Oregon Modeling Improvement Program

1
Oregon Modeling Improvement Program
TLUMIP Stage 3Progress and Work Ahead
Rick Donnelly, PB Consult Inc.
February 18, 2004
2
Agenda

• First generation model
• Second generation model
• Current issues
• Converting code to parallel processing
• Transitional model
• Workload
• Urban model integration
• Testing using applications
• Discussion and questions
• Break

85 minutes
3
First generation model
4
Client perspective

• Why?
• Model methods (outside Portland Metro area) were
  very outdated
• Could not meet new state and federal mandates
• Could not provide needed information
• Losing ability to effectively participate in
  decision-making process

5
The first generation models
Economic growth
Person Product Flows
Economic Model
LocationModels
TransportModels
Transport Costs
Consumption Costs

• Economic Model determines growth of state
  economy
• Location Models allocate production and
  transactions
• Transport Models estimate demand and allocate
  trips to routes
• Model components are linked in space and time

6
First generation model flow

• Difference in consumption costs at export
  stations affect amount exported (5 year lag).
• Final demand passed to location model which
  distributes production and spatial interactions
  (no lag).
• Dollar flows are converted to functional flows
  (tons) by transport category (regular and heavy
  goods) (no lag).
• Interzonal transport costs affect activity
  locations and spatial interactions (5 year lag).

costs
costs
Location
Location
Location
Model
Model
Model
costs
costs
costs
flows
flows
flows
Transport
Transport
Transport
Model
Model
Model
2005
2000
2010

• Transport model converts functional flows into
  trips and assigns them to combinations of
  operators (regular truck, heavy truck, alternate
  truck) and routes based on relative costs
  (multinomial logit).

7
Second generation model
8
Impetus for second generation models

• Establish fully integrated statewide model
• Explicit representation of economy, land use, and
  transport
• Linkages to environmental analyses and
  performance indicators
• Build on lessons learned from Gen1 modeling work
• Fit into OMIP framework
• Systems thinking approach

• Key criteria
• Flexible geographic scale
• Truly integrated components
• Hybrid formulation
• Dynamic activities
• Static economy
• Activity-based models
• Agent-based micro-simulation
• Tight consistency
• Affordable and tractable

9
Model structure

• Economic and demographic (ED)
• Production allocation and activity interaction
  (PI)
• Household allocation (HA)
• Land development (LD)
• Person travel (PT)
• Commercial travel (CT)
• Transportation supply (TS)
• Utilities

Spatial activity
Transport
Aggregate
Microsimulation
10
Appeal of microsimulation

• Flexibility in aggregation
• Shift burden from wetware to hardware
• Increased computational burden
• Reduced model complexity
• Permit more complete accounting
• Facilitate explicit treatment of influences
• Non-linearities
• Finer resolution
• Higher fidelity
• Enable sensitivity variation as source of
  dispersion

11
Treatment of space Macro scale
Nested zone systems Alpha (2,984 zones)Beta
(748 zones)Gamma (121 zones)Counties (36
counties)Regions (4 economic)
Collar zones
Model boundary

12
Willamette Valley
13
Treatment of space Micro scale
street
beta zone
arterial
freeway
centroid
alpha zone
cell
14
Utility signals

• Utilities (and generalized costs) used in
• Logit models
• Hazard duration models
• Ordered utility models
• Rules
• Parameterized technical coefficients
• Used for
• Allocations in aggregate components
• Probabilities in Monte Carlo selections (in
  microsimulated components)

15
Economic-demographic (ED) framework
16
Production interaction (PI) scale
Production allocation and activity interaction
(PI)
Operates at the beta zone level
17
Beta zones
748 Beta zones (609 in Oregon, 139 outside)
18
Production, exchange consumption
19
Lumber light industry
Activity, light industrial floorspace, SCTG 26
(wood products) price
20
Light lumber industry
Activity, skilled wages
21
Household allocation (HA) structure
22
Household allocation (HA) structure
Leave household
Income
Syntheticpopulation
n
Deaths
y
Join newhouseholds
New house-hold pool
n
Householdmove
y
Income
Updatesyntheticpopulation
Movingpool
Jobs
23
Location choice model structure
24
Household transition example
Example household 1
autoownership
incomeage
locationsex
empl.status
studentstatus
occupation
education
interval t
interval t2
25
Household transition example (Continued)
Example household 2
autoownership
incomeage
locationsex
empl.status
studentstatus
occupation
education
interval t
interval t2
26
Land development (LD) scale
Operates at the 30 m2 grid cell level
27
Considering transitions cell by cell
commercial
residential low
residential medium
industrial
vacant
demolish
addition/renovation
no change
medium density
28
Transition model structure
29
LD development

• Draws on first generation (UrbanSim) structure
• Some utility coefficients used directly
• Others reproduce transition rates
• Lane County
• American Household Survey
• Where UGB effects enter the model

30
Person travel (PT)

• Travel by each household member considered
  separately
• Generate list of trips for typical weekday
• Monte Carlo assignment of characteristics or
  states
• Behavioral (from choice probabilities)
• For each person in each household
• Assign activity pattern and duration based on
  zonal attributes and specific sensitivities
• For each home- and work-based tour
• Use utilities based on zonal attributes and
  specific sensitivities
• Assign primary destination zone
• Assign tour mode choice
• Determine intermediate stop(s) and location(s)
• For each trip on each tour
• Assign start time and link
• Assign end time and link

31
PT model structure flow
HA
synthetic households
day pattern model
activity duration
destination choice
tour mode choice
Intermediate stops
trip mode choice
start stop times
TS
multi-class assignment
32
Commercial travel (CT)

• Use intersector flows (annual ) from PI to
  depict origins and destinations by commodity
• Use a microsimulation process to generate
  discrete shipments of tours
• Capture important dynamics
• Trans-shipment
• Trip chaining
• Package those tours for network assignment
• Resemble reality

33
CT model structure and flow
PI
O/D flows by mode commodity
Select trans-shipment location
Trans-shipment rates
Interzonal distances
Shipment size distributions
Translate O/D ? discrete shipments
Allocate to origin, destination points
I/O make use tables
Firm locations
Carrier and vehicle class distributions
Assign to carrier vehicle classes
Average load weights
Tour optimization
Dwell times, shift limits
Interzonal travel times
Package for assignment
TS
Multi-class assignment
34
Transport supply (TS)

• For auto trips
• Frank-Wolfe static user equilibrium assignment at
  alpha or beta zone level
• Microassign each trip
• Assign to origin and destination link
• Calculate minimum cost path
• Utilities based on link attributes and
  person-activity-specific sensitivities
• Work once through entire trip list

• For transit trips
• Determine path probabilities through network
  components
• Consistent with tour mode
• Monte Carlo path assignment
• Work once through entire trip list

35
Current status

• Were pushing the envelope
• Work in progress
• Critical issues remaining include
• Microsimulation with heterogenous agents
• Behavioral dispersion successful
• More consideration of what is emerging
• Object-oriented perspective
• Run times straining practicality
• Parallel processing (a la DAF)
• Assess simplifying the models
• Systemic calibration ahead
• Temporal dynamics
• Emergent behavior

36
Implementation view
Linux cluster
ED
AO
PI
Ricks ThinkPad
Databasecluster
HA
6 GB
LD
60 GB
PT
CT
TS
37
The cluster

• Ad hoc cluster
• File and SQL server
• 4-way Pentium III Xeon server with 2 MB L2 cache
• 4 GB memory
• 100 GB disk space
• 8 number crunchers
• Dual 2.8-3.2 GHz Pentium IV processors with 533
  or 800 MHz FSB
• 2 GB memory
• 73 GB disk
• Wired together with isolated gigabit ethernet
  network

38
(No Transcript)
39
User interface
PC user interface
Receive decodeuser input
Sets and holdssystem state
Servlet (Controller)
User Action
Java serverpage (View)
SystemResponse
Updates context-aware display
Web container (runs on node 0 of Linux cluster)
Web browser (running onany machine anywhere)
TCP/IPnetwork
40
Current issues
41
Current issues

• Opportunities
• Converting code to parallel processing
• Transitional model
• Workload
• Urban model integration
• Testing using applications

• Constraints
• Still in development
• Cross-sectional (static)
• Dynamic (empirical)
• Parallel processing
• Learning curve
• Hardware
• Wetware

42
Distributed application framework (DAF)

• Conceptual basis
• Extends OOP approach ? objectstasks
• Consists of tasks, queues, and messages
• Very light-weight
• Completely implemented in Java (portable)
• Implementation
• Provides the connection between the TLUMIP model
  components and the web-application
• Instantiates objects that monitor and report on
  the model state
• Uses message queues to hold model output for
  display by the web-application
• Much more difficult than anticipated!

43
Transitional model
Current Gen2 model
Transitional model
ED
Nextinterval
tHA
tLD
Spatialactivity
PI
Transport
PT
CT
a.m. peakmid-dayp.m. peakoff-peak
DA
TS
TS
TS
TS
44
Urban integration
Summary integration
Model coupling
Statewidemodel
SE control totals externalflows
Transport costs disutilities
Urban travel model
Possible today
Possible using transitional Gen2 models
45
Urban integration
Complete integration
Increased spatial detail in area of interest
Capability unknown
46
Workload

• Existing commitments
• Finish Gen2 model
• Finish transitional model
• OTP Update
• Regional planning study for Jackson County
• Linn/Benton cross-county commute
• Embellishing Clark County coverage
• Fourth TLUMIP symposium
• Continued PSU support

• Pending commitments
• OTIA3 analyses
• Columbia River Crossing
• Metro 2060 Plan
• Lane COG 2050 Plan
• On the horizon
• Mid-Willamette Valley COG 2050 Plan
• Areawide sustainability test case

47
The team
ODOT staff
Bill Upton, Brian Gregor, Brian Dunn, Becky
Knudson, Jan Shearer, Ed Arabas
Consultant team
Applications like the BLS have been some of the
most important activities of the team
Peer review panel
Julie Dunbar, Kim Fisher, Bob Gorman, Frank
Koppelman, Gordon Shunk, David Simmonds, Michael
Wegener