1. Integrated Dynamic/AB Models: Getting Real Discussion
Integrated ABM-DTA Model: First Application Experience and Lessons Learned
Peter Vovsha, Jim Hicks, Matt Stratton, Chrissy Bernardo, Rosella Picado (WSP)
Rebekah Anderson, Greg Giaimo (ODOT)
Guy Rousseau (ARC)
Robert Tung, Vassilis Papayannoulis (Metropia)
Integrated Dynamic/AB Models: Getting Real Discussion
May 15, 2017

2. Complete regional microsimulation model (ABM+DTA)
Major Strands of Evolution of Urban Travel Models & Network Simulations
Aggregate 4-Step Trip-Based models (since 1980th)
Disaggregate Tour-Based models (since 2000th)
Activity- Based Models (ABMs) (since 2010th)
Complete regional microsimulation model (ABM+DTA)
Aggregate Static Assignments (since 1980th)
Sub-area and corridor Traffic Simulations (since 1990th)
Regional Dynamic Traffic Assignment (DTA) (since 2000th)
Generations of Travel
Demand Models
Generations of Network
Simulation Models

3. Construction of ABM-DTA integrated system

4. Construction of ABM-DTA integrated system: Step 1/Q
Fixed transit and NM LOS skims or static assignments or DPTA in future?
Transit, NM LOS
ABM
Initial highway LOS
Start with static LOS skims or preliminary DTA with initial trip roster?
List of person tours & trips
Car occupancy?
Randomized departure time?
Individual VOT or classes?
External trips, trucks?
Roster of vehicle trips
DTA

5. Construction of ABM-DTA integrated system: Step 1/A
Fixed transit and NM LOS skims
Transit, NM LOS
ABM
Initial highway LOS
Start with static LOS skims
CT-RAMP2 design includes consistent translation of person trips into vehicle trips w/ continuous departure time
Simpler ABMs require post processing
External trips, trucks require one-time processing
List of person tours & trips
Roster of vehicle trips
DTA

6. Construction of ABM-DTA integrated system: Step 2/Q
Initial highway LOS
Transit, NM LOS
ABM
How to feed back LOS for potential trips that have not been simulated?
List of person tours & trips
How to resolve schedule inconsistencies?
Roster of vehicle trips
Individual trajectories
DTA

7. Construction of ABM-DTA integrated system: Step 2/A
Initial highway LOS
Transit, NM LOS
ABM
How to feed back LOS for potential trips that have not been simulated?
List of person tours & trips
iSAM
Internal loop
Roster of vehicle trips
Individual trajectories
DTA

8. Construction of ABM-DTA integrated system: Step 3/Q
Initial highway LOS
Transit, NM LOS
ABM
LOS manager
ADIT
List of person tours & trips
External loop
iSAM
Internal loop
Roster of vehicle trips
Individual trajectories
DTA

9. Integration Layer Components
External Loop 1 w/mining individual trajectories LOS
Internal Loop 2 w/individual Schedule Adjustment Module (iSAM)

10. Essence of each Loop External Loop 1: Internal Loop 2:
Generates activity patterns & schedules
Uses individualized LOS through trajectory mining
Internal Loop 2:
Simulates activity patterns
Adjusts schedules for realistic trip chain loading
Uses individual trajectories
Evaluates “stress” measures

11. Data Input and Output for iSAM
Planned Individual Schedule
Input:
Simulated Travel Time
Output:
Adjusted Individual Schedule
Planned departure time
Planned departure time
Adjusted departure time
Expected travel time
Simulated travel time
Simulated travel time
Trip 1
Preferred arrival time
Adjusted arrival time
Planned activity duration
Adjusted activity duration
Planned departure time
Planned departure time
Adjusted departure time
Expected travel time
Simulated travel time
Simulated travel time
Trip 2
Preferred arrival time
Adjusted arrival time
Planned activity duration
Individual schedule adjustment and stress evaluation
Adjusted activity duration
Schedule adjustment parameters

12. Schedule adjustment example (ARC)

13. Schedule adjustment example (ARC)
SOV/free
SOV/paid

14. Schedule adjustment example (ARC)
HOV2/free
HOV3/free

15. Schedule adjustment example (ARC)
Walk
Walk to local transit

16. Schedule adjustment example (ARC)
Destination-work
Origin-work

17. Schedule adjustment example (ARC)
Destination-school
Origin-school

18. Schedule adjustment example (ARC)
Destination-discretionary
Origin-discretionary

19. Daily Volume Differences

20. 1x30 iters without iSAM

21. 2x15 iters with iSAM

22. Learning about Space from Individual Trajectories (Dynamic Choice Set)
One implemented trip provides individual learning experience w.r.t. multiple destinations [Tian & Chiu, 2014]
Destination
Origin 3 2 1 4
Moreover, people use trips to certain destination to learn more about the space. Each simulated trajectory provides learning experience with respect to multiple OD pairs that greatly improves the coverage. Efficient on-the-fly mining of the bank of individual trajectories and dissecting them into sub-trajectories is one of the essential components of the new integrated model.
Instead of storing infinite OD skims we store individual trajectories and mine them efficiently.
OD pairs covered:
OD,
O1, O2, O3, O4, 1D, 2D, 3D, 4D
12, 13, 14, 23, 24, 34

23. Indexing Schema and Search
Trip
Node 1 (loading point 1):
Trajectory ID=1, nodePlacement=5
Trajectory ID=2, nodePlacement=3 ….
Node 2 (loading point 2):
Trajectory ID=2, nodePlacement=8
Trajectory ID=3, nodePlacement=10
Origin MAZ
Destination MAZ
Origin loading points (nodes)
Destination loading points
Trajectories containing origin node
Trajectories containing dest. node
Trajectories containing origin node and destination node in the right order: origNodePlacement<destNodePlacement
Extract sub-trajectories meeting matching criteria
Best representative sub-trajectory

24. Matching Levels (Excluding Intra-MAZ trips)
TOD
Temporal
Spatial
Occupancy
VOT 1
AM, PM
5 min
OMAZ, DMAZ
All MD
EA, NT 2
15 min
EA, MD, NT 3
OTAZ, DTAZ 4
60 min 9
No match found / skims used

25. If multiple trajectories found: Chose the min discrepancy function
Find the closest simulated individual trajectory (sub-trajectory) by the penalty function:
Occupancy discrepancy × Weight=0.5 (if relevant)
VOT discrepancy, $/h × Weight=0.1 (if relevant)
Departure time discrepancy, min × Weight=0.05
Same origin TAZ, different MAZ × Weight=0.1
Same destination TAZ, different MAZ × Weight=0.1
Trajectory age penalty = (iter_ABM-iter_DTA-1) × 0.1

26. Trajectory Coverage Stats
TOD
Aggregation level 1 2 3 4 9
Total
Before 6
83.0%
12.5%
0.2%
0.1%
4.2%
100.0%
6-10
61.3%
5.6%
19.5%
7.6%
5.9%
10-15
93.4%
5.7%
0.7%
15-19
66.6%
17.2%
6.2%
After 19
92.0%
6.9%
0.0%
0.9%
77.7%
6.1%
9.6%
3.6%
3.0%

27. Travel Time Differences by aggLevel: Trajectory-Skim, min

28. Travel Time Differences by TOD: Trajectory-Skim, min

29. Example of Analysis of Time Budgets

30. Mode Choice Switch between Skim-Based run and Trajectory-Based Run (MORPC)
mode/skims
Mode/trajectories
1=SOV
2=HOV2 driver
3=HOV3 driver
4=HOV passenger
5=Walk to local bus
6=PNR local bus
7=KNR Local Bus
8=Walk to express bus
9=PBR express bus
10=KNR express bus
14=Walk
15=Bike
16=Taxi
17=School bus
153,858 - 93 99 16 13 2 9
200 11
39,470 22 4 5 1 58 6
33,943 3 21 7
84,368
119 17 15
136 23 88 10 20
4,783
535
222
211 69 78
21,642
968
1,676 8
208 53
12,482
C10 FHWA Webinar, November 1, 2016

31. Conclusions Deep integration of ABM and DTA is feasible:
Already practical for regions under 1M
Many additional new avenues
Moving towards AgBM
Runtime is an issue:
Integration layer adds only a little
DTA and ABM constitute major time-taking components, especially DTA

32. Contact(s) Peter Vovsha, PhD Jim Hicks, PhD
Assistant Vice President, WSP
Systems Analysis Group
Jim Hicks, PhD
Principal Software Architect