TSIS-CORSIM Did You Know?

An HOV lane can be defined to be closed to all traffic, which could simulate a lane blockage, and it can also be open to all traffic, which could model normal operation. Other than that incidents are the only way to simulate lane blockage. Incidents can last longer than a time period. The user specifies the start of the incident and its duration. The incident can start at any time (up to 9999999 sec.) and last for any duration (up to 99999 sec.). Incidents are specified in the first time period, but they are not tied in any way to the first time period or any other time period. The incident arrays are dimensioned to 100, so the max number of incidents for the entire simulation is 100. They could all be used in a single time period or spread out through the simulation. An incident cannot change its behavior over time, but an incident of one type can be followed by an incident of another type in the same location, which can be followed by another incident, and so on, to represent changing behavior over time. There can be parallel incidents to represent a blockage in one lane and rubbernecking in adjacent lanes. Queuing information can be indirectly obtained from detectors or data stations upstream from the blockage.

If on-ramp vehicles are having trouble entering the mainline, or if on-ramp vehicles are reaching the end of the acceleration lane, this can be handled by increasing the percentage of drivers who will cooperate with a merging vehicle. The default value is only 20%. This parameter works with the anticipatory lane change logic, to allow vehicles to merge more smoothly.

The Erlang distribution causes vehicle arrivals to be random when coming out of entry nodes. Once inside the network, platoon dispersion between intersections is affected by Lane Change settings and the Free-Flow Speed Distribution, which in TRAFED can be adjusted under Network > NETSIM Setup. Heavy source and sink node volumes tend to eliminate platooning, and cause random arrivals at intersections.

CORSIM allows bus stations to hold up to 6 buses. Buses are moved through the station in a first-in-first-out manner. TRAFED allows the user to modify routes and schedules easily, and CORSIM reports MOEs for buses and bus routes.

CORSIM can model pedestrians in three ways: 1. Coding pedestrian volume intensities. Higher pedestrian volumes cause vehicles to slow down more when making right-turns. 2. Pedestrian actuated signals. This is only applicable to actuated controllers, although actuated signals can be made to "act" pre-timed by invoking max recall. 3. Mean discharge headway (saturation flow rate). Mean discharge headways can be increased to account for pedestrians as recommended by the Highway Capacity Manual procedures.

Curvature, elevation and friction coefficient are only used to determine a safe upper limit to the freeflow speed on a freeway link. If that safe speed is less than the input link freeflow speed, the safe speed will be used instead, which would affect travel speed. If the input speed is less than the safe speed, curvature will have no effect. CORSIM will issue a warning if it reduces the link freeflow speed because it would exceed the safe speed. The calculation of safe speed is described in the CORSIM Reference Manual.

U-turns and left-turns can only be simulated from a shared lane when actuated control is used. To model shared u-turns and left-turns at a pre-timed controller, it may be preferable to code the signal as actuated, but specify equal maximum and minimum green times to emulate pre-timed operation.

All vehicles that are in queue when the signal turns green are tagged as candidates for phase failure. If a tagged vehicle fails to discharge before the signal turns red, it is counted as a phase failure. Vehicles that enter the queue after the signal turns green are not tagged as candidates for phase failure. In the case of a delayed left-turn vehicle in a through lane, it is not considered to be in queue, so it would not be considered for phase failure. A through vehicle waiting behind a left turn vehicle is also not considered to be in queue and would not be considered for phase failure.

CORSIM adjusts a vehicle's acceleration ability based on grade, so trucks will not accelerate as fast going uphill as they will on level ground. However, CORSIM doesn't allow a different freeflow speed for different vehicle types, which would be useful in modeling the downhill grade where trucks would drive more slowly to avoid braking problems.

Although this is not realistic and recommended for most areas, FRESIM is capable of being calibrated to simulate freeway capacities above 3000 vehicles per lane per hour. A sample input file (called "Max Capacity") is available on the McTrans web site, to demonstrate this technique.

The sample case CORSIM City (distributed with TSIS) includes a toll booth with two separate facilities in parallel. One uses a yield sign and the other uses a stop sign, but they could use pre-timed signals with different timing plans. There is also a freeway link that bypasses the toll booth, to represent prepay systems. Other users have simulated toll booths with a freeway feeding into a small NETSIM section that had several parallel links. Each link had a fixed-time signal with short cycle times that represented the delays associated with the booth operation. The freeway resumed on the other side of the NETSIM section.

Auxiliary lanes can be used to augment the number of through freeway (FRESIM) lanes. In the CORSIM City example case, there is a section of freeway in that network that has 8 lanes. There needs to be an on-ramp to get the auxiliary lane started and an off-ramp where it terminates, which is why the sample case has both of those. However, there can be many "basic freeway segment" links in the middle, having 8 through lanes and no ramp connections. In addition to the CORSIM City example case that is distributed with TSIS, another sample input file (called "Full Aux") is available on the McTrans web site, to demonstrate this technique.

The TRAFVU animation module is capable of displaying networks containing a minimal set of input data. Even though it would not be possible to run CORSIM simulation without certain additional inputs, TRAFVU is capable of displaying networks that only contain record types 1, 2, 3, 4, 5, 11 or 19, 170 and 210. A sample data file illustrating these minimal inputs (called "Minimal.trf") is available for downloading from the McTrans website.

FRESIM considers two types of lane changes, "mandatory" and "discretionary." Mandatory lane changes are due to responses to geometric objects and the assigned exit. When a vehicle detects that it needs to make a lane change to get into the lane leading to its exit, or detects that it needs to get out of a lane that leads to an off-ramp that is not its assigned exit, or when it detects something like a lane drop or an incident ahead it makes a mandatory lane change. The risk the driver is willing to accept to perform the lane change is related to several factors, including the distance to the geometric object. When it doesn't need to make a mandatory lane change it might consider making a discretionary lane change. A discretionary lane change can be performed when a vehicle is traveling at a speed below its desired speed and it appears that making a lane change would allow it to increase its speed. The advantage of staying in the current lane versus moving into another lane is calculated using the car following logic and comparing the current leader's speed and headway versus the speed of the other potential leader and the headway in that lane.

CORSIM reports spillback on a link when the back bumper of the last vehicle in any lane on the link is within the intersection at the upstream end of the link, traveling less than 3 feet/second, with a leader who is traveling less than 5 feet/second. The width of the intersection at the upstream end of the link is determined by counting the number of lanes on the cross links at that point and multiplying that number by 12. Opposing left turn pockets are assumed to be facing each other, so they are only counted on one of the links.

CORSIM utilizes three unique random number seeds (record type 2 entries 4, 17, and 18) to control traffic decisions and operations within NETSIM and FRESIM. Entry 4 is used to randomize vehicle headways when vehicles enter either NETSIM or FRESIM. This entry has no effect on results when the uniform distribution is used, but does effect results when the normal or Erlang distributions are used. NETSIM uses entry 17 for some traffic decisions, and to assign driver types plus vehicle types. The driver types are used by CORSIM to simulate varying levels of aggressiveness and decision-making, whereas the vehicle types are used to simulate varying levels of performance between passenger cars and heavy vehicles. FRESIM uses entry 18 for all traffic decisions, and to assign driver types plus vehicle types. NETSIM uses entry 18 for some traffic decisions, and to determine pedestrian activities.

Certain input parameters can be used to calibrate traffic performance at unsignalized intersections in CORSIM. These parameters include the start-up lost time (record type 11) for calibrating follow-up time, acceptable gaps in near-side cross-street traffic (record type 142), and acceptable gaps in far-side cross street (record type 143).

CORSIM reports the number of vehicle trips on each link. For the purposes of counting trips, vehicles that come from source nodes or that exit at sink nodes only count as half a trip because they only traveled half of the link. Vehicles entering from a source node travel the downstream half and vehicles exiting at a sink node travel the upstream half of the link. When CORSIM reports the number of vehicles discharged it reports all of the vehicles that reach the end of the link and get discharged onto the next link. It doesn't matter if they entered the link from a source node or if they entered the link from an upstream link.

The car-following sensitivity (CSF) factors are actually headway terms used in the Pitt car-following equation. When the factors are larger the separation between vehicles will be larger. Increasing the distance between vehicles might tend to allow more opportunities for lane changes, but on the other hand, increasing the sensitivity factors dictates that individual drivers will want to maintain a larger separation behind their leader, which will cause them to reject more lane change opportunities. Greater driver aggressiveness is reflected though smaller car-following sensitivity factors. That is reflected in the default values. A type 1 driver has a default CSF of 1.35 seconds, and a type 10 driver has a default CSF of .35 seconds.

In TRAFED, there is a "Minimum Drawn Radius of Curvature" for drawing purposes only on the Link (Surface or Freeway) Properties dialog Graphics page. This only controls what TRAFVU will use to draw the link. It does not affect speed at all. The "Radius" on the Freeway Link Properties dialog General page does affect speed but does not affect drawing curvature.

When simulating oversaturated conditions, CORSIM users often notice that queue spillback from the major street will effectively block all traffic from moving on the minor street. The simulation results become overly pessimistic because in the real world, cooperative drivers will often allow minor street vehicles to pass through (or join) the oversaturated queue. CORSIM contains a default value for the probability of joining spillback. When the simulation results become overly pessimistic due to queue spillback, the user can calibrate the probability of joining spillback in order to achieve more realistic results. The probability of joining spillback may be coded within the TRAFED graphical input editor contained within TSIS, or may be coded "manually" on record type 141.

Delay and travel time per vehicle values (from the cumulative summary report) account for vehicles that remain on the link when the time period ends. When oversaturated conditions prevail, this should allow for more accurate delay and travel time results.

By default, CORSIM reports or output files contain a corresponding input data report, or input "echo". However, the input data file may be modified so as to suppress the input echo. A request to suppress the input echo may be specified within the TRAFED graphical input editor, or may be coded manually on record type 210.

Unlike NETSIM, FRESIM uses a gravity model to determine the origin and destination of individual vehicles, based on the mainline volume and entering/exiting ramp traffic. Origin-destination data can be used to override the internally determined vehicle paths. However, when coding this data, freeway origin-destination pairs cannot be separated by NETSIM sections. For two nodes to be an acceptable origin-destination pair there must be a path from the origin to the destination that includes FRESIM links only.

Although the default vehicle arrival distribution is uniform, random arrivals may be requested in the input file. CORSIM has difficulty simulating random vehicle arrivals when the input volume is extremely low. CORSIM uses a distribution to specify entry times within each one-minute interval. When the entry volume is extremely low the process breaks down and becomes a constant distribution. With 10 vehicles per hour (vph) the per minute entry volume is one-sixth of a vehicle, so no vehicles enter for 5 minutes and then one vehicle enters during the sixth minute. Low-volume driveways can optionally be modeled using a time-varying source node. Instead of specifying 10 vph for the entire hour, 0 vph can be specified for most of the hour along with a few short intervals with 20 or more vph, so that the total number of vehicles emitted over the hour is 10.

It is possible to calibrate the behavior of left-turn sneakers and left-turn jumpers within the CORSIM input data. Calibration of these parameters can be important when attempting to achieve consistency with other analyses involving permitted left-turns. In CORSIM terminology, left-turn sneakers are referred to as left-turn "laggers".

As with TRANSYT-7F, user-defined link lengths cannot be too short. Link lengths and free flow speeds must not allow a vehicle to completely skip over any link during one second of simulation. If a vehicle is capable of skipping over a link completely, this can potentially compromise simulation results or cause fatal errors.

In certain cases, it is possible to increase freeway and ramp capacities by lowering the desired free flow speed on ramp links. Indeed, lower speed limits are often observed on short ramp links in the field. This buys time for the drivers to make better decisions.

In order to obtain accurate speed and volume results on congested freeways, one solution is to change car following parameters for vehicles traveling on those links. Link-specific car following sensitivity multipliers cause vehicles to follow the vehicle in front of them closer than they would using the network-wide sensitivity factors. Threshold speed and distance for anticipatory lane changing for on-ramps can be changed to minimize anticipatory lane changing that would contribute to the weaving. Moving an off-ramp warning sign location farther upstream can lengthen the weaving zone area and give vehicles more distance to make the required lane changes.

The TSIS interface is capable of disabling the large animation files (e.g. the “.TSD” files) generated by CORSIM. This allows for faster running times on the computer, and potentially prevents the large animation files from clogging the hard drive. The obvious disadvantage of this strategy would be the inability to view dynamic animation, although the basic network geometry can still be viewed within TRAFVU even without animation files present.

The default arrival pattern for entry node vehicle generation is the uniform distribution. Two other arrival distributions, normal and Erlang, are available in order to model random arrivals. A special case of the Erlang distribution is the negative exponential distribution, i.e. Erlang with a parameter of 1.0. When the negative exponential distribution is requested, the result is actually a “shifted” negative exponential distribution, because vehicle separation is prohibited from falling below a specific minimum value. This shifted negative exponential distribution allows the program to closely replicate random, Poisson vehicle arrivals on external links. These non-default arrival distributions may be requested within the input file.

If the input file is correctly designed, CORSIM is capable of simulating two intersections governed by one controller. A sample file that illustrates this technique is available for downloading from the McTrans web site.

TRAFVU can be launched before CORSIM has completed simulation. However, if the TRAFVU animation “catches up” with CORSIM, a warning message will appear.

By defining bus routes and calibrating bus (vehicle type) characteristics, it is possible to simulate links that can only be used by trucks or taxis.

CORSIM networks contain no vehicles at the beginning of a run. As the first seconds are simulated, vehicles are emitted onto the network from entry and source nodes. The time required to fill the network with traffic is referred to as the initialization period. Since the initialization period does not accurately represent the conditions to be modeled, no statistics are gathered during this period. A check is made at the end of every time interval for equilibrium, i.e. the end of initialization. Equilibrium is assumed when the number of vehicles in the network is within 8% of the number of vehicles in the network during the previous time interval, and within 12% of the number of vehicles in the network during the second previous time interval. In the CORSIM output file, this information is reported in the section called “Initialization Statistics”. One example of a network not reaching equilibrium would be a relatively large network with a relatively short initialization time. In this situation, at the end of the short initialization time, vehicles will have entered the entry nodes and will have begun to filter in toward the middle of the network. If the initialization time is too short, then they haven't had enough time to reach the middle of the large network yet, so any statistics collected from the first few minutes of simulation will be unrealistic. Not only do vehicles need enough time to reach the middle of the network, they also need enough time to fill in the exit links that lead out of the network. If the exit links are not fully initialized by normal traffic flow patterns, the results can also be unrealistic.

The TRAFVU animation module contains some useful static graphics functionality. Double-clicking on a signal indication launches a static graphics dialog box containing information about the signal settings. Double-clicking on a vehicle generates a dialog box that shows current information about that vehicle. If a link is highlighted, then clicking on the MOE button on the edge of the screen provides access to tables and graphs that provide additional information about link performance.