Princeton University Prospect Eleven Dec. 6, 2005 Towards Real-Time MinETA Routing Tuesday, December 6, 2005 RPI, Troy, NY Alain L. Kornhauser Professor,

Dec. 6, 2005

Princeton UniversityProspect Eleven

Towards Real-Time MinETA Routing

Tuesday, December 6, 2005

RPI, Troy, NY

Alain L. KornhauserProfessor, Operations Research & Financial Engineering

Dec. 6, 2005


Finally… Convergence of

GPS

Real-Time Decisions that substantially improve the Bottom line of Carriers & Shippers

& Robust OS

& Memory

Real-Time Mobile Information

& Wireless

&

& Processing

Optimal Real-Time Management

& Control of Mobile Assets

Dec. 6, 2005


1 6

1

1 1

2

2

2

3

3

3

4

3

5 5

6

6

8

8

9

6

•PROBLEM: How to get from A to B

•Many Paths, Each with a Different Value to the Decision Maker

Company Fundamental

4

Dec. 6, 2005

Princeton UniversityProspect Eleven Including Real-Time Information:

Concepts

Peak Hour Characteristics & Return to

Normalcy250

350

450

550

650

750

850

950

1050

0 10000 20000 30000 40000 50000 60000 70000 80000

Function

Real Data

580

680

780

880

980

1080

1180

1280

0 10000 20000 30000 40000 50000 60000 70000 80000

Series2

Series1

During Peak Hours, Traffic Patterns Remain at a relatively constant distance to Historical Estimate

There will be a time at which traffic patterns will return to free flow conditions

Moorland - Downtown

Burleigh - Zoo

Dec. 6, 2005


•Method of “smoothing” a time series of observations•Most recent observations are given a high weight and previous observations are given lower weights that decrease exponentially with the age of the observation

Including Real-Time Information: Concepts

Exponential Smoothing

310)1(11

tSyS ttt

10)1(

10)1(

11

111

bSSbbSyStttt

tttt

10)1(

10)1(

10)1(

11

111

ISy

I

bSSb

bSIy

S

Ltt

tt

tttt

ttLt

tt

Single

Double

Triple

Dec. 6, 2005


Including Real-Time Information: Solution

•During Peak Periods:

•Adaptation of Double Exponential Smoothing

•Trend is the Trend of the Historical Estimate

•Observation weighted with Most Recent Estimate + Slope for Smoothed Estimate

•Forecast done by adding trend to most recent estimate

}1,0{ parameters smoothing ~ }{

functionparameter 10 Estimated)(

:

)()(

:

)()()1(

:

11

111

n

nnnn

nnnnn

t

where

tt

Forecast

tt

Smoothing

SS

SXS

Dec. 6, 2005


Including Real-Time Information: Solution

•During Non-Peak Periods

•Adaptation of Double Exponential Smoothing

•Trend is decay to free flow Conditions

}1,0{ parameters smoothing ~ c},,{

3Chapter in estimatedfunction parameter 10)(

:

)0()1(

:

)0()1()1(

:

1

11

n

nn

nnn

t

where

cc

Forecast

Smoothing

SS

SXS

Dec. 6, 2005


250

450

650

850

1050

1250

1450

0 10000 20000 30000 40000 50000 60000 70000 80000

Smoothing

Function

Real Data

(c)

(b)

(a)

250

450

650

850

1050

1250

1450

0 10000 20000 30000 40000 50000 60000 70000 80000

Smoothing

Function

Real Data

0.0

0.2

0.4

0.6

0.8

1.0

0 10000 20000 30000 40000 50000 60000 70000 80000

Weights

Figure: Empirical testing of forecasting algorithm. (a) Forecast from most recent observation. (b) Weighting on most recent observations. (c) Realization of travel time data.

Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Real-Time Congestion Dependent Sat/Nav

•250 Volunteers using CoPilot|Live commuting to/from RPI

• CoPilot continuously shares real-time probe-based traffic data

• CoPilot continuously seeks a minimum ETA route

“Advance” project Illinois Universities

Transportation Research Consortium

The late 90s

Conducted it’s version of the abandoned “Advance” project

Link

&

Dec. 6, 2005


Trip Length Distribution11,272 Trips; 4,211Trip-miles

020

4060

80100

120140

1 18 35 52 69 86 103 120 137 154 171 188

Distance in tenths of miles

# T

rip

s

Dec. 6, 2005


Dec. 6, 2005


Record & Send: Mi, Mj, Tj-Ti

Dec. 6, 2005


Dec. 6, 2005


Cell Phones as Traffic Probes:A Way to Get Started?

Dec. 6, 2005


Cell Probe Technology• Part of general trend away from fixed sensors toward vehicle-

based information• Reflects frustration with high costs and slow pace of

deployment for traditional sensors• Characteristics:

– Low cost– full regional coverage– performance-based, and – self sufficient business model

Dec. 6, 2005


Cell Probe Technology• Practical success requires more than cell phones• Cell phone movement based on cell location and “hand-offs”

from one cell to another• Pattern recognition techniques filter out data from those not

on the highway• Then traffic algorithms generate travel times and speeds on

roadway links• Cell phones need to be turned on, but not necessarily in use• Full regional systems in place in Baltimore, Antwerp, and Tel

Aviv = 4,600 miles

Dec. 6, 2005


Cell Probe Technology

GSM

SampleObserverSampleSample

ObserverObserver

Cell

Cell

Cell

Directionof travel

GSMGSM

SampleObserverSampleSample

ObserverObserver

Cell

Cell

Cell

Directionof travel

Dec. 6, 2005


Cell Probe Privacy

Speeds on road links

Personal cellular

position data

Estimotion sample

observer

Estimotion traffic

situation

ITIS publishing systems

Cellular network operator

Cellular network operator ITISITIS

Firewall

Speeds on road links

Personal cellular

position data

Estimotion sample

observer

Estimotion traffic

situation

ITIS publishing systems

Cellular network operator

Cellular network operator ITISITIS

Firewall

Dec. 6, 2005

Princeton UniversityProspect Eleven Handset 49, part 1

Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005


Dec. 6, 2005

Princeton UniversityProspect Eleven Handset 49, part6

Dec. 6, 2005

Princeton UniversityProspect Eleven Handset 49, full trip

Dec. 6, 2005

Princeton UniversityProspect Eleven Handset 49, full trip

Dec. 6, 2005


Path-Finding Drive Tests

Handset 47Handset 47 Handset 52Handset 52

Handset 49Handset 49 GPS TrackGPS Track

(b)(a)

(c) (d)

Dec. 6, 2005


Possible Applications• Travel Times

– For message signs; web sites• Performance measures

– Include arterial network– “Top 10” routes– TTI-type reports

• Operations planning– Special events– Work zone management– Evaluation of actions

• Safety– Focus on problem areas and assessments

• Port/intermodal access• Local/regional web sites• Statewide coverage

Dec. 6, 2005


Applications• General Planning and Management

– Regional congestion management– Archived data supports system analysis, ”average day”

information, long-range planning– Integrated regional or corridor management– Plan for “extreme” or special events– Homeland security applications – no-notice evacuations– Rapid evaluation of alternatives– Work zone management– Rural planning and operations– Traffic volume estimates -- future

Dec. 6, 2005


Applications (2)• Performance Measurement

– System performance in near real time

– Reliability measures – critical from user’s perspective (travel time index, planning time index, etc.)

– Performance-based systems – information for operators, users, and the public

– Congestion management – support for HOT lanes and other finance alternatives

– Economic value from partnerships with business – the DOT a part of just in time delivery

Dec. 6, 2005


Applications (3)• Travel Demand and Air Quality Modeling

– Today – Validate travel demand and Mobile6 models– Tomorrow – origin/destination data– Tomorrow – New model development: activity-based and beyond

• Safety– Analysis and prediction– Targeted deployment of safety personnel

• Communication– Public participation – real data on congestion– Near real-time data – web, PDA, 511– Premium 511 service

Dec. 6, 2005


Applications (4)• Freight Operations

– Web- or cell-based distribution of roadway information– Individual dynamic routing recommendations based on congestion– Travel time prediction to improve asset utilization

• Freight Analytics– Strategic analysis of freight movement for congestion mitigation– Origin/destination data to examine flows and set priorities– Support for cost/benefit and alternatives analysis

Dec. 6, 2005


Safety Example• Operational tool

– Assign patrol cars to road segments based on:• Average speed – Z percent above speed limit

• Trigger points – X percent of traffic more than Y percent above speed limit

– Identify trends and historical patterns

– Short-term forecasts

• Evaluation tool – near real-time– Assess what worked and how well

– Statistical analysis of patterns

Dec. 6, 2005


Safety ExampleRoad # 5

0

20

40

60

80

100

120

0 5 10 15 20 25

Hours

Km

/h

5290

4920

4918

Speed Limit

Documents

Princeton University Prospect Eleven Dec. 6, 2005 Towards Real-Time MinETA Routing Tuesday, December 6, 2005 RPI, Troy, NY Alain L. Kornhauser Professor,