harrington by ashrafp

VIEWS: 62 PAGES: 19

									Update and Validation of the M-NCPPC TRAVEL/2 Model


                                by

                  Thomas K. Harrington, AICP
    Maryland-National Capital Park and Planning Commission
                    Silver Spring, Maryland

                               and

                          Luc Deneault
                        INRO Solutions*
                        Montréal, Québec
        *Currently working with Groupe Conseil LOCTRANS INC.



                     For Presentation at the

      14th Annual International EMME/2 User’s Conference
                         Chicago, Illinois

                     October 20 –22, 1999




                                                               1
                   Update and Validation of the M-NCPPC TRAVEL/2 Model


ABSTRACT

The Maryland-National Capital Park and Planning Commission (M-NCPPC) in Montgomery
County, MD has been a long-time user of EMME/2 in the development and application of the
county’s travel demand model, called TRAVEL/2. Some of the innovative features of this model
include: estimation of P.M. peak period travel (generation, distribution, and mode choice) rather
than daily travel, modeling of non-motorized trips, modeling of intersection delay, and extensive
use of feedback loops. Two recent projects have significantly improved the performance and
usability of the TRAVEL/2 model.

The first project involved the refinement of the TRAVEL/2 procedures to take full advantage of
EMME/2 macros and software enhancements. Application of the model was converted from
UNIX script language to EMME/2 macro language. Model improvements included use of multi-
class assignment procedures to model HOV trips, revisions to volume-delay functions, modeling
bimodal (park-and-ride) trips using matrix convolutions, and use of the successive average
method for convergence of congested auto times in the feedback loop.

After completing the model refinement project, the second major effort involved the validation of
the model for a new base year. The validation study showed that the revised TRAVEL/2 model
matches observed data reasonably well for individual model components as well as for overall
measures such as screenline crossings.

These recent model improvements will be of interest to agencies that need to update their models
without the resources to perform a full-scale model development and re-calibration effort.


ACKNOWLEDGEMENTS

The authors wish to acknowledge the efforts of the project teams that completed the work
described in this report. The Model Update Study was conducted by INRO Solutions Inc.
including Luc Denault and Alain Audette, with special assistance from Dr. Michael Florian. The
Model Validation Study was conducted by BMI Inc. including Phil Shapiro, Donald Stephens, and
Daniel Goldfarb. The M-NCPPC staff involved in these studies include Eric Graye, Donald
Ostrander, Yetta McDaniel, and Richard Hawthorne.

Contact Information:

Thomas K. Harrington
Transportation Planner
Maryland-National Capital Park & Planning Commission
8787 Georgia Avenue
Silver Spring, MD 20910, U.S.A.
(301) 495-4519
harringtont@mncppc.state.md.us




                                                                                                    2
Introduction
The Montgomery County Department of Park & Planning has used travel demand models since
the mid-1970’s for administering the county’s Adequate Public Facilities Ordinance, developing
area master plans, and supporting transportation planning studies. An EMME/2 license was
obtained in 1986. The current travel demand model, called TRAVEL/2 was developed from 1991
to 1994.

Transportation planning agencies rarely have the resources to perform a full-scale model
development and re-calibration effort. Instead, often an incremental approach is used to refine
and update procedures that take advantage of advances in computer software and hardware and
the latest state-of-the-art modeling procedures. Model improvements have been undertaken to
respond to the evolving analytical needs brought on by changes in federal (TEA-21), state (Smart
Growth Initiative), and local legislation (County Ridesharing Goals).

Two recent projects have significantly improved the performance and usability of the TRAVEL/2
model:

(1) Model Update Study was conducted to streamline and improve the application of TRAVEL/2
    procedures. The improvements took advantage of the latest EMME/2 features and macro
    language
(2) Model Validation Study was conducted to compare travel model results with observed data
    to verify model performance. While the model components were not fully re-calibrated, some
    adjustment of model parameters was performed based on validation results.

This report describes the TRAVEL/2 model, and provides an overview of both the model update
and model validation studies.


Background
Montgomery County Department of Park and Planning is a unit of the Maryland-National Capital
Park and Planning Commission (M-NCPPC). M-NCPPC is a bi-county planning agency with
responsibility for a regional system of parks and the preparation and administration of a general
plan for the physical development of most of the non-incorporated portions of Montgomery and
Prince George’s counties, the two Maryland suburban counties adjacent to Washington, DC.

Montgomery County is a member of the Metropolitan Washington Council of Governments
(MWCOG). The Metropolitan Planning Organization (MPO) for the Washington, D.C. region is
the MWCOG Transportation Planning Board. In contrast with M-NCPPC’s county-level focus,
MWCOG’s model is used to analyze regional travel conditions and to perform air quality
conformity analysis.

As shown in Figure 1, Montgomery County is located north of Washington DC and south of
Baltimore. The county is the most populous jurisdiction in Maryland, home to roughly 825,000
residents making up 308,000 households. There are approximately 500,000 jobs in the county. It
is also one of the wealthiest counties in the U.S. with a median household income of $68,500.

As shown in Figure 2, the County is bisected by two interstates: the Capital Beltway (I-495),
providing for east-west travel, and I-270, providing for north-south travel in the county. There are
a number of North-South major highways fanning out from the District of Columbia into the
county. There is a more limited number of East-West arterials connecting suburban activity
centers. The County is served by 12 stations of the Red Line of the Metrorail system. Bus
service is provided by regional Metrobus and the Montgomery County Ride-On bus sytem.
Commuter rail connects 11 stations in Montgomery County with Union Station in downtown D.C.



                                                                                                    3
4
5
Overview of TRAVEL/2
TRAVEL/2 is a regional transportation planning model with a focus on Montgomery County. The
model region for TRAVEL/2 encompasses the combined Washington and Baltimore metropolitan
areas as shown in Figure 1 – Location Map. Trips made to/from neighboring jurisdictions,
particularly Washington D.C., make up a significant portion of county travel and must be
accounted for. There are 677 TAZ’s in the model region covering 15 counties, with a population
of 6.25 million and 4 million jobs. Montgomery County has a much more detailed zone system
than the remainder of the region with 318 TAZs. The existing transportation network is
represented in EMME/2 with 18,500 directional links and 5565 nodes.

A detailed description of the model development process, structure, and parameters is contained
in TRAVEL/2 A Simultaneous Approach to Transportation Modeling, Draft Report (February
1993). Data sets from the MWCOG Household Travel Survey, Montgomery County Planning
Department (MCPD) Trip Generation Studies, MCPD Census Update Survey, MCPD Travel Time
and Delay Study, Metrorail Passenger Surveys, and MCPD traffic counts database were used to
estimate, calibrate, and validate the model.

Three unique aspects of the model include:

   P.M. peak period travel (generation, distribution, and mode choice). Travel demand is
    forecast for a three-hour peak period rather than for daily travel. The P.M. peak period was
    chosen because local studies found it to contain the most widespread congestion, and the
    highest percentage of the rapidly-growing non-work trip purpose.
   Inclusion of non-motorized (walk & bike) trips (generation, distribution, and mode choice).
    Although bike/walk trips are not assigned to the transportation network, the model can be
    used to test the impact of land use and transportation policies on the share of non-motorized
    trips.
   Equilibration using a feedback loop that takes congested times from assignment for use in
    distribution and mode choice. Improvements to the feedback loop process are described in
    the next section.

Travel patterns in the P.M. peak period tend to be more complex than in the A.M. peak period,
with travelers much more likely to stop off for other purposes on the commute home from work.
TRAVEL/2 accounts for these linked work trips as a separate trip purpose. There are seven trip
purposes: Work-to-Home (unlinked), Work-to-Other (linked), Other-to-Home (linked), Home-to-
Work, Home-to-Other, Other-to-Home (unlinked), and Other-to-Other. Direction is explicitly
identified (i.e. no productions and attractions) because of the use of a specific time period.

TRAVEL/2 follows the traditional four-step/UTPS approach, but with significant enhancements
and additions. The inputs to the model include the transportation network, land use activity,
demographics, and public policies. The primary model steps are shown in the flowchart in Figure
3:




                                                                                                  6
                                          FIGURE 3
                                      TRAVEL/2 Flowchart

                   Initialization Steps
            Set iteration counter at 1
            1. Initial set up
                           Free flow speeds and capacities on links
                           Inputs of bimodal trips
                           Volume-delay functions on approaches
                        and cycle-length on intersections
                           Transpose vectors
            2. Auto ownership (LOGIT)                                              optional
            3. Trip generation                                                     optional
            4. Initial all-or-nothing auto assignment to generate auto LOV and HOV skims
            5. Mode share defaults                                                 optional

                   Begin Demand Loop
            Incrementation of iteration counter
            5. Transit skims (Walk connect)
            6. Transit skims (Auto connect)
            7. Mode choice model (LOGIT)                                           optional
            8. Destination choice model (gravity)
            9. Departure time choice model (LOGIT)
            10. Computation of auto connect to transit matrices (Park& Ride, Kiss& Ride)
            11. Computation of sum peak hour trip matrices
            12. Full equilibrium multi-class auto assignment
            13. Network interim reports [optional]                                 optional
            Evaluation of stopping criteria
            14. Assignment of « averaged » flows on « skim » scenario
            15. Computation of intrazonal time
                   End Demand Loop

                   Closing Steps
            16. Determine auto connect/walk connect to transit trip matrices
            17. Walk connect and auto connect transit assignments
            18. Final reports




NOTE: Steps listed as optional may be skipped based on the user selected run parameters.
However, these steps are skipped only when default values from a previoius run are used instead.




                                                                                               7
Model Update
The new release of TRAVEL/2 (officially Release 3 and referred to as TRAVEL/2.3) has generally
improved upon the format and implementation of the model. Model structure and parameters
used previously in Release 2 (now referred to as TRAVEL/2.2) remain primarily intact. Unlike
TRAVEL/2.2, which uses AWK scripts and FORTRAN programs to execute a number of modeling
procedures, TRAVEL/2.3 is fully implemented using the EMME/2 macro language and file
structures. In addition, a new user interface is used to set up model run parameters either
interactively or in batch mode.

The changes made in TRAVEL/2.3 are documented by INRO Solutions in two reports: Revised
Elements in the Implementation of the Model and User Manual of EMME/2 Macros (June 1998).
Key differences in the revised TRAVEL/2.3 model as compared to the TRAVEL/2.2 model include:

1.      Modeling of auto connect to transit (bimodal) trips, i.e. park-n-ride facilities

2.      Use of multi-class assignment procedures to properly model HOV facilities and truck
        prohibitions.

3.      Revision of the volume-delay functions

4.      Definition/implementation of the demand feedback loop.

These refinements are described in more detail in the following sections.


Modeling Bimodal (Park-and-Ride) Trips

TRAVEL/2.3 uses an approach to model bimodal, or park-and-ride trips, that was developed by
Heinz Speiss and makes use of Module 3.22 - Matrix Convolutions. The previous release of
TRAVEL/2 did have a sophisticated park-and-ride procedure that was implemented with a
FORTRAN progam. However, the current approach is implemented fully within EMME/2 and uses
parking lot capacities in the process of determining the transfer zone.

The park-and-ride procedures are applied to work trips (from work to home in the P.M.) and
involve the following steps:

(1)     Macro t.parkrideimp computes auto-connect to transit impedance matrix (uses
        legimped.mac, an INRO shared macro developed by Heinz Spiess):
         Identify intermediate transfer zone (from list of possible park-and-ride lots)
         Matrix convolutions used to combine 1st leg (transit) with 2nd leg (auto) times
         Auto time weighted by 3 so that transfer zone is chosen closer to home.
         Parking cost is also allocated based on best intermediate zone.

(2)     Mode choice model computes shares for auto-driver to transit (ADT) and auto-passenger
        to transit (APT) modes.

(3)     Macro t.bimodaldem implements logit parking choice model with explicit capacities using
        parkride.mac (INRO shared macro developed by Henz Spiess).
         For every O-D pair, this Logit model allocates an « utility » to each parking lot based
            on the impedance of the first leg and second leg of the bimodal trip. The number of
            trips that use a parking lot k is the based on the ratio of the utility of parking lot k to
            the sum of utilities of all the parking lots of the model. The mathematical formulation
            of the algorithm that solves this problem considers the capacities of the parking lots.



                                                                                                          8
            t.bimodaldem computes the auto-driver connect to transit demand, multiplying its
             mode share by the sum of the work to home trips and work to home linked trips. It
             also computes the auto-passenger connect to transit demand, by multiplying its mode
             share by the sum of the work to home trips and work to home linked trips.
            t.bimodaldem runs the parkride.mac macro to generate the auto portion of auto-driver
             connect to transit trips (second leg demand matrix) and the transit portion of auto-
             driver connect to transit trips (first leg demand matrix).

During the validation process, application of the park-and-ride procedures resulted in some
unlikely auto-connect transit trips. For example, transit trips between two suburban activity
centers, not currently connected by transit, were created in which the transfer point was located
close to the work end of the trip, so that the auto leg of the trip was much longer than the transit
leg. In order to prevent this from occurring, a constraint matrix is created that caculates whether
O-D pairs met certain criteria including:

        Auto-Connect Transit Leg > 15 minutes
        Auto-Connect Transit Leg < Walk-Connect Total Transit Time
        Auto-Connect Auto Leg < SOV Auto Time
        Auto-Connect Auto Leg < 1/3 of the ACT Total Time

Auto-connect to transit mode shares are calculated only for O-D pairs that meet each of the four
criteria.


Modeling HOV Facilities and Truck Prohibitions

TRAVEL/2.3 allows for more flexibility and refinement in the modeling of the HOV facilities.
Although currently there are only HOV-2 lanes in Montgomery Country, the model is flexible
enough to test four possible HOV configuration options:
                         no HOV facilities;
                         HOV 2+ facilities only;
                         HOV 3+ facilities only;
                         HOV 2+ facilities and HOV 3+ facilities

Based on the configuration selected, the TRAVEL/2.3 model is flexible enough to :
  1.         apply the proper computations of the mode choice model and trip distribution models
  2.         handle the creation of the appropriate demand matrices;
  3.         manage the preparation and computation of the appropriate multi-class assignment of
             the auto demand matrices.
Highway links are coded with an auto mode code « c » and auxilliary auto modes « 1 », « 2 » and
« 3 » to specifiy the occupant classes allowed. For example, a general use highway lane would
be coded with modes « c123 », while an HOV-2+ facility is coded as « c23 ».
Multi-class assignment can also be used to model heavy trucks prohibitions on minor roadways.
Auxilliary auto mode « s » is associated to the heavy trucks trips. Prohibitions are coded by
removing mode « s » on the minor links where the heavy trucks are not allowed.

Depending on the user specification, the auto assignments computed by the M-NCPPC
transportation model will always assign from three to five classes of vehicles simultaneously: one
to three classes of autos depending on the configuration of the HOV network, the light and
medium trucks class and the heavy trucks class.




                                                                                                       9
Revision of the Volume-Delay Functions (VDFs)

A number of refinements were made to the volume-delay functions including
               Reduction of the number of functions, combining classes with similar speed and
                capacity assumptions.
               Use of a logistic volume-delay function to replace the BPR function used on links
                outside of Montgomery County (to eliminate excessive delay on sketch-level
                network).
               Use of a link-based approach to model intersection delay at all signalized
                intersections in the county.

In addition, during the validation study, new freeflow speeds and capacities were applied.

Links Outside Montgomery County

Test runs of TRAVEL/2.2 showed that the auto assignment was not converging properly to an
equilibrium solution. After 100 iterations of auto assignment, the target relative gap of 1% was not
achieved. A review of the network identified a number of links with unrealistically high travel
times, especially on links in the Baltimore area. This suggested that the sketch-level network
used outside Montgomery County is too coarse to handle the demand generated by the large
TAZs used for surrounding counties.

The selected solution to this problem was to use a logistic function on links outside the county that
would set a threshold that speeds could not drop below. The « s » shape functions generate
more realistic travel times when the assigned volume exceeds the capacity of the link

The logistic vdf have the following form:
                                                      ß×(v-c)/c           ß×(v-c)/c
                                  T=TO+(TS- TO) × ø               /1+ ø
where :
                TO: the time that corresponds to the free flow speed;
                TS: time that corresponds to the congested (bottom) speed;
                v : link flow;
                c : link capacity;
                ß : parameter of the vdf that defines the slope of the curve between the free flow
                 and congested link speeds ; a high value of ß generates a steep curve and a low
                 value of ß generates a smooth curve.

If the link volumes produced by the model are unrealistic (which is likely to occur in areas of the
network that do not constitute the object of extensive study and analysis) this type of function will
ensure that the delays associated to the volumes are more realistic. The minimum speed of
major links was set at 6 minutes/mile or 10 mph (vdf 50-58) on highways and major arterials and
at 8 minutes/mile or 7.5 mph (vdf 61-86) on minor arterials.

For example, a freeway with a capacity of 2100 vehs/lane-hour and freeflow speed of 60 mph
would have the VDF:

Time=length*(60/60+ 5 / (1+5 / ((volau +1)/(2100*lanes))^5))




                                                                                                    10
Modeling Intersection Delay

The presence of intersections is a major factor of delay and congestion for trips during the peak
periods. In some dense urban areas, the intersections will have more impact on the delay than the
link itself. Therefore, it is more realistic to estimate the travel time due to the intersection
separately from the actual travel time on links between intersections.

The previous release of TRAVEL/2 calculated intersection delay based on detailed intersection
data using the 1985 Highway Capacity Manual (HCM) procedures, and predicted turning
movements for 517 of the county’s 750 signalized intersections. The process required numerous
inputs such as the number of turning lanes and signal phasing that were subject to change and
difficult to maintain. While this HCM approach is appropriate for sub-area studies, it is not
practical for regional forecasts. The revised process requires fewer inputs and can be
implemented for all signalized intersections in the county.

The approach implemented in TRAVEL/2.3 is an adaptation of the Chicago Area Transportation
Study (CATS) strategy. This approach is much less data consuming, it does not even require the
characterization of the phases of a cycle (identification of the turning movements, number of
phases etc..). It also relies on simpler computation operations.

CATS uses the Webster equation to model intersection delays in their EMME/2 transportation
model. This intersection delay has uniform and incremental components. This equation is the
result of regression analysis performed for a range of cycle lengths and green time to cycle length
ratios.

The intersection delay is computed on the approach links of the designated intersections. On
these links, the approach delay is added to the travel delay on link.

Here follows an example of the definition of a volume delay function modified for the computation
of the intersection delay on an approach link :
                       vdf = (60 * length / A) * (1 + (volau / (B * lanes)) ^ 4)
                            + ((6.0*volau/(B * lanes)+2.7*(volau/(B * lanes))^8+ul2).min.ul3)

The first part of this function is a regular BPR volume-delay function where A is the free flow
speed and B is the capacity per lane of the link. The second part is actually the intersection delay
where ul3 is the cycle length (1.5 or 1.0 minute) and ul2 is a fix term of the intersection delay
function equal to :
             ul2 = ((( 3.4 – 0.39*(G*ul3*60) + 0.35*ul3*60 – 4.5 – 7.3*G) .max. 0) / 60)

where G is the green phase duration for this approach. The expression of the fix term is
compared to zero to avoid - possible - negative values.

This intersection delay is defined such that it will never exceed the cycle length. It is also adapted
from the intersection delay function presented in the theory to estimate intersection delays in
seconds rather than minutes.

In TRAVEL/2.3, these functions are implemented using the same capacity for the end-node of a
link (intersection) as for the link itself since it is assumed that the capacities associated to the
links in the model account for the presence of traffic lights or of any other type of intersection
delay.




                                                                                                       11
Feedback Loop Structure

TRAVEL/2.3 has adopted the link averaging method as described in a presentation made at the
Tenth Annual EMME/2 Users’ Group Conference by Walker et al. of Portland Metro. The Portland
paper demonstrated that, using different performance criteria, the link averaging method was
shown to converge faster than two other approaches.

The link averaging method, averages the equilibrium flows of the auto assignments as follows :
                                       k          k       k-1
                                      v = ½ volau + ½ v

The demand model implemented is an iterative model in which demand matrices are estimated
until one of two stopping criterias of the demand loop is satisfied. At each iteration, auto and
transit demand matrices are estimated and assigned on their respective networks. Then, following
a « successive average » strategies, the flows that result of the auto assignment are weighted
with those of the previous iterations and assigned on a separate scenario. The auto trip times
matrix that results of this « dummy » assignment and the transit impedance matrix obtained from
the transit assignment are then used to re-estimate the demand matrices of each mode
considered. The « successive average » strategy « smoothes » the flows assigned on the auto
network and accelerates the convergence to a good solution. At the end of each iteration of the
demand loop, the convergence of the solution is evaluated.


         The two stopping criterias of the demand loop are :
            maximum number of iterations;
            relative difference between the Vehicle Hours Traveled (VHT) resulting from auto
             assignments on consecutive iterations of the demand loop.




                                                                                                 12
Model Validation
After completing the model refinements described in the previous section, a study was undertaken
to evaluate and improve the model’s ability to replicate existing travel, also know as model
validation. The base year against which the model results were compared was updated to 1998.
While the model components were not fully re-calibrated, some adjustment of model parameters
was performed based on validation results.

The first step of this study involved assembling validation data from a number of sources. Figure
4 displays four different types of data sources including:

Travel surveys: A regional household survey was condicted in 1994 and this can be used to
obtain estimates of trips generated as well as trip length distribution. Unfortunately, the sample
size of this survey (and most regional travel surveys) was not sufficient to obtain detailed origin-
destination flows within Montgomery County. The sample would have been even more limited if
only P.M. peak trips were used. Instead, the 1990 Census Transportation Planning Package
(CTPP) data provided a better sample of work trips within the county. In order to use the CTPP,
trips were factored to reflect 1998 conditions during the P.M. peak period.

Traffic Counts: The county has maintained a fairly extensive database of traffic counts, although
the counting program has been more limited in recent years. Traffic counts are primarily
intersection turning movements for the A.M. and P.M. peak hours on arterials. In addition, tube
counts were available for the Interstate roadways. Traffic counts from 1994 through 1998 were
assembled and reviewed for the extent of the coverage. One future source of traffic counts will
be the county’s Advanced Traffic Management System (ATMS) that automatically collects counts
using loop detectors.

Transit Ridership Counts: Rail boardings and alightings for each station in the county were
availble from the transit agency. The ridership counts did include the mode of arrival/departure so
that estimates of the auto-connect to transit shares for each station could be calculated. Bus
ridership counts were also available and were aggregated by line and for corridors.

Travel Time Data: A 1994 survey contained travel times on selected arterials obtained using the
floating car approach. These times were useful for reasonableness checks, although the limited
sample of roadways and the lack of multiple survey times reduces their reliablity. Future surveys
of travel times will be needed given the importance of travel time in the equilibration/feedback loop
process.

The study has taken a top-down approach to model validation. Figure 5 illustrates the steps used
to validate the model. Checks are applied for each of the individual model components, before
assessing the overall results of the model. This approach ensures that modeling error is not
propagated between steps in the model chain. Travel survey data provided the primary source
for assessing the trip generation, distribution, and mode choice components of the model. Traffic
and transit counts were used to assess assignment results.

In order to evaluate county-wide travel patterns, a set of screenlines were developed. Auto
volumes on the links crossing these imaginary lines are aggregated to determine how close the
model replicates actual traffic patterns. The screenlines selected for Montgomery County are
shown in Figure 6. and capture most of the key north-south and east-west movements. Some
smaller cutlines were also used to assess corridor forecasts.




                                                                                                   13
Figure 4




           14
Figure 5




           15
Figure 6




           16
Validation targets were based in part on Federal Highway Administration guidelines that set the
maximum error by functional type:

Freeways within 7%
Principal arterials within 10%
Minor Arterials within 15%
Other within 25%

Table 1 lists the 1998 observed and simulated traffic volumes for the P.M. peak hour at 12
screenlines. The overall target for screenline volumes was to match counts within 10%. The
results show that most screenlines are within this range. In particular, assigned volumes crossing
the major north-south screenlines of the Beltway and Montrose/Randolph Road are both only 3%
over the traffic counts.

Table 1
Comparison of Observed and Simulated Screenline Volumes
1998 P.M. Peak Hour

                                                                                Ratio of
No.    Screenline                       Count               Assigned Vol        Assn/Count
1      Potomac River                    16,305              15,593              0.96
2      D.C. Line                        30,122              34,497              1.14
3      P.G. County Line                 39,837              42,467              1.07
4      Frederick County Line            8,547               8,409               0.98
5      Howard County Line               9,389               10,347              1.10
6      Beltway                          52,707              54,326              1.03
7      Montrose Rd/Randolph Rd          53,948              55,874              1.03
8      Norbeck Rd                       18,490              21,180              1.15
9      Shady Grove Rd                   29,496              35,639              1.21
10&    Seneca Rd/Darnestown             22,026              24,419              1.11
11
12     Rock Creek                       19,914              19,529              0.98




                                                                                                  17
Next Steps
The model improvement program is a continuous process and model validation will need to be
performed as additional refinements to the process are made. Three areas that will continue to
demand attention in the future include:

   GIS/Travel Model Linkages:            There is a continuing need to interface between the M-
    NCPPC data that is stored in ARC/Info and the data formatted for EMME/2. GIS is cuurently
    used to develop some of the zonal inputs to the model, and for displaying final results. The
    development of a comprehensive database of transportation facilities, traffic counts and
    speed data, and other travel data will be critical for the continued update and maintenance of
    the travel demand model.

   Coordination with Metropolitan Planning Organizations: As a county planning agency,
    M-NCPPC depends on the work of the two regional MPO’s, MWCOG and Baltimore
    Metropolitan Council, for collecting regional land use forecasts and the regional transportation
    network. To avoid duplication of data collection and model development, M-NCPPC will
    benefit from greater consistency between the regional and county modeling practices.

   Development of A.M. Peak and Daily Analytical Tools:           While TRAVEL/2 is an excellent
    tool for assessing the P.M. peak conditions, there is a need to forecast A.M. and daily travel
    conditions. The addition of other time periods may require some changes to the current
    model components, particularly trip generation.




                                                                                                  18
References

Barton-Aschman Associate, Inc. and Cambridge Systematics, Model Validation and
Reasonableness Checking Manual, Travel Model Improvement Program, February 1997.
Chicago Area Transportation Study, Procedures for Multiple Time Period Daily Highway
Assignments, Memorandum from Ronald W. Eash to Interested Parties, July 17, 1997.
INRO Consultants, EMME/2 User’s Manual Release 9, August 1998, Montreal.
Ismart, Dane, Calibration and Adjustment of System Planning Models, U.S. DOT Federal Highway
Administration, Publication No. FHWA-ED-90-015, December 1990.
Levinson, D.M. and A. Kumar. Multimodal Trip Distribution: Structure and Application. In
Transportation Research Record 1466, TRB, National Research Council, Washington, D.C.,
1994, pp. 70-77.
Levinson, D.M. and A. Kumar. Integrating Feedback into the Transportation Planning Model:
Structure and Application. In Transportation Research Record 1413, TRB, National Research
Council, Washington, D.C., 1992, pp. 70-77.
Levinson, D.M. and A. Kumar. Specifying, Estimating and Validating a New Trip Generation
Structure: A Case Study of Montgomery County, Maryland. In Transportation Research Record
1413, TRB, National Research Council, Washington, D.C., 1992, pp. 107-113.
Levinson, David M., TRAVEL/2 Flow Chart, Montgomery County Planning Department,
Transportation Planning Division, Transportation Modeling Group, 1995, Silver Spring.
Levinson, David M., TRAVEL/2 Model : User’s Guide, Montgomery County Planning Department,
Transportation Planning Division, Transportation Modeling Group, 1991, 1994 update, Silver
Spring.
Montgomery County Planning Department, TRAVEL/2 Draft Report, A Simultaneous Approach to
Transportation Modeling, Montgomery County Planning Department, Transportation Planning
Division, Transportation Modeling Group, 1993, Silver Spring.
Spiess Heinz, A Logit Parking Choice Model with Explicit Capactities, EMME/2 Support Center,
1996, Aegerten.
Tremblay Pierre, Appréhension de la congestion routière dans un modèle statique de simulation,
le cas du Modèle de Transport de la Région de Montréal (MOTREM), Revue de l’AQTR, Hiver
1993, pp. 12-21, Montréal.
Walker R., Pederson C. and Higgins S., Equilibrating the Input and Output Impedances in the
Demand Modeling Process, Portland Metro, prepared for the Tenth Annual EMME/2 Users’ Group
Conference, November 1995, Portland, Oregon.
Webster F.V and Cobbe B.M., Traffic Signals, Road Reseach Laboratory, Ministry of Transport
Road Research, Technical Paper No. 56, 1966.




                                                                                               19

								
To top