1. Introduction to Activity-Based Modeling
Joshua Auld
Transportation Research and Analysis Computing Center
Argonne National Laboratory
November 28, 2011

2. Travel Demand Modeling Current and Future Applications
Roadway planning / new construction
Decreasing funding for new roads
Environmental/Air quality impacts
More important after ISTEA, SAFETEA-LU
Conformity analysis
Travel Demand Management
Large collection of strategies
Increase efficiency of system (TSM)
Change user behavior (TDM)
Congestion pricing, ride-share program, etc.
ITS / Operations

3. Activity based model theory
Travel is derived from participation in activities
Not accounted for in 4-step models
Time and space between the activities generates travel
Activity participation is modeled at household/individual level
Microsimulation model
Individual’s activity participation constrained by:
Time availability
Location
Institutional characteristics (operating hours, etc.)
Household considerations
Activity patterns are generated by individuals which satisfy these constraints and meet some other criteria

4. Activity based model theory (cont.)
Figure shows a daily activity pattern graph (for travel in 1-dimension)
Vertical lines represent activities
Diagonal lines are travel episodes
Explicit representation of time of occurrence for all travel episodes, linked to associated activities
ABM generates an activity pattern for modeled individuals
Know when and where they are traveling at all times
Home
Shopping
Work
Lunch
Hagerstrand (1972) – Time geography – tracing of human activity through time-space subject to the constraints placed on those activities (biological, institutional, spatio-temporal). Gave rise to the time-space prism (which is another form that some ABMs take – assumes fixed activities, and flexible activities are scheduled in the available prism).
This representation above is a major advance over four step modeling. There is no gravity modeling/calibration procedure to estimate flows between zones – flow is known precisely from the results of the activity pattern.
Home

5. Activity based model theory (cont.)
It is the goal of an activity based model to develop an activity schedule that:
Satisfies all of the given constraints, and
Satisfies some schedule optimization criteria
The models take as inputs:
Individual/household attributes
Environment attributes (land-use, activity locations, transportation networks, etc.)
Attributes and actions of other individuals
Then use these inputs to model a series of choices:
What activities to schedule
When to schedule them
Where to schedule them
Who to go with
How to get there
How long to stay
How these choices are modeled depends on the type of model used
Not all models directly estimate all of the choices, and all of the models estimate them in different ways and in different orders.

6. Approaches to Activity Based Modeling
Two general approaches to modeling activity participation choices
Econometric models
Rule-based models
In econometric type models:
Models are usually tour based – select activity tours from predefined set
Utility maximization to model pattern formation
In rule-based models:
Computational process models or other derived decision rules
Bottom-up approach to schedule building

7. Example of a nested logit model for decision making
Econometric models
Decisions modeled using discrete choice
Usually nested logit models
Model sequence:
Number and type of tours -> Stops in each tour -> mode choice for tour, etc.
Example of a nested logit model for decision making
This example logit decision structure first determines the number of maintenance stops generated, assigns them to household individuals, then allocates a vehicle for each stop (activity).
This is the first stage of a two-stage model, where the second stage then uses a NL model to form tours for each individual.
Source: Wen and Koppelman (2000)

8. Econometric models (cont.)
Econometric models are generally tour-based
Model the number and purpose of activity tours
Chose from a set of pre-defined tours
Examples:
Bowman-Ben Akiva (1996) and derivatives (DAYSIM)
PB Consult models (MORPC, NYMTC, etc.)
Jakarta Model - Yagi and Mohammadian (2008)
Wen and Koppelman (2000)
Tours are pre-defined, since the choice set with no predefined tours would be too large – combinatorial problem
Ex. 6 activity types – how many different patterns can be created?
Source: Wen and Koppelman (2000)

9. Criticisms of econometric models
Unrealistic behavioral assumptions
Utility maximization in decision making by individuals
Artificially restrict activity scheduling to predefined choices
Can not represent full range of scheduling behavior
Reduces number of choices to be modeled, i.e. combinatorial problem
No consideration of dynamics (full day selected at one time)
Limitations on time-scale of analysis
Usually discretized to time-of-day periods
See the Garling, Kwan paper for critique of utility maximization in scheduling.
Basically, the choice set is too large, so in reality people develop heuristics. Do not actually maximize their utility.
EXAMPLE: Mode Choice to work: My wife has the car and there’s no parking at work, so I’ll take the train. This is a choice heuristic, rather than calculating utility over every choice facet and arriving at optimal solution.
These models do not work well to capture the process of scheduling – which may be important in their responsiveness to policy changes.

10. Rule based model overview
Create activity schedules based on heuristics
Computational Process Models or other derived rules
Attempt to model the underlying process of activity scheduling
Examples:
SCHEDULER (Garling 1989)
AMOS (Pendyala 1995)
ALBATROSS (Arentze, Timmermans 2000)
TASHSA (Roorda, Doherty, Miller 2005)
ADAPTS (Auld and Mohammadian 2009)

11. Computational process models
Production system (Newell, Simon 1972) for modeling decision making behavior
Rules as condition-action pairs
Describe how information relating to a choice is searched
Choice made depends on current information acting on set of rules
Allows incorporation of learning
Representation of process of decision making
Production systems often modeled as
Decision trees (ALBATROSS)
Heurstic rules (SCHEDULER, TASHA)
In operational models, none of the models are true CPMs, since learning is not included in peoples behavior. This was left as future refinements of the operational models. These are simplifications of true CPMs, due to limited modeling data.

12. Data requirements
Rule-based activity scheduling models have extensive data needs:
Synthesized population
Socio-demographic characteristics of individuals
Synthesized city/environment
Activity locations/operating hours
Roadway and transit networks
Land use variables
Activity diary/scheduling data
Activity participation/generation rates
Conflict resolution rules
Planning process data
Executed schedules for model validation
It may be possible that activity diary data will not be needed depending on the results of transferability analysis.

13. ALBATROSS model overview Arentze and Timmermans (2000).
Rule-based model for predicting:
Activity participation, location, timing, duration, party and travel mode choice
Includes household, institutional and spatio-temporal constraints
Decision rules modeled using CHAID decision trees at each step
Derived from activity survey data
Model attempts to simulate daily activity schedule creation for individuals
Long-term decisions considered fixed
Household interactions modeled
Starts with an assumed schedule skeleton
Represents routine, fixed activities
Considered the highest priority
Sequentially attempts to add new activities in order of assumed priority
Created to model travel in the Netherlands.

14. ALBATROSS model Scheduling process (continued)
Ovals represents decision steps
In ALBATROSS:
A skeleton schedule first constructed containing routine activities
Activity agenda is created
Finally, activities added to the schedule from the activity agenda
Source: Arentze, Timmermans (2000).

15. TASHA model overview Roorda, Doherty, Miller (2005)
Activity scheduling model for Toronto area
Based on a 1996 household travel survey
Activities generated based on travel survey
Attributes assigned using choice models
Entirely rule-based
Activities fit to Project Agendas, then Schedule
Activities added to schedule based on assumed rules
Heuristics represent schedule adjustment, conflict resolution
Decision rules not well represented in the model
I.E. mode choice and activity location do not figure into the scheduling process

16. TASHA model Scheduling process
Generate activity episodes
Activities generated based on trip diary data
Frequency, start time and duration chosen from feasible portion of probability distributions
Fill each project agenda
Activities added to respective project agendas to form provisional schedules within each project
Add activities to persons schedule
Activities added to schedule in order of priority
Priority assumptions: Work/School > Other > Shopping and Joint activities > Individual activities
Adjust activities until conflicts removed
Generate activity
Fits in agenda No
Yes
Add to project agendas
Next activity
Next activity chosen is the highest priority activity remaining in project agendas.
Activity episodes are generated using probability distributions for various groups classified by age, employment status, occupation, gender, etc. based on the type of activity
Number of activities generated of each type is chosen from the activity frequency distribution
Frequency-start time joint PDF and start time-duration PDF.
Fits in schedule No
Yes
Add to schedule

17. SIMTRAVEL Integration of OpenAMOS ABM with MALTA DTA
Prism-constrained activity-travel scheduling (PCATS) – Kitamura et al. (1997)
Fill in prism gaps sequentially until no time-remaining between fixed activities
Source: SimTRAVEL website -

18. Research Gaps in ABM Simplification of scheduling process
Priority rules, fixed order of choices
Short-term, diary data used
Most rule-based models implemented using 1-2 day diary data
Limited integration with traffic simulation
Static assignment for fixed time-periods
Ad hoc interoperability between model systems
Activity Generation/Planning/Scheduling Dynamics???

19. Scheduling Order Example
A) Impulsive Shop - Preplan Eat Out
B) Preplan Shop - Impulsive Eat out
Before Change
Before Change
After Change
After Change
This shows an example where the original activity pattern is the same in both situations (A and B) – home->shop->lunch with friends->home
The only difference is that in A the shopping trip is impulsive on the way to lunch with friends.
In B the lunch is impulsively decided on during the shopping trip – calls friends while out shopping.
So patterns are the same only planning is different.
Then there is a change in policy (or something else) which causes the store to be unavailable – observe different responses
In A, the shopping trip is just skipped and person goes to lunch as planned. (7 trips reduced to 6)
In B, the shopping trip is diverted to a different available store, not near restaurant so no trip with friends is planned (7 trips reduced to 2).
Point is that policies, system changes, etc. may have different impacts depending on planning context

20. ADAPTS - Introduction ADAPTS activity-based model:
Simulation of how activities are planned and scheduled
Extends concept of “planning horizon” to activity attributes
Time-of-day, location, mode, party composition
Fits within overall framework of integrated land-use transportation model
Constraints from long-term simulation (land-use model)
Combined with route choice and traffic simulation
Core concept:
Universal set of activity planning / scheduling processes represented by heuristics and/or models
Outcomes constrained by local context

21. ADAPTS Simulation Framework
Information Flow
Simulation Flow
Initialize Simulation
Initialize World
Synthesize Population
Generate routines
For each timestep
Household Schedule
Household Planning
Household Memory
Land Use
Institutional
Constraints
This slide shows an overview of the simulation process, so after the simulation is initialized (note that we are currently using 15min timesteps for 4 weeks of activity scheduling)
At each timestep:
Household activities are generated/planned/scheduled/executed (as shown in the next slide)
Individual activities are generated/planned/scheduled/executed (again in next slide)
Any activities which are executed are output to the traffic assignment
Results of traffic used to update network knowledge for next timestep
Go to next timestep
Individual Schedules
Network LOS
Individual Memory
Individual Planning
Social Network
Write Trip Vector
Traffic Assignment

22. ADAPTS Planner/Scheduler
At timestep t
Handles at each timestep:
Generation
Planning
Scheduling
Each step can occur at different times for same activity
Core of the framework is the Attribute Plan Order Model
Activity Generation
Generate new activity
Yes
Attribute Planning Order model No No
Update existing activity(s)
Set Plan Flags:
(Ttime,Tloc, etc.)
Yes
Activity Planning
t = Ttime
t = Twith
t = Tloc
t = Tmod
Time-of-Day
Party
Destination choice
Mode Choice
This slide is the most important in the presentation
Stress the use of three core steps in the simulation – Generation, Planning and Scheduling, as opposed to just Generation and Scheduling as usually done (where the attributes are determined at generation time)
Allows for more behaviorally realistic responses
Planned Activity Schedule
Activity Scheduling
Resolve Conflicts
Yes
Conflict Resolution Model
Decision
Logical test
Model
Simulated events No
Execute activity
Executed Schedule

23. Activity Plan Order Model
Decision Example:
Activity Generation
Activity Plan Order Model
Plan Work Location
T1 Plan new activity
Ttime
Tloc
Twho-with
Tmode
Ttime
Plan time-of-day
Tloc
Plan location
Twho = Tmode
Plan mode and who-with
Texec
Execute Activity
Ttime
Tloc
Tmode/who
Texecute
Simulation Time
Time:
Shop
Time: ?
Loc: ?
Mode: ?
At Home
Time: 12:00 AM – 8:00 AM
Loc: Home
Mode: None
Work
Time: 8:00 AM – 4:00 PM
Loc: HOME
Mode: None
Work
Time: 8:00 AM – 4:00 PM
Loc: ?
Mode: ?
Shop
Time: 4:00 – 5:00
Loc: Mall
Mode: Auto ?
Schedule ?

24. References
Arentze, T. and H. Timmemans (2000). ALBATROSS – A Learning Based Transportation Oriented Simulation System. European Institute of Retailing and Services Studies (EIRASS), Technical University of Eindhoven.
Auld, J. A., and A. Mohammadian (2009). Framework for the development of the Agent-based Dynamic Activity Planning and Travel Scheduling (ADAPTS) model. Transportation Letters: The International Journal of Transportation Research, 1 (3),
Bowman, J.L. and M.E. Ben-Akiva (2001). Activity-Based Disaggregate Travel Demand Model System with Activity Schedules. Transportation Research Part A. 35, 1-28.
Kitamura, R., C. Chen, C., and Pendyala, R.M. (1997) Generation of synthetic daily activity-travel patterns. Transportation Research Record, 1607,
Pendyala, R.M.; R. Kitamura, A. Kikuchi, T. Yamamoto, S. Fujii (2005). Florida Activity Mobility Simulator: Overview and Preliminary Validation Results. Transportation Research Record: Journal of the Transportation Research Board, No. 1921,
Roorda, M.J., S.T. Doherty and E.J. Miller (2005). Operationalising Household Activity Scheduling Models: Addressing Assumptions and the Use of New Sources of Behavioral Data. Integrated Land-use and Transportation Models: Behavioural Foundations, M. Lee-Gosselin and S.T. Doherty (eds), Oxford: Elsevier, pp
Yagi, S. and A. Mohammadian. (2008). Modeling Daily Activity-Travel Tour Patterns Incorporating Activity Scheduling Decision Rules, Transportation Research Record: Journal of the Transportation Research Board, No. 2076, TRB, National Research Council, Washington D.C., pp

25. Thank You!