Assessment and Refinement of Real-Time Travel Time Algorithms for Use in Practice

Assessment and Refinement of Real-Time Travel Time Algorithms for Use in Practice

Project Goals

Determine best approach for travel time estimation for real-time applications

Recommend algorithm Midpoint Coifman

Provide statistical analysis so performance of algorithm is understood under different conditions (free-flow, congestion, incidents(?))

Provide confidence in travel time estimations

Task 1: Impact of Various Factors on Travel Time Estimation Investigate impact of several factors on travel time estimation Detector Spacing Algorithm Data Quality Highway geometry

Today: Initial results on Detector Spacing and Algorithm Very preliminary results on Data Quality

Deliverable: Full results at next meeting (Nov) Note: Expansion and extension of Task 1 in work order

Task 2: Ground Truth Data Collection Ground Truth Collection to be done by

consulting company $5000 budget for data collection

Initial set of runs in October/early November Select corridors and try to finalize plan today

Analyze data from runs by early January Second set of runs Jan/Feb 2007 Deliverable: Initial Collection done by Nov 10,

2006

Task 3: Sensitivity Analysis

What input parameters are algorithms sensitive to? Reveal biases the algorithms may have to different

parameters Include study of work using Kalman filters (most

recent ITS seminar) Real-time and deals well with dirty data

Survey other algorithms proposed and in use Deliverable: Presentation/Memorandum Nov 10,

2006

Future Tasks

Task 4: Algorithm Refinement Technical Memorandum due Dec 1, 2006

Task 5: Detailed Comparative Study of Algorithms Technical Memorandum due March 23, 2007

Task 6: Draft Final Report Due May 18, 2007

Task 7: Final Report Due June 15, 2007

Current Work

Travel Time Estimation Algorithm Comparisons Coifman Algorithm Midpoint Algorithm (ODOT algorithm)

Quantification of Travel Time Estimation Error Detector Spacing Data Quality Road Geometry Algorithm

Algorithm Comparisons

Travel time estimates from archived loop data Coifman algorithm

Four different scenarios Midpoint algorithm

Two different scenarios

Probe vehicle data Probe cars TriMet bus data

Variety of traffic conditions Congested vs. Free Flow Incidents

Free Flow Conditions

292.00

293.00

294.00

295.00

296.00

297.00

298.00

299.00

300.00

17:03 17:05 17:07 17:09 17:11 17:13

Time

Mile

po

st (

mi.)

Probe

Coifman u/s

Coifman d/s

Midpoint

Incident Conditions (Congestion)

0.00

1.00

2.00

3.00

4.00

5.00

6.00

7.00

8:28 8:32 8:36 8:40 8:44 8:48 8:52 8:56 9:00

Time

Mil

ep

os

t (m

i.)

Probe

Coifman u/s

Coifman d/s

Midpoint

Large Detector Spacing

293.00

294.00

295.00

296.00

297.00

298.00

299.00

300.00

8:11 8:13 8:15 8:17 8:19 8:21 8:23

Time

Mil

ep

os

t (m

i.)

Probe

Coifman u/s

Coifman d/s

Midpt

Travel Time Estimation Errors Coifman u/s

R2 = 0.4329

R2 = 0.3177

0

20

40

60

80

100

120

0 0.5 1 1.5 2 2.5 3 3.5

Detector Spacing (mi.)

Tra

ve

l Tim

e E

sti

ma

tio

n E

rro

r (s

ec

)

Uncongested

Congested

Error vs. Detector SpacingCoifman d/s

R2 = 0.2269

R2 = 0.2357

0

10

20

30

40

50

60

70

80

90

100

0 0.5 1 1.5 2 2.5 3 3.5

Detector Spacing (mi.)

Tra

ve

l Tim

e E

sti

ma

tio

n E

rro

r (s

ec

)

Uncongested

Congested

Error Vs. Spacing contd….Midpoint

R2 = 0.546

R2 = 0.4607

0

20

40

60

80

100

120

0 0.5 1 1.5 2 2.5 3 3.5

Detector Spacing (mi.)

Tra

ve

l Tim

e E

sti

ma

tio

n E

rro

r (s

ec

)

Uncongested

Congested

Data Quality

0

1

2

3

4

5

6

7

8

9

10

W/O DQ With DQ

RM

SE

Coifman u/s

Coifman d/s

Midpoint

Loop Detectors On I-84

Indicates WB detectors

33rd Ave (mp 2.1)

Detector Locations on US 26

EB detectors

Skyline, mp 71.37

26 @ 405, mp 73.62

Data Quality Flags

Data is flagged as invalid if it meets any of the following criteria (adapted from TTI criteria) 20 second count > 17 Occupancy > 95% Speed > 100 MPH Speed < 5 MPH (probably being removed) Speed = 0 and Volume > 0 Speed > 0 and Volume = 0 Occupancy > 0 and Volume = 0

Data quality is determined (in part) by percentage of 20-second readings for which a detector fails one of the above tests

Ground Truth Collection

Two Phases (Pilot Phase, Final Phase) Phase 1: Soon (October/early November) Phase 2: January/February

Focus on only two corridors in initial phase Second phase may add additional corridors Initial Number of Runs (my calculations show

~50 runs for 5% error at 95% confidence) Start with 20 runs/corridor Getting quotes from several firms

Ground Truth Data Collection Corridor Selection Criteria (Adapted from Sue

Ahn’s criteria for SWARM project) Must have moderate level of recurrent congestion Require reasonable loop detector spacing to

ensure good evaluation of algorithms Ideally detectors have high data quality Construction Schedule – avoid times/areas when

there is construction

Detector Locations I-5 S of Downtown

Detector Locations - 217

I-5 N Wed, Oct 4, 2006

traffic flow

I-5 S, Wed, Oct 4, 2006

traffic flow

217 N, Wed, May 17, 2006

traffic flow

217 S, Wed, May 17, 2006

traffic flow

I-205 N, Wed, Oct 4, 2006

traffic flow

I-205 S, Wed, Oct 4, 2006

traffic flow

How Good is Good Enough?

Source: Travel Time Data Collection for Measurement of Advanced Traveler Information Systems Accuracy (Toppen, Wunderlich) June 2003, MTS Systems

> 5% accuracy, limited benefit

Below this line, commuter is better off using historical experience (13%-21% accuracy)

Data is for Los Angeles

What do you want?

What are your expectations for the project? What is a ‘good enough’ estimate?

Maximum allowable error? Is assumption 8%-10% accuracy ‘good enough’

OK? Should this be investigated more? Can we prioritize recurring congestion over

incidents? Which corridors are a priority to you?

So we can concentrate on those corridors (probe vehicle data  collection etc.)

I-84 (East and Westbound) Limited number of loop detectors and poor data quality

I-405 (North) Relatively short (≈ 3.5 miles) and limited loop detectors

I-405 (South) This freeway corridor is relatively short (≈ 3.5 miles), lightly congested during peaks

US-26 (East and Westbound) Was under construction – what is data quality like on 26?

OR217 Northbound Sue had problems with the queue location – when are we getting detectors again?

OR217 Southbound Looks pretty good – when are detectors going to be turned on?

I-205 Northbound Looks pretty good. When are new loop detectors going in?

I-205 Southbound This corridor is lightly congested during the peak periods. The speed remains above 40 mph throughout the entire corridor.

I-5 Upper-section Northbound Poor data quality

I-5 Upper-section Southbound Poor data quality??

I-5 Lower-section Southbound A recurrent bottleneck is located near the Wheeler Ave. on-ramp. The resulting queue, however, usually propagates only 2 – 3 miles

upstream. A queue that forms near Wheeler Ave. often overrides the upstream bottleneck near Columbia Blvd (in the upper-section of I-5). In

this case, the entire queue propagates upstream of the Interstate bridge, where loop detector data are not available to PSU. I-5 Lower-section Northbound

There are several of sections along this corridor where the spacing of adjacent loop detectors is very large. 2.5 miles between Terwilliger Blvd. and Macadam Ave., 3 miles between Nyberg Rd. and Stafford Rd.

In terms of loop detector spacing, ORE 217 southbound and I-205 northbound show relatively small average spacing (≈ 0.7 and 1.1 miles respectively) as well as smaller maximum spacing (< 2 miles) compared to the other two candidate corridors. Hence, measurements from the loop detectors on these two corridors will provide better assessment of freeway conditions and their dynamics.

I-5 Lower Northbound 217 Southbound I-205 Northbound I-205 Southbound

Implemented February, 2006 November, 2005 December, 2005 December, 2005

Length of study section 17 miles 7 miles 19 miles 19 miles

Number of loops 51 24 46 46

Number of on-ramps (with loops) 16 12 9 18

Level of congestion: pre SWARM (duration, queue length, low speed)

(2-3 hrs, 6 miles, 25-35mph) (2-4 hrs, 4-6 miles, ~25mph) (2-3 hrs, 5 miles, ~30mph) (2 hrs, 4-6 miles, ~35mph)

Level of congestion: post SWARM (duration, queue length, low speed)

(2-3 hrs, 6 miles, 25-35mph) (2-4 hrs, 4 miles, ~25mph) (2-3 hrs, 5 miles, ~30mph) (2 hrs, 3-5 miles, ~40mph)

Queue contained within corridor? (pre SWARM, post SWARM)

AM: (Yes, Yes) PM: (Yes, Yes)

AM: (Yes, Yes) PM: (Not clear, Not clear)

AM: (Not clear, Not clear) PM: (Not clear, Not clear)

AM: (Yes, Yes) PM: (Yes, Yes)

Coverage of loop detectors (miles/loop station)

1.14 (max: 3.1) 0.74 (max: 1.2) 1.1 (max: 1.9) 1.46 (max:4.3)

Data quality (Avg % good readings, Min %)

(94.2, 21.8) (99.2, 98.9) (98.0, 94.8) (98.3, 85)

No. of Loops < 90% 3-7 0 0 1-2

Construction schedule Late summer of 2006