1. Illinois Statewide Travel Demand Model Technical Approach
Joshua Auld, Behzad Karimi, Zahra Pourabdollahi, Kouros Mohammadian
October 10, 2014
Illinois Department of Transportation

2. Outline Introduction Methodology Data Collection Network Development
Long distance travel
Freight
External/Rural
Data Collection
Network Development
Simulation System
Future Tasks 2

3. Introduction 3

4. Statewide Models
Statewide models increasingly common as states struggle with complex transportation issues that must be studied from a system perspective
Portion of travel not modeled directly in MPO models:
34% of all person miles traveled from long-distance trips (NHTS 2009)
Freight trucks account for 10% of vehicle miles traveled (NTS 2011)
Provide the opportunity to evaluate the entire system in an integrated framework
Results in better understanding of travel behavior across the state
Cover area beyond MPO borders - crucial for future land use development
Help with better planning of transportation services for all modes
More efficient infrastructure planning and management for the
Demand management, supply management, safety, economic development, land-use 4

5. Statewide Models
Applied for projects ranging from local traffic to multi-state corridor studies. The range of applications include:
Statewide Plan: as a common ingredient to many components of a plan including transportation systems analysis, scenario analysis, economic benefits and environmental analysis.
Local Planning: The model could be used for local planning in the smaller rural communities.
Long Distance Corridors: The integration of long distance travel and freight movements can make the comparison of alternatives in a corridor possible.
Support for local and regional models: the statewide model can be used to find external trips, truck trips and other modes to be used in the smaller model.
Our study will focus on items 3 and 4. 5

6. UIC’s Project
MPOs across the state have developed sophisticated transportation demand models that are used to determine travel demand
Such models could greatly benefit from more complete long-distance personal travel and truck freight movement data.
Researchers at University of Illinois at Chicago (UIC) examine statewide long-distance travel and truck freight movement within the state and parts of the larger region that directly impact our state.
“Long Distance Travel” is defined similar to the National Household Travel Survey as those trips that are at least 100 minutes (approximately 75 miles or more).
MPOs can use the model to modify their existing TDM to more accurately depict travel and freight behavior, thus having better information and more efficiently plan or manage transportation system. 6

7. Current Status of Statewide Modeling
Current status of statewide models (Alan Horowitz) 7

8. Current Status of Statewide Modeling
Technical approaches used in statewide:
Traffic count and growth factor (e.g. Montana)
Four-step model
Activity- and Tour-based microsimulation model (e.g. Ohio and Oregon)
Cost to develop the model highly depends on the technical approach:
$25,000 for South Carolina to millions of dollars for states of Ohio and Oregon
considerable portion of spent costs and time in tour- and activity-based models goes to data collection efforts
It costs $3,500,000 for Ohio to collect needed data for the being revised model 8

9. Current Status of Statewide Modeling
Time to develop the model also depends on the technical approach:
6 months for Traffic count and growth factor model
8 years for integrated tour-based model
Statewide models are moving toward a more detailed zone systems and networks.
The second generation of Oregon statewide model, called SWIM2, has over 3,000 zones and over 53,000 links and it is while SWIM1 had only 125 zones and 2,000 links. 9

10. Model Development Through Collaboration
Given the size / scope of this work collaboration necessary
Collaboration on model development:
Argonne National Laboratory computing resources, network editing and simulation software
UIC: freight modeling, activity-based modeling survey implementation
MPOs: network and land-use information, demand model results
Result of this work should be useful to many agencies
IDOT for long distance planning purposes
External and freight trips for local MPO models 10

11. Current Status of Statewide Modeling
Market Segmentation is crucial to deal with heterogeneity:
Short- and Long-distance trips
Trip purposes
Combination of trip purposes in short- and long-distance trips can be very different
Freight and Passenger 11

12. Current Status of Statewide Modeling
Threshold distance between long and short distance trips varies from 50 miles (e.g. Oregon) to 100 miles (e.g. California)
Based on the following chart, 75 miles for Chicago and 50 miles for other part of Illinois was selected as the distance threshold
Data Source: NHTS 2001 12

13. methodology
A World-Class Education, A World-Class City

14. ILSTDM Development Methodology
Four primary components:
Long-distance passenger travel model
Freight model
Local travel demand
MPO results where available
Default activity-based model for other areas
Visitor & pass-through trips
A World-Class Education, A World-Class City

15. ILSTDM Methodology 15 1. Long-Distance Travel Model 2. Freight Model
3. MPO and External
4. Other Local
Population Synthesis
FirmSynthesis
OD Tables
Population Synthesis
Trip Frequency
(Generation)
Supplier Selection
Diurnal Curves
Activity Generation
Trip Distribution (Location Choice)
Shipment Size
Transims Convert Trips routine
Destination Choice
Mode Choice
Mode Choice
Mode Choice
Long-distance trips
Freight trips
Local/Visitor trips
Local trips
5. Network Simulation 15

16. Long-distance travel modeling
A World-Class Education, A World-Class City

17. Long-Distance Travel Modeling
Simulation of trips over 50/75 miles
Important for statewide planning
Accounts for a significant portion of trips on interstates and state highways
All trips on intercity bus, rail and air
Long distance travel simulated for all residents of Illinois and neighboring counties
Estimated using econometric activity-based model covering all primary modes
A World-Class Education, A World-Class City

18. Long-Distance Travel Model Framework
Primary inputs:
Census data (ACS and 2010 SF1
TAZ Land use data from MPOs
Congested Network skims
Person and intercept survey results
Five inter-related models
Generation, Distribution, Mode choice connected through logsums.
Conditional time-of-day choice
Population synthesis using PopSyn program developed for CMAP
Time of day choice
Long-distance Travel

19. Trip Generation
ZINB count regression models:
Gives annual work/non-work trips
Utilizes logsums from destination choice models
Estimated using weighted person travel survey results
Party size choice model
Ordered logit model for 1, 2, 3+
Annual trip counts by household then used as input to daily trip realization model

20. Trip Generation Discussion
Factors associated with higher trip rates:
Males, whites and high-income (all)
More vehicles (all)
More children (non-work trips)
Employment accessibility (work)
Destination log-sum (non-work)
Decreased trip rates:
Larger households (work )
Cultural accessibility (non-work)
Factors associated with higher zero trips
Low income (work)
Factors associated with lower zero trip probability:
Larger households and households with children (all)
College educated and male (work)
Employed and married individuals (non-work)

21. Destination Choice Models
Two-level destination choice model:
Region-choice utilizes TAZ choice logsum
20 regions (including external regions)
TAZ choice in region
Sample of regional TAZs
Uses mode choice logsum for TAZ

22. Destination Choice TAZ results
Factors increasing TAZ utility:
Increased population and employment
Cultural and recreational opportunities
Nearby employment
Access to university and recreational areas
Higher mode choice logsum
Factors decreasing utility:
Higher surrounding population
Higher zonal average income
Variable
Coefficient
t-stat
p-value
Population
0.117
4.20
Employment
0.597
18.32
Cultural Area
1.63
1.54
0.12 *
Major Recreational Area
1.21
3.40
Avg. Household Income
-0.094
-4.45
Population Accessiblity
-0.287
-2.02
0.04
Recreational Accessibility
0.024
1.68
0.09
Retail Accessibility
0.103
1.49
0.14
Total Employment Accessibility
0.177
6.93
University Accessibility
0.035
4.18
Mode Choice Logsum
0.951
4.00
Model Fit Statistics
Number of observations:
828
Likelihood ratio test:
866.6
Rho-square:
0.176

23. Mode Choice Models Two levels of MNL models:
Main mode choice – depends on access/egress logsums
Access / egress mode choice
Estimated using weighted SP/RP survey data
Modal constants calibrated to observed survey distribution

24. Time of Day Choice Model
Moving to daily travel model makes TOD component significant
Estimate segmented time-of-day choice for each long distance trip
Implemented using multinomial logit conditional on other choices
Uses data collected from household travel survey 24

25. Freight modeling 25

26. Freight Model Methodology
Firm Synthesis
Introducing individual decision-makers
Supplier Selection
Determining trade relationships/supply chains
Shipment Size
Using an iterative proportional fitting model
Mode Choice
Modal split between truck and rail
FAME Framework
Firm Synthesis
Supply Chain Formation
Logistics Decisions
Shipments Forecasting
Network Analysis 26

27. Freight Model Framework
Geographical Scale
National Scale: Domestic freight flows
Zone System (333 zone)
Township level zones in the Chicago area (118 zone)
County level zones in rest of Illinois (95 zone)
FAF zones in the rest of US (120 zone)

28. Freight Model Framework
Decision-making agents
Firms : the decision-maker units
Producer/Receiver of goods
Form supply chains
Specify logistics choices
Firm-types : a group of firms with the same
industry type
employee size
geographic location in the zoning system

29. Freight Modeling Framework
National Agent-Based Freight Model Framework
Economic Activity
Firm Synthesis Model
Zone System
CBP Data
Zoning
System
IO Accounts
Socio-
Economic
Factors
Freight Generation Model
List of Firms with Their Characteristics
Industry-
Commodity
Crosswalk
Economic Activity Data
Commodity Production Consumption Rates
IO accounts/ Industry-commodity crosswalk
Supplier Selection Model
Supplier Evaluation Model
Establishment Freight Survey
Annual Commodity Flow
(firm-to-firm)
Logistics Choices
Shipment Size / Frequency Choice Model
Main Mode Choice Model
Establishment Survey
Shipping Chain Configuration
(direct/non-direct shipping chains)
GPS data gathering
Number of Stops per Chain Model
Stop Type Model
Access/Egress Mode Choice Model
Interview Survey (specialists)
Vehicle Choice Model
Simulated Individual Shipments
Network Analysis
Empty Trucks / Backhauling
Network Assignment
Transportation Performance Measures

30. Economic Activity Overview
Zone System
CBP Data
Economic Activity
Firm Synthesis Model
IO Accounts
Zoning
System
Industry-
Commodity
Crosswalk
Socio-
Economic
Factors
Freight Generation Model
List of Firms with Their Characteristics
Economic Activity Data
Commodity Production Consumption Rates

31. Firm Synthesis and Freight Generation
7,687,522 business establishments Classified into 70,116 firm-type groups
Freight Generation Model
Commodity-industry crosswalk
Firm level production/consumption rates
Make-Use commodity-industry crosswalks
Number of establishments in zone
Size of establishments (employee size)
Data
Input-Output Accounts (BEA, 2013)
Freight Analysis Framework (FAF)
Commodity Flow Survey (CFS)
Synthesized Firm-types
Firm-Type: (17)
Zone
Number of Establishments
Menard County
NAICS
Construction of Buildings
Employee Size
1-19 employee

32. Logistics Choice Modeling Overview
Logistics Choices
Commodity Production Consumption Rates
Supplier Selection Model
Establishment Survey
Annual Commodity Flow
(firm-to-firm)
GPS data gathering
Supplier Evaluation Model
IO accounts/ Industry-commodity crosswalk
Establishment Freight Survey
Interview Survey (specialists)
List of Firms with Their Characteristics
Shipping Chain Configuration
(direct/non-direct shipping chains)
Shipment Size / Frequency Choice Model
Main Mode Choice Model
Vehicle Choice Model
Number of Stops per Chain Model
Stop Type Model
Access/Egress Mode Choice Model
Simulated Individual Shipments

33. Supplier Evaluation and Selection Model
A two-step modeling framework
Multi-criteria supplier evaluation model
To take into account decision makers’ opinions
To calculate suitability score for each potential supplier
Multi-criteria supplier selection optimization model
Maximize total suitability score of selected suppliers
Minimize total logistics costs
Meet the production capacity of suppliers and cover total demand of buyers

34. Shipping Chain Configuration Model
Shipping chain/Distribution channel/Transport chain
Modeling Approach
Rule-based decision tree clustering method
Growth method: Exhaustive CHAID algorithm
Number of intermediate stops in a chain & type of facility at each stop
Chemical manufacturing
Nonmetallic mineral product manufacturing
218
3 KT
Chemical and Pharmaceutical Products
One stop at a Distribution
Shipping Chain:
Center
Mode :Truck
Shipment size: 4K ~ 30K lbs
Actual weight: lbs
Annual frequency: 204

35. Network Analysis Framework
Network Assignment
Logistics Choice Models
Transportation Performance Measures
Empty Trucks / Backhauling
Simulated Individual Shipments
Network Analysis

36. External trips 36

37. External / MPO Trip Models
1995 American Travel Survey (ATS) used to compute base year long-distance trip distribution for U.S.
Iterative proportional fitting procedure updates base trip distribution to 2010 using Census data
Generate OD table for model zoning system
Combined with MPO OD tables
Converted to individual trips using Transims ConvertTrips utility + diurnal distribution assumptions
External trips in Gravity Model formulation to include sensitivity to network changes 37

38. Local travel 38

39. 4. Local Travel Model: ADAPTS ABM
Activity-based scheduling process model:
Bottom-up approach to activity-travel pattern formation
Activities generated, planned and scheduled dynamically
Planning process is explicitly modeled
Operationalized using multiple scheduling process surveys
Integrated activity-travel microsimulation
Dynamic, multi-day activity-travel simulation
Activities planned, scheduled and executed in single framework
Fully agent-based: all aspects implemented as individual agent behaviors – including routing and travel simulation
Currently Implemented in POLARIS model framework
Estimated based on Chicago-region data – not adjusted to rural area

40. Local Travel Model Overview
Preprocessing
Each Planning Time-step (5-min intervals)
In continuous time
Read Data and Scenario
Household Activity Generation
Generation Model
Destination Choice
Population Synthesis
Individual Activity Generation
Planning order model
Timing Choices
Routine and Preplanned Activity Scheduling
Check Activity Schedule
Activity Scheduling
Mode Choices
Gather Pre-trip info
Party Choice
Modify plans
Get Route
Schedule Departures
Simulation

41. Data collection 41

42. Household Survey
Collects trip data for long-distance trips at household level
Similar to American Travel Survey 1995 – conducted as part of NHTS
Collects:
Trip frequency
Travel modes
Trip type
Party composition
Used to estimate trip frequency models
Dependent on household characteristics
Destination characteristics
Mode characteristics
Approximated logsums (accessibility-based)
Can be combined statistically with NHTS and ATS to extend sample 42

43. Instrument Design Introduction
Verify correct contact information
Household roster and demographic information
Demographics for respondent
Demographics for other household members
Housing information
Trip screening questions
Number of trips in the past 12 months
Trip count by quarter, mode, purpose, party size
Commuting trips
Trip detail for last trips (work and non-work)
Start date, duration
Origin, destination, station access
Travel Modes
Purpose

44. Household Survey Long distance trip purpose Long distance Mode Choice44

45. Household Survey
Departure Time-of-day choice
Start time 45

46. Household Survey
HH Income
HH Income 46

47. Freight Data Sources Publicly Available Data Survey Data
County Business Patterns
Industry input-output accounts
Commodity Flow Survey
Freight Analysis Framework
Survey Data
UIC establishment survey, 1st wave (2009)
UIC establishment survey, 2nd and 3rd waves ( ) 47

48. UIC Establishment Survey (2010-2011)
Data Collection Method
telephone introductions
blast campaigns
web crawling

49. UIC Establishment Survey (2010-2011)
Survey Design
Participants: logistics or shipping managers of firms
Three major parts
Characteristics of the business establishment
Attributes of five most recent shipments
Contact information

50. UIC Establishment Survey (2010-2011)
Survey Results
Approximately 219,000 contacts nationwide
657 establishment surveys
970 useable shipment survey forms
1st wave
2nd wave

51. Data Acquisition from MPOs
Network
Zoning
Land Use
TDM Results
Chicago ü
Springfield
Champaign
St. Louis
Bloomington
Peoria
Quad-Cities
Danville
Decatur
Kankakee
Dekalb
Rockford
Dubuque 51

52. network development 52

53. ILSTDM Network Development
Approximately 90,000 links
Data sources:
MPO models
FAF2 Network
Illinois HPMS (IRIS)
Argonne Chicago network 53

54. Combining networks Widely varying networks depending on source:
1-way vs 2-way links
Missing capacity, lanes, speeds
Disconnects (especially in HPMS)
Very little traffic control information
Develop custom Python scripts to combine networks
Generate estimates for speed, capacity, etc. when not provided
Import all networks into a common database format
Sqlite open-source DBMS with Spatialite extensions
Compatible with Argonne Network Editing software 54

55. Network Editor
Adding a missing link
Correcting connectivity 55

56. Synthesized Intersection Controls
Generate signal/stop information using Transims IntControl program
Estimates signal/sign warrants given link connectivity, link capacities, link type and area type
Different outcomes by area type (Chicago, St.Louis, Other urban or rural) and primary/secondary street
Timing/phasing are then estimated using the warrants and a given signal type, cycle length, etc. 56

57. ILSTDM Zone System Development
5800 zones in 20 regions
Region (Superzone) Organization
TAZs
Census Tracts
Chicago
Rest of Northern Illinois
Chicago Suburbs
Rest of Central Illinois
St. Louis
Rest of Southern Illinois
St. Louis Suburbs
Counties
Champaign
Iowa - Neighboring
Springfield
Wisconsin - Neighboring
Quad Cities
Indiana - Neighboring
Peoria
Missouri - Neighboring
Bloomington
Kentucky - Neighboring
Decatur
States
Danville
Northeast (states)
Northwest (states)
Southeast (states)
Southwest (states) 57

58. Socio-Economic and Land Use Data Collection
Many areas lack required zone data
Process to generate employment and land use:
Extract employment by Zip Code from Zip Code Business Patterns
Plot Illinois Land coverage for ILGS, showing developed areas
Transfer zip code employment to developed areas through overlay (assumes equal distribution of jobs throughout the developed area in each zip code
Transfer employment from developed area shapes to zones through overlay
Aggregate to county level and perform IPF to match county totals
Follow same process for Census of Governments data
Land use overlay from ESRI points of interest/landmarks dataset 58

59. Simulation system 59

60. Simulation-Based Dynamic Traffic Assignment Model
Trips enter traffic simulation either:
Directly: Long-distance model, Freight model, ADAPTS
Indirectly from OD tables converted to trips by Transims
Network simulation contains:
Route Choice Model
En-route Switching Model
Traffic Control Model
Mesoscopic Traffic Simulation Model
Currently implemented in POLARIS agent-based simulation framework

61. Route choice model
One-Shot Assignment using prevailing travel information
Averaged experienced travel times in last interval (e.g. 5 minutes)
Travel times are output from traffic simulation model
Current implemented route choice model is pre-trip route choice model with enroute replanning
Pre-trip route choice model is for pre-trip users who use the travel time information based on current traffic conditions to find a shortest path from his/her origin to destination. (e.g. using google map to compute shortest path considering traffic at that time)
Shortest path algorithm: individual link-based A-Star Algorithm that takes care of delay at turn movement
Enroute replanning: travelers can switch routes at decision points based on experienced travel times using bounded rationality model
This solution maintains an approximation of an instantaneous equilibrium

62. Traffic Simulation Model
Newell’s Simplified Kinematic Wave model
Using cumulative curves
Capturing queue formation, spillback, and dispersion
Capturing shock wave
Adhering to the fundamental diagrams
Output:
Network flow pattern
Cumulative vehicles at upstream and downstream of a link
Vehicle trajectory (enter time and exit time of each link)
Network performance
Time-dependent link travel time by turn movement

63. POLARIS Simulation Environment63

64. Simulation Model Process
Population Synthesis
FirmSynthesis
OD Tables
Population Synthesis
Trip Frequency
(Generation)
Supplier Selection
Diurnal Curves
Activity Generation
Trip Distribution (Location Choice)
Shipment Size
Transims Convert Trips routine
Destination Choice
Mode Choice
Mode Choice
Mode Choice
Long-distance trips
Freight trips
Local/Visitor trips
Local trips
Network Simulation
Network Skims 64
Check Convergence
Results

65. Next steps
A World-Class Education, A World-Class City

66. Next Steps Estimate models using statewide data
Trip Generation, destination choice with new regions/zones, mode choice
Develop and estimate Time-of-day Models / diurnal distributions
Implement models in POLARIS simulation framework
Local travel ABM already implemented
Integrate long-distance travel, freight travel models
Model calibration and validation
Run model in calibration iteration scheme
Match to ground counts, survey observations where possible
Policy Scenario Analysis
Work with stakeholders to implement scenarios of interest for testing 66

67. THANK YOU!
Illinois Department of Transportation 67