Intelligence in traffic simulation model: Modeling congested network

Abstract

Traffic simulation tools are becoming more popular as complexity and intelligence are growing in transportation systems. The need for more accurate and intelligent traffic modeling is increasing rapidly as transportation systems are having more congestion problems. Although traffic simulation models have been continuously updated to represent various traffic conditions more realistically, most simulation models still have limitations in overcapacity congested traffic conditions. In traditional traffic simulation models, when there is no available space due to traffic congestion, additional traffic demand may never be allowed to enter the network. The objective of this paper is to investigate one possible method to address the issue of unserved vehicles in overcapacity congested traffic conditions using the VISSIM trip chain. The VISSIM trip chain is used for this analysis as it has the advantage of holding a vehicle without eliminating it when traffic congestion prevents its entrance onto a network. This will allow the vehicle to enter when an acceptable gap becomes available on the entry link. To demonstrate the difference between the simulation using standard traffic input and the trip chain method, a sample congested traffic network is built and congested traffic scenarios are created. Also, simulations with different minimum space headway parameters in the priority rules are analyzed to model congested traffic conditions more realistically. This will provide the insight about the sensitivity of the model to this parameter. Based on the analysis conducted it is concluded that, with appropriate calibrations, the trip chain feature in VISSIM has the potentials to be useful in modeling overcapacity congested traffic conditions more realistically.

Keywords

Traffic simulation environments traffic congestion modeling intelligence of traffic simulation simulation analysis network simulation

1 Introduction

Traffic analysts are increasingly utilizing traffic simulation as a problem solving tool for transportation system analysis [1 –7]. With the development of real-time sensors and communication technologies, traffic simulation can provide users real-time traffic information [8 –14] and future traffic condition predictions [15 –19]. The goal of the traffic analyst is to better describe and illustrate real world traffic and to develop efficient traffic operation plans [1 –10]. However, most traffic simulation models have limitations in modeling overcapacity demand conditions [2 –4]. For example, when traffic in a congested area “fills” the available roadway space most simulation models do not allow additional vehicles to enter the system, even though unserved demand exists at that time. Unfortunately, demand not allowed into a system during one time period is typically not retained to be released into the system when capacity becomes available. The simulation essentially ignores the unserved demand, in some cases simply creating error messages stating all vehicles are not generated as the implemented discharge rate is smaller than the input flow. When this occurs, the total number of vehicles simulated does not match the vehicle demand input by the analyst and realistic operational statistics are not obtained.

The intent of this paper is to explore one possible means to address the issue of “lost” entering vehicles during congested conditions. The simulation tool used for this effort is VISSIM and the particular feature utilized is the ability to create trip chains, which allow the user to specify departure times and origin/destination points for individual vehicles. The advantage of using trip chains is that when congestion on the entry link hinders the generation of an additional vehicle, the vehicle is not lost, it is instead held until an acceptable gap exists on the entry link. Therefore, as long as the simulation model is run for a sufficient time period (this time may be significantly longer that the originally assumed analysis period) the total number of vehicles that enter at a particular entry point will equal that input by the traffic analyst.

To demonstrate the use of trip chains in the analysis of overcapacity scenarios a sample network was constructed in VISSIM. In the presented example we consider the analysis of a driveway intended to serve a new development on a busy roadway. This paper will explore the results obtained for this driveway under overcapacity conditions, both with and without trip chains. Also, simulations with different VISSIM priority rule parameters in overcapacity conditions are analyzed, allowing for an initial investigation of the potential sensitivity of modeling results to different priority rules.

2 Background

VISSIM is a discrete, stochastic and time step based microscopic simulation model developed to model urban traffic and public transit operations [20]. The model is a useful tool for the evaluation of various alternatives based on transportation engineering and planning measures of effectiveness. VISSIM models individual vehicles using a psycho-physical driver behavior model developed by Wiedemann [20–21]. The underlying concept of the model is the assumption that a driver can be in one of four driving modes: free driving, approaching, following and braking. The model was originally developed at the University of Karlsruhe, Germany during the early 1970s.

2.1 Trip chain

Sample network is created (Fig. 1). The dynamic assignment module within VISSIM allows the traffic analyst to specify traffic demand using origin-destination matrices. Within this module, it is also possible to specify the traffic demand using trip chains. In contrast to origin-destination matrices, trip chains are assigned in an FKT file (Fig. 2) and provide the simulation with detailed individual vehicle travel plans, including vehicle number, vehicle type, origin zone number, departure time, destination zone number, activity number, and minimum stay time.

Fig. 1

Illustration of sample network.

Fig. 2

Trip chain file.

For example, in Fig. 2, vehicle number 1 is an auto (vehicle type 100), is assigned to depart from origin point 1 at 1,810 seconds from the start of the simulation, and will travel to destination point 2. (A model may have multiple origin/destination points, each of which is defined during the model construction.) Vehicle 1 will then stay at destination 2 for 100 seconds. While not shown in Fig. 2, further extensions of the vehicle 1 trip to additional destinations may be specified, creating a “chain” of trips. Similar to vehicle 1, it is seen that vehicle 2, also a type 100, is assigned to leave origin 1 at 1,820 seconds after the start of the simulation and travel to destination 2, again with a stay time of 100 seconds. Vehicles 3, 4, 5, and so on are defined in a similar manner. It is important to note that the departure time is stated as the assigned time. If the link the vehicle is intended to depart on is congested and there exists no space for the vehicle to enter, the vehicle will be held until capacity exists. Should multiple vehicles be held at an originating point they will be released in a first-in-first-out (FIFO) manner.

2.2 Sample network

Figures 1 and 3. illustrate the VISSIM network utilized for this study. The network consists of a one-way 3-lane road (Main Street) with 4 signalized intersections (intersections of Main St. with First St., Second St., Third St., and Fourth St.) The scenarios under consideration assume that a new development is proposed on Main Street at an approximately mid-block location between intersections with Third St. and Fourth St. The driveway from the proposed development is assumed to be stop controlled. Each of the four signalized intersections operates a two-phase, 90 second cycle. Table 1 lists the signal timing for each intersection. For this sample network each roadway link is taken to be 880 ft long with a 100 ft driveway on the proposed development, the vehicle fleet is assumed to be 100% autos, and the desired speed is 30 mph. Main Street is a one-way 3-lane road and each side street is a one-way 2 lane road.

Fig. 3

Sample network coded in VISSIM.

Table 1

Signal timing

Intersection No.	Phase	Movement Served	Phase Length
1, 2 &3	Φ1 (East / West)	East / West	75 sec
	Φ2 (North / South)	North / South	15 sec
4	Φ1 (East / West)	East / West	45 sec
	Φ2 (North / South)	North / South	45 sec

2.3 Vehicle inputs and scenarios

As stated previously the objective of this analysis is to consider the performance of the proposed driveway under overcapacity conditions. In order to adequately model the impact of overcapacity conditions it is necessary to model the periods before, during, and after the overcapacity demands are experienced. This allows for a modeling of the impact of increasing congestion, the congested period, and the return to non-congestion. For this study this was achieved by modeling an overcapacity traffic demand followed by an under capacity demand on Main Street and the four cross streets (Table 2).

Table 2
Traffic inputs

Link No. Direction Simulation Time Vehicle Inputs

1 Eastbound 0 –3600 sec 2,000

3600 –7200 sec 1,000

2 Southbound 0 –3600 sec 360

Left on Main Street 3600 –7200 sec 180

3 Northbound 0 –3600 sec 360

Right on Main Street 3600 –7200 sec 180

4 Southbound 0 –3600 sec 300

Left on Main Street 3600 –7200 sec 180

5 Northbound 1800 –3600 sec 100

Right on Main Street 0 –1800 &3600 –7200 sec 0

Link No.	Direction	Simulation Time	Vehicle Inputs
1	Eastbound	0 –3600 sec	2,000
		3600 –7200 sec	1,000
2	Southbound	0 –3600 sec	360
	Left on Main Street	3600 –7200 sec	180
3	Northbound	0 –3600 sec	360
	Right on Main Street	3600 –7200 sec	180
4	Southbound	0 –3600 sec	300
	Left on Main Street	3600 –7200 sec	180
5	Northbound	1800 –3600 sec	100
	Right on Main Street	0 –1800 &3600 –7200 sec	0

In the first 60 minutes, representing the peak hour period, twice as many vehicles are generated over the network compared to the second 60 minutes simulation time period, resulting in each simulation run being 2 hours. From the proposed development on Main Street between intersection 3 and 4, a demand of 100 vehicles is generated from simulation time interval 30 minutes to 60 minutes (i.e. a traffic flow rate of 200 veh/hr is experienced for 30 minutes).

To compare the simulation results in a congested network, four scenarios were chosen. Scenario one is constructed based on standard traffic flow inputs and scenarios two to four are built using the trip chain method, with different space headway parameters in the priority rules of VISSIM (Table 3).

Table 3

Scenarios

Scenarios No.	Traffic Input of the Proposed Development	Parameters in Priority Rules
1	Standard Vehicle Input	Default (Min. Space Headway: 16.4ft)
2	Trip Chain	Default (Min. Space Headway: 16.4ft)
3	Trip Chain	Min. Space Headway: 10.0ft
4	Trip Chain	Min. Space Headway: 30.0ft

Scenario two utilizes the same default space headway parameters as scenario one. Scenarios three and four explore the sensitivity of the results to minimum space headway in VISSIM. In some sense minimum space headway may be thought of as a measure of driver aggressiveness.

The minimum space headway represents the gap in feet, at low speed, between two vehicles that a vehicle entering from a side street would be willing to accept to join the traffic stream by entering between the two vehicles. The shorter this gap the more aggressive the driver, i.e. the smaller the gap the driver will need to “push” their way into the traffic steam. Scenario three represents more aggressive drivers than found in scenarios one and two by using a shorter minimum space headway to enter Main Street. Scenario four models less aggressive drivers by using a longer minimum space headway to enter Main Street.

3 Simulation run

After the simulation run for each scenario was completed, the on-screen animation and model outputs were reviewed. To obtain unbiased results, each scenario is replicated 5 times with the same set of random number seeds (79, 295, 536, 712 and 10101).

A snapshot of congestion on Main Street from scenario 2 at time 3,730 seconds is depicted in Fig. 4. To estimate travel time of each vehicle on the proposed driveway, two data collection points are placed at the beginning and end of the driveway.

Fig. 4

Simulation snapshot (Scenario 2 at time 3,730 seconds).

A comprehensive quantitative comparison is conducted to explore the difference between constructing a model based on standard traffic flow inputs and the trip chain method and to investigate the sensitivity of the analysis to different minimum space headway parameters.

4 Results and analysis

For the scenarios utilizing trip chains to assign vehicles to the network, the time the vehicle is first assigned to enter the network is listed in the FKT file. If the driveway the vehicle is intended to depart on is congested, the vehicle is held until it finds a space to enter. For example, people who want to leave the parking lot of the proposed development (i.e. enter the simulated network) have to wait in the parking lot until there is space available in the driveway. In this paper, the time difference between the assigned departure time of trip chain and actual time the vehicle is able to enter the network is defined as departure delay time.

The departure delay time accounts approximately 90% of total travel time from the proposed development to Main Street (Fig. 5) in the three scenarios utilizing trip chains. As there is no accounting for assigned departure time in scenario one, the departure delay time information is not taken into account in the scenario one simulation analysis.

Fig. 5

Average travel time comparisons.

Table 4 presents the total number of vehicles generated and the time when the 100th vehicle from the proposed development entered the driveway. Due to the congestion on Main Street, shown in Fig. 4, vehicles from the proposed driveway can’t readily enter Main Street during the peak hour (i.e. the first hour of the simulation run). This results in the congestion of the driveway such that additional vehicles can’t enter the driveway from the proposed development.

Table 4

Vehicle generation (average of five simulations with different random seeds)

Scenario No.	No. of vehicles generated from the proposed development	No. of vehicles which failed to be generated	Time when the 100^th vehicle from the proposed development entered the driveway
1	19 vehicles	81 vehicles	–
2	100 vehicles	–	4,909sec
3	100 vehicles	–	4,809sec
4	100 vehicles	–	4,968sec

As a result, scenario one generated only 19 vehicles from time 30 minutes to 60 minutes from the proposed development onto the driveway, and only 14 of 19 vehicles succeeded in entering Main Street by the 60 minutes simulation time. (Recall that the demand from the proposed development is 100 vehicles over the half hour from minute 30 to minute 60.) As scenario 1 utilizes traffic flow as an input (200 veh/hr during the time 30 minutes to 60 minutes) no additional vehicles are assigned to leave the proposed development after the 60 minute mark. This means that 81 vehicles are never assigned to the network and are essentially lost, with their delay, travel, and contribution to continued congestion never taken into account.

Similar to scenario one, scenarios two, three and four are unable to release the full 100 vehicle demand from the proposed development during the minute 30 to minute 60 periods. Unlike scenario one, in scenarios two, three and four VISSIM holds the unserved vehicles until entering capacity exists on the driveway after the peak hour, even though the demands are generated during the peak hour. For example, in scenario two, the 100th vehicle was assigned a generation time of 3,600 seconds into the simulation (i.e. the end of the peak hour), however it actually entered the driveway at 4,909 seconds, 21.8 minutes after the assigned departure time.

Simulations with shorter and longer minimum space headways for the priority rules are tested in scenario three and four. Scenario three represents drivers who utilize smaller gaps to enter Main Street and in scenario four drivers who require larger gaps. The impact of reducing the minimum space headway in scenario three is to increase the number of available acceptable gaps in the Main Street traffic stream. Thus, more vehicles are able to enter Main Street and thereby more vehicles are able to enter the driveway from the proposed development during the peak period. It is seen in Tables 4 and 5 that for scenario three the 100th vehicle enters the simulation earlier and that the number of vehicles that enter the simulation during the peak period is greater. The opposite trend is seen in scenario four where the less aggressive drivers are utilized.

Table 5

Number of vehicles entering the proposed development driveway (average of five simulations with different random seeds)

Scenario No.	1st 30 min	2nd 30 min	3rd 30 min	4th 30 min
1	–	19 vehicles	–	–
2	–	21 vehicles	79 vehicles	–
3	–	27 vehicles	73 vehicles	–
4	–	11 vehicles	89 vehicles	–

Figure 6 shows the total travel time from the proposed development to Main Street for each vehicle in all four scenarios, for each replicate run. It is seen that the travel time (and subsequent) delay experienced by the vehicles in scenarios two, three and four are significantly greater than that experienced by the those in scenario one, due to the inclusion of time spent waiting to enter the network. It is also clearly seen that each replicate of scenario one processed less than 20% of the actual demand, whereas the other scenarios using the trip chain method processed the entire demand. Figure 6 also illustrates the performance differences of the three trip chain scenarios, with the least aggressive merging behavior resulting in the longest travel times and the most aggressive merging behavior resulting in the shortest travel times.

Fig. 6

Travel time for individual vehicle with different minimum space headway average travel time comparisons.

5 Conclusion

Modeling overcapacity conditions is a challenge for traffic analysts. The intent of this paper is to explore one possible means to address traffic conditions where demand exceeds capacity and to avoid the issue of “lost” demand. The VISSIM trip chain feature is used for this analysis as it has the advantage of not eliminating a vehicle when congestion hinders its entrance onto a network, instead holding the vehicle until an acceptable gap exists on the entry link. Therefore, the total number of vehicles simulated equals the vehicle demands input by the analyst, allowing for an accurate analysis of overcapacity demand scenarios.

Also, using the trip chain feature it is possible to determine departure delay time, which is defined as the time difference between the “desired” assigned departure time and actual time the vehicle is able to enter the network. The paper also provides a limited demonstration of the sensitivity of the simulation model to differing minimum space headway parameters in the priority rules.

This result implies that a small change in parameters of the priority in VISSIM can result in significant differences in simulation results under overcapacity conditions. Careful calibration of these factors should be conducted as part of any modeling effort.

As a consequence of this analysis it is concluded that, with appropriate calibrations, the trip chain feature in VISSIM is a potentially useful method to provide realistic modeling of various overcapacity conditions.

Footnotes

Acknowledgments

This work was partially supported by NRF-2017R1D1A1A09000606 and NRF-2020R1A2C1011060 of the Korean government.

References

Fellendorf

and Vortisch

, Validation of the Microscopic Traffic Flow Model VISSIM in Different Real-World Situations, in Transportation Research Board Annual Meeting, National Research Council, Washington, D.C., 2001.

Bloomberg

and Dale

, A Comparison of the VISSIM and CORSIM Traffic Simulation Models On A Congested Network, Transportation Research Record 1727, TRB, National Research Council, Washington, D.C., (2000), pp. 52-60.

Gomes

, May

and Horowitz

, A Microsimulation Model of a Congested Freeway using VISSIM, Transportation Research Record 1876, TRB, National Research Council, Washington, D.C., (2004), pp. 71-81.

Anderson

, Suh

, Guin

, Hunter

and Rodgers

, Accounting for composite travel time distributions within a traffic stream in determining Level-of-Service, Journal of Intelligent & Fuzzy Systems 36(2) (2019), 955–965.

Yang

and Koutsopoulos

, A microscopic traffic simulator for evaluation of dynamic traffic management systems, Transportation Research Part C: Emerging Technologies 4(3) (1996), 113–129.

Suh

, Henclewood

, Greenwood

, Guin

, Guensler

, Hunter

and Fujimoto

, Modeling pedestrian crossing activities in an urban environment using microscopic traffic simulation, Simulation: Transactions of the Society for Modeling and Simulation International 89(2) (2013), 213–224.

Chu

and Recker

, Micro-Simulation Modeling Approach to Applications of On-Line Simulation And Data Fusion. University of California, Berkeley. 2004.

Lee

and Chandrasekar

, A Framework for Parallel Traffic Simulation Using Multiple Instancing of a Simulation Program, Intelligent Transportation Systems Journal 7 (2004), 279–294.

Suh

, Henclewood

, Guin

, Guensler

, Hunter

and Fujimoto

, Dynamic data driven transportation systems, Multimedia Tools Applications 76 (2017), 25253–25269.

10.

Henclewood

, Suh

, Guin

, Guensler

, Fujimoto

and Hunter

, Real-time data-driven traffic simulation for performance measure estimation, IET Intelligent Transport Systems 10(8) (2016), 562–571.

11.

Junchaya

and Chang

, Exploring real-time traffic simulation with massively parallel computing architecture, Transportation Research Part C: Emerging Technologies 1(1) (1993), 57–76.

12.

Hunter

, Kim

, Suh

, Fujimoto

, Sirichoke

and Palekar

, Ad Hoc Distributed Dynamic Data-Driven Simulations of Surface Transportation Systems, Simulation: Transactions of the Society for Modeling and Simulation International 85(4) (2009), 243–255.

13.

Klefstad

, Zhang

, Lai

, Jayakrishnan

and Lavanya

, A distributed, scalable, and synchronized framework for large-scale microscopic traffic simulation, Intelligent Transportation Systems. Proceedings (2005), pp. 813–818.

14.

Wang

, Lee

and Cheu

, Dynamic Routing Decisions for Commercial Vehicle Operations in Real-Time Traffic Conditions, Transportation Research Record 1882. 2004.

15.

Suh

, Hunter

and Fujimoto

, Ad hoc distributed simulation for transportation system monitoring and near-term prediction, Simul Model Practice and Theory 41 (2014), 1–14.

16.

Park

, Kim

, Lee

, Park

and Suh

, Vision-based Surveillance System for Monitoring Traffic Conditions, Multimedia Tools and Applications 76(23) (2017).

17.

Suh

, Mitigating Initialization Bias in Transportation Modeling Applications, Proceedings of Engineering and Technology Innovation 3 (2016), 34–36.

18.

Suh

, Park

and Kim

, Traffic Safety Evaluation Based on Vision and Signal Timing Data, Proceedings of Engineering and Technology Innovation 7 (2017), 37–40.

19.

Suh

, Park

and Kim

, Traffic Safety Evaluation Based on Vision and Signal Timing Data, Proceedings of Engineering and Technology Innovation 7 (2017), 37–40.

20.

PTV. VISSIM User’s Manual, Karlsruhe, Germany, 2012.

21.

Wiedemann

, Simulation des Verkehrsflusses. Schriftenreihe des InstitutsfürVerkehrswesen, Universität (TH) Karlsruhe. 1974.