ACCESS TO DESTINATION: ESTIMATION OF ARTERIAL TRAVEL TIMES

1
ACCESS TO DESTINATION ESTIMATION OF ARTERIAL
TRAVEL TIMES

• Gary A. Davis
• Henry Liu
• Hui Xiong
• University of Minnesota

2
Accessibility The Basics
3
Two Interesting Questions

• If land use remains constant, but congestion
  increases, accessibility will decline
• But, if land use becomes more compact,
  accessibility may improve, even in the face of
  congestion
• 1. Has this happened in the Twin Cities?
• 2. Can we monitor accessibility, in the same way
  we monitor congestion?

4
Information and Issues

• Required Network model of Twin Cities region
• Required Distribution travel objectives in Twin
  Cities
• Required Estimated travel time on road links,
  both freeways and arterials
• Issue While surveillance system provides data
  for estimating freeway travel times, data on
  arterial travel is limited
• Objective of this project Produce plausible
  estimates of arterial travel times

5
Estimating Arterial Travel Times A Short
Tutorial

• 2 Possible Sources of Travel Time Information
• From a prediction model
• From a sample of actual measurements

6
A Simple Prediction Model

• t t01x/c
• t travel time
• t0 free-flow travel time
• x demand flow
• c capacity

7
Example Estimation Problem

• Prediction Model
• t0 5 minutes, x 1900 vph, c 1800 vph
• t.p 511900/1800 10.3 minutes
• Sample Data
• t16 min, t217 min, t312 min
• t.bar (61712)/3 11.7 minutes
• Which should we use?

8
Combining Estimates

• Suppose
• t.real is (unknown) actual mean travel time
• t.p is normally distributed
• mean t.real, variance 16
• t.bar is normally distributed
• mean t.real, variance 9
• Optimal estimate
• t.opt (9/25)(10.3) (16/25)(11.7) 10.7

9
BASIC PROBLEM How to Combine Available
Information to Obtain Best Estimates of Travel
Time on All Links
Loop Detector
Spot Speed
ATR
Traffic Count
Delay Study
10
HIERARCHICAL BAYES APPROACH

• Stage 1 Parametric models gives predicted travel
  times from
• Capacity, Length, Free-flow speed, Volume, Signal
  Timing?
• Stage 2 Update predicted travel time using
  link-specific data
• Spot speed, Delay study, Volume count
• Stage 3 Update predicted travel time using data
  from neighboring links
• Spatial analysis problem
• Areal model approach

11
OVERVIEW OF PHASE IINETWORK WIDE ESTIMATION OF
ARTERIAL TRAVEL TIME

• Required Tasks
• Develop Network Models of Twin Cities Arterial
  System for Target Years (1995, 2000, 2005)
• Design Database for Storing Historical Arterial
  Data
• Link characteristics
• Link volumes
• Signal Timing
• Spot speed/Intersection delay/travel time
• Acquire and Store Historical Data
• Acquire/Modify/Develop Spatial Analysis Software
• Compute Estimates of Travel Time

12
PHASE I WORKPLAN

• Task 1 Review Existing Data Sources
• Task 2 Identify Candidate Link Performance
  Models
• Task 3 Design Link Model Evaluation
• Task 4 Evaluate Links Performance Models

13
One Piece Parametric Models

• BPR Function
• Conical Delay Function

14
Simplest Two-Piece Model

• t running time signal delay
• L length
• v running speed
• C signal cycle length
• g duration of green phase
• s saturation flow

15
First Pilot Study
16
First Pilot Study Results

• Green BPR model
• Black Conical model
• Red HCM model
• Blue Singapore model
• Pink Skabardonis-Dowling Model

17
First Pilot Study Conclusions

• 1. Floating Car Method
• Sensitive to relative fraction of stopping vs
  non-stopping runs
• Difficult to determine fraction which gives
  unbiased estimate
• 2. Stop Speed plus Intersection Delay Method
• Requires 3 technicians, when video vantage point
  not available
• 3. Parametric Models
• HCM best reproduced SSID measures
• Not conclusive because of low V/C ratio

18
Second Pilot Study License Plate Matching
19
Second Pilot Study Procedure

• Observers at each end of link record last 3
  characters of license plate numbers
• Computer program adds time-stamp to each record
• Computer program matches license plate numbers
  and computes travel times
• Video camera records traffic during license plate
  study
• Ground truth travel times extracted from video

20
Distributions of Observed Travel Times
21
Second Pilot Study Conclusions

• 1. License plate matching can produce reasonable
  sample sizes with only two observers
• 2. License plate matching biased towards
  over-representing non-stopping vehicles
• 3. Use of license plate method will require
  correction to remove bias

22
Proposed Correction Method

• Assume travel time distribution is a normal
  mixture with two components
• Use EM algorithm to estimate means of two
  components from license plate sample
• Use ground-based video to obtain unbiased
  estimate of mixing fraction

23
Testing Results
24
Testing for Classification Accuracy

• Suppose for travel time ti there were n1
  non-stopping vehicles and n2 stopping vehicles
• The probability of an observation being from a
  non-stopping vehicle is
• Use binomial test to check if n1 and n2 are
  inconsistent with estimated Pi

25
Results of Binomial Tests
26
Field Study Data Collection

• Sample of 55 arterial links drawn using 1995
  planning model
• During Summer 2006, each sampled link visited by
  2 observers during AM or PM peak period
• License plate sample plus ground video obtained

27
Field Study Data Reduction

• Use EM algorithm to estimate component means at
  for each location
• Use video to obtain unbiased estimate of
  stopping/non-stopping factions
• Compute estimated mean travel time for each site
• Compute confidence interval for mean travel time

28
Field Study Model Testing

• 1. For each site, obtain additional data needed
  for parametric models
• volume, capacity, signal timing, progression
  factor, saturation flow
• 2. For each site, for each model, compute
  predicted travel times
• 3. For each model, identify whether or not
  predicted travel time falls within confidence
  interval

29
Phase I The End

• Questions?