Investigation of Sources of Congestion at the t th Hampton Roads - - PowerPoint PPT Presentation

Investigation of Sources of Congestion t th at the Hampton Roads Bridge Tunnel (HRBT)

Mecit Cetin, Ph.D. Filmon G. Habtemichael, Ph.D. Khairul A Anuar

1

Khairul A. Anuar

Outline Outline

• 1. Introduction
• 2. Purpose and scope of the project
• 2. Purpose and scope of the project
• 3. Data sources
• 4. Exploratory data analysis

5 Quantifying incident‐induced delay

• 5. Quantifying incident induced delay
• 6. Identifying bottleneck locations
• 7. Conclusions

2

Introduction Introduction

• HRBT connects the cities of Norfolk and Hampton (I‐64)

p ( )

– Two lanes per direction, shorter tunnel clearance in the WB direction (14’6” EB and 13’6” WB)

HRBT ff f h i d

• HRBT suffers from heavy congestion due to:

– Demand >> Capacity – Incidents

3

Purpose and scope of the project Purpose and scope of the project

• 1. Identifying specific sources of congestion at

the HRBT

• Estimate the impact of delay along the HRBT, and
• Estimate the impact of over‐height vehicles and traffic

p g incident

2 Evaluate the spatial and temporal

• 2. Evaluate the spatial and temporal

characteristics of traffic congestion

Id tif ifi b ttl k l ti d

• Identify specific bottleneck locations, and
• Determine the capacity of those bottlenecks

4

Data sources Data sources

Provided by VDOT and INRIX: Year 2013

• 1. Incident records (two different databases)

( )

• Incident time, clearance time, type of incident ….

2 INRIX speed and travel time data

• 2. INRIX speed and travel time data
• Probe‐based speed and travel time for segments on

HRBT HRBT

• 3. Continuous count stations
• Volume speed and occupancy aggregated at 5 &15
• Volume, speed and occupancy aggregated at 5 &15

minutes

5

Matching incident databases Matching incident databases

T d t b (B/T C id l )

• Two databases (B/T Corridor only)

– VA Traffic DB:

• All incidents impacting traffic including disabled vehicles, crashes, B/T stoppage events
• 2,556 B/T stoppage events: Don’t know the cause?

– Response Time DB:

• Detailed log for responders and events (more records than VA Traffic DB)

Th f B/T i h i h k d b i

• The cause for B/T stoppage events given, e.g.: over‐height truck, debri, etc.
• Match the B/T stoppage events in VA Traffic to Response Time DB

– Time stamps, travel direction, duration

• Out of 2,556 events, 1,280 matched confidently

– 64% of the matched events classified as “over‐height truck”, 20% as “disabled vehicle”, 10% as “crashes” … disabled vehicle , 10% as crashes … – The rest 1,276 were then proportionally distributed using the proportions

• btained from the matched ones

6

Exploratory data analysis – incident data Exploratory data analysis – incident data

Incident frequency by category

7

Exploratory data analysis – delay at HRBT Exploratory data analysis delay at HRBT

Median travel time by day of the week (WB) y y ( )

8

Exploratory data analysis – delay at HRBT Exploratory data analysis – delay at HRBT

Median travel time by day of the week (EB) y y ( )

9

Exploratory data analysis INRIX data Exploratory data analysis ‐ INRIX data

10

Methodology adopted Methodology adopted

A il bl h i f if i i id i d d

• Available techniques for quantifying incident‐induced

delay (IID)

1 D t i i ti i th b d 1. Deterministic queuing theory‐based 2. Shockwave theory‐based 3. Simulation‐based, and 3. Simulation based, and 4. Statistical procedures‐based

Thi j t St ti ti l d

• This project: Statistical procedures
• Data driven
• Estimate IID by examining similar traffic patterns and
• Estimate IID by examining similar traffic patterns and

establishing reference traffic profile under “normal” traffic conditions

11

Methodology adopted Methodology adopted

• Previously, researchers estimated IID by grouping traffic based

y, y g p g

• n day‐of‐the‐week, time‐of‐the‐day and daily volume of

traffic.

• Can be misleading:

– Similar total daily volumes but different profiles

250 300 WB)

• Sep. 11, 2013 (DT = 34207 veh)

150 200 h in 5 minutes (W

• Oct. 13, 2013 (DT = 34376 veh)

50 100 Count of veh

• Aug. 31, 2013 (DT = 34904 veh)

12

0:00 0:50 1:40 2:30 3:20 4:10 5:00 5:50 6:40 7:30 8:20 9:10 10:00 10:50 11:40 12:30 13:20 14:10 15:00 15:50 16:40 17:30 18:20 19:10 20:00 20:50 21:40 22:30 23:20

Methodology adopted Methodology adopted

• To estimate IID, a reference travel time profile needs to be

p

• established. This was done by examining similar traffic

patterns.

The question is:  H t t bli h

30 35

An incident happens and travel time increase more than Incident affected subject travel time profile

 How to establish reference profile?  How to select d d

20 25 me (min)

usual Delay due to demand Incident- induced delay p

candidate or similar profiles?  What weighting

10 15 Travel tim

Delay due to demand (IID) I id f

mechanism should be adopted?

5

Queue due to incident clears and traffic is back to normal

Incident-free reference travel time profile

13 7:00 7:35 8:10 8:45 9:20 9:55 10:30 11:05 11:40 12:15 12:50 13:25 14:00 14:35 15:10 15:45 16:20 16:55 17:30 18:05 18:40 19:15 19:50 20:25 21:00 21:35 22:10 22:45 23:20 23:55

Measuring similarity between traffic profiles g y p

14

How to establish the reference profile?

1 D t i hi h TYPE f t ffi fil t

How to establish the reference profile?

1. Determine which TYPE of traffic profiles to use

• Volume‐based (flow at upstream and downstream sensors),
• Travel time‐based, and

,

• Hybrid of volume and travel time‐based

2. Identify CANDIDATE (similar) traffic profiles y ( ) p

• Day‐of‐the‐week‐based,
• K‐nearest neighbor‐based, and
• Cluster analysis based
• Cluster analysis‐based

3. Apply weighted summation to establish a REFERENCE profile

• Equal weight
• Equal weight,
• Inverse distance (Shepard’s method), and
• Rank‐based (Rank Exponent method)

15

Which TYPE of traffic profiles to use? Which TYPE of traffic profiles to use?

16

How to identify CANDIDATE traffic profiles? How to identify CANDIDATE traffic profiles?

• Day‐of‐the‐week‐based
• Grouping days of the week by season of the year
• d

f h k b

• Grouping days of the week by ADT
• K‐nearest neighbor (K‐NN)‐based, and
• Select the nearest K data points
• K was in the range from 1 to 10

Cl t l i b d

6 7

• Cluster analysis‐based

3 4 5 Height

• Ward’s minimum variance was used

1 2

• Optimum number of clusters was 13

17

Establishing the REFERENCE profile Establishing the REFERENCE profile

18

Experiments Experiments

19

C l d b d Calendar based methods not so good. K‐NN method with k=6, β 0 90 and α 0 10 β = 0.90 and α = 0.10 provided the least prediction error

Delays at the HRBT Delays at the HRBT

Percentage of total delay and volume of traffic

18%

14% 16% 13%

14% 16% 18% Percentage of Total Delay

9% 8% 13% 8% 9% 9% 9% 9% 8% 9%

10% 12% Total Delay Percentage of Volume of traffic

4% 8% 7% 8% 8% 7% 4% 8% 7% 8% 9% 8% 8% 8%

4% 6% 8%

4% 4% 3%

0% 2% 4% Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

21

Total delay by day of the week Total delay by day of the week

22

Sources of total delay Sources of total delay

C h Debris Other Disabled Crash 8% 1% Other 4% Disabled Vehicle 7% Over‐ height 8% Demand 72%

23

Breakdown of incident induced delay Breakdown of incident‐induced delay

V hi l Mainten ance 2% Wide 2% Vehicle Fire 1% Other 3%

WB

Wide 4% Vehicle Fire 1%

EB

Over- height Debris 3% Hazmat 3% Over- height 27% Maintena nce 1% 4% Other 13% g 28% Vehicle Accident 15% 3% 27% Debris 7% Hazmat 3% Disabled Vehicle Multi- Vehicle A id Disabled Vehicle Vehicle Accident 13% 7% Vehicle 23% Accident 20% Vehicle 25% Multi- Vehicle Accident 13% Accident 6%

24

Cost of delay at the HRBT Cost of delay at the HRBT

Constants used (Schrank et al., 2012 and Caltrans, 2011)

25

Annual cost of delay at the HRBT corridor Annual cost of delay at the HRBT corridor

26

Identifying bottleneck locations Identifying bottleneck locations

N‐curves and speed heat‐maps were used to identify the p p y location of bottlenecks and their capacities

27

Inside or at the entrance of the tunnel

Congestion inside the tunnel the tunnel Congestion before the tunnel entrance the tunnel entrance

28

Flow rates observed inside and at the entrance

• f the tunnel

EB EB

29

Flow rates observed inside and at the entrance

• f the tunnel

WB WB

30

Vehicle classes Vehicle classes L 1 EB Lane 1 EB

31

Vehicle classes Vehicle classes L 2 EB

Very few trucks

Lane 2 EB

Very few trucks in Lane 2

32

Queue discharge flow rates by travel lanes Queue discharge flow rates by travel lanes

• EB direction

33

Potential improvements in capacity Potential improvements in capacity

• Keeping congestion outside of the tunnel can improve

h h b d throughput by 21% in EB direction

• If right lane operates at the same level as left lane during the

k h th th h t f th i b 15%

If trucks are t i t d d i

peak hours, the throughput can further increase by 15%.

restricted during peak‐hours If congestion is kept outside the tunnel

34

Conclusions Conclusions

1. Annual congestion cost at the HRBT is 1.13M vehicle‐hours

• r $33 2M
• r $33.2M

2. Turnaround of over‐height trucks contributes the most to incident‐induced delay ($2.58M ) y ($ ) 3. Even though commercial vehicles constitute 3.1% of the HRBT traffic, they incur about 10% of total delay cost 4. Bottleneck capacity varies significantly depending on their locations (especially in EB) 5. Throughputs of left and right lanes are significantly different 6. Keeping congestion outside the tunnel and restricting trucks d i k h i ifi tl i t ffi during peak‐hours can significantly improve traffic

• perations at the HRBT

35

Acknowledgements Acknowledgements

VDOT

• VDOT

– Chris Moore, Jessie Neal, Rusty Fitzhugh, Thomas Schinkel Brooke Kordas Neil Reed Stephany Schinkel, Brooke Kordas, Neil Reed, Stephany Hanshaw, Michael Griffin, Frank Shearman, Oliver Rose, Robert "Bud" A. Morgan, Morris O. "Pete" P D C k Ad J k K C d C th Pearson, Dwayne Cook, Adam Jack, Ken Coody, Cathy McGhee.

• HRTPO

HRTPO

– Keith Nichols

• ODU

ODU

– Michelle Allen, Ozhan Unal, Seth Schipinski, and Gulsevi Basar

36

Transportation Research Institute www.tri-odu.org www.tri odu.org

Questions? Questions?

37