ITS - Intelligent Transportation Systems Report ITS Home Page
Table of Contents

Appendix B. Examples of Model Interfaces

This Appendix contains examples of interfaces that have been developed to assist in the integration of different types of models outlined in the AMS Methodology.  The specific examples apply to steps followed in interfacing travel demand models and microsimulation models.  These are just examples of interfaces developed for specific projects – they should not be construed as the recommended approaches to interface different types of models for AMS.  These examples are provided here for illustrative purposes.

Create Subzone Trip Table

This section outlines the steps involved in creating a subzone trip table suitable for use with a traffic microsimulation model.  First, a daily trip table is created by identifying the outline of the analysis subarea and running the full-scale travel demand model to identify the origins and destinations of link flows traversing the subarea boundaries.  The zones outside the subarea are aggregated to entry and exit points along the cut-line (external stations) and the zone and network geography inside the subarea are refined to match the traffic microsimulation network.  The daily trip table is assigned to the subarea network.  The daily assigned volumes are used to determine the time of day the trips occur.

Daily trip tables are used as the input to the time of day analysis because in most travel models peak period trip tables are often based on static factors derived from base-year surveys.  Most often they are not sensitive to peak-spreading due to congestion that is likely to occur in future year analysis.

Window the Analysis Subarea

Decide where the cut-lines will be for the analysis subarea.  The subarea should include alternate routes for congested corridors.  Links that cross the cut-lines should be identified in the traffic assignment step of the travel demand model as “select links.”  This will ensure that the origins and destinations for traffic on these links are saved to be used when aggregating the zones outside the study area.  If the analysis area is very large, it may help to identify the select links by geographically intersecting a polygon of the subarea with the road network.

Run the Travel Demand Model

Run the travel demand model in full to estimate travel demand and link volumes for the base year.  Additionally, the origins and destinations for the “select links” will be saved and used in the next step for aggregating zones outside the subarea.  The output from this step is the daily trip table by trip purpose as well as the origins and destinations by trip purpose for each of the links crossing the subarea boundary.

Aggregate Zones Outside Subarea

Trips with an origin or destination that lie outside the windowed subarea need to be allocated to external stations along the subarea boundary.  Trip origin and destinations should be aggregated to the external station that corresponds to the link that was used to cross into the subarea.  The output from this step is the daily trip table windowed to the subarea.  If possible, separate trip tables should be maintained by vehicle type and trip purpose.

Add detail to Subarea Zone System

Traffic analysis zones (TAZs), represented in the network by centroid nodes, are the origins and destinations of all trips.  Centroids can also be thought of as the sources and sinks of vehicles in the models.  All centroids (and their corresponding TAZs) should be numbered the same in all model networks and represent the same geographic area.  Centroids are connected to the network by centroid connector links that determine the point where traffic is loaded on the network.  Load points should be consistent for both the macroscopic and microscopic models.

Traffic microsimulation models are very sensitive and require a detailed treatment of traffic loading points onto the network.  Travel demand models using static traffic assignment are not as sensitive and therefore often have multiple access points aggregately represented by a single centroid connector and loaded on to a single point on the network.  Therefore, it is necessary to distribute the assigned trips from a single macroscopic-level zone in the travel demand model among all the access points used in the traffic microsimulation model.  To maintain consistency between the travel demand model and the traffic microsimulation model, each loading point that exists in the microsimulation model should correspond to one zone/centroid in the travel demand model.  Additionally, the zone/centroid numbering schema should also be the same between the models.  Therefore, many of the Travel Demand Model zones will need to be broken up into subzones.  There are two methodologies that can be employed depending on the availability of data (or they can be used in tandem).

In the first methodology, trip tables can be split from zones to subzones based on zonal characteristics.  For example, the number of trips originating from the zone can be allocated to each subzone based on the proportion of zonal houses in each subzone and the number of trips terminating in a subzone can be determined by the proportion of employment that exists in that subzone relative to the other subzones contained in that zone.  If little or no socioeconomic data are available at this detailed level, then subzone areas can be used to allocate trips from zones to subzones.  However, using the relative area can produce illogical results for some trip purposes and where the areas are not proportional to the density of land use (i.e., a field and a lake with one house next to a subdivision).

The second methodology is to run traffic assignment on the regular zones (not the subzones), but with several centroid-connectors attached to each centroid – one for every subzone.  The trips can choose where to load on to the network and the proportion of trips using each centroid connector can be used to distribute trips amongst the subzones.  This method makes the most sense to use for zones where there is a lot of internal connectivity and vehicles have many options about where to load on to the external street network.

The end product from this step is the daily trip table – the number of trips originating and terminating in each subzone and external station by trip purpose and vehicle type.

Zone Interface

Two levels of communication between the macroscopic and microscopic model must be accomplished in this step.  First, the macroscopic model must emulate the number and location of centroids and load-points (centroid-connectors) of the microscopic model.  Second, the vehicle demand between origins and destinations must be transferred from the macroscopic model to a format that can be read by the microscopic model.

The macroscopic model can import the geography of the microsimulation model network.  However, often human judgment must be used as to how to break up the zones into subzones.  This will likely include the use of aerial photos to determine logical split points.

Vehicle demand predicted by the macroscopic travel demand model is represented by the following characteristics and can traded between the demand model and traffic microsimulation model in a columnar data format for easy importation to the traffic microsimulation model:

Ideally, the vehicle trips would be stratified by (and have separate columns for):

Add Detail to the Subarea Network

Traffic microsimulation models often have very detailed networks compared to macroscopic travel demand models:  both the number of links that are coded, as well as the number of attributes on the link.  Additionally, traffic microsimulation models have very detailed representations of signalized intersections.  Often this level of detail is not possible in the macroscopic travel demand model because of the sheer size of the network and the required amount of data entry.  By mapping the two networks spatially, a correspondence can be developed between them.

Network Interface

A relationship between the network characteristics in the microsimulation model and the demand model will facilitate the information flow both for feedback (level of service parameters), as well as reasonableness checking.

Links

Links in both the macroscopic and microscopic model networks are referred to by a unique i (link origin), j (link destination) combination (in some model platforms the letters A and B are used).  They have the following characteristics at a minimum that should be commonly defined between the models:

Several other variables exist in the microsimulation network and may be used to help determine the capacity used in the demand model.  They include the following:

Additionally, the following level of service attributes will be added:

The level of service attributes will be used to feed back information to the macroscopic model.  The macroscopic model can use this to predict peak spreading.

Nodes

Node numbers in the subarea travel demand model should match the node numbers in the traffic microsimulation model.  This will enable the LOS characteristics of links to be easily transferred from the microsimulation model to the demand model.  Nodes all have the following characteristics:

If a node is an intersection then it will have numerous characteristics that are grouped by turning movement, described in the next section.

Turning Movements

There are five attributes that define a turning movement:

There are several attributes that characterize a turning movement:

Assign Subarea Vehicle Trips to Subarea Network

Once the daily subzone trip table and the subarea network have been created, assign the daily subzone trips to the subarea network.  Once the time period profile is determined, an hourly trip table should be assigned to the subarea network to make sure there are no links or turning movements over capacity.  Further refinements to the network including turn-movement capacity constraints may be necessary if problems arise.

Time-of-Day Model

Choice models that produce trips by time period are not as common in practice, but use traditional logit choice estimation techniques to apportion trip tables by purpose to various time periods.  Choice models spread the number of trips that occur in the peak period based on an assessment of congestion, level of service, purpose and socioeconomic or density variables.

The objective of the time-of-day choice models is to provide sensitivity to traveler’s temporal decisions with respect to socio-demographic and trip characteristics.  This sensitivity to temporal decision-making is expected to have significant impacts on forecasting results, as peak period travel is more likely to be occurring in saturated conditions.  Fixed time period factors provide realistic estimates of peaking characteristics under current conditions, but are not sensitive to changes in travel behavior as congestion increases or demographics shift.

The time-of-day choice models are applied to produce probabilities that trips will occur in different discrete time periods.  These probabilities are then applied to trip tables for each purpose to produce trip tables by time period and purpose.  This process is very similar to how mode choice models are estimated and applied.  The sum of the resulting time period trip tables will equal the total daily trips.

Capturing the variations in travel by time of day is essential to predicting transportation system performance to congestion pricing and ITS technologies, and air quality impacts of the transportation sector.  This is also necessary to predict traffic volumes at very disaggregate time periods and thereby replicating the reality of traffic assignments accurately.  This is critical to integrate travel demand or planning models to simulation models.  A vast amount of transportation research has been conducted to study travel demand by time of day.  Much of this research has been limited to observing trends in service usage, such as vehicular volumes and the number of person trips.  While important to understanding past and present usage patterns, these types of studies are less valuable for predicting future travel by time of day given changes in transportation service availability, quality, and policy.  Possibly the behavior least accounted for in travel forecasting is “peak spreading” (e.g., persons rescheduling their travel from daily periods of high demand to the portions of the day where travel takes less time and is more reliable).  Travel surveys and other monitoring activities have documented the correlation between decreasing service quality (congestion) and longer peak periods.  Also, many planning agencies need to test the effectiveness of policy initiatives specifically targeted at shifting travel demand to off-peak periods.

The Matrix Varigator (Simons, C., 2006, I-285 Matrix Variegator:  A Practical Method for Developing Trip Tables for Simulation Modeling from Travel Demand Modeling Inputs, presented at the 85th Annual Meeting of the TRB, Washington D.C., January 2006.) approach is a trip table refinement procedure that applies a unique temporal distribution to each O-D pair, where appropriate temporal distributions are based on the amount of congestion that is present between each pair.  This approach has been applied at the corridor level with some reasonable success.  The procedure assumes that the degree of peak spreading that is likely to occur between any O-D pair depends on the amount of congestion that is present along the shortest travel path for each O-D pair.  The distributions used here are approximations based on a previous study (Margiotta, R., H. Cohen, and P. DeCorla-Souza, 1999, Speed and Delay Prediction Models for Planning Applications, Sixth National Conference on Transportation Planning for Small- and Medium-Sized Communities, Spokane, Washington.) that developed a set of temporal distributions that varied by the ratio of the daily volume to hourly capacity (AADT/C).  These distributions were manually estimated as a simple means of moving demand from peak hours to off-peak hours as congestion increases.

Another method that facilitates peak spreading (Volpe National Transportation Systems Center, 1994, IVHS Benefits Assessment Model Framework. Final Report, Cambridge, Massachusetts.) but on a systemwide basis has been implemented by the Volpe National Transportation System Center (VNTSC) within a modeling framework applied in evaluating Intelligent Transportation Systems (ITS).  This peak spreading approach considers the systemwide excess travel demand and delay and distributes excess travel demand between the individual travel hours that comprise the peak period.  This approach is neither link-specific nor trip-specific, which is one of its serious limitations.  Also, since it was designed to model the travel impacts of ITS deployment, it assumes that a significant amount of travel information is available to travelers and thus the traveler’s temporal response to congestion can be modeled on a systemwide basis rather than on a trip-specific or link-specific basis.

One of the most essential modeling component to the analysis, modeling and simulation methodology is the time-of-day choice model that provides sensitivity to traveler’s temporal decisions with respect to sociodemographic, travel conditions, and cost of travel.  This sensitivity is needed to effectively evaluate ITS and pricing strategies and improve forecasting results.  So in the time-of-day choice models, the inclusion of more temporal details or time periods will make the models more sensitive to congestion pricing.  Most of the prior time-of-day choice modeling studies considers time as a discrete variable, that is, the various time choices are represented by several temporally contiguous discrete time periods such as a.m. peak period, off-peak period and p.m. peak period.  There are several drawbacks of using such an approach to model time-of-day choice. (Bhat, C. R., and J. L. Steed, 2002, A Continuous-time Model of Departure Time Choice for Urban Shopping Trips, Transportation Research Part B (36), pp. 207-224.) The use of discrete time periods requires a pre-determined partitioning of the day into time intervals, the characteristics of which may or may not be the same in the future.  This might preclude the analyses of potential future congestion pricing strategies during time periods which are smaller than those used in the base year.  Also, the discrete choice structure considers the time points near the boundaries of intervals as belonging to one or the other of the aggregate time periods.  But in reality, two closely spaced time points on either side of a discrete interval boundary are likely to be perceived as being similar rather than as distinct alternatives.  So either many finer discrete time intervals have to be specified to obtain a reasonable time resolution, which might not be very practical as this will involve estimating many parameters, or a distinction should be made between adjacent discrete time periods.

CS recently completed an FHWA research project on time-of-day models that resulted in a methodology for time-of-day choice models that for trip-based models and another for activity-based models.  These were tested and validated in case studies in Denver and San Francisco.  The trip-based time-of-day modeling method was applied to a pricing scenario in the Denver region. Tolls were assumed on a (currently toll-free) 20-mile section of a circumferential freeway.  Tolls were highest in the two peak periods (0.2 to 3.5 hours long), with lower tolls in shoulder periods (1 to 3.5 hours) and lowest tolls in the off-peak periods.  The time-of-day choice method estimated trips by time of day for half-hour periods.  The application of the model for this scenario showed a modest amount of peak spreading resulting from the implementation of the period-based tolls.

The tour-based time-of-day modeling method was applied to a pricing scenario for downtown San Francisco.  The time-of-day choice method estimated trips by time of day for half-hour periods. A hypothetical $4.00 toll was applied for all auto trips entering downtown San Francisco during the a.m. peak period (6:00 to 9:00).  Although it is impossible to separate all of the effects of the pricing, it is apparent that the largest effect appears to be on mode choice.  About 20 percent of the reduction in downtown trips is due to people choosing not to travel downtown at all.  About 70 percent of the total, is due to changes is mode, and about 10 percent of the reduction appears to be due to time-of-day shifts.  These results seem reasonable, as many downtown travelers, such as commuters to work, may not have the flexibility to change their times of travel.

For the Washington State DOT, CS updated the time-of-day choice models by dividing the five main periods (a.m. peak, midday, p.m. peak, evening, and night) into 30-minute subperiods, in order to model peak-spreading behavior (Kuppam, A. R., M. L. Outwater, M. Bradley, L. Blain, R. Tung, and S. Yan, 2005, Application of Time-of-Day Choice Models Using EMME/2 – Washington State DOT Congestion Relief Analysis, presented at 19th International EMME/2 User’s Group Conference, Seattle, Washington, October 19-21.) In addition to auto travel time variations between periods, the model has also been structured in such a way that it will be sensitive to auto travel cost differences between periods, for instance to emulate time-of-day-specific congestion pricing.  The new time-of-day choice models were estimated for eight trip purpose/direction combinations, using a new set of 32 alternatives.

Outputs of Time-of-Day Choice Model

The time-of-day choice models produce the choice probabilities that measure the magnitude or the ratio of vehicle trips made in a time interval to vehicle trips in the given base period, which is usually a day.  These probabilities or ratios are applied to vehicle trip tables after the trip distribution modeling step.  Based on the number of time-of-day choice models, the trip tables can be broken into the following categories:

previous