Transportation leadership you can trust.

presented to

ITS Georgia

presented by

Richard Margiotta, PrincipalCambridge Systematics, Inc.

October 5, 2009

Developing and Predicting Travel Time Reliability

2

Overview

Defining reliability

Measuring reliability

Predicting reliability

Tie this to the current SHRP 2 Project L03: Analytic Procedures for Determining the Impacts of Reliability Improvement Strategies

3

What is Travel Time Reliability?

Definition: A consistency or dependability in travel times, as measured from day to day and/or within different times of day

Travelers on familiar routes learn to “expect the unexpected”

• Their experience will vary from day-to-day for the same trip

Reliability “happens” over a long period of time

• Need a history of travel times that capture all the things that make them variable

4

Averages don’t tell the full story

Jan. Dec.July

Traveltime

How traffic conditions havebeen communicated

Annual average

Jan. Dec.July

Traveltime

What travelers experience

Travel times varygreatly day-to-day

What theyremember

5

Communicating the Benefits of Improvements

When Mn/DOT’s ramp meters were turned off ( “before period”) in 2000:

22-percent increase in average travel times

91-percent decline intravel time reliability

Traveltime

Before After

Avg. day

Small improvement inaverage travel times

Larger improvement intravel time reliability

Reliab.

Before After

Worst dayof month

6

Reliability has costs!

Variability in travel times means that extra time must be planned for

In other words, travelers have to leave earlier – they build in a BUFFER to their trip planning, or suffer the consequences

These extra costs have not been accounted for in traditional economic analyses of transportation improvements

7

Reliability has costs (cont.)

Planned extra time at least as costly as regular travel time

Some studies place the Buffer’s costs at 1-6 times higher than average travel time

Some trips will still exceed the Buffer – late penalties

Some trips will take much less than the Buffer – early arrival penalties

Reliability (or the lack of it) just says that travel times are inconsistent/variable – it doesn’t tell you why!

Measuring Reliability

9

A Model of Congestion and Its Sources

n = Source of Congestion= Source of Congestion

Base DelayBase Delay(“Recurring” or “Bottleneck”)(“Recurring” or “Bottleneck”)

PhysicalPhysicalCapacityCapacity

……interacts withinteracts with…… DemandDemandVolumeVolume4

Event-RelatedEvent-RelatedDelayDelay

TotalTotalCongestionCongestion

Daily/SeasonalDaily/SeasonalVariationVariation

SpecialSpecialEventsEvents

PlannedPlanned

……determinedetermine……EmergenciesEmergencies2 31

……lowers capacitylowers capacityand changes demandand changes demand……

Traffic ControlTraffic ControlDevicesDevices

Roadway EventsRoadway Events

WeatherWeather

IncidentsIncidents

WorkWorkZonesZones

5

6

7

…can cause…

…can cause…

…can cause…

10

SHRP 2 Project L03: The Data Challenge

Reliability is defined by a long history – at least a year – of travel times (a distribution)

• Implies that automated equipment is the only feasible method of data collection, but...

• Automated equipment not deployed everywhere

So, how can enough empirical data be collected to study the effect on reliability?

• Tap existing data sources as much as possible

• Supplement with data purchased from private vendors

• Rely on a cross-sectional predictive model

11

Analysis Data Set

Traffic DataTraffic Data Incident Incident DataData

Weather Weather DataData

Incident Incident ManagementManagement

Geometric Geometric CharacteristicsCharacteristics

VolumesVolumes

SpeedsSpeeds

DemandDemand

Traffic Traffic StatisticsStatistics

By By Time SliceTime Slice

Section Section Reliability Reliability MeasuresMeasures

Section Section Traffic Traffic

CharacteristicsCharacteristics

Agency Agency GeneratedGenerated

Traffic.comTraffic.com NWS NWS Hourly Hourly

ObsObs

• Service Service PatrolsPatrols

• PoliciesPolicies

• CapacityCapacity• BottleneckBottleneck• Ramp Ramp

MetersMeters

Analysis Data Set

12

I-405 Northbound, Seattle, 4-7 P.M.

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

160,000

180,000

200,000

14 16 18 20 22 24

Travel Time (in Minutes)

Number of Trips

On-Time at Mean + 10% = 85%

On-Time at Mean + 30% = 99%

P10

Median

Mean P90 P95

Buffer Index Buffer Index = = 0.190.19

Skew Statistic Skew Statistic = = 2.022.02

Planning Planning == 1.391.39Time IndexTime Index

Misery Index Misery Index = = 1.481.48

13

Atlanta, I-75 NB, I-285 to SR-120, 2007, Mid-Day

0

1,000,000

2,000,000

3,000,000

4,000,000

5,000,000

6,000,000

5 10 15 20 25 30 35

Travel Time (minutes)

Number of Trips

Av erage TTI = 1.024

95th Percentile = 5.69 minutes

Buffer Index = 0.009

Skew Stat = 1.337

Free Flow Trav el Time = 5.5

minutes

Mean = 5.6 minutes

14

Orlando (Signalized), US-441, PM Peak

0

100

200

300

400

500

600

4 6 8 10 12 14 16 18 20

Travel Time (min)

Fre

q

Avg Speed = 25.5 mphPTI = 2.187Buffer Index = 0.465Skew Statistic =1.676 % on time @40mph = 0.4%

Avg Travel Time= 10.7 min

P95 Travel Time= 15.7 min

15

Travel Time History: D.C. to GW Bridge

16

Travel Time History: Richmond to Philadelphia

17

Influence of Trip Start Time: Test Trip #1

18

D.C. to GW BridgeThanksgiving Holiday Travel

19

Trends in Reliability: Atlanta Study Sections

All Sections

2006 2007 2008

Travel Time Index 1.720 1.800 1.585

Average Travel Time 10.03 10.49 9.22

95th Percentile Travel Time 14.27 15.15 13.60

Buffer Index 0.399 0.428 0.451

80th Percentile Travel Time 11.87 12.40 10.99

Skew Statistic 1.186 1.196 1.308

VMT Change +0.6% -2.1%

Predicting Reliability

21

Project L03 Before/After Studies

Urban freeway study sections revealed 17 before/after conditions:

• Ramp meters – 4

• Freeway service patrol implementation – 2

• Bottleneck improvement – 3

• General capacity increases – 5

• Aggressive incident clearance program – 2

• HOT lane addition – 1

22

SR-520 Ramp Metering

Peak Period: 6:00 – 9:00

Seattle, WA

Before After%

Change

Reliability Metrics

Travel Time Index 1.87 1.66 -11,2%

Buffer Index 0.32 0.31 -3.1%

Planning Time Index 2.46 2.17 -11.8%

Other locations show similar reports (5-11% reduction in PTI)

23

Capacity Addition: Peak Period Comparison

I-405: add 1 GP lane to 2 existing GP + 1 HOV lanes

Travel Time Buffer Planning

Period Index Index Time Index

Before (2007) 2.6 31% 3.4

After (2009) 1.5 44% 2.2

(-42.3%) (-35.2%)

I-94: add 1 GP lane to 2 existing

Travel Time Buffer Planning

Period Index Index Time Index

Before (2001) 1.6 52% 2.4

After (2005) 1.1 28% 1.4

(-31.2%) (-41.7%)

24

1

1.5

2

2.5

3

3.5

4

4.5

5

5.5

1 1.5 2 2.5 3 3.5

Average TTI

95

th %

ile

TT

I

25

Statistical Modeling

Results show that all reliability measures defined in the study can be predicted as a function of average Travel Time Index

Allows reliability prediction from a wide variety of other methods/models that predict the average TTI

• Except that our TTI includes the effect of all sources; models predict recurring-only

• Analysis shows Overall TTI is 15-20% > Recurring Only TTI

26

Statistical Modeling (cont.)

Both average and 95th%ile TTI can be predicted as a function of:

• “Critical” demand-to-capacity ratio− Most significant factor

− Highest d/c ratio of individual segments on the section

• Incident lane-hours lost (minimal work zones in data)

• Hours where rainfall >= 0.05”

RMSEs ~ 20%

27

Congestion by Source: A Simple Analysis with Atlanta Data (peak period)

Identified days where incidents and precipitation occurred

• Recurring only……………………….. 47%

• Incident……………………………….. 35%

• Precipitation…………………………. 10%

• Incident + Precipitation……………. 8%

28

A More In-depth Look at Congestion by Source: Seattle Preliminary Findings

Volume is the primary factor in congestion and the effect of any given type of disruption

Congestion only forms when disruption is big enough to reduce capacity below demand

Once congestion forms in the peak period, the effects linger until the end of the peak period

Disruptions in the leading shoulder of a peak have larger/longer effects than those in the peak or trailing shoulder

29

Probability of Being in Congestion: Rain Versus No Rain I-90 Westbound From Issaquah to Bellevue

30

Comparison of Mean Travel Times With and Without the Influence of Incidents. I-5 Northbound Through the Seattle Central Business District

31

Percentage of Delay By Type of Disruption Influencing That Congestion: Seattle

32

Implications of Project L03 Findings

Volume (demand) is a major determinant of reliability and total congestion

• Determine base congestion and how severe events will be

• Volume can be used to determine when / where incident response vehicles are deployed

• Demand management strategies are a major reliability mitigation strategy

Early AM benefits are lower than late midday benefits

• Problems in midday can cause big evening congestion

From a congestion relief perspective, this suggests

• more emphasis on middle of day

• less emphasis early and late