Trip Distribution

1
KENTUCKY MODEL USERS GROUP
Trip Distribution Trip Assignment Calibration
/ Validation
October 29, 2004
2
Outline of Presentation
Trip Distribution Friction factors Trip
lengths Trip Assignment Methodology Parameters C
alibration / Validation Tests Adjustments Tools
3
Outline (cont.)

• Information based on
• Recent Kentucky Models Simpson Co.,
  Radcliff/Elizabethtown,
• Kentucky Statewide Traffic Model (KYSTM) Update
• MPO Models in Georgia
• Discussion will mainly focus on urban modeling
• Will discuss a few topics as they relate to
  Kentucky Statewide Traffic Model

4
Trip Distribution

• Process of creating trip tables from trip
  generation data

5
Trip Lengths

• There are significant differences between trip
  length distributions for trip purposes
• Compare to survey data if available
• Trip length distributions can also be derived
  from the National Household Travel Survey (NHTS)

6
Trip Lengths in the KYSTM

• The new KYSTM considers long distance and short
  distance trips separately
• Therefore, trip lengths are very important
• Short Distance Trips Trip Lengths / TLFDs
  derived from the NHTS
• Long Distance Trips Trip Lengths obtained from
  the 1995 American Travel Survey

7
Friction Factors

• Friction Factor Functions
• Exponential function
• Inverse power function
• Combined/gamma function
• Discrete function

8
Friction Factor Functions

• A few thoughts
• Exponential function and power function have one
  parameter for calibration combined/gamma
  function has two
• The greater the number of parameters, the easier
  it is to obtain a closer fit with the sampled /
  observed trip length distribution
• Generally, the combined/gamma function or a
  well-developed friction factor look-up table fits
  the observed TLD
• The bulk of the representational and policy
  relevance advantages of the gravity model lies in
  the friction factor function

9
Issues

• The gravity model is the most widely used
  technique for trip distribution, including large
  MSAs like Atlanta, Houston, Dallas, etc
• But the gravity model can have problems with
  distributing work trips in some circumstances
• The gravity model is unable to distinguish
  between types of workers and assigns a worker to
  the closest work opportunities
• Example Too many trip production ends in a
  low-income residential zone (like areas within or
  close to CBD) are linked to trip attraction ends
  in a high-income work zone (like CBD).

10
Issues (cont.)

• Are there options to replace the gravity model?
• Destination choice models that can have household
  socioeconomic data incorporated as one or more of
  the variables
• However, to estimate a good destination choice
  model, the survey data has to be very good
• This could make the model not cost-effective

11
Intra-Zonal Travel Times

• Generally, two approaches are available for
  estimating intra-zonal travel time
• Average of travel times from the subject zone to
  the nearest several zones (TransCAD, TP)
• 2. Determined by the size of the zone and
  intra-zonal speed (QRS II)
• Some studies have shown that the second approach
  is more accurate than the first

http//ntl.bts.gov/DOCS/images/CAS/CAS11.GIF
12
K Factor

• Generally, K factors are used to correct serious
  imbalances in the model results, such as effects
  caused by CBDs and natural barriers (such as
  rivers, mountains, railroads, etc)
• Can reflect some conditions that are not captured
  by the model or difficult to formulate.
• Problem with using K factors
• 1) lacking of understanding of what they
  represent and
• 2) problem of changes over time and
• 3) masking of fundamental errors in trip
  distribution.
• Using travel time penalties are preferred to
  using K factors nowadays. But K factors are
  still widely used in many models.

13
Trip Assignment

• All-or-Nothing Shortest path only (capacity is
  not an issue)
• Capacity Restraint Iteration between
  all-or-nothing and travel time calculations, but
  does not converge. Results often vary after each
  iteration.
• User Equilibrium Iterative assignment process in
  which travelers improve their travel time by
  using other paths (considers capacities)
• User Equilibrium is most common in models
  developed by WSA in Kentucky

14
BPR Curve

• Standard parameters

a 0.15

• b 4.0
• NCHRP 387 proposes

a 0.05 for signalized facilities and 0.2 for
all other facilities b 10.

• According to report, the new parameters fit the
  real situations better then the standard values

15
BPR Curve (cont.)

• For QRS II

Source QRS II 6.0 Manual, AJH Associates
16
Calibration / Validation
The terms calibration and validation are
often (mistakenly) intertwined Calibration A
process of adjusting model parameters in order to
match simulated traffic volumes with observed
traffic counts Validation Calculations used to
verify model statistics Validation should be
conducted after each step not just at the end!!
17
Validation Tests

• Count / Volume
• Total
• Functional Classification
• Screenline
• Percent Root-Mean Square Error (RMSE)
• R2
• Post-Processing Tools available to calculate
  quickly

18
(No Transcript)
19
Model Adjustments

• There are many adjustments that can be made
  during calibration
• Trip Rate Adjustments
• Trip Distribution Parameters
• Revisit centroid connectors
• Turn Penalties (prohibitions, penalties, etc)

20
Screenlines

• Screenlines can provide statistics that often
  tell the true story of traffic assignments
• Example - Radcliff / Elizabethtown MPO model

• Screenlines were used to quantify travel between
  Radcliff and Elizabethtown
• Also used to verify traffic flows between Hardin
  and Meade Counties

21
Feedback Loops

• The primary goal of introducing feedback to a
  four-step process modeling process is to reach an
  overall or partial equilibrium within the
  forecasting system
• This is so that the inputs and outputs of each
  step of the process are reasonably consistent
  with one another.
• The most commonly adopted application is the
  feedback between trip distribution and traffic
  assignment.
• Successful implementation of this feedback will
  result in a trip distribution model that
  determines trip interchange patterns using
  zone-to-zone travel time that are consistent with
  the final loaded speeds of assignment.

22
Feedback Loops (cont.)

• The most common current practice is to use
  free-flow travel times to conduct trip
  distribution.
• This may not reflect capacity-constraint
  conditions
• There may exist significant differences between
  the travel times used for trip distribution and
  those that output from the assignment.

23
Feedback Loops

• Recent WSA Applications
• Dalton, GA MPO model
• Hinesville, GA MPO model
• Method of Feedback Method of Successive
  Averages with Equilibrium Assignment
• Provides equal weight to each previous
  iterations equilibrium assignment results.

24
Feedback Loops

• Convergence criteria (Both have to be met)
• Criteria 1 Less than 5 of O-D pairs have peak
  hour speeds that change by more than 5, and
• Criteria 2 Average change in peak hour link
  volumes is less than 5

25
Feedback Loops (cont.)

• Impacts of feedback
• System-wide average link speed is increased
• Average travel time is decreased
• Average trip length is decreased
• Average v/c ratio is decreased and
• Total VMT is decreased.
• Average model run time increased from about 30
  seconds (no feedback) to 120 seconds for the two
  models mentioned above.

26
Select Link

• A great function to answer the question
• Where did the trips come from? Where are they
  going?
• If used properly, can provide many answers to
  interested parties that may not be technically
  knowledgeable about travel demand models

27
Recommendation / Conclusion

• There are many options available for
  distribution, assignment, calibration, etc.
• The use of each model should be considered
  throughout model development to ensure the
  techniques applied are most appropriate for that
  use
• It is important to review the model output after
  each step