedouros (2001) also found that a small percentage of vehicles had a negative start-up lost time on left-turns due to anticipation of the upcoming signal indication change by observing at the cross-street signal heads. Bonneson (1992) also found that headways for left-turn movements were variable by turn radius confirming independent findings by TRRL (Kimber, et al., 1986). Specific findings for Single-point Urban Interchanges (SPUI) intersections indicate that headways are shorter and start-up lost times longer than traditional intersections.

Some microsimulation systems do allow input of turning speed by maneuver (AIMSUN, TEXAS, among others), but it is not clear that headways vary with turn radius in the models as Bonneson indicates may be true in the field. Another issue, raised by Lin and Cooke, is that not all drivers in the first position of the queue stop exactly at the stop bar. Some drivers encroach into the crosswalk, and others stop up to 10 to 15 feet before (Lin and Cooke, 1985). No systems are noted to model such variability of stopping location. The operational effects (e.g., delays, speeds, interaction with pedestrians) of such variability have not been reported in detail.

Car-Following Regimes

Most typically in existing simulation systems, queue discharge is considered to be covered under the general regime of car-following, where the first vehicle in the queue reacts to the signal change with the same reaction time as if the vehicle was reacting to a leading vehicle. This is true of most of the simulation systems reviewed, including AIMSUN, Paramics, VISSIM, Integration, SIGSIM, MITSIM, HUTSIM, and TEXAS. One study on Integration discusses the implementation of such a phenomena using a virtual vehicle placed at the end of each lane on a link during a red signal indication (Van Aerde, et al., 1996). Some simulations do include modifications to the normal car-following/lane changing operations for queue discharge that are notable.

NETSIM allows the first vehicle in the queue to have a start-up lost time (delayed reaction to the signal change) that is different than the normal car-following reaction time (FHWA, 2002). These start-up lost time parameters are driver-type specific. In addition, NETSIM allows different queue discharge headway by driver type to be applied on a link-by-link basis.

SIGSIM uses the general Gipps equations for car-following around intersections and during queued conditions. To represent typical driver slowing near intersections, the user is allowed to input a reduced desired speed within 50 meters upstream of the stop line and five meters beyond the stop line (Sha’aban, et al., 2003). Note that this phenomenon could also be modeled in most other fully-featured commercial systems by configuring additional links with reduced speed limits.

Sha’aban, et al. (2003) indicate that using the standard Gipps equations for queue discharge results in saturation flows of 2400 vplph (vehicles per lane, per hour) – much higher than typical discharge rates observed empirically. For queue formation on freeways, the CARSIM simulation was developed to address congested conditions that INTRAS (now FRESIM) could not represent well (Benekohal and Treiterer, 1988). CARSIM included start-up lost times (above and beyond the normal reaction time delays of drivers), modified deceleration rates for following vehicles in congested conditions, and variable reaction times by driver type. In addition, CARSIM assumed that drivers with shorter reaction times will also wait less (in start-up lost time) than drivers with longer reaction times. Other than this CARSIM development, there is not much literature discussing any differences or modifications applied to model (or allowing the user to calibrate) the Start-Up Lost Time (SULT) phenomenon for uninterrupted flow facilities in major simulation systems.

2.3.3 Evaluation

Empirical evidence indicates that start-up lost times are observable at intersections, and are not simply the result of the driver’s reaction time from perception to action. Most simulation systems do not have representation of start-up lost times other than as normal driver reaction times. While average queue lengths are typically considered to be valid by existing discharge modeling at intersections, the variability of queue lengths in congested conditions should be studied further.

The CORSIM approach of allowing start-up lost times to be configured by location is a promising one, although it exposes a level of configuration (estimation and calibration) that the user does not typically have information to include, or would have to study in the field to assign adequately. The modification of SULT values by location must be correlated to site factors that the engineer or analyst can identify without extended field study. The issue with validation of an enhanced start-up lost time model is then to identify the effect of variable start-up lost time on the prediction of queue length variability by simulation systems. Weather conditions have the most effect on variability in signal discharge rates (e.g., controlled by normal car-following), but it is unknown if start-up lost times are also affected by weather. The extensibility of a location- and driver-type-based SULT model was rated highly. The NGSIM program could provide guidance on the choice of SULT distributions or providing functional relationships of influencing factors (geometrics, driver populations, etc.) to variability in SULTs. Table 2.3 presents the evaluations of queue discharge models.

Table 2.3 Evaluation of Queue Discharge Models