Act Now: An Incremental Implementation of an Activity-Based Model System in Puget Sound

1
Act NowAn Incremental Implementation of
anActivity-Based Model System in Puget Sound

• Presented to
• 12th TRB National Transportation Planning
  Applications Conference
• May 19, 2009
• Presented by
• Maren Outwater, PSRC
• Chris Johnson, PSRC
• Mark Bradley
• John Bowman
• Joe Castiglione

2
PRESENTATION OVERVIEW

• PSRC model development strategy
• Activity-based models
• Activity generator technical approach
• Model calibration validation
• Model application

3
PROJECT CONTEXTPSRC MODEL DEVELOPMENT

• Short-Range
• Expand time periods
• Expand purposes
• Expand modes
• Calibrate

• Mid-Range
• Develop activity-based travel demand model
• Replace land use models
• Integrate economic, land use, activity-based
  models
• Benefit-Cost Analysis Tool
• EPA MOVES/Mobile models

• Long-Range
• Dynamic traffic assignment
• Continuous time
• Weekend
• Scenario evaluation tool

4
4-STEP MODEL LIMITATIONS

• Insensitive to
• Interactions among trips, tours (trip chains)
• Interactions among persons in HH
• Aggregation biases
• Demographic / market segmentation
• Temporal
• Spatial
• Unable to answer key policy questions
• Insensitive in trip generation to pricing and
  climate change policies

5
ACTIVITY-BASED MODELS ADVANTAGES

• Better policy sensitivities
• Broader
• More behaviorally accurate
• Consistency
• Within person-day of travel
• Across persons in a household
• More detailed information
• Travel choices
• Impacts on travelers

6
ACTIVITY-BASED MODEL PROJECTS IN THE U.S.
7
AN INCREMENTALAPPROACH

• Replace parts of trip generation with
  activity-generator
• Integrate with current and new models
• Build upon PSRC model design, enhancement and
  development efforts
• Implement quickly

8
PSRC MODEL SYSTEM
9
INTEGRATE W/ CURRENT MODEL

• Land Use Allocation (Urbansim)
• Synthetic population
• Usual workplace location
• Zonal Data
• Distribution

10
KEY FEATURES

• Policy Sensitivity
• Transportation
• Land use
• Induced/suppressed demand (accessibility via
  logsums)
• Broader set of HH and individual attributes
  incorporated
• Transition to full activity-based model

11
ACTIVITY PURPOSES

• Work
• Usual other
• School
• By age group
• Escort (pick up / drop off)
• Shopping
• Personal business
• Meal
• Social / recreational

12
ESTIMATION

• 2006 HH Survey
• Processed into tours, trips, activity patterns
• Expanded, re-weighted
• Discrete choice logit models
• Vehicle availability
• Out-of-home activity purposes
• Number of primary tours
• Number of work-based tours
• Number, sequence, purpose of intermediate stops

13
IMPLEMENTATION

• Microsimulation models
• Household vehicle availability
• Person activity generation
• Stochastic application for all HHs / persons in
  synthetic sample
• Initially in Delphi, translated to Python
• Integration into overall model runstream

14
ACCESSIBILITY MEASURESMODE DESTINATION CHOICE
LOGSUMS

• Pre-calculated by Activity Generator
• Mode choice logsums
• Based on existing trip-based mode choice models
• Segmented by purpose, income, auto availability
• Used in destination choice modes
• Destination choice logsums
• Activity Generator uses destination choice models
  to pre-calculate mode/destination accessibility
  logsums for residence zones.
• Re-calculated at beginning of each global
  feedback iteration

15
SYNTHETIC POPULATION

• Synthetic population input to vehicle
  availability and activity generator model
• Produced by Urbansim (also predicts usual work
  locations)
• Based on 2000 Census PUMS
• Distributions regionally controlled
• Household size (1,2,3,4)
• Household workers (0,1,2,3)
• Household income (lt30K, 30K-60K,
  60K-100K,gt100K)
• 3.45 million regional residents

16
SYNTHETIC POPULATIONCALIBRATION VALIDATION
17
VEHICLE AVAILABILITY

• Predict number of motorized vehicles used by
  household (own, lease, other)
• 0,1,2,3,4
• Key inputs
• HH attributes
• Home-work mode choice logsums
• Usual work location accessibility information
• Residence location accessibility information
• Vehicles vs. potential drivers

18
VEHICLE AVAILABILITYCALIBRATION VALIDATION

• Observed data 2006 PSRC Household Survey

19
DAY PATTERN MODEL

• Jointly predicts for each person
• Number of tours by purpose
• Occurrence of additional stops by purpose
• Allow substitution between making additional
  tours and additional stops
• Balance between person-day-level and tour-level
  sensitivities
• Example Shopping
• Good access to stores -gt spread shopping across
  multiple stops and multiple tours
• Poor access to stores -gt concentrate shopping
  within fewer stops

20
DAY PATTERN MODEL

• Key inputs
• HH attributes
• Person attributes
• Residence land use and accessibility
• Workplace land use and accessibility
• Utility components
• Purpose-specific
• More tours and stops, regardless of purpose
• Purpose interaction effects
• Tours and tours
• Tours and stops
• Stops and stops

21
DAY PATTERN MODEL

• Exact number of tours by purpose
• Number and purpose of work-based subtours
• Number and purpose of intermediate stops
• Usual workplace location vs other work location

22
INTEGRATION WITH4-STEP PROCESS

• Activity generator replaces parts of trip
  generation step
• Integrated into model system run stream as an
  executable
• Activity generator outputs are converted to trip
  arrays for use in subsequent use in distribution,
  mode choice, assignment

23
INTEGRATION WITH4-STEP PROCESS

• Activity-based model outputs converted to
  trip-based model trip purposes
• HB Work
• HB School
• HB College
• HB Shop
• HB Other
• NHB Work simple origin choice models predict
  production end
• NHB Other simple origin choice models predict
  production end

24
ACTIVITY GENERATORCALIBRATION VALIDATION

• Goals
• Replication of key aspects of travel
• Reasonable regional network assignment results
• GPS-adjusted targets
• Under-reporting of trips in HH survey
• HH subsample vehicle-based GPS

25
ACTIVITY GENERATORGPS ADJUSTMENTS

• Adjust for under-reporting of travel
• Limitations
• Vehicle-based trips and HHs only
• Missing purpose information
• Model developed to predict probability that given
  type of trip was missing
• Binary logit
• Based on HH and trip attributes
• Probability converted into adjustment factor
• Factors constrained

26
ACTIVITY GENERATORGPS ADJUSTED TRIPS
27
ACTIVITY GENERATORTRIP GENERATION vs. ACTIVITY
GENERATION
28
ACTIVITY GENERATORCALIBRATION VALIDATION
29
ACTIVITY GENERATORCALIBRATION VALIDATION
30
MODEL APPLICATIONTRANSPORTATION 2040

• Regional Transportation Plan update
• Integrated model system
• Puget Sound Economic Forecasting model
• Urbansim
• Activity Generator-enhanced 4-step model

31
TRANSPORTATION 2040ALTERNATIVES

• Alt 1 Existing system efficiency
• Alt 2 Capital improvements
• Alt 3 Core network expansion and efficiency
• Alt 4 Transportation system management
• Alt 5 Accessibility and reduced carbon emissions

32
TRANSPORTATION 2040ALTERNATIVE INVESTMENTS
33
TRANSPORTATION 2040EVALUATION CRITERIA

• Mobility
• Finance
• Growth Management
• Economic Prosperity
• Environmental Stewardship
• Quality of Life
• Equity

34
TRANSPORTATION 2040VEHICLE AVAILABILITY
35
TRANSPORTATION 2040ACTIVITY GENERATION
36
TRANSPORTATION 2040VEHICLE AVAILABILITY
ACTIVITY GENERATION
37
CONCLUSIONS

• Activity generator can replace trip generation in
  a 4-step model
• Data requirements comparable to traditional trip
  generation
• Can be implemented and calibrated quickly and
  efficiently
• Provides enhanced model sensitivities, though
  effects were modest