Main Page

From MSTM
Revision as of 06:55, 24 February 2015 by Mradovic (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

THE MARYLAND STATEWIDE
TRANSPORTATION MODEL (MSTM)

This wiki site is an "unofficial" site designed to facilitate and promote collaboration, research and public outreach of the Maryland Statewide Transportation Model. This site is not funded by the Maryland State Highway Administration (SHA), and comments and discussions may not necessarily reflect those of SHA or the Maryland Department of Transportation (MDOT).

Overview

The national capital region is experiencing higher growth in traffic, and new developments have spread farther from central cities causing increased demand for transportation services in developing areas and placing strains on what once were rural road networks. Efficient planning is required to understand the interactions between these changing land use patterns and traffic and to be able to develop strategies which will mitigate the effects of growth. Baltimore Metropolitan Council (BMC) and Metropolitan Council of Governments (MWCOG) are the two Metropolitan Planning Organizations (MPOs) in the region which currently have transportation models. The travel demand models of BMC and MWCOG are well suited for their respective jurisdictions<ref>Member Jurisdictions of the Baltimore Metropolitan Council</ref> <ref>Member Jurisdictions of the Metropolitan Washington Council of Governments</ref>

However, there are issues which must be addressed in the context of a multi-state region. These include; (1) the interaction of travel on the boundary between the two MPOs, (2) modeling of transportation in regions outside the MPO boundaries such as Western Maryland or the Eastern Shore of the Chesapeake Bay and (3) estimating the impact of travel which passes through the multi-state area, particularly freight travel. These issues can be partially addressed by MPO models (or in some cases not addressed at all). To fully address the issues it requires a broader view supported by multi-state analytic procedures.

It is estimated that Maryland's population will grow by approximately 1.1 million by year 2040. Employment is also growing, and during this same timeframe it is expected that approximately 0.4 million jobs will be added. State agencies such as the Maryland Department of Transportation and the Maryland Department of Planning are proactively working to accomodate this growth in the most efficient manner, whil minimizing the strain on the transportation system. After Several years of developmetn, calibration and validation, the Maryland Statewide Transportation Model (MSTM) has been integrated into many SHA's policies, programs, and projects (providing an analytical framework to inform decision makers). MSTM is a multi-layer travel demand model working at national, statewide and urban levels to forecast and analyze key measures of transportation system performance

MSTM provides comprehensive demand modeling & forecasting capabilities in and around Maryland for:

  • corridor status
  • freight movement
  • transportation/land use charges
    Overview of MSPM Application
  • impacts of external factors and policies
  • long-range plan and project prioritization
  • transportation system performance measurement * scenario analysis
  • support for MPOs and local initiatives

The model privides an analytic framework that can facilitate Maryland's policies and address the State's transportation challeneges.

As Maryland continues to grow, providing a high quality, reliable transportation system will be vital. Careful Planning and data driven processes are critical in order to ensure safety, mobility, and multi-modal choices for the movement of people and goods. The Maryland Statewide Transportation Model (MSTM) provides the analytical framework that can facilitate Maryland policies and transportation challenges.

Background

The Maryland State Highway Administration (SHA) has developed a statewide transportation model that (1) will allow consistent and defensible estimates of how different patterns of future development change key measures of transportation performance, and (2) can contribute to discussion and other evaluation tools that address how future transportation improvements may affect development patterns.

The Maryland Statewide Travel Model (MSTM) is by design a multi-layer model working at a Regional, Statewide and Urban level (Figure 1-1). The Regional Model covers North America, the Statewide Model includes Maryland, Washington DC, Delaware and selected areas in Pennsylvania, Virginia and West Virginia, and the Urban Model which serves to link for comparison purposes only, the urban travel models where they exist within the statewide model study area, for instance by connecting MSTM with the Baltimore Metropolitan Council (BMC) Model or the Metro Washington Council of Governments (MWCOG) Model.

This documentation is a User’s Guide focusing on the implementation of the Regional and the Statewide Model components. Past and future efforts strive to compare MSTM model results to MPO models and data at the Urban level. Every level is simulated to study travel behavior at an appropriate level of detail. The interaction of the three levels potentially improves every level by providing simulation results between upper and lower levels. All MSTM assignment of the travel demand occurs at the Statewide level.

At the Statewide Level, there are The 1588 Statewide Model level Zones (SMZs) that cover Maryland, Delaware, Washington DC, and parts of New Jersey, Pennsylvania, Virginia and West Virginia (Figure 1-2). The 151 Regional Model Zones (RMZs) cover the full US, Canada, and Mexico. RMZs are used for the multi-state commodity flow model and the long distance passenger model only and are eventually translated into flows assigned to networks and zones at the Maryland-focused (SMZ) level.

Model Steps

A crucial input for the model is travel time on the network. Initially, congested travel times were assumed based on free-flow speed, link length, area type and facility. Congested travel times were an exogenous input that did not change with congestion. To overcome this shortcoming, a feedback loop was implemented that uses travel times calculated by the assignment and feeds them back into trip generation.

Statewide Model Zones (SMZs)

Transit skimming and transit assignment are not included in the feedback loop, as these two processes do not affect highway travel times, nor do transit travel times change with congestion. As these two transit modules are computationally relatively intensive, excluding them from the feedback accelerates a model run.

The initial skim values are calculated using free-flow travel time. All subsequent modules use these skim matrices. After the assignment has been completed, skim matrices are recalculated using the travel times generated in the assignment. To avoid oscillating model results, the new highway skims are not used directly but rather averaged with the previous skim values. By using the average between the previous skim values and the recalculated skim values, changes happen more gradually and the model is able to converge more quickly.

Trip Generation

Person trip generation follows the same basic approach as the BMC model and encompasses the same trip purposes. The trip production component was updated to use household characteristics and trip rates derived from 2007-2008 HTS data and more recent Census data. The trip attrac-tion component is based on linear regression equations derived from the same household survey data.

Iterative Proportional Fitting (IPF)
MSTM person trip generation model uses trip production and attraction rates by household size (SIZ) by income (INC) and households workers (WRK) by income (INC). Since the SMZ data only provides households by income (see Section 4), a pre-generation step is applied to generate these joint distributions for the scenario year. An iterative proportional fitting (IPF) process combines the SMZ household data for the scenario year as marginals with joint-distribution seeds (from 2000 Census PUMS) to create households by SIZ and INC and households by WRK and INC at the SMZ level for a specified scenario year.


Non-Motorized Trips

The Maryland Statewide Transportation Model (MSTM) generates motorized trips only. Walk and bike trips are generated by trip generation, but shall not be included in trip tables for subsequent modules.

Non-Motorized Share
(HBW-1 trips)


A certain share of trips is dropped before trip productions and attractions are fed into the destination choice model. Previously, the MSTM model applied Weibull functions to estimate the non-motorized shares by area type and purpose. Plotting these shares showed unexpected patterns, which affect trip origins, mode choice and the assignment results. To mitigate the impact, non-motorized shares were averaged across counties. This resulted into reasonable patterns non-motorized shares, however, the was a steep border effect were two neighboring zones in different counties may have very different non-motorized shares, while all zones within one counties were treated as being equal in terms of non-motorized shares.


In this phase, the 2007 Household Travel Survey was used to estimate the non-motorized share by zone. A multiple regression was set up to analyze the impact of various measures of densities and accessibilities on non-motorized shares at the zonal level.

Trip Distribution/Destination Choice

The destination choice model predicts the probability of choosing any given zone as the trip attraction end. The model was estimated in a multinomial logit form using the ALOGIT software. These models are preceded by the trip production models, which forecast the number of productions by zone for different trip markets, chiefly identified by purpose and household income level. The destination choice models include mode choice logsums, distance terms, zonal employment, household characteristics and region geographic characteristics. The destination choice formulation is used for all purposes except for Home Based School (HBSCH), which uses a gravity formulation.

Estimation Dataset
The combined household travel surveys (HTS) in the MWCOG and BMC regions constitute the backbone of the estimation dataset. No travel behavior data is available for people residing outside of these two metropolitan areas. Information about trip characteristics obtained from the household survey includes trip production and attraction location, purpose, household income and auto ownership and departure time. While the surveys provide considerably more detail about trip-makers and their households, the models are limited to the attributes forecasted by the trip production models. Mode choice logsums and distance skims from the current version of the statewide model provide the trip impedance information. In addition, various terms identifying the region where the trip starts or ends were developed. These terms identify the metropolitan area (Washington DC or Baltimore) and the area type (CBD, Urban, Suburban, Other), as well as whether a bridge crossing is required.

Since there are a large number of destination alternatives, it is not possible to include all alternatives in the estimation dataset. A sampling-by-importance approach was used to choose alternatives sets for each trip. Each trip record was duplicated 10 times and different choice sets with 30 alternatives each were selected based on the size term and distance. This approach is nearly statistically equivalent to selecting 300 alternatives as the choice set of each trip, once a sampling correction term is applied in estimation.

Mode-Choice

Person trip mode choice is an adaptation of the most recent BMC nested logit mode choice mod-el, shown in Figure 6 1 . The modes defined in Section 4.2, Consolidated Network Develop-ment, were aggregated into these nests. The figure indicates the modes and sub-modes that are incorporated in the model. Rail includes LRT and Metro and the Commuter Rail (CR) includes AMTRAK services as well as MARC commuter rail. All local bus services are included under the Bus and express bus and commuter bus services are included in the ExpBus modes.

Mode-Choice Model Structure

                                    Person
                _______________________|_____________________
               |                                             |
              Auto                                       Transit(b)
            ___|___                          ________________|_______________
           |       |                        |                                |
          SR(b)   DA                      Walk                             PnR(b)
         ___|___                  __________|_________             __________|_________
        |       |                |      |      |      |           |      |      |      |
        SR2   SR3+(b)           Bus    Rail(b)  CR(b)  Exp(b)    Bus   Rail(b) CR(b)  Exp(b)

(b) = carries bias coefficient

Mode choice is based on generalized utility functions for auto and transit travel. Separate utilities were developed to represent peak and off-peak conditions. Home-based work trips and Non-home based work trips are based on peak period travel characteristics while other purposes are based on off-peak characteristics. Auto utilities for each auto mode include driving time and cost, terminal time and parking costs at the attraction end, and tolls. Transit utilities for each transit mode include walk and drive-access times, initial wait time, in-vehicle time, and transfer time. Bias constants or mode specific constants are included as indicated in Table 6 4 and Table 6 5 below which list all the variables included in the utility expression for each mode and sub-mode.

Mode-Choice
Validation

The mode split model has been calibrated to resemble the mode split observed in the survey. As no independent data were available, a true validation of mode split was not possible. Instead, a comparison of survey data and model results shows that the mode split model was calibrated to resemble observed travel behavior. The following figure compares survey and model results for every trip purpose. Given that the statewide model covers a highly heterogeneous study area with parts that have excellent transit service and other parts with almost no transit access, the comparison shows a reasonable picture.





Regional Person Model

A long-distance model called Nationwide Estimate of Long-Distance Travel (NELDT) has been implemented to cover long-distance travel. The model was presented at the Transportation Research Forum, an exchange with international researchers helped to further advance the model design.

Regional Trips
50+ miles

This new person long-distance model that is now implemented for MSTM covers all trips traveling a one-way distance of 50 miles or more. In other words, this model handles External-External, External-Internal, Internal-External and Internal-Internal long-distance trips. The following graphic shows the 50 mile range around downtown Baltimore and downtown Washington DC Trips between the two metropolitan areas are within the 50 mile radius, and therefore, covered by the short-distance model. Other trips that exceed the 50 mile range are simulated by NELDT.
In 2001/2002, the Federal Highway Administration conducted the National Household Travel Survey (NHTS) [10], which collected data on both daily and long-distance travel within the U.S. [11]. The survey consisted of 69,817 telephone interviews conducted from March 2001 to May 2002. Respondents were asked about their daily travel patterns (short distance) as well as any travel within the past 28 days where the furthest destination was 50 miles or more away from their home (long distance). This data set offers a rich source of information for long distance trips by all modes of transportation within the U.S. A total of 45,165 (raw count) long distance data records are available. In 2010, FHWA published a new NHTS conducted in 2009 [11]. This time, however, interviews focused on daily traffic only, without a special survey for long-distance travel. From this dataset, a total of 28,246 records (raw count) with trip length over 50 miles are available. An analysis of available data records shows that the smaller number of records and the different survey format makes these data unusable for long-distance travel in Indiana. While the NHTS 2002 asked people about their long-distance travel in the last 28 days, the NHTS 2009 asked about trips in a 24h period. As a consequence, long-distance travel is underrepresented in the NHTS 2009.

Table of NHTS 2002 long-distance records of Maryland Residents

Maryland 202 New York 27 Abroad 14 Florida 6 SouthCarolina 4 Missouri 2 Michigan 1
Pennsylvania 103 Delaware 25 North Carolina 27 Massachusetts 6 Washington 4 Arkansas 1 New Mexico 1
Virginia 78 West Virginia 20 Ohio 7 Arizona 5 Colorado 1 Hawaii 1 Tennessee 1
DC 43 NJ 219 California 6 Nevada 4 Georgia 2 Iowa 1 Texas 1

TOTAL RECORDS: 593

Table 7 1 summarizes the number of NHTS records for Maryland by destination state. While the number of records is relatively small for travel demand modeling, this area is represented in the NHTS fairly well in comparison to other parts of the country. Particularly neighboring states, which are of most interest to traffic flows to and from Maryland, are fairly well represented. Air travel data are published by the Bureau of Transportation Statistics based on ticketed passen-gers [12]. These data provide a ten percent sample of ticketed passengers between all U.S. air-ports, distinguishing between passengers changing flights and passengers having their final des-tination at one airport. Data are available by quarter, and to ensure compatibility with the NHTS data, air travel data was retrieved for 3/2001, 4/2001, 1/2002 and 2/2002.

Regional Truck Model

FAF Region Processing













Highway Assignment

Speed Feedback Loop

A crucial input for the model is travel time on the network. Initially, congested travel times were assumed based on free-flow speed, link length, area type and facility. Congested travel times were an exogenous input that did not change with congestion. To overcome this shortcoming, a feedback loop was implemented that uses travel times calculated by the assignment and feeds them back into trip generation.


Transit skimming and transit assignment are not included in the feedback loop, as these two processes do not affect highway travel times, nor do transit travel times change with congestion. As these two transit modules are computationally relatively intensive, excluding them from the feedback accelerates a model run. The initial skim values are calculated using free-flow travel time. All subsequent modules use these skim matrices. After the assignment has been completed, skim matrices are recalculated using the travel times generated in the assignment. To avoid oscillating model results, the new highway skims are not used directly but rather averaged with the previous skim values. By using the average between the previous skim values and the recalculated skim values, changes happen more gradually and the model is able to converge more quickly.


Figure 10 2 shows the convergence of the feedback loop by iteration. The x-axis shows the iteration, and the y-axis shows the percent root mean square error (%RMSE) between the skim values of two subsequent loop iterations. If the %RMSE is 0, the skim values did not change when using the speed of the latest two assignments. A non-zero %RMSE indicates that the resulting speed of the assignment has been different from the speed used to calculate the skim values. The blue line shows the %RMSE of a model setup in which the speed of the latest assignment is used to calculate the skim values directly. The %RMSE is reduced continuously over the first six iterations, and then starts oscillating around 12%. As a test, 75 iterations were run and the %RMSE did not improve much over 12%; therefore, the graphic only shows the first 16 iterations. The red line in Figure 10 2 shows the convergence of a feedback implementation, in which the revised skim matrices are averaged with the skim matrix from the previous iteration. The model does not oscillate and reaches convergence after a couple of iteration. The latter version has been implemented in the MSTM model.

Model Applications

Even during the early development stages of MSTM, the focus remained on how this tool will inform SHA processes and provide the analytical framework for data-driven transportation solutions.


In the Fall of 2013, the MSTM underwent a review from a panel of national experts. This "Peer Review" was conducted by FHWA, as part of their TMIP program. The focus of this review was to lay the groundwork for future enhancements and developments of the MSTM. The goal of the review was to maintain the mission of the MSTM and its role in integrating with other tools in providing the necessary analytical framework for data-driven decision-making.

Part of this discussion was identifying several real-world "case-studies" that are facing our region. These transportation challenges are ones that are not easily addressed and will require coordination among various agencies and integration of several analytical tools


Ongoing Development

Population Synthesizer

Auto Ownership Model

Related Research

Analytic DTA

The Maryland State Highway Administration (SHA) has developed the Maryland Statewide Transportation Model (MSTM) to analyze traffic issues throughout the state including traffic in rural areas, freight movements and travel between the Baltimore and Washington Metropolitan areas. Like most statewide models it relies on a static assignment procedure for four time periods and does not account for dynamic network conditions. In addition, like other static models, all travel is assigned to one time period and is assumed to move from origin to destination, completing the trip, within that time period.

An issue facing the MSTM and other statewide models is how to account for the dynamic conditions of the network and to understand the impact of these conditions on trips which span multiple time periods.

A typical method for dealing with temporal network conditions is the Dynamic Traffic Assignment (DTA). The DTA moves vehicles on a second by second basis, accounting for intersection delays, signalization and other network conditions. However the DTA requires detailed network information on signal location, timing and in some cases intersection configuration. In the MSTM with the large number of links (more than 167,000) and large number of signalized and unsignalized intersections to be accounted for, the tasks of data collection and validation would be daunting if not impossible. In addition, current software and computing capability the application of a DTA to a network the size of the MSTM would require massive computational capability and computer time. Yet, with these problems, the need to understand temporal issues remains.

As a way of improving the understanding of daily network conditions but avoiding the problems of detailed data collection, validation and computer run time, the Federal Highway Administration and the SHA have provided support to the University of Maryland to develop a proof of concept to test whether a variant of DTA method can improve the understanding of dynamic conditions without the detailed data collection and calibration typically required by most DTA applications. In this study we track individual vehicles but move them at fifteen minute time intervals rather than second by second and use the BPR functions to update link delays. This work is in progress and the project is in the validation phase. Preliminary results suggest that this method provides better estimates of dynamic traffic conditions, enables tracking of individual vehicles, and allows for the rerouting of vehicles in anticipation of congested conditions, while avoiding the need for detailed intersection information. The method also allows for computation within a reasonable amount of time. The method can not only be used as a planning tool but can also be used to provide improved input into small area simulations or detailed DTA applications.

Sources

<references />