1. Greater Golden Horseshoe Model
Peter Kucirek
Transportation Modeller
M2: Activity-Based Models
May 15th, 2017

2. Model area The Greater Golden Horseshoe Region Clients:
Lake Ontario
Toronto
Niagara Falls
Hamilton
Kitchener Waterloo
Peterborough
Barrie
The Greater Golden Horseshoe Region
Clients:
Ministry of Transportation (MTO), Systems Analysis and Forecasting Office (SAFO)
Metrolinx (formerly Greater Toronto Transit Authority), Planning and Policy

3. At a glance Activity-based long-term choice models
Activity-based daily tour model
Trip-based mode choice
Capacity, crowding, and reliability transit assignment
Microsimulated urban truck model
A “renaissance model”: robust implementation in several different areas, but not specialized in any one.

4. Model stats Model built on the Emme Modeller platform
67% Python code, 32% FORTRAN code, 1% Other
10-40 hour run time, depending on convergence criteria
8 million microsimulated persons
400,000 microsimulated firms
3,119 zones
389 matrices
5 time periods

5. Model Structure

7. Model pre-processors
Population synthesis (PopSyn3)
Firm synthesis

8. Model initialization
Initial traffic assignments from seed demand matrices
Initial transit assignments from seed demand matrices
Can be optionally skipped if run previously

9. Passenger demand model
Accessibility measure generation

10. Passenger demand model
Accessibility measure generation
Long-term choice

11. Passenger demand model
Accessibility measure generation
Long-term choice
Tour generation

12. Passenger demand model
Accessibility measure generation
Long-term choice
Tour generation
Trip generation

13. Mode choice model
7 trip purposes:
Home-based work
Home-based university
Home-based other
Home-based market (shopping)
Home-based school
Work-based other
Non-home based
Includes station choice sub-model

14. Transit assignment
More on this later

15. Commercial vehicles model
Truck tour generation
Vehicle and tour purpose
Tour start time
Tour-building procedure:
Next stop purpose
Next stop location
Stop duration

16. Traffic assignment
Multiclass traffic assignment with 13 user classes (10 auto, 3 truck)
2 levels of convergence:
“Coarse” applied during model iterations
“Fine” applied at the end of the run for analysis
Peak spreading model (used only in forecasting)

17. Optional, special market models
Airport passenger model
Covers airport workers, resident and nonresident business and leisure travel
Special events model
Large scale sports events such as the Pan-Am games
Visitors model
Tourism

18. Looping structure and entry points
Model step
Model initialization
Passenger demand model
Mode choice model
Transit assignment
Traffic assignment

19. Looping structure and entry points
Model step
Cold start
Model initialization
Warm start
Passenger demand model
Half start
Mode choice model
Transit assignment
Traffic assignment

20. Looping structure and entry points
Model step
Loop level
Cold start
Model initialization
Warm start
Passenger demand model
Half start
Mode choice model 1
Station capacity restraint
Transit assignment
Traffic assignment

21. Looping structure and entry points
Model step
Loop level
Cold start
Model initialization
Warm start
Passenger demand model
Half start
Mode choice model 1 2
Transit route choice
Transit assignment
Traffic assignment

22. Looping structure and entry points
Model step
Loop level
Cold start
Model initialization
Warm start
Passenger demand model
Half start
Mode choice model 1 2 3
Mode equilibrium
Transit assignment
Traffic assignment

23. Looping structure and entry points
Model step
Loop level
Cold start
Model initialization
Warm start
Passenger demand model
Second-order effects
Half start
Mode choice model 1 2 3 4
Transit assignment
Traffic assignment

24. Mode choice nesting diagram
Auto Modes
Transit Modes
Nonmotorized Modes
Walk
Bike

25. Mode choice nesting diagram
Auto Modes
Drive alone
2-person shared
3-person shared
No Toll
Toll
No Toll
No HOV
No Toll
HOV
Toll
No HOV
Toll
HOV

26. Mode choice nesting diagram
Transit Modes
Local Bus & Streetcar
Rapid Bus
Premium Bus
Subway
Commuter Rail
Walk Access
Drive Access
Walk Access
Transit Access
Kiss & Ride
Park & Ride
Subway Access/Egress S1 S2 S1 S2 S3 S4

27. Capacity, Crowding, and Reliability Transit Assignment

28. CCRTA background
Originally estimated for Los Angeles and re-calibrated for Toronto
Takes into account “real” impacts of transit delay e.g. boarding delay due to full vehicles
Takes into account “psychic” impacts of discomfort due to unavailability of seats or crush loads
Consists of three components in the transit assignment and mode choice models:

29. Capacity
Extra waiting time from being unable to board a vehicle due to crowding
Vehicle bunching
Image credit: AARON LYNETT / TORONTO STAR

30. Crowding
Captures discomfort from crowded vehicles, seated vs. standing vs. crush capacity
Image credit: Laura Pedersen/National Post

31. Reliability
Captures effects of un-reliability of travel time between stops due to traffic interference
Image credit: Brad Ross

32. CCRTA results: transit volumes

33. CCRTA results: extra added wait time

34. CCRTA results: crowding penalty

35. CCRTA results: link unreliability

36. Traffic assignment results PM Peak

37. Model Implementation

38. Model interface Implemented as a Python package
Embedded in the Emme Python interpreter
Dual interface:
Graphical interface
Scripting interface
GUI (Modeller)
REPL (Jupyter)
Model API
Model components
Common
Models
Controllers
Main

39. Graphical User Interface (GUI)

40. Application programming interface (API)

41. Run management
Model parameters are 100% fully described in a JSON-based configuration file
Hierarchical, human-readable file format
Uses a standard syntax, permitting third-party analysis tools
Parameters located in one place rather than spread across multiple files
Different runs can be compared easily
Model utility parameters described as expressions

42. Model configuration example (auto ownership)
{ “1 adult”: [ "{‘No car’: -12.2, '1 car': 0, '2 cars': -2.2, '3 cars': -4.9, '4+ cars': -6.0}", "(hh.dwelling < 4) * {'No car': -2.2, '2 cars': 1.3, '3 cars': 2.1, '4+ cars': 2.5}", "(hh.persons.sum(age < 16) > 0) * {'No car': -0.3, '1 car': 0.1, '3 cars': -0.6, '4+ cars': -0.6}", ] }
Market segment name
Constants
Choice name
Coefficient
Coefficients on dwelling type
Household expression
Coefficients
Coefficients having children in the household

43. Model applications (on-going)
High occupancy toll lanes study
Dynamic tolling study
Regional express rail study
Regional transportation plan update

44. Contacts Peter Kucirek, M.A.Sc. Bill Davidson Transportation Modeller
Systems Analytics and Policy
Bill Davidson
Dunbar Transportation Consulting