Appendix A. Additional Options for ATIS Evaluation in ICM/AMS
Karl Wunderlich, Noblis
The current proposal to deal with traveler information centers on a sensitivity-testing approach manipulating a series of parameters in traffic simulation to reflect various proportions of travelers changing mode, trip departure, and route choice. The advantage of such an approach is that it is relatively simple to implement with most traffic simulation models and avoids the need to develop a complex travel behavior model. Aggregate system-level performance measures can be directly obtained from simulation outputs. The disadvantage of this kind of simple approach is that it is likely to be too coarse to identify the effects of incrementally improved traveler information services. Further, since the sensitivity must be performed on trip decisions (e.g., varying mode choice by origin-destination) it cannot clearly differentiate between service users (who may use information to confirm that sticking with usual choices is preferable) and non-users (who may divert or make other uniformed decisions based on unexpected conditions).
Alternative, complementary methods could be considered to provide a more robust representation of traveler behavior. This appendix describes one possible approach using software and techniques developed as a part of the Heuristic On-Line Web-Linked Arrival Time Estimation (HOWLATE) development program over the last six years [1, 2, 3, 4, 5, 6, and 7].
HOWLATE resources could be brought to bear for ATIS (Advanced Traveler Information Services) impact analysis in several ways, depending on several key factors:
- Type(s) of traveler information services to be evaluated;
- Targeted market segments of these services (e.g., commuters vs. tourists);
- Projected market penetration for various traveler information services;
- Quality and availability of archived data; and
- Capability of a selected traffic simulation to represent pre-trip and en-route choices.
This appendix presents two options for utilizing HOWLATE resources for improved ATIS modeling in ICM. The options are presented in ascending order of complexity. It should be stressed that these options represent potential complementary or enhanced analyses within the simulation-based ICM AMS methodology and not a replacement to the methodology.
Option 1: HOWLATE With Archived Data
The first option is a complementary analysis based on archived travel time data and traveler information content (Figure A.1), and corresponds to the traditional HOWLATE evaluation method used in previous studies. No direct link or interaction with a traffic simulation is required. This option assumes that quality travel time data can be obtained for all modes and integrated with an archive of synchronous traveler information content. Further, the average error in the travel time data has been measured and its distribution known. Finally, a file describing multimodal network structure in terms of links and decision nodes must be available. This network file is not as complex as a simulation network file, it is just a representation of the various routes and modes a traveler must consider getting from trip origin to trip destination. Detailed characteristics such as link capacity or number of lanes are not required. However, for each link in the network, there must be a corresponding travel time available describing the actual travel time to traverse the link, every five minutes. For best results, a minimum of 100 days of synchronous travel time and ATIS content are required. This large set of days is broken into two mutually exclusive datasets, one for establishing usual travel choices for regular commuters in the corridor and one for the evaluation of the benefit of the traveler information services under a representative set of congestion conditions.
Figure A.1 HOWLATE with Archived Data
In the first analytical step in HOWLATE, synchronous data for the training period is fed into the Travel Habituation Module (THM), which identifies usual mode, departure time, and route choice for all of the possible origin-destination-time of arrival triplets in the network. For large metropolitan networks, this may entail upwards of 100,000 triplets in the network considering 15-minute target time of arrival increments. For each triplet, the THM finds the time of departure, mode and route with the minimum generalized disutility that results in on-time arrival at some specified threshold (typical values: 85 to 95 percent). Generalized disutility is based on work by Small et al. [8], and is a monetized combination of travel time, late schedule delay (minutes late), and early schedule delay (minutes early). HOWLATE does not currently take into account fares or tolls, but this can be extended with minimal effort. These usual choices, along with a file describing average travel times on the multimodal network are carried forward to the evaluation phase.
In evaluation, days from the archive are fed into the yoked study simulator for analysis. Here, each triplet is associated with a simulated non-user (control) subject who ignores information and sticks to the usual choices every day. Using the archive of travel time data, the yoked study simulator simply adds up the elapsed retrospective time accumulated by the non-user along each link in the time-dependent network. This implies that arriving at an intermediate node later in the rush period may cause even more delay as recurrent congestion builds on the latter portions of the trip. A simulated traveler information user (experimental) subject is also associated with each triplet, who will react to provided information based on a behavioral archetype. The archetype can be scripted to consider only certain choices (e.g., no early departures, but will consider a change to mode) and consider certain types of information (e.g., pre-trip consultation of a congestion map).
Using the rules from the archetype and the ATIS content, the yoked simulator assembles a travel experience and time of arrival at the trip destination based on conditions actually observed on all the links in the trip. Pre-trip and en-route choices are generated as simulator moves forward in time. Regardless of the type of user, trip decisions and trip outcomes are recorded for each day for each of the many possible triplets. In previous HOWLATE studies, the ATIS users deviate from usual choices at rates proportional to the amount of unexpected variation in conditions, and quite often arrive more consistently on time with smaller travel budgets than non-users. Comparisons can be made between non-users and ATIS users on the individual yoked pair by day, or aggregated in a post-processor.
Because no traffic simulation is involved in the process (only simple re-creation of conditions from an archive), the HOWLATE engine is computationally efficient. Roughly 5,000 simulated yoked trials can be conducted every second on a low-end PC. This allows the quantification of ATIS user benefits across large networks, and also allows for the rapid evaluation of multiple archetypes. Major HOWLATE analyses have involved the processing of several hundred million simulated yoked trials.
Option 1: Limitations
There are several limitations to the use of HOWLATE with archived data (Option 1) for ICM, however. First, Option 1 is a purely retrospective approach and cannot be tested in future conditions significantly different from the current state of operations (e.g., instituting congestion pricing for the first time). It also assumes that significant archives of travel time data exist for the corridor. Second, Option 1 is valid for prospective (future) ATIS services only if the number of ATIS users is small enough that resulting congestion patterns are unaltered. For the evaluation of current services, this is not an issue, even if market penetration is high (e.g., the HOWLATE study on broadcast radio traffic reports). As a rough rule of thumb, market penetrations under 6 percent generally do not result in significant changes in congestion patterns.
Option 1: Advantages
The advantage of Option 1 is that it is a relatively low-cost, high-payoff extension to a corridor analysis if travel time and other data are already being collected for the purposes of performance management or traffic simulation modeling. Results of the HOWLATE study can be used to guide a more precise depiction of user response than simple range testing. For example, if benefits for a pre-trip service are geographically concentrated non-uniformly for longer trips in the corridor that have a good alternative, then user density can be non-uniformly distributed over the network in a more realistic fashion. The simulation analysis is enhanced not only by non-uniform distribution of traveler information usage, but also providing a resource of experiential data to draw on within the simulation analysis when trip decisions must be made in response to traveler information (e.g., modeling of variable message signs or broadcast traffic reports).
Option 2: HOWLATE With Simulation(Soft Feedback)
In this option (graphically illustrated in Figure A.2), individual modules of the HOWLATE software are utilized to incorporate more realistic traveler behavior in response to traveler information with limited feedback to a traffic simulation. This option is more complex than Option 1 because HOWLATE is applied iteratively in conjunction with a traffic simulation, but allows for higher market penetration ranges (6 to 25 percent) to be evaluated. The traffic simulation must be jointly calibrated with the travel habituation module so that there is at least a modicum of consistency between usual route selection and operational conditions. This option is less complex than Option 1 in terms of archived data, however – all the travel time can be provided from a traffic simulation. This option is only valid with simulated ATIS content, which can be generated from the simulation outputs using relatively simple rule sets (e.g., speeds between 30 mph and 40 mph are coded as yellow on the congestion map, for example). Another key requirement is that the traffic simulation is run through a set of scenarios that reflect the range of conditions typically seen in the corridor (that is, combinations of variations in travel demand, incident patterns, and weather impacts). Because the THM needs a range of conditions over which to find a preferred usual trip departure time, mode and route choice, these conditions and their probability of occurrence must be provided as input.
Figure A.2 HOWLATE with Simulation – Soft Feedback
The THM finds a set of choices and these are input back into the traffic simulation. If the set of selected usual routes is relatively stable from iteration to iteration, then we can declare a rough state of convergence (denoted in Figure A.2 with a “C”). Note that we have a soft convergence criterion, that is, the use of traveler information may have impact in particular scenarios (say, a major incident), but the market penetration rate is low enough that these relatively rare events do not change habitual behavior significantly.
Once some kind of acceptable convergence criterion is reached, then the stable conditions (travel times by scenario) and usual decisions (habitual decisions) can be used as input to the yoked study simulator to identify outcomes by scenario. Output can be processed using the existing post-processor to generate aggregate performance measures by user vs. non-user. In addition, the traffic simulation outputs themselves will provide system-level performance measures.
Limitations
The technique is untested, so there is technical risk associated with its implementation. Further, there is a technical risk that even the proposed “soft” feedback loop will not result in convergence. The TRANSIMS effort has struggled for several years to force “hard” convergence in large metropolitan networks based on a single simulation run representing normal conditions with near-zero indifference thresholds. Soft feedback is more tolerant, but does not guarantee convergence. Triplets with significant flow rates may have to be broken up into lower flow components and fitted with different disutility distributions to avoid unstable bang-bang control issues. This observation is valid for both ATIS users and non-users since the decisions they make influence the congestion conditions. The complex feedback at the simulation level must also be reconciled with feedback loops to other regional models, which may result in the need for multiple convergent solutions at the simulation level. The representation of en-route choice consistent with the assumptions of the THM will depend on the flexibility and capability of the individual traffic simulation used in the analysis.
Advantages
The advantage of such an approach is that if a convergent solution can be identified, then the system and user-level benefits from ATIS can be assessed in whatever range of conditions the ICM program wants to explore.
Summary and Recommendations
This appendix presents two options for incorporating analytic assets from the HOWLATE traveler information impacts evaluation effort into ICM analysis. The goal of providing these options for consideration is to provide a more robust analysis of traveler information impacts for ICM, a recognized need in the program and in the field of traffic simulation.
Given the technical risks associated with Option 2, it is unclear that an effort to pursue this unproven method in the early phase test corridor would be wise. However, if the archived data exist for the corridor sufficient to conduct a more traditional HOWLATE analysis (Option 1) then such an effort may be quite valuable and could be conducted in the timeframe. The results could be used to demonstrate capability in ATIS analysis, which would be helpful to the ICM effort as a whole. Depending on success, need and a longer timeframe to explore the technical risks, the more complex Option 2 may be a useful notion to reserve for consideration in later phases of the project.
References
Wunderlich, K., M. Hardy, J. Larkin, and V. Shah, On-time Reliability Impacts of Advanced Traveler Information Services (ATIS): Washington, D.C. Case Study, prepared by Mitretek Systems for FHWA, 2001 (Document #13335 on the ITS Electronic Document Library, http://www.its.dot.gov/welcome.htm).
Jung, S., J. Larkin, V. Shah, A. Toppen, M. Vasudevan, and K. Wunderlich, On-time Reliability Impacts of Advanced Traveler Information Services (ATIS), Volume II: Extensions and Applications of the Simulated Yoked Study Concept, prepared by Mitretek Systems for FHWA, March 2002. (EDL #13630).
Shah, V., S. Jung, J. Larkin, A. Toppen, M. Vasudevan, and K. Wunderlich, Accuracy and Coverage Implications for Advanced Traveler Information Services (ATIS) Benefits, Volume III: Extensions and Applications of the Simulated Yoked Study Concept, prepared by Mitretek Systems for FHWA, May 2003. (EDL #13859).
Vasudevan, M., K. Wunderlich, A. Toppen, and J. Larkin, Making the Most of Limited Data in Evaluating Advanced Traveler Information Systems by Experimental Resampling, In Transportation Research Record 1897, pp. 1-8, 2004.
Vasudevan, M., K. Wunderlich, J. Larkin, and A. Toppen, Comparison of Mobility Impacts on Urban Commuting: Broadcast Advisories Versus Advanced Traveler Information Services, In Transportation Research Record 1910, pp. 38-45, 2005.
Shah, V., K. Wunderlich, and J. Larkin, Time Management Impacts of Pre-Trip ATIS: Findings from a WashingtonD.C. Case Study, In Transportation Research Record 1774, pp. 36-43, 2001.
Vasudevan, M., and K. Wunderlich, Quantifying Serenity Benefits from Use of Advanced Traveler Information Services (ATIS), prepared by Mitretek Systems for the FHWA, October 2004.
Small, K., R. Noland, and D. Lewis, Valuation and Travel-Time Savings and Predictability in Congested Conditions for Highway User-Cost Estimation, NCHRP Report #431, National Academy Press, Washington, D.C., 1999.