DVRPC has developed an eight-year plan for improving the travel forecasting tools at its disposal. As part of developing this plan, a number of meetings were held between DVRPC's Office of Modeling and Analysis, staff from other DVRPC departments, as well as other coordinating agencies. DVRPC identified a number of improvements and modeling features it would like to have. Part of the plan, of course, is the development of the activity-based model.
The rest of this section discusses key modeling needs identified through these meetings and courses of actions. In addition, this section describes the plan for development of the DVRPC activity-based model.
In planning for model improvements, DVRPC identified a number of areas in which they would like to spend resources to evaluate alternative approaches or make improvements to modeling techniques. These items are listed below.
While some of the modeling needs and objectives described above will require additional modeling tools, many of the needs will be, at the very least, partially addressed when DVRPC's activity-based model is completed. The development plan for the ABM is described in the next section.
DVRPC's overall plan for ABM development is to transfer the DaySim ABM implementation from the Puget Sound Regional Council (PSRC), integrate it with Visum (for operations like network skimming and assignment), and validate the model. Currently, the model development plan calls for reestimating the models using very similar or identical specifications to the PSRC model. In other words, it is desired to limit the amount of time spent evaluating alternative model specifications, and instead to simply update parameter values from the PSRC model.
The overall ABM structure is shown in figure 4‑1. The system will use a population synthesizer to simulate characteristics and locations of households, DaySim models the "within region" travel of residents, and additional nonresident and commercial trips are added to the trip tables prior to network assignment. Visum is used for network assignment and network skimming processes and Python scripts are used to connect the different modeling components.
Figure 4‑1: Activity-Based Model System Architecture
(Source: DVRPC Presentation to Peer Review Panel, October 29, 2014.)
DaySim, the demand model, considers two primary components of travel behavior. The first is mobility choices of each household. The mobility choice modeling components are shown in figure 4‑2. These models simulate long-term choice dimensions of households, including usual work, school, and parking locations, usual travel methods for household members, and household auto ownership. A key input to these models are disaggregate and aggregate tour logsums, which serve as accessibility measures for different household choices.
Figure 4‑2: DaySim Mobility Choice Models
(Source: DVRPC Presentation to Peer Review Panel, October 29, 2014.)
The second travel behavior component considered by DaySim is the travel-day choice, shown in figure 4‑3. Unlike mobility choices, these travel choices are short term and could easily change from one day to the next. These models include an overall day pattern (which determines the type and number of activities each household member engages in over the travel day), tour destination, time of day, and mode choices, and intermediate trip and stop-level models of destination, time of day, and mode. As shown in figure 4‑2 and figure 4‑3, the two modeling components communicate via disaggregate and aggregate logsums.
Figure 4‑3: DaySim Travel-Day Choice Model Structure
(Source: DVRPC Presentation to Peer Review Panel, October 29, 2014.)
The overall model will be implemented at the Census block level, rather than traffic analysis zones, which the trip-based model relied upon. This will reduce the spatial aggregation error of the ABM, relative to the trip-based model.
The mode choice model components will explicitly consider the choice of toll or no toll routing for auto travelers (rather than relying on Visum to handle in traffic assignment). In addition, the mode choice model will rely on Visum's path selection process to determine transit submodes, thereby making transit submode choice unnecessary in the mode choice model itself. The mode choice model also will utilize different value of time (VOT) user classes. This will improve the model's sensitivity to transit policy changes, since low VOT travelers may be most sensitive to such changes, but cannot be adequately modeled using average VOTs for all travelers. However, this will increase computing time, memory requirements, and overall model complexity.
The model development plan calls for using 12 time-of-day period skims for auto. This includes 5 skims for each of the 2 peak periods, plus 1 midday skim and 1 overnight skim. This should enable DaySim to better model peak spreading and time-of-day pricing. For transit, only 5 skim periods are planned and 4 transit assignments.
Since DVRPC plans to transfer many (or all) of the model specifications used in the PSRC implementation of DaySim, the model development plan calls for extensive calibration work to ensure the models are matching local observed data appropriately. A number of validation measures have been proposed to ensure system traffic and transit counts are accurate as well, including a back-casting exercise. In addition, sensitivity tests will be applied to check the responsiveness of the model system to inputs and changes in policy variables.
