Project Meeting 2022.03.22 - ActivitySim/activitysim GitHub Wiki

Presentation: Disaggregate accessibilities.pptx

Accessibilities in the Current Model

  • Three types of accessibility measures in the current ActivitySim model:
    • Origin-destination (mode choice) accessibilities
    • Origin-based (destination choice) accessibilities -> Phase 7 work will focus on these origin-based accessibilities
  • Units are "utiles" (measuring utility), or relative cost per utility. These can be converted to times/cost to be compared to other things. For example, take total utility and scale by cost coefficient in the mode choice model to convert the accessibilities to dollars, or by time coefficient to convert to time equivalents.
  • Parameters are asserted.
  • Shortcoming include:
    • Requires a special set of transit skims
    • Does not take into account important decision-maker attributes, mode-specific variables, emerging mobilities, or other employment segmentations.

CT-RAMP Accessibility Calculator

  • 50 different accessibilities calculated
  • Segmented by activity type, period, mode combination, auto sufficiency
  • Computationally intensive for regions with lots of zones

Proposed Update

  • Build “prototypical” synthetic population – controlled synthetic population with controlled attributes
  • Use restart mode to generate tours, build input chooser tables, save the logsums
  • Merge logsums with actual synthetic population (based on proximity or match algorithm)
  • 9 types of households for each zone
  • Suggestions to reduce run-times
    • Could you bypass the logsum calculations for those locations where a particular mode option isn’t available?
    • Develop a pre-process check to make sure those types of households exist in each zone in the actual synthetic population. However, if the purpose is to calculate accessibilities for land use integration, all the segmentations may be needed.
  • Details and Complications
    • Need to increase destination sampling for mode choice logsums, testing required
    • Need to consider how to make these changes with reasonable run-times, especially for models with microzones
    • Need to consider how much detail is needed
      • Maybe keep household types by TAZ but distribute them across the MAZs
      • Use some households at the edge of the TAZ and some near the middle
  • Next Steps
    • Work out details
    • Propose software enhancement design
    • Propose segmentations