presentation to the travel forecasting subcommittee ron milone and mark moran

TPB Travel Forecasting Model, Version 2.3:

Specification, Validation, and User’s Guide (Draft Report)

Presentation to theTravel Forecasting Subcommittee

Ron Milone and Mark MoranMetropolitan Washington Council of Governments

National Capital Region Transportation Planning Board (COG/TPB)

July 18, 2008

Ver2.3model_tfsPres_2008-07-18e.ppt

The Ver. 2.3 Regional Travel Model: A presentation to the Travel Forecasting Subcomm., 7/18/08 2

Overview of the presentation

• Overview of the Ver. 2.3 travel model• Overview draft documentation• Model release schedules• Detailed description of what is new in the Ver.

2.3 travel model– Truck models– Mode choice

• Summary of model performance• Conclusions and next steps


Overview of theVer. 2.3 travel model

• Base used for development: Ver. 2.2 travel model• Major enhancements

– Updated of the medium and heavy truck models– Replacement of the 5-choice, sequential, multinomial logit mode

choice model with a 15-choice, nested-logit mode choice model and the ability to conduct transit assignments

• Calibration years:– 2002 for NL MC– 2005 for truck model

• Validation years:– 2002 (transit)– 2005 (highway)– 2030 (for reasonableness)


Yet, some things remain the same: Ver. 2.3 vs. Ver. 2.2

• Both models share:– Modeled area (2,191 TAZ, 6,800 sq. mi.)– Trip-based, aggregate, four-step model– Four trip purposes (HBW, HBS, HBO, NHB)– Commercial vehicle model– Queuing delay function applied to freeways &

ramps– Ability to model both fixed-price and variably

priced toll facilities (e.g., toll roads, HOT lanes)


Overview of thedraft documentation

• Ver. 2.3 model is in draft and so is the documentation

• Report is divided into three main sections– Model specification and validation– User’s guide– Technical appendices

• Some sections of the documentation are unfinished, especially in the user’s guide


Model release schedules

Jan 2007 Jun 2008

Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun

Schedule for the Ver. 2.2 travel model

Jan-07 - Jun-07Sensitivity testing & model adjust.

01/19/07Ver. 2.2 model released w/ draft documentation

Oct-07 - Jan-08AQC 2007 CLRP

03/01/08Final documentation released

Jul 2008 Dec 2009

Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Proposed Schedule for the Ver. 2.3 travel model07/18/08

Ver. 2.3 model released w/ draft documentation

Aug-08 - Dec-08Sensitivity testing & model adjust.

Mar-09 - Jun-09AQC 2009 CLRP

Jan-09Release revised documentation

Apr-08 - Jun-08AQC 2008 CLRP

01/18/08Released 2nd draft documentation

Jan-09Make final adjustments to model

Sep-07Final revisions to model

2008Re-run of

AQC 2008 CLRPJuly-Nov

Ver. 2.3 draft documentation

Focus of this presentation:

What’s different between Ver. 2.3 and its predecessor, Ver. 2.2


Ver. 2.3: Revised medium and heavy truck models

• Motivation:– Previous truck models had not been updated recently– Air quality planning: Trucks are major contributors to

nitrogen oxides (NOx) and airborne particulates

• Difficulty:– Collecting observed travel patterns of trucks

• Consequently, many MPOs use either “borrowed” truck models from comparable regions or truck models comprised of default parameter values taken from the literature



• TPB initiated a project in FY-2005– To develop a new commercial vehicle model

(implemented in the Ver. 2.2 model) and – To revise the existing truck models using

locally collected data (in the Ver. 2.3 model)

• A consultant (Bill Allen) was retained to support this multi-year effort, which involved the development of an innovative truck modeling approach that obviated the need to collect operator-based information



• TPB travel models have historically distinguished two types of trucks:– Medium

• Definition: Single unit, two-axle vehicles with six tires (i.e., vehicles with dual rear wheels)

• Examples: Recreational vehicles (RVs) or medium-duty delivery trucks

– Heavy• Definition: Any multi-unit vehicle with three or more axles• Example: Tractor trailers

• These designations have been retained



• Structure– Trip generation and trip distribution models

• Results in daily truck trips

– Time-of-day process, before traffic assignment, results in AM, PM, and off peak

– Medium & heavy truck trips are assigned to the highway network as a separate and distinct market (trip table)

– More details to be found in Bill Allen’s presentation (next)



• Ver. 2.3 traffic assignment procedures have been modified, as a result of the new truck models– Ver. 2.2: Loading of five trip tables

• 1) SOVs and commercial vehicles; 2) 2-occupant HOVs; 3) 3+ occupant HOVs; 4) Total trucks - heavy and medium; 5) Air passenger vehicles

– Ver. 2.3: Loading of six trip tables• Same as Ver. 2.2, except medium and heavy trucks are now loaded

separately• Drawback: 6 trip tables means longer model run times• Advantage: Affords TPB staff the opportunity to explore the option of

treating medium and heavy trucks as varying passenger car equivalents (PCEs) in the assignment process. This option is but one of many that will be evaluated in the coming months.


Ver. 2.3: Nested-logit (NL) mode choice model

• A 15-choice, NL mode choice model (replaces the 5-choice, sequential, multinomial-logit mode choice model)

• Includes– Three auto modes

• Drive alone, shared ride 2, and shared ride 3+– Four transit modes

• Commuter rail, all bus, all Metrorail, and combined bus/Metrorail– Three modes of access to transit

• Park and ride (PNR), kiss and ride (KNR), and walk

Highway

DriveAlone

SharedRide

SR 2 SR 3+Comm.

RailBus/

MetrorailAll

BusAll

Metrorail

WalkAccess

PNRAccess

K&RAccess

Transit

Comm.Rail

Bus/Metrorail

AllBus

AllMetrorail

Comm.Rail

Bus/Metrorail

AllBus

AllMetrorail



• Although not explicitly listed, the NL MC model can also model light rail transit (LRT), bus rapid transit (BRT), and street car.

• TPB NL MC model was based on an earlier NL MC model developed by AECOM for WMATA in 2004-2005.

• We have retained some features of the AECOM/WMATA model and have changed some features:– Table on next slide compares the two NL MC models


Ver. 2.3: TPB NL MC vs. AECOM NL MC

Item AECOM/WMATA NL MC COG/TPB NL MC Travel modes and nesting structure

15 (3 auto, 12 transit) modes; 3 levels of nests

Same

Model application code AEMS Fortran program Same How the mode choice model is applied

As a post process to the regional travel model

Within the speed feedback loop of the travel model

Trip purposes 3 (HBW, HBS/O, NHB) 4 (HBW, HBS, HBO, NHB) Types of travel skims 2 (AM peak per. and off-peak per.) Same Number of mode choice models

6 (HBW AM, HBW OP, HBS/O AM, HBS/O OP, NHB AM, and NHB OP)

4 (HBW AM, HBS OP, HBO OP, and NHB OP)

Geographic market segmentation

7 superdistricts; 20 production/attraction interchanges

Same

Revised transit access coding

Many enhancements to transit access coding

TPB adopted most of these enhancements

Data used for calibration

2002 WMATA rail survey; 2000 regional bus survey; Boarding counts for express bus and commuter rail

Same

Calibration approach Calibrated by AECOM for 6 models applied as a post process

Re-calibrated by TPB staff for 4 models; Applied in the speed feedback loop



• Calibration data (assembled by AECOM; used by both agencies):– 2002 land use and networks– Control data

• 2002 WMATA Metrorail survey• 2002 published transit boarding counts by operator

– Other data• 2000 Regional Bus Survey• 2000 Census journey-to-work data

• This contrasts with the Ver. 2.2 MNL MC model:– Calibrated with the 1994 Household Travel Survey– Validated to 2000 Census Transportation Planning Package

(CTPP) data



• Definition of HOV trips has changed– SMNL (Ver. 2.2 and earlier):

• “HOV trips” coming out of the mode choice model referred to only those that use HOV facilities for a substantial portion of their trip.

• “LOV trips” included both drive-alone and carpools (provided the carpools did not use a preferential HOV facility).

– NL (Ver. 2.3):• LOV refers to only the drive-alone trips. • HOV refers to all shared-ride trips, irrespective of whether

they use an HOV facility or not



• The new NL MC model is applied using a Fortran program named AEMS.EXE, which replaces the COGMC.EXE program that has been used for years– AEMS is completely parametric, i.e., all

characteristics for any given MC model are specified in a control file.

– Characteristics represented in the control file include• nesting structure, market segmentation, utility/disutility

functions, and the values of coefficients and constants. – AEMS can handle models with any nesting structure

and up to 15 choices (a newer version can handle up to 18)



• Geographic market segmentation:– Seven superdistricts were

combined to make 20 production/attraction “market segments”

• There is one nesting constant for each market segment and each travel mode

• Calibration consists of estimating these nesting constants

• Automated calibration routine: CALIBMS

Production AttractionArea Area

1 DC DC core2 DC VA core3 DC Urban DC, MD, VA4 DC Suburban MD, VA5 MD urban DC core6 MD urban VA core7 MD urban Urban DC, MD, VA8 MD urban Suburban MD, VA9 VA core/urban DC core

10 VA core/urban VA core11 VA core/urban Urban DC, MD, VA12 VA core/urban Suburban MD, VA13 MD suburban DC core14 MD suburban VA core15 MD suburban Urban DC, MD, VA16 MD suburban Suburban MD, VA17 VA suburban DC core18 VA suburban VA core19 VA suburban Urban DC, MD, VA20 VA suburban Suburban MD, VA



• Revised transit access coding– The consolidation station file/database

• Six new fields/variables added

– Sidewalk links and zonal walk links– Zonal auto-access links– Station transfer links– Zonal percent walk to transit calculations


Ver. 2.3 NL MC choice model:Consolidation station file/database

Sequence # Mode

Access Distance Code PNR? Active? Station Name Centroid TAZ

Rail Node

Parking Node

Bus Node 1

Bus Node 2

Bus Node 3

Bus Node 4

Parking Capacity

X Coordinate

Y Coordinate

Peak Parking Cost

Off-Peak Parking Cost

Peak Shadow Price

Off-Peak Shadow Price

First Year

81 M 1 Y Y Shady Grove 2331 482 7301 7501 19019 19020 5253 1265919 528848 400 50 1990

82 M 2 Y Y Rockville 2332 479 7302 7502 19017 7605 659 1270919 516348 400 50 1990

83 M 2 Y Y Tw inbrook 2333 416 7303 7503 19013 19014 1075 1278519 507948 400 50 1990

84 M 2 Y Y White Flint 2334 405 7304 7504 19016 991 1280319 503348 400 50 1990

85 M 2 Y Y Grosvenor 2335 403 7305 7505 19012 650 1283219 494948 400 50 1990

86 M Y Medical Center 2336 346 7306 3054 0 1284819 484948 0 0 1990

87 M 2 Y Y Bethesda 2337 344 7307 7507 3048 500 1285519 479748 500 50 1990

88 M Y Friendship Heights 2338 204 7308 9140 20401 0 1287919 471348 0 0 1990

89 M Y Tenleytow n 2339 207 7309 9117 20403 0 1289519 466648 0 0 1990

90 M Y Van Ness-UDC 2340 113 7310 9153 0 1294219 465148 0 0 1990

91 M Y Cleveland Park 2341 117 7311 9156 0 1295619 462248 0 0 1990

92 M Y Woodley Park-Zoo 2342 117 7312 9163 20501 0 1297319 458248 0 0 1990

93 M 9 Y Dupont Circle 2343 46 7313 8901 8905 0 1299919 452648 0 0 1990

94 M 9 Y Farragut North 2344 17 7314 8440 20416 0 1301119 450348 0 0 1990

95 M 9 Y Metro Center 2345 19 7315 8912 8919 0 1304319 448448 0 0 1990

96 M 9 Y Gallery Place 2346 23 7316 8955 20118 0 1306119 448448 0 0 1990

97 M 9 Y Judiciary Square 2347 26 7317 8474 0 1307519 447948 0 0 1990

98 M 9 Y Union Station 2348 634 7318 8656 8654 7601 0 1310319 448248 0 0 1990

99 M 3 Y Y Rhode Island Ave 2349 146 7319 7519 9422 350 1313319 456848 350 50 1990


Ver. 2.3 NL MC choice model: Sidewalk links and zonal walk links• In Ver. 2.2 and earlier

models, there was a walk network, used for transferring from one transit line to another, in downtown DC and downtown Silver Spring, MD

• In Ver. 2.3, there is a sidewalk network in almost the entire modeled area

• Regional sidewalk network (Mode 13) is generated using a new Fortran program WALKACC.EXE

• WALKACC.EXE also generates zonal walk-access-to-transit links (Mode 16).


Ver. 2.3 NL MC choice model: Zonal auto-access links

• Fortran program AUTOACC4.EXE is used to generate auto-access-to-transit links. Auto access links (Mode 11) are a function of multiple criteria:

– Orientation toward downtown (defined as the Ellipse);

– A backtracking penalty and a prohibition of crossing the Potomac River (except for trips from Loudoun County to MARC commuter rail);

– A maximum link distance, which is a function of station type (e.g., terminal vs. non-terminal) and transit mode; See next slide.

– Manually specified overrides (e.g., pentagon.prn, which lists zones that should be connected to the Pentagon with KNR-access, or slugging)


Ver. 2.3 NL MC choice model: Zonal auto-access links

MaximumConnect.

Access LengthMode Dist. Code (miles)Metrorail station PNR 1 15Metrorail station PNR 2 5Metrorail station PNR 3 3Metrorail station PNR 0 3Commuter rail station PNR 1 15Commuter rail station PNR 2 10Commuter rail station PNR 0 5Bus PNR 1 5Bus PNR 0 3BRT/Street car PNR 1 5BRT/Street car PNR 0 3LRT PNR 1 5LRT PNR 0 3

Acc Dist Code Interpretation

1 End-of-the-line station (e.g., Shady Grove Metro)

2 Intermediate station (e.g., Rockville Metro)

3 PNR close to a CBD (e.g., Rhode Island Ave. Metro, Fort Totten)

0 Only KNR-access links generated (e.g., Braddock Road, National Airport, Clarendon)

9 Metrorail sta. in use, but no PNR/KNR access (e.g., Dupont Circle, Farragut North, Metro Ctr.)

8 Pentagon Metro Sta., allows for very long KNR links, to represent “slugging” (informal carpool)


Ver. 2.3 NL MC choice model: Station transfer links

• Station transfer links are walk links connecting– Stations and sidewalks (Mode 12)– Stations and bus service (Mode 12)– Stations and PNR lots (Mode 15)

• These links are generated automatically from data in the consolidated station file.

• For PNR-station transfer links (Mode 15), the walk time is a function of parking capacity and cost,

– Impedance (Mode 15) = (walking time from PNR to sta.) + (parking cost, converted to time @ $10/hr)

– walking time from PNR to sta. = function of lookup table, e.g., <= 500 spaces = 2.0 min; 500-1000 spaces = 2.5 min; etc.

– For details, see memorandum from Manish Jain – AECOM, “MWCOG network coding guide for Nested Logit Model.” February 2008.

• Mode 15 links are generated by the Fortran program PARKER.EXE. • Mode 12 links are generated by the Fortran program STAPROTP_V1.EXE


Ver. 2.3 NL MC choice model: Zonal percent walk to transit

• Zonal percent walk is the percent of a zone’s area that lies within walking distance to transit service

– Short walk: ≤ 0.5 miles– Long walk: > 0.5 miles and ≤ 1.0 mile– (Previous: Short ≤ 1/3; Long ≤ 1.0)

• The following walk designations are used:

– Short walk to Metrorail;– Long walk to Metrorail;– Short walk to AM transit;– Long walk to AM transit;– Short walk to off-peak transit;– Long walk to off-peak transit.

• A GIS procedure is used to develop point buffers around transit stop nodes and then overlay these point buffers with zone (TAZ) boundaries.


Ver. 2.3 travel model:Summary of model performance

• General observations on 2005 simulation– Ver. 2.3 Est/Obs VMT = 1.03 for the MSA

• By contrast, Ver. 2.2 was 1.00

– Ver. 2.3 total vehicle trips ≈ 1.5% higher than Ver. 2.2– Ver. 2.3 total VMT ≈ 4.0% higher than Ver. 2.2– Ver. 2.3 transit ≈ 3.0% higher than Ver. 2.2

• Details on calibration results, model performance can be found in the draft report


Conclusions

• Version 2.3 travel model is in DRAFT (beta release)

• Evaluation will proceed in the coming months


Next steps

• Begin sensitivity testing• Possible refinements to the model• Move from Cube TP+ ver. 4.1 to Cube

Voyager ver. 5.1• Implement a procedure for the applying

the transit constraint through the regional core

• Consistent networks• HOT lane evaluation

presentation to the travel forecasting subcommittee ron milone and mark moran

Documents

regional travel model

tpb travel forecasting

travel modelbase

model release schedulesthe

new commercial vehicle

nestedlogit mode choice

borrowed truck models

previous truck models