Spatial and Temporal Feasibility Validation for Activity Schedules and Destination Choices

[amsamimi]

Problem

ActivitySim currently does not provide a built-in mechanism to verify that simulated activity schedules are spatially and temporally feasible once activity timing and destination choices have been made.

As a result, the model can generate activity patterns where the required travel between consecutive activities exceeds what can reasonably be achieved within the available time budget. While these cases may represent a very small proportion of all schedules, they have a significant impact on VKT.

This issue was identified during the Transport for NSW (TfNSW) MVP ActivitySim implementation developed with support from RSG and WSP.

Observed Example

The following tours were generated for a single individual:

Start	End	Tour Category	Tour Type	Origin	Destination	Distance to Destination (km)
10	28	non_mandatory	othmaint	3319	988	142
30	33	non_mandatory	othmaint	3319	1198	156
35	41	non_mandatory	eatout	3319	3297	7

This schedule contains:

• A maintenance tour approximately 142 km from the home location.
• A second maintenance tour approximately 156 km from the home location shortly afterwards.
• An eat-out tour scheduled only a few time periods later.

Even under uncongested conditions, completing these activities sequentially within the allocated schedule is highly unlikely.

Additional Examples

TfNSW has observed other similar cases, including:

• School destinations located more than 300 km from the person's home location.
• Discretionary activities allocated to destinations requiring several hours of travel despite limited available time in the schedule.
• Daily activity patterns where cumulative travel requirements exceed what is physically achievable within the day.

Current Mitigation

To reduce the frequency of extreme destination choices, TfNSW has implemented a soft-capping mechanism within destination choice models. The soft cap reduces the probability of selecting destinations at excessive distances from the origin and has been very effective in limiting some unrealistic outcomes.

[amsamimi]

@asiripanich @dhensle, please add any additional comments if you think anything is missing.

[i-am-sijia]

Re: School located more than 300 km from home. This is likely a result of simulation-based shadow pricing. When all attractive zones close to home are fully registered, a small portion of students may have to, unfortunately, go farther. See related issue #815

Re: Non-mandatory destination located too far from home, causing unrealistic schedule. For TfNSW, I estimated a version of non-mandatory tour destination choice model with a person's maximum remaining time window within the regular travel schedule as an explanatory variable. This specification increases the distance penalty when less time remains after scheduling mandatory and joint tours, reducing (but not eliminating) the likelihood of assigning distant destinations that create unrealistic schedules. However, this would not help the three NM tours example Amir listed, because all NM tour destinations are simultaneously chosen. Also, implementing this specification in ActivitySim would first require addressing issue #1079.

Re: Spatial and temporal feasibility in general. This is a known challenge in ABMs. The ActivitySim roadmap discusses several aspects of this issue, including response to schedule stresses (see Section 4.13).