TransModeler’s lane-changing model is used for large-scale simulation of route choice and requires careful parameter tuning. In large network simulations, lane-changing parameters directly affect lane selection. A common issue is that on a multi-lane approach to an intersection, one lane becomes heavily queued while other available lanes remain nearly empty. This is a driving-behavior parameter setting issue. The following case analyzes how to diagnose and resolve it.
1. Problem description
During a district-level simulation, a user observed that at the north approach (Link25) the four left-turn lanes showed three lanes nearly unused while the leftmost lane had 100% occupancy. Downstream links and lane connectivity were checked and found to be correct. Increasing left-turn demand did not change the behavior: vehicles continued to queue exclusively in the leftmost lane.
2. Cause analysis
Path analysis showed that after turning left vehicles continue straight through Link25 → Link26 → Link15 → Link16 → Link11100136 (red path). Because the downstream links involve multiple changes in lane counts and Link11100136 has a bus-only lane on the rightmost lane, the number of required lane changes depends on the departure lane chosen at Link25. Vehicles departing from the leftmost lane require the fewest lane changes (one), while those departing from the second, third, etc., lanes require more lane changes.
Because the lane-change count heavily influences the path-choice cost, the leftmost lane becomes the optimal route in the model, which does not match observed behavior. This is tied to the mandatory lane-change model parameter called critical distance. Downstream lanes only enter the lane-choice evaluation if the vehicle’s distance to the next segment exceeds the critical distance parameter. Each additional required lane change increases the effective distance needed to perform those lane changes. Variations in driver behavior across regions are represented by adjusting the critical distance distribution.
3. Simulation principles
Discretionary lane changing (DLC)
Before considering discretionary or mandatory lane changes, a vehicle checks the current and adjacent lanes to determine which lanes meet driving conditions. A lane may be considered ineligible for various reasons, for example:
- There is an obstacle between the current lane and the target lane.
- Lane changes from the current lane to the target lane are prohibited (solid line), and the driver complies with the rule.
- The lane is blocked by an accident or stalled vehicle.
- The lane is closed with a lane-use sign.
- Toll booths on the lane are closed.
- The lane restricts access to certain vehicle types (for example, HOV-only).
- The vehicle is within the approach critical distance for an exit or intersection and the target lane does not connect to the vehicle’s path.
DLC is considered when the vehicle is generally dissatisfied with its current speed and no mandatory lane-change condition applies. If the vehicle is not at or near its desired speed and lane-change behavior is not restricted, it will consider discretionary lane changes.
Mandatory lane changing (MLC)
As with DLC, the model first determines the set of eligible lanes using the same rules. However, MLC is not modeled as a selectable alternative: various conditions create a need to change lanes and, in absence of alternatives, the lane change becomes mandatory. Examples include:
- The vehicle is within the critical distance of a path split and is not currently in a connected lane.
- An accident or stalled vehicle is blocking the current lane.
- A lane-use "closed" sign is in effect.
- The current or downstream lane is reserved for other vehicle types.
- The vehicle lacks an electronic toll permit but is on a lane with a downstream toll booth.
- A lane drop is occurring (lane connectivity deviation less than 1).
Look-ahead
Throughout the simulation, TransModeler maintains an approximate downstream "sketch" of each vehicle’s path. This sketch represents the driver’s perception of the downstream path and is determined by the look-ahead range, measured in time. The look-ahead distance is computed by assuming the vehicle travels each link at its free-flow speed for the look-ahead time. For example, with a 90-second look-ahead, the driver considers maneuvers needed roughly 90 seconds ahead. Lane-change decisions are made to accommodate required operations within the look-ahead horizon.
For long links, a reasonable look-ahead time may not cover the downstream distance where maneuvers are required. To avoid long links interfering with look-ahead, TransModeler enforces a minimum number of downstream links considered. If this is uncommon or if computational efficiency is prioritized, reducing the minimum number of links can help.
When looking ahead, the model also accounts for lane drops. If the connectivity deviation alpha for a lane connection is less than 1, a proportion of drivers (1 ? alpha) will attempt to force out of the lane before it ends. Connectivity deviation models situations such as lane drops or lanes that are less desirable. Because drivers may not perceive these conditions until close to the lane end, a parameter controls the response distance at which connectivity deviation begins to influence mandatory lane-change decisions.
Critical distance
As a vehicle looks ahead, it identifies critical points on its path where a lane change is required (turns, ramp exits, other splits). Any lane-change decisions made upstream within the critical distance of these points are treated as mandatory. Even if the vehicle is currently in a suitable or connected lane that matches the required path, staying in the current lane would be overridden to ensure the vehicle is positioned correctly at the critical point, unless special circumstances such as a blocking incident exist. Critical distance varies among drivers: more aggressive drivers may allow themselves to approach nearer to the critical point before forcing a lane change, while more cautious drivers will make mandatory lane-change decisions farther upstream.
Critical distance can also vary by facility type. On high-speed facilities, critical distances are typically longer because decisions and actions must be taken over a longer distance. On low-speed urban streets, critical distances are shorter.
Parameter tables define distributions of critical distances for drivers on highways and urban streets. The table specifies the percentage of drivers assigned to each critical distance value for the general fleet (non-transit). A second parameter table defines the distribution for transit vehicle drivers. The images below show the TransModeler 6.1 parameter-table interface.
Downstream lanes are considered for lane-change choice only if the distance to the next link exceeds the critical distance drawn for that driver. Each additional required lane change effectively multiplies the needed look-ahead distance. For example, if the critical distance D0 is 400 m (drawn randomly for a driver), then one lane change requires 400 m, two lane changes require 800 m, three lane changes require 1,200 m, and so on. In the case above, the total distance along Link25 → Link26 → Link15 → Link16 → Link11100136 is about 600 m, which is only sufficient to plan for one lane change.
4. Solutions
1) Extend the link
Extending Link25 downstream in the model allowed left-turn vehicles to distribute across all four left-turn lanes, resolving the issue. In this case, Link25 was at the simulation boundary, so extending the link was reasonable. If the actual physical link is only the shorter length, extending the model link is not appropriate.
2) Set entry-lane usage at origins (Entry Lane at Origin)
The Entry Lane at Origin parameter applies to origin links and specifies the utilization proportion of different lanes at the link start. This is similar to setting centroid-to-node connection proportions in macroscopic models. Use the Parameters > Parameter Markers Toolbar and select Entry Lane at Origin to set lane usage for different vehicle types.
After applying entry-lane usage, the left-turn lanes were utilized, but analysis of the subsequent maneuvers showed that straight and right-turn vehicles were present on the left-turn lanes, which did not match observed behavior. This indicates the issue was not fully resolved.
3) Modify critical distance parameters
Reducing the global critical distance parameter can improve the situation on the current link but may produce unrealistic driving behavior elsewhere. A localized approach is to use Parameter Markers to shorten the critical distance for Link25 only. For large networks, manually applying localized adjustments can be labor-intensive.
4) Preferred method: TransModeler 7.0 parameter
TransModeler 7.0 added an "Impact from additional lanes" parameter to address this issue more directly. Let alpha denote this parameter. The minimum distance D required for N lane changes is computed as:
D = D0 * (1.0 + alpha * (N - 1))
where D0 is the base critical distance. For example, with alpha = 1 and D0 = 400 m (typical default for some user populations), three lane changes would require 1,200 m. With a smaller alpha, the additional distance per extra lane change is reduced. Users in the Chinese market might select alpha values smaller than 1.0, for example 0.25 or even 0. With alpha = 0.25 and D0 = 400 m, three lane changes would require only 600 m. The alpha parameter reflects aggressive versus conservative driving tendencies and can also vary by vehicle class.
After introducing this parameter and tuning it for the local driving behavior, the instantaneous simulation display showed more realistic lane usage.
