PLACE IN rte-rum BOX to remove this checkout from your record.
To AVOID FINE return on or before date due.
MAY BE RECAUJED with earlier due date if requested.

 

I
1

DATE DUE I
C " r

//

T351!

1 DATE DUE

DATE DUE

 

,r

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1/96 WWI-$659.14

 

 

Development of Freeway
Incident-Based Congestion Measures

By

Mohammed A. Saif

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Civil and Environmental Engineering

1998

ABSTRACT

DEVELOPMENT OF FREEWAY INCIDENT-BASED
CONGESTION MEASURES

By

Mohammed A. Saif

This research project presents the development of incident-based congestion
measures to be used in testing the effectiveness of an ITS deployment program in
reducing off-peak incident-based congestion. These measures are used to identify the
location and quantify the duration of incident-based congestion on the freeway system as
it exists today. These measures are also used to estimate aggregate delay and the
characteristics of queues caused by incidents on various segments of the freeway
network. To meet these objectives, three measures were established. The first measure
quantifies individual station incident-based congestion. The second is a measure of the
aggregate impact of individual incidents across freeway segments. And finally, the third
measure is an index to quantify system-wide incident-based congestion.

These measures were tested on traffic data obtained from the metropolitan Detroit
freeway system. The estimated system-wide off-peak incident-based congestion produces
an average of 2.5 minutes of incident-based congestion per station per day. The
congestion ranged from a low of 0.4 to a high of 10.5 minutes of delay per day over the

12 segments included in the study.

ACKNOWLEDGMENTS

I would like to express my appreciation and gratitude to my advisor and
committee chairman Dr. William C. Taylor. I am greatly thankful for his remarkable
guidance, patience, kindness and insights throughout my study at Michigan State
University. The completion of this work is deeply indebted to his support.

My appreciation and gratitude are also extended to Dr. Virginia Sisiopiku, Dr.
Thomas Maleck, Dr. Thomas Wolff and Dr. Ramamoorthi, R.V. for serving as members
of the guidance committee and for their assistance.

The financial support of Umm Al-Qura University for my graduate study is
sincerely appreciated. I am also grateful to the Michigan Department of Transportation
for sponsoring the research contained in this document.

A special thank you goes to the support staff of the Department of Civil and
Environmental Engineering, Linda Steinman, Laura Taylor and Mary Wiseman for all of
their administrative assistance.

My appreciation and gratitude are also extended to my colleagues at Michigan
State University for their help and support during my school years, although I can not
refer to them all here.

Finally, my accomplishments would not have been possible without the love and
support of all members of my family. I am especially indebted to my father Dr. Ahmed
Mohammed Noor Saif and my mother Fatima Saif Squer for granting me their
unconditional love and support. I would also like to extend my deepest gratitude to Lisa

Rumisek for her understanding, love and support.

iii

TABLE OF CONTENTS

LIST OF TABLES ............................................................................................................. vi
LIST OF FIGURES .......................................................................................................... viii
CHAPTER 1
INTRODUCTION .............................................................................................................. 1
1.0 Description of the Congestion Problem ................................................................. 1
1.1 Project Location and Description .......................................................................... 2
1.2 Benefits of the ITS Deployment ............................................................................ 3
1.3 Statement of the Problem ...................................................................................... 5
1.4 Objectives and Scope of the Research ................................................................... 5
1.5 Research Methodology .......................................................................................... 6
CHAPTER 2
LITERATURE REVIEW .................................................................................................. 8
2.0 Introduction .......................................................................................................... 8
2.1 The Congestion Problem ...................................................................................... 8
2.1.1 Measures of Congestion ................................................................................. 9
2.1.2 Types of Congestion .................................................................................... 12
2.1.3 Measures to Alleviate Traffic Congestion ................................................... 14
2.2 Characteristic of Traffic Incidents ...................................................................... 15
2.2.1 Incident Definition ....................................................................................... 15
2.2.2 Incident Types .............................................................................................. 15
2.2.3 Cost of Incidents .......................................................................................... 16
2.2.4 Incident Rates ............................................................................................... 17
2.2.5 Behavior of Traffic in the Presence of an Incident ...................................... 19
2.2.6 Impact of Incidents on Capacity .................................................................. 20
2.2.7 Incident Duration and Delay ........................................................................ 21
2.3 Prior ATMS/ATIS Evaluation Studies ............................................................... 26
2.3.1 Compass (Toronto FTMS Evaluation Study) .............................................. 26
2.3.2 INFORM
(Information for Motorists, New York FTMS Evaluation Study) ............... 27
2.3.3 The Korean FTMS Evaluation Case ............................................................ 29
2.4 The Role of ITS in Managing Incident-Based Congestion ................................ 30
2.4.1 Incident Detection and Verification ............................................................. 31
2.4.2 Incident Response and Clearance ................................................................ 33
2.4.3 Electronic Incident Management and Control Strategies ............................ 33
2.5 Summary ............................................................................................................. 35

iv

CHAPTER 3

METHODOLOGY .......................................................................................................... 36
3.0 Introduction ........................................................................................................ 36
3.1 Description of the Database ................................................................................ 37
3.2 Establishment of the Incident-Based Congestion’s Speed Threshold Value ...... 39
3.3 Incident-Based Congestion ................................................................................. 41

3.3.1 Criteria for the Start and End of Incident-Based Congestion .................... 42
3.3.2 Development of the Incident-Based Congestion Periods ............................ 43
3.4 Clustering Station Incident-Based Congestion Periods ...................................... 46
3.4.1 Criteria for Clustering Station Incident-Based Congestion Periods ............ 49
3.4.2 Development of the Clustering Technique .................................................. 51
3.5 Incident Queue Length ........................................................................................ 58
3.6 Categorizing Segment Incident .......................................................................... 59
3.7 Segment Incident-Based Congestion Index ........................................................ 62
3.8 Summary ............................................................................................................. 64

CHAPTER 4

ANALYSIS AND RESULTS .......................................................................................... 65
4.0 Introduction ........................................................................................................ 65
4.1 Description of the study segments ...................................................................... 65
4.2 Processing the traffic data ................................................................................... 66

4.2.1 Processing a single traffic data file .............................................................. 69
4.2.2 Integration of multiple traffic output files ................................................... 71
4.3 Incident-based congestion index ........................................................................ 78
4.3.1 Station incident-based congestion index ...................................................... 78
4.3.1.1 Station incident-based congestion index on WB I-94 .................................. 78
4.3.1.2 Station incident-based congestion index on EB 1-94 ................................... 85
4.3.1.3 Station incident-based congestion index on NB M-IO ................................ 87
4.3.1.4 Station incident-based congestion index on SB M-lO ................................. 90
4.3.2 Segment monthly variation of incident-based congestion index ................. 92
4.3.3 Segment variation of incident-based congestion index ............................... 99
4.3.4 System monthly variation of incident-based congestion index .................. 101
4.4 Incident frequency and queue length ................................................................ 102
4.5 Analysis of traffic volume ................................................................................ 110
4.6 Summary ........................................................................................................... 119

CHAPTER 5

SUMMARY AND CONCLUSIONS ............................................................................ 120

APPENDIX A ................................................................................................................ 124

REFERENCES .............................................................................................................. 130

Table 2.1

Table 3.1

Table 3.2

Table 3.3

Table 4.1

Table 4.2

Table 4.3

Table 4.4

Table 4.5

Table 4.6

Table 4.7

Table 4.8

Table 4.9

Table 4.10
Table 4.11
Table 4.12
Table 4.13

Table 4.14

List of Tables

Incident rates ................................................................................................ 18
Format of the traffic detector input file ........................................................ 38

Sample of a header record, and functioning, reconstructed, and

malfunctioning records ................................................................................ 39
Categories of incident queue lengths ........................................................... 59
Segment description ...................................................................................... 67
Months and dates of traffic data sets ............................................................ 69

Single file of the duration of incident-based congestion index for

1/13/97, AM ................................................................................................. 70
Duration of incident-based congestion 1/13/97 ........................................... 72
Single file output of incident frequency and queue length for

1/ 13/97 AM .................................................................................................. 73

Single file output of factored incident frequency and queue length for

1/13/976, AM ............................................................................................... 74
Factored incident frequency and queue length for 1/13/97 ......................... 75
Duration of incident-based congestion for January 1997 ............................ 76
Incident frequency and queue length for January 1997 ............................... 77
Average monthly station incident-based congestion index ......................... 79
Average winter station incident-based congestion index ............................. 80
Average spring station incident-based congestion index ............................. 81
Average summer station incident-based congestion index .......................... 82

Incident-based congestion index for the different months and segments 83

vi

Table 4.15 Incident-based congestion index and number of daily minutes of congestion

per station for the different directions of the freeways .............................. 101
Table 4.16 Incident frequency and queue length for the different months .................. 104
Table A] Summary of the before and after comparisons ........................................... 129

vii

List of Figures

Figure 1.1 South eastern Michigan ITS-ATMS/ATIS deployment program .................. 4
Figure 2.1 Behavior of traffic under an incident ........................................................... 20
Figure 3.1 Criteria for the start and end of station incident-based congestion .............. 44
Figure 3.2 Flow chart for the station incident-based congestion algorithm .................. 47
Figure 3.3 Duration of congestion on WB [-94, Mon. 7/24/95 ..................................... 48
Figure 3.4 Clustering successive stations of incident-based congestion (Case 1) ........ 50
Figure 3.5 Clustering periods of incident-based congestion belonged to the same station

(Case 2) ........................................................................................................ 50
Figure 3.6 Clustering malfunctioning stations (Case 3) ................................................ 5]
Figure 3.7 Flow chart for the first pass algorithm ......................................................... 53
Figure 3.8 F low chart for the second pass algorithm .................................................... 54
Figure 3.9 Flow chart for the third pass algorithm ........................................................ 56
Figure 3.10 Incident samples ........................................................................................... 57
Figure 3.11 Upstream and downstream parts of the random variable ............................. 58
Figure 3.12 Flow chart for the queue length of incident-based congestion .................... 60
Figure 3.13 Flow chart for the queue length subroutine ................................................. 61
Figure 3.14 Flow chart for the station minutes per segment ........................................... 63
Figure 4.1 Study segments on the map of Detroit ......................................................... 68
Figure 4.2 Seasonal variation in the incident-based congestion index on WB I-94 ..... 86
Figure 4.3 Seasonal variation in the incident-based congestion index on EB [-94 ....... 88
Figure 4.4 Seasonal variation in the incident-based congestion index on NB I-94 ...... 9]

viii

Figure 4.5 Seasonal variation in the incident-based congestion index on SB I-94 ....... 93
Figure 4.6 Monthly variation of the incident-based congestion index for

segments 1, 2, and 3 (WB I-94) ................................................................... 95
Figure 4.7 Monthly variation of the incident-based congestion index for

segments 4, 5, and 6 (EB I-94) .................................................................... 96
Figure 4.8 Monthly variation of the incident-based congestion index for

segments 7, 8, and 9 (NB M-10) .................................................................. 97

Figure 4.9 Monthly variation of the incident-based congestion index for

segments 10, 11, and 12 (SB M-10) ............................................................ 98
Figure 4.10 Segment variation of the incident-based congestion index ........................ 100
Figure 4.11 Monthly variation of the networks incident-based congestion index ........ 103

Figure 4.12 System-wide incident frequency and queue length for

the months of Dec -96, Jan-97, and Feb-97 .............................................. 105
Figure 4.13 System-wide incident frequency and queue length for

the months of Mar —97, Apr-97, and May-97 ............................................ 106
Figure 4.14 System-wide incident frequency and queue length for

the months of Jun —97, Jul-97, and Aug-97 ............................................... 107

Figure 4.15 Average monthly variation in incident frequency

per mile per segment .................................................................................. 108
Figure 4.16 System-wide monthly variation in incident frequency per mile ................ 109
Figure 4.17 Average monthly incident frequency by queue length .............................. 111

Figure 4.18 Station incident-based congestion index versus average

monthly off-peak volume ........................................................................... 113

ix

Figure 4.19 Incident-based congestion index versus volume by direction

and freeway (1-94) ...................................................................................... 114
Figure 4.20 Incident-based congestion index versus volume by direction

and freeway (M-10) ................................................................................... 115
Figure 4.21 Incident-based congestion index versus average month

off-peak volume ......................................................................................... 116
Figure 4.22 Incident-based congestion index versus average segment

off-peak volume ......................................................................................... 116
Figure 4.23 Relationship between incident-based congestion index

and the average daily off-peak station volume .......................................... 117

Figure 4.24 Relationship between the incident-based congestion index, congested

volume, and average total volume. ............................................................ 118
Figure A] Annual sampling plan ................................................................................ 125
Figure A.2 Seasonal sampling plan .............................................................................. 127

Chapter 1

Introduction

1.0 Description of the Congestion Problem

The growth in urban traffic congestion has become a serious problem in all large
metropolitan areas and a cause of discomfort for millions of motorists. Traffic congestion
costs the American people an estimated $100 billion each year in the form of lost
productivity (1). Congestion has been estimated to be increasing at a rate that will result
in 7.3 billion gallons of wasted fuel, and over $50 billion in user cost per year by the year
2005. By then, freeway delay has been projected to have increased by between 300 and
500 percent over the 1985 levels (2’ 3’ 4).

Although traffic congestion is not a new problem for motorists of the inner cities,
it has spread to include the suburban areas. Several factors contributed to the rapid
growth of traffic congestion in the United States. First, the number of registered vehicles
has increased in a disproportionate rate to the population growth. Second, since the near
completion of the Interstate system in the early 1970’s, construction of new highway
facilities has slowed considerably. Third, because of increased access to the automobile
and the suburban migration of both businesses and residents, a higher percentage of
commuters now drive instead of using public transit. This change in commuting patterns
in addition to the preceding factors has increased congestion on the nation’s local streets
and highway systems (5). Traffic congestion forces traffic to reduce speed and sometimes
causes traffic idling, where tons of pollutants are emitted into the air causing a major

environmental problem.

To improve the quality of the nation’s traffic facilities and have an efficient and
environmentally sound transportation system, Congress passed the Interrnodal Surface
Transportation Efficiency Act of 1991 (ISTEA) (6). Included in this bill was funding for
Intelligent Transportation Systems (ITS) with the potential to transport individuals and
goods safely and in a more efficient state-of-the-art transportation system.

To alleviate traffic congestion on the metropolitan Detroit freeways, the Michigan
Department of Transportation (MDOT) and the US Department of Transportation in
conjunction with private firms has planned, developed, and implemented projects using
technologies commonly referred to as Intelligent Transportation Systems. Among those
projects was the ATMS/ATIS (Advanced Traffic Management Systems/Advanced
Traveler Information Systems) deployment program.

Michigan State University has carried out a comprehensive evaluation study to
investigate the effectiveness of this ITS deployment program. The study involved an
assessment of congestion prior to the ITS deployment, an analysis of the effectiveness of
changeable message signs and highway advisory radio; and a study of changes in the
frequency, duration, and location of incident-based congestion. This study addresses the

effectiveness of the system in reducing incident-based congestion.

1.1 Project Location and Description

The system deployed augmented an existing freeway monitoring and control
system covering 32.5 miles of the urban freeway network. The combined system covers a
total of 180 miles. CCTV cameras, machine vision and induction loop detectors,

changeable message signs (CMS), ramp meters, and highway advisory radio (HAR) were

installed as components of the ITS deployment program. Rockwell International designed

and built the system that covered selected freeway segments of I-75, I-94, I—96, [-275, I-

696, M-10, and M-39. Figure 1.1 illustrates the system that existed prior to the expansion

(known as SCANDI) and the newly instrumented system. The type and location of the

various components of the ITS deployment program are shown in this figure. The

program became fully operational in September 1998.

1.2

Benefits of the ITS Deployment
The benefits MDOT anticipated from the ITS deployment program included:
Providing the Michigan Intelligent Transportation Systems (MITS) Center in Detroit
the capability to monitor traffic and congestion through the use of mainline traffic
detectors (loops and machine vision), ramp metering and video surveillance;
Providing traveler information via highway advisory radios (HAR) and changeable
message signs (CMS);
Providing the Michigan DOT the means to detect and verify incidents on selected
corridors in a timely manner;

Providing traffic operations personnel sufficient data to respond to incidents and to
transmit traffic and congestion information to motorists so they can select or modify
their travel plans;

Providing the data required to effectively manage mainline work zones, calculate
mainline volume demand, and predict traffic flow patterns for special events, planned
work/constructions zones, and other special conditions.

Improving the safety of the area’s transportation system;

 

 

 

Existing System

M-[O (8 Mile Rd to Jefferson)
l-94 (Wyoming to Kingsville)
[-75 ([-94 to [4 111. St)
1-375 (1-75 to Jefferson)

 

 

 

 

Newly Instrumental System

[-75 from Lepeeer. tel-94
M-S9 (Silvadome net)

[-696 from [-96 to [-94

M-[O from 8 mile to [-696

[-96 from Beck Rd. to [696

1-75 fiunGreeanto [4th.St.
1-94 from Belleville Rd. Wyoming
[-94 from Kingsville Rd to [0 mile Rd.
1-96ll-21S from [4596 to [VI-[4
1-275 from M-[4 to [-94

[-96 from l-Z‘IS to 1-75

M-39 t'mm M-[O to [-94

 

 

 

 

 

 

 

 

 

 

 

 

 

Total Miles 32.5
it,”
a
a.“ . A
. a. o o I z
. 3' g
a loll-l :
g a“:

 

 

 

 

 

 

 

 

 

TotalMlleelfl
i J
!r .. a.
.g ' a .6:
a: I
- 1 l /
J
’ I
U, .

 

 

 

 

 

Figure 1.1 South eastern Michigan ITS-ATMS/ATIS deployment program

0 Reducing energy and environmental costs associated with traffic congestion; and

finally,

0 Enhancing present and future productivity.

1.3 Statement of the Problem

Over the past 10 years, research, development, and implementation of ITS
technologies has expanded in fields such as information processing, communications,
control, and electronics to serve transportation needs. ITS technologies can provide many
benefits, one of which is to improve the ability to manage recurring and non-recurring
congestion through more efficient use of the existing or improved infrastructure. ATMS/
ATIS is expected to reduce traffic demand following an incident by providing motorists
with information on delay due to congestion and alternate routes, which will increase the

rate of diverted traffic and reduce the gap between demand and reduced capacity.

1.4 Objectives and Scope of the Research

There were three objectives for this research project. The first involved
identifying and quantifying the frequency, duration, and location of incident-based
congestion. The next objective was to estimate delay and the characteristics of queues
caused by incidents on various segments of the freeway network. The third objective was
to evaluate the net reduction in the impact of incidents attributed to the expansion project.
To meet these objectives, the following tasks were identified and addressed:
0 Developing a technique to measure the changes in congestion;

0 Collecting historical data on the SCANDI system before the ITS deployment;

0 Analyzing data to determine the frequency, duration, and location of incident-based

congestion prior to the ITS deployment;

0 Developing an index to quantify the level of incident-based congestion.

1.5 Research Methodology

This research project focused on the congestion caused by incidents during off
peak periods. The morning and evening rush hours were considered to be from 6:30 AM
to 9:30 AM and from 3:30 PM to 6:30 PM, respectively. Congestion that occurred
during this time was included in another study. This study utilized traffic volume and
speed data obtained from the Michigan Intelligent Transportation Systems (MITS)
Center. The data included minute by minute average values of volume, occupancy, and
speed as well as an indication of the number of lanes and the number of lanes reporting
data at various locations on the freeway network. Data was gathered on the detector
stations within the SCANDI system for one month of 1996 and eight months of 1997.

A speed threshold value was set to define congestion, and incidents were flagged
based on this value. To determine the duration of an incident, a criterion was developed
to determine an incident’s start and end times. Next, the queue length parameters of
incidents that covered more than one station were estimated. The incidents were then
divided in different groups based on their queue length, and their total impact was
summed for various segments of the network for each month. The network freeways were
divided into segments of various lengths to help identify areas with high incident rates, as
well as to reduce the effect of malfunctioning detectors on the outcome of this research.

This division will also allow easier before and after comparisons to determine whether

the ITS deployment had different effects on incident frequency and duration on the

different segments of the network.

Chapter 2

Literature Review

2.0 Introduction

A literature review was conducted to obtain the current state-of-the—art related to
the research study described in chapter 1. The literature search was performed to
accomplish three primary goals. First, the current knowledge related to traffic congestion
and traffic incidents was explored. Second, literature was used to develop an
understanding of how similar problems were approached in the past, and to help develop
new ideas to resolve the problem under investigation. Third, three ATMS/ATIS
evaluation studies were reviewed to determine the potential role of ITS in managing
incident-based congestion.

The following sections of the literature review have been divided into the main
areas of traffic research related to this study. They include research conducted on traffic
congestion and it’s related topics followed by traffic incident characteristics and
measures of incident duration and delay. The third topic presents similar evaluation
studies cited in the literature. The fourth part describes the role of ITS in reducing

incident-based congestion. Finally, a summary of the literature review is included.

2.1 The Congestion Problem
Because of the close relationship between traffic congestion and traffic incidents,

the literature that is related to traffic congestion was included in the literature review.

The relevant topics; measures of congestion, types of congestion, and measures to

alleviate congestion are presented in the following subsections.

2.1.1 Measures of Congestion

Congestion is defined as a “condition of overcrowding on a roadway caused by
demand exceeding capacity which is manifested by high densities, low volumes, low
speeds, stop-and go driving, increased delay and a high rate of rear-end collisions
occurring upstream of the bottleneck” (7" Congestion is described as a phenomenon that
is measured by using different variables in equation formulas to describe the extent,
severity, and duration of congestion. One type of measuring congestion uses indicators
that are related to the level of congestion as well as the probable cause of congestion.
Examples of possible indicators include vehicle miles of travel per lane mile of roadway,
and population density. Another type of measure uses variables that are descriptors of
the effects of congestion. Vehicle delay, congestion duration, and average travel speed
are all examples of variables that describe the effects of congestion (5). The following
paragraphs present the two types of measuring congestion.

The Highway Capacity Manual (HCM) (8) adopted the Level-Of-Service (LOS)
concept to represent a range of operating conditions. The LOS of a facility is determined
by traffic variables such as vehicle density and volume to capacity ratio (v/c), depending
on the facility type. The HCM describes traffic behavior at Level Of Service D as
unstable with limited space to allow for minor disruptions. The average travel speeds at
this level are greater than (46, 42, 40) mph for the design speeds of (70, 60, 50) mph

respectively. Capacity is reached at the border of LOS D and E with an average travel

speed of 30 mph. At LOS E, traffic operates under an extremely unstable condition
where all space is utilized and a slight disturbance is sufficient to cause a breakdown in
flow with deterioration to LOS F. Congested traffic conditions fall into the LOS F range
that is used to define forced or breakdown flow. In this range, traffic demand exceeds the
capacity of the roadway.

The General Accounting Office (GAO) (9) study found that on Urban Interstate
Systems, the total travel delay is relatively insensitive to changes in VIC between 0.77 and
0.99. This range covers LOS C, D, and E for the design speeds of (70, 60, 50) mph. The
study also included the definition of congestion used by several metropolitan planning
organizations, local governments, and the California Department of Transportation. The
study considered the road as congested once v/c and the average travel speed were near
1.00 and 35-mph respectively.

CALTRANS attempted to better describe the duration of congestion by using the
number of hours of LOS F “0). For example, LOS F2 represents 2 hours of LOS F. They
believed that the combination of the v/c ratio and the duration of congested operation
enhance the LOS concept alone and account for the peak spreading that is common in
many urban areas. They found this improved measure relatively easy to calculate,
interpret, and communicate.

An analysis technique developed by Lindely (3’4) to measure congestion used
Highway Performance Monitoring System (HPMS) data, traffic distribution patterns, and
the Highway Capacity Manual to calculate fieeway travel delay. He defined the
congestion severity index as the total freeway delay in vehicle hours per million vehicle

miles of travel. Then he ranked urban freeway systems according to the congestion

10

severity index. He also developed a methodology to calculate delay caused by incidents
using an accident database of breakdown types and rates. In Lindley’s calculations, the
congestion threshold was defined at a We ratio of 0.77 or higher (LOS D or worse) during
1 or more hours per day.

Lomax and Christiansen (1”

investigated the use of several variables as indicators
of areawide congestion levels. Among those presented as possible indicators were traffic
volume per lane, percentage of freeway lane miles of ADT greater than 15,000, K-factor,
and peak hour travel distance per lane mile. They calculated these possible indicators for
1975 to 1980 for five urban areas in Texas. Their study concluded that VMT per lane
was perhaps the most reliable indicator and developed a congestion standard that
combined weighted values for freeway and principal arterial street VMT per lane mile.
Subsequent research by Lomax et al. resulted in the development of a roadway
congestion index (RC1) (5’ ’2). The indicator of daily VMT (DVMT) per lane mile for
both freeways and principal arterial streets was weighted and normalized in the index’s
equation. They ranked major US. urban areas according to the RC1 value. Their chosen
threshold value for congestion was at a v/c ratio of 0.77 or higher and correlated to ADT
per lane mile values for freeways and principal arterial streets through basic assumptions
about traffic characteristics.

Thurgood ('3) defined congestion based on the LOS dropping from E to F as
determined by speed measurements. He used speed as a traffic parameter for the reason

that it is easy and relatively low in cost to measure, and because most agencies prefer to

use speed to measure congestion.

11

Gall and Hall “4) proposed logic to distinguish between incident and recurrent
congestion. The logic consisted of two major steps. The first step classified traffic
operations on a freeway facility into four states on the bases of volume and occupancy
graphs. States 1 and 2 were separated from states 3 and 4 by an occupancy of 25 percent.
A minimum discharged volume of 1,920 vehicles per hour per lane was used to separate
states 3 and 4. States 1 and 2 were separated by a non-linear line constructed by using
detector data that was considered to represent uncongested operations based on speed
above 40 miles per hour and occupancy less than 25 percent. The second step of the
logic used the information obtained from step 1 to determine the cause of congestion.

Cottrell “5) developed a lane-mile duration index (LMDIF) as a measure of
recurring freeway congestion in urbanized areas. The LMDIp represents a summation for
an urban area of the congested freeway lane miles multiplied by the respective duration
of LOS F. The use of the duration of LOS F service in intermediate calculations of
LMDIF is similar to Caltrans reporting of LOS F2. Cottrell chose LOS F as the

congestion threshold, which he correlated to an AADT/C value of 9, where C is the

capacity.

2.1.2 Types of Congestion

Congestion can be divided into recurring and non-recurring congestion, and
congestion due to special events. Recurring congestion is repetitive in nature and occurs
regularly during peak periods when traffic demand exceeds capacity. Peak period
congestion occurs daily and is expected to occur with some predictability and regularity

in both effect and duration “6). Distinguishing between recurring and non-recurring

12

congestion might be achieved by examining the type of bottleneck or by calculating

travel time “7).

Recurring congestion is most often associated with a permanent
geometric bottleneck, such as a heavily used entrance ramp, a horizontal curve in the
roadway, a steep grade or the termination of a lane (7). In addition, travel time builds up
slowly in this type of congestion. Recurrent congestion has received considerable
attention during the past 30 years, leading to the development of freeway ramp control
strategies that have shown their effectiveness in reducing recurrent peak hour congestion
([8).

Non-recurring congestion is unpredictable and is usually associated with
unscheduled and essentially random events such as incidents that may cause a reduction
in the freeway capacity below the level of demand. Environmental changes such as
heavy rain, ice, snow, fog, etc., might also fall into this category. Not all incidents result
in significant delay, however lane blocking incidents during the high traffic demand
create queuing on the freeway, which can be a serious traffic hazard to uninformed
motorists and may lead to secondary accidents. Adverse weather conditions may reduce
capacity in the case of major storms including times when partial or total freeway
closures might be necessary. Travel time in this type of congestion changes suddenly “6).

Special events such as ballgarnes and parades often generate large volumes of
traffic that is relatively predictable in magnitude and space. Traffic planners can predict
the effects of many special events from their historical data and manage congestion by
diverting traffic to alternate parallel routes. Operational studies have shown that
managing traffic during special events could result in a substantial reduction in

congestion and delay ('9’ 2°).

13

2.1.3 Measures to Alleviate Traffic Congestion

The historical solution to satisfy the growing traffic demand was through new
freeway construction and/or freeway-widening programs. However, as these new
freeways were built or lanes were added, the added capacity was filled almost as soon as
the freeways were opened to traffic. Therefore, less expensive alternatives were sought,
among which the Intelligent Transportation Systems (ITS) was chosen as a promising
technique to reduce congestion.

When a capacity-reducing incident occurs on the freeway, the optimal control
strategy is to immediately intercept freeway traffic before it reaches the reduced capacity,
and to prevent additional vehicles from entering the freeway at upstream entrance ramps.
Vehicles involved in the incident must be removed as quickly as possible and the demand
must be redirected to areas of available capacity in the freeway corridor. Additionally, to
reduce secondary accidents, drivers approaching the queue area should be warned of the
slow traffic “6).

(”3123"”) showed that a significant number of motorists would take an

Surveys
alternate route around congestion if they had prior knowledge of the location and
magnitude of the congestion. They often cite the need for more accurate information as
being their primary factor on deciding whether to divert or not. For motorists to react to
the information, they must believe that the information accurately reflects the current

(24)

travel conditions . Therefore, operators of freeway management systems must have

information that is both accurate and current.

14

2.2 Characteristics of Traffic Incidents

To effectively manage traffic incidents and reduce their impact on freeway
operations, many researchers felt the need to quantify the characteristics of the traffic
incidents and determine the factors influencing their occurrence. Incident characteristics
can be described in terms of incident rates, incident duration, and incident severity.

These characteristics are presented in the following subsections.

2.2.1 Incident Definition

Many authors have defined incidents. Judycki (25) defined an incident as “a
random event such as an accident, stalled vehicle, or a spilled load that causes a
reduction in the capacity of the roadway “. Guiliano (26) described an incident as “any
occurrence that aflects roadway capacity, either by obstructing travel lanes or by
causing gawkers block”. Another definition was stated by Urbanek and Rogers (27’ as “a
spill, breakdown, accident, or any other extraordinary event that causes congestion and
delay by restricting normal trafiic flow”. These definitions suggest that random events

that may disturb the normal flow of freeway traffic may be considered as incidents

causing a temporary bottleneck where one would not normally occur.

2.2.2 Incident Types

Incidents have been categorized by type as accidents and other incidents (26). The
reported relative frequencies of incident types are quite diverse. Accidents were cited as
high as 49% (28) and as low as 6% (29) of the total observed incidents. A higher

percentage of accidents result in lane blocking incidents due to the fact that vehicle

15

disablements often allow the driver to leave the traveled way immediately, or to reach the

shoulder before the vehicle is completely disabled (28).

2.2.3 Cost of Incidents
In 1993, traffic accidents claimed 40,115 lives and injured an additional three

million people in the United States (I)

. In a metropolitan freeway network, at least one
lane is expected to be blocked per mile of freeway 7.2 percent of the time because of
accidents, 6.7 percent of the time as a result of stopped vehicles, and 27.8 percent of the
time due to maintenance activities 00). Arceneaux and Mikhalkin have determined that
almost 60% of the vehicle-hours lost on an urban roadway system is a result of non-
recurring congestion (3 ”. Freeway incidents in urban areas are recognized as a disruption
problem that accounts for more than 50 percent of all freeway congestion (32).

On urban freeways, non-recurrent congestion caused by incidents was noted to be
responsible for as much motorist delay as the recurring congestion caused by a geometric
bottleneck (33). A study conducted by the California Department of Transportation
(CALTRANS) has estimated that more than 50 percent of all delay to motorists on the
freeway systems are the direct result of an incident (3").

In Texas, freeway incidents in 1990 were the cause of over 440,000 vehicle hours
of delay, costing the motorists approximately $2.2 billion. In Houston, over $1.5 billion
are lost annually due to congestion and over 59% of this cost is due to delays and excess

fuel consumption resulting fiom incidents. On a per capita basis, the cost of congestion

in terms of delay and fuel consumption is estimated to be approximately $440 per person

16

annually, and total congestion cost per registered vehicle is $695 annually when the cost
of insurance is considered (34).

Another study (Robinson, 1., 1990) estimated that the impacts of incidents in
terms of hours of delay, wasted fuel consumption, and excess road user costs have

(35)

increased 5 fold over levels experienced 10 years prior to the study Goolsby found

that a stalled vehicle caused a delay of 1,610 and 2,940 vehicle-hours for a one lane and

two lane closure, respectively (36).

2.2.4 Incident Rates
Incident rates are location dependent. Differences are due to factors such as road
geometry, level of traffic demand, weather, grade, and shoulder availability (37). Incident

rates are generally expressed as incidents per hour per lane-mile (285638)

, incidents per
million vehicle-kilometers (MV K) (38‘ 39‘ 40), or daily rate per freeway segment (26). Table
2.1 shows incident rates from five different locations in the United States. The first three
rows shown in the table were lane-blocking incidents of all types observed via closed
circuit television. The substantially higher rate of the third row is most likely due to the
absence of shoulders on the Bay Bridge.

Guiliano ‘29

conducted a study to describe incident patterns and analyzed incident
duration as a function of incident characteristics on a 12-mile section of the I-10 freeway
located in Los Angeles, California. The MO freeway has 5 lanes in each direction over

most of its length. Accidents were found to be most frequent on weekdays. The daily

average weekday rate was 3.1 accidents per million vehicle miles and the weekend daily

17

rate was 2.2. Friday had the highest daily rate of 3.8, while the lowest daily rate of 2.0

occurred on Sunday.

An estimate of incident rates in Houston was 0.68 incidents per 100 million

vehicle-kilometers (MVK). This is equivalent to a major incident every 147 MV K

Table 2.1 Incident rates.

(42)

 

 

 

 

 

 

 

Unit of Measure Location Rate
Detroit,
Lane-blocking incidents per hour per lane-mile Michigan 0.014
Gulf Freeway,
Lane-blocking incidents per hour per lane-mile Houston, Texas 0.010
Oakland Bay Bridge,
Lane-blocking incidents per hour per lane-mile California 0.036
Toronto,
Incidents per million-vehicle-miles (40) Canada 19
San Francisco,
Incidents per million-vehicle-kilometers (4') California 65
Houston Freeways,
Fatal accident rate per 100 MV Km (42) Texas 0.708

 

 

 

 

 

Frank DeRose Jr. (28) performed a study to determine the frequency, duration and
character of the freeway traffic incidents in addition to the proportion of accidents to
incidents in Detroit. He indicated that accidents were about twenty five percent of the
total number of incidents. The number of incidents increased during the periods of high
traffic demand when the capacity was exceeded. He calculated the rate of incidents per
mile as well as total incidents per million vehicle-miles for the total 3.2 mile study
section. He found that the rates per million vehicle-miles were 11.9 and 8.7 incidents for
the 4 and 3 lane sections respectively, and the average daily incidents per mile rate was
0.69 and 0.51 incidents for the 4 and 3 lane sections respectively. Climatic conditions

showed higher frequencies at low temperatures, rain, snow, and wet pavement. The

18

average duration of 927 incidents was 6.14 minutes for accidents, 4.94 minutes for
vehicle disablement and an over all average of 5.24 minutes. The average duration of the
incident was defined as the time the incident occurred to when the incident was removed

to the shoulder or when the freeway resumed movement.

2.2.5 Behavior of Traffic in the Presence of an Incident

When an accident occurs on a high-volume freeway, most likely a queue will
form at the location of the incident. The queue and its resulting congestion then begin
backing upstream from the site of the bottleneck. Whitson (43) has presented volume-
density plots of freeway operations in Houston during an incident, which clearly illustrate
this upstream progression of the queuing area. The frontal boundary of this queue as it
moves upstream is called a shock wave and commonly travels at speeds of 10-20 mph
during moderate to heavy traffic conditions. Whitson also noticed that there was a
second wave moving downstream from the incident location. This wave explains the

change that occurs downstream of the incident, from normal flow to a much lighter flow.

Figure 2.1 is a graphical illustration of freeway traffic behavior upstream and
downstream of an incident location when the incident partially or totally blocks the
freeway. Upstream from the incident, the characteristics represent traffic conditions
moving at normal speeds and normal density k... The area located immediately upstream
of the incident represents the congested area where vehicles are queuing and traveling at
low speeds and experiencing high density kq. The region immediately downstream of the

incident reflects traffic flowing at a metered rate with low density and slightly higher

19

speeds than the normal flow. The far downstream area from the incident represents
traffic flow similar to the area far upstream of the incident with normal traffic flow and

kn (44). This behavior continues until the incident is completely removed

normal density
and the queued vehicles upstream of the incident are completely discharged. Wu] and W,“

represent the speed of the upstream and downstream shock waves respectively.

Incident location

 

 

Congested Queue
Density (kq) War

 

 

 

*
Normal
Density (kn) Normal Density (kn)
Metered Density (km)

 

 

 

 

 

Direction of Traffic Flow

Figure 2.1. Behavior of traffic under an incident.

2.2.6 Impact of Incidents on Capacity

Merrel E. Goolsby (36) found that incidents created a reduction in flow
disproportionate to the physical reduction in roadway width. He noticed that an incident
removed to the shoulder on a three-lane freeway reduced capacity by one third due to the
gawkers block phenomenon, even though no physical obstruction exists; a single lane

blockage reduced capacity by 50 percent, even though the physical reduction is only 33

20

percent; and a two-lane blockage reduced capacity by 79 percent. Lindley (3) determined
the reduction in capacity due to an incident as a function of total lanes and the number of
blocked lanes. He found that the amount of capacity reduction is greater than the
percentage of blocked lanes to the total number of lanes. He also found that shoulder
disablement seems to have little or no effect when the total number of lanes is more than
two.

McShane et a1. (45) presented an example to illustrate the effect of capacity
reduction on the volume to capacity ratio. They considered three different values for We
ratios and then they simulated the losses in capacity due to an incident by changing these
three values by different percentages. They illustrated that decreasing capacity by 10%
can change freeway operation from a functional system to an oversaturated system

depending on the demand level at the capacity reduction time.

2.2.7 Incident Duration and Delay

Traffic demand has a significant impact on incident duration and its associated
delay. As long as demand is below the reduced capacity, all vehicles passing the incident
will experience no delay. On the other hand, once demand exceeds capacity a queue will
build upstream of the incident causing excessive delay to motorists. The rate at which the
queue builds depends on the difference between demand and the reduced capacity. The
research dealing with incident duration is presented first followed by a discussion of
incident delay.

The highway capacity manual (46) divides incident duration into four intervals.

The first interval represents the time between the occurrence and verification of an

21

incident. The second interval represents the time between the incident verification and
the arrival of the police or towing vehicles to clear the incident. The third phase
represents the time between the arrival of these vehicles and the clearance of the incident.
The last phase represents the recovery time or the time required for traffic conditions to
resume normal operation. Effective incident management can reduce incident duration by
minimizing the detection, response, and clearing time of an incident and reduce the
recovery time. The primary factor influencing incident duration is the traffic demand that
can be controlled by diverting and restricting demand upstream of the incident site by
using ramp metering, changeable message signs or any other diverting technique.

The impact of incidents on freeway operations was investigated by Merrel E.
Goolsby (36) by relating frequency, duration, and flow passing freeway incidents in order
to estimate the magnitude of motorists delay resulting from incidents on the Gulf
Freeway in Houston. In two years, 1,154 accidents were studied. They found that the
average accident required 19 minutes from the time of reporting to removal fiom traffic
lanes and an additional 26 minutes for the police investigation.

Guiliano (26) used three different data sets to develop a statistical model to
estimate incident duration based on incident characteristics. She developed two separate
models, one for all incidents and the other for accidents. In these two models, she used
qualitative categories representing incident type, lanes closed, time of day, accident type,
and truck involvement as independent variables. One of her most important findings was

the highly significant effect of truck involvement in accident duration, where the

presence of a truck in an accident causes a substantial increase in total duration.

22

Several methods have been developed to estimate incident delays. These methods
can be classified into three types. The first method is based on queuing analysis (46’ 47).
The second method is based on shock wave analysis (48’ 49’ 5°). The third method is based

on freeway traffic simulation (5‘52).

The research cited regarding these methods is
presented in the following paragraphs.

The Federal Highway Administration (FHWA) funded research in the late 1970’s
to develop guidelines and recommendations to help highway departments, police
agencies, and other organizations select, plan, design, and implement low-cost-measures
to deal with incidents that cause freeway congestion. The research was published in
reports presenting an overview of the nature and magnitude of the freeway incident

management problem and summarized possible solutions (53)

. The reports also included
analytical procedures to estimate traffic delay and congestion and assess the tradeoffs in
cost-effectiveness among many alternative measures.

Morales (47) summarized the basic analytical procedures presented in these reports
and developed a model for computing delay, time to normal flow (TNF), and maximum
queue (me) caused by freeway incidents. He quantified delay based on a graphical
representation of cumulative arrival and departure curves and calculated the crunulative
vehicle hours of incident delay. He divided the congested time period into smaller time
intervals during which demand and/or capacity is assumed to be constant, resulting in
linear arrival and departure curves at the incident bottleneck. The method assumed that
the initial demand was less than the capacity of the roadway section. The queue length

was calculated as the algebraic difference between the cumulative demand and the

cumulative departures at any time (t).

23

The highway capacity manual method uses the same approach as Morales (47’

method with an important modification that considers cases of incidents occurring during
the peak period of congestion with initial demand exceeding capacity. The two methods
are widely used by practitioners and researchers to estimate incident delay.

Garib et el. (54) presented two regression models for estimating freeway incident
congestion and a third model for predicting incident duration. They used two sets of data
from Interstate 880 in Alameda County, California. The first set of data was collected
using a fleet of moving observers during morning and evening rush hours reporting on
observed incidents and their characteristics. The other set of data was obtained from
freeway loop detectors. They used multiple regression analysis to develop incident delay
models using these two sets of data. The first delay model calculates delay as a function
of incident duration, traffic demand, and capacity reduction. These are the exact variable
used by Morales (47). Their second model predicts the cumulative incident delay as a
function of incident duration, number of lanes affected, and number of vehicles involved.
The last model was a statistical model to predict incident duration as a function of
number of lanes affected, number of vehicles involved, truck involvement, time of day, ,
police response time, and weather condition.

Messer et al. (48) used the kinematics wave theory of Lighthill and Whitharn to
develop a method for predicting individual travel times on the freeway during incident
conditions. They divided the time space plane during incidents into areas representing
four different traffic flow conditions: normal flow, queue flow, metered flow, and
capacity flow. The boundaries of these areas were defined by linear shock waves, and

the speed of each shock wave was derived assuming a linear speed-density relationship

24

developed by Greenshields (55). They applied the method to four incidents that occurred
on the Gulf Freeway in Houston. They found that two-thirds of the observed travel times
were within 10 percent of the predicted travel times. The linear Greenshield’s speed-
density model results in parabolic curves for volume-speed and volume-density plots.
The major problem with using the parabolic curves is that if they do not match the actual
conditions in regions upstream of the incident then significant errors result in calculating
the wave speeds.

Wirashighe (49) applied the shock wave theory of Lighthill and Whitham to
develop formulas independent of any particular macroscopic theory of traffic flow for
calculating individual and total delay upstream of incidents. The formulas are based on
areas and densities of regions representing different traffic conditions that are mainly
congested and capacity regions that are formed by shock waves in the time-space plot.

We-Min Chow ‘50) performed a study of traffic performance in the presence of an
incident using shock wave and queuing analysis under the assumption that traffic density
is not a firnction of time and a unique flow-density relationship exists. He derived
formulas to calculate total delay, which he found yielded identical results. His study
proposed two methods; the first was to compute the duration of time to discharge stored
vehicles after the incident is removed. The second was to calculate delay in terms of the
difference between total travel time under the incident and non-incident cases. He
concluded that if he had used a time dependent flow-density relationship which is more
realistic, then the two methods would have given different results.

Wicks and Lieberman (5') developed INTRAS, a microscopic freeway traffic

simulation model designed for freeway corridor simulation. INTRAS was enhanced and

25

the new version called FRESIM has been used in the past few years. Both versions share
the same fundamental structure based on car-following theory. Cheu et el. (56) used
INTRAS and data from a Los Angeles Freeway and concluded that INTRAS may
underestimate the occupancy during free flow conditions in the recovery periods after
incidents. In addition, they indicated that the car-following theory equation used in
INTRAS gave satisfactory results in general, but failed to produce high volume and
occupancy values that match collected field values in their study site. INTRAS and
FRESIM can be used to estimate incident congestion by simulating the freeway with and

without the incident and calculating the difference in vehicle hours of travel.

2.3 Prior ATMS/ATIS Evaluation Studies

Three FTMS evaluation studies were cited in the literature and are presented
briefly to illustrate the significance of F TMS implementation in improving the quality of
freeway operations. The first study was conducted in Toronto, Canada to evaluate
COMPASS and the second study was performed in New York to evaluate INFORM.

The last case was performed in Korea.

2.3.1 COMPASS (Toronto FTMS evaluation study)

The Ontario Ministry of Transportation implemented a Freeway traffic
management system (F TMS) called COMPASS in 1980 to improve the quality of traffic
on Freeway 401 in the Toronto area. The system included CMS, CCTV, and loop
detectors. The primary goal of COMPASS was to reduce congestion by balancing traffic
between the expressway and collector lanes. This balancing was intended to manage the

demand in an efficient way and ensure that any spare capacity on either roadway

26

becomes available to all traffic as needed. The parallel roadway system allowed vehicles
to bypass congestion and incidents by moving to the adjacent roadway. The COMPASS
system calculates travel time between two transfer points, and a balancing message is
displayed when a significant change is observed. Speed was used as a single parameter
to calculate link travel times.

Masters et el. (57) performed a before and after study by using volume as the
control parameter to ensure that the before and after data sets were suitable for
comparisons. Only a 5% variation was allowed in the study. Incidents with duration less
than 5 minutes were ignored. The findings included a reduction in secondary accidents
by 70% on the Burlington Skyway as a result of quick detection, response and removal of
incidents and a lowered congestion level due to reduced time of lane blockage and
diversion resulting from the pre-warning to drivers. Moreover, average speeds on the
Skyway were increased between 7% and 18.6%, and the travel time was reduced by the

same amount.

2.3.2 INFORM (Information for Motorists, New York FTMS evaluation study)
INFORM is a corridor traffic management system designed to improve the
utilization of the existing freeway facilities. It contains two major freeway facilities, the
Long Island Expressway (LIE) and the Northern State Parkway/Grand Central Parkway
(N SP/GCP) in Long Island New York. The system also includes a number of crossing
and parallel arterials and freeways forming a total of 128 miles of controlled freeway.
Parallel freeways such as the LIE and the NSP/GCP offered an opportunity to divert. The

system consists of electronic monitoring, variable message signing, ramp metering, a

27

limited number of CCTV, road side citizens band radios, and arterial streets intersections
under INFORM control.

Steven and Perez “8)

evaluated INFORM using field data and perception surveys.
The measures of effectiveness used to assess INFORM were; vehicle miles of travel;
average speeds; ramp delays; motorists perceptions; and effectiveness of changeable
message signings.

The system used passive message signs, and their effect was observed during
incident conditions by monitoring changes in the volume of upstream off-ramps. The
operators used the changes in ramp volumes as an indication of how effective the
messages were. For a typical incident and using passive messages 5 to 10 percent of
mainline traffic in the INFORM corridor was diverted over several upstream off-ramps.
Volume on these ramps was increased by 40 to 70 percent for up to three upstream off-
ramps.

Speeds were compared for metered and non-metered periods in both the AM and
RM. Morning peak period freeway speeds for the March 1990 metering increased 3 to 8
percent over speeds for March 1990 non-metering and 13 percent over speeds for spring
1987. For the RM. traffic speed, no change in speed was observed for the metered and
no-metered conditions. Delay savings for the peak period incidents were analyzed and
the annual delay savings for the incident related effects of CMS were estimated.
Estimated delay savings for the peak period incidents analyzed ranged up to 1,900

vehicle-hours. The estimated annual delay saving for the incident-based effects of the

variable message signs was 300,000 vehicle-hours.

28

Residents were surveyed in the INFORM area to obtain their opinion of the
usefulness of the information displayed, and whether they changed their routes in
response to the messages. Twenty five percent of the motorists viewed INFORM as a
helpful system and 45 percent thought that INFORM helped once in a while. 45 percent
of the drivers stated that they sometimes change their route in response to the sign
messages and 25 percent never changed their route in response to these messages. They
were also asked about their choices once they encountered an on-ramp red signal and the
time they spent waiting at red signals. Fifteen percent of those who encountered an on-
ramp red signal indicated that they used the service road or other roadway to avoid
waiting. Moreover, the survey included the resident’s opinions on CMS and ramp
metering.

Their over all findings were that the system produced delay savings. There were
however, cases of arterial breakdown during heavy diversion. The residents expressed a

positive attitude on CMS and mixed feelings over ramp metering.

2.3.3 The Korean FTMS Evaluation Case

Korea initiated a Freeway Traffic Management System (FTMS) in 1992 and it
went through an operational test in February 1995. The system covered 318 kilometers
of freeway segments of Kyoung-Bu and Chung-Bu express highways that connect Seoul
and Daejon.

Kyungsoo et el. (58)

presented a paper describing the Korean FTMS evaluation
study. They performed quantitative and qualitative analyses to assess the benefits of the

system. Their quantitative analysis included the calculation of the reduction in recurring

29

and non-recurring delay and estimated the monetary benefit from delay reduction
achieved by the installation of the FTMS. Reduction in recurring and non-recurring
delay due to the FTMS installation was estimated to save 65 and 9,320 million US dollars
respectively from the time the system became operational through the year 2004. For the
qualitative analysis, they conducted a survey to obtain the motorist’s opinion and level of
satisfaction over the system’s performance. Their study showed that 65 percent of the
freeway users acknowledged the effectiveness of the system, 95 percent of the motorists
were satisfied with the accuracy of the information provided, and 68 percent requested
information on alternative routes. The overall outcome of the study showed a reduction

in delay as well as an encouraging acceptance from the motorist.

2.4 The Role of ITS in Managing Incident Based Congestion

In recent years, ITS technologies have been incorporated with incident
management systems to assist in alleviating non-recurrent congestion in most
metropolitan areas in the United States. Incident management is defined as a planned and
coordinated procedure developed to eliminate the impact of incidents on traffic
operations as quickly as possible. It involves collective efforts of human, mechanical,
and electronic resources to serve different purposes. These purposes include detecting
and verifying the incident and assessing its magnitude as well as identifying what is
needed to restore the facility to normal operation. It also helps to provide the appropriate
response in the form of control strategies, and information dissemination (40). The
following subsection reviews incident detection methods, incident response and

clearance, and finally, electronic incident management and control strategies.

30

2.4.1 Incident Detection and Verification

Rapid detection is one of the most important elements in incident management.
The sooner an incident is detected, the faster the response to initiate its removal. It has
been noted that emerging Intelligent Transportation Systems technologies offer a promise
to substantially improve detection capabilities and reliability. Technologies available for
detecting incidents range from low-cost non-automated methods to sophisticated
automated surveillance techniques.

The effectiveness of automatic detection technologies depends in part on the type
of algorithm used to analyze the detector data. Three parameters generally used to
monitor the performance of an incident detection algorithm are detection rate, detection

time, and false alarm rate (62).

Detection rate is defined as the percentage of the total
number of incidents that are detected by the computer algorithm. Detection time is
defined as the time between when an incident occurred and the time it is detected by the
algorithm. The false alarm rate is generally used to provide an indication of how many
times the algorithm incorrectly flags an incident.

There is a general relationship that exists between detection rate, false alarm rate,
and detection time. With most incident detection algorithms, the false alarm rate
increases as the detection rate increases. Also, the false alarm rate increases as the
detection time decreases. This is because as the sensitivity of the algorithm is adjusted to
detect less severe incidents more quickly, minor fluctuations in traffic can trigger the
algorithm to signal that an incident is present, when in fact no incident exists on the

freeway. False alarms can be tolerated in order to achieve higher detection sensitivity, as

long as they are not too frequent and can be ignored by the system operator. Evaluation

31

of false alarm rates is based on their frequency over a given time period. An algorithm
may be reported as having a very small false alarm rate percentage but yield a fairly high
number of false alarms because the total number of such checks made during a given
time period is so high.

The five general categories of incident detection algorithms that rely on volume,
speed, and or occupancy data include the following: (63)

0 Comparative.

0 Statistical.

- Time-series/smoothing.

- Traffic and theoretical models.

0 Advanced incident detection techniques

Comparative or pattern recognition algorithms compare traffic parameters at a
single detector station or between two detector stations against predetermined threshold
values that define when incident conditions are possible. Statistical algorithms use
statistical techniques to determine whether the observed detector data differ statistically
fi‘om historical or defined incident conditions. Time series and smoothing algorithms
compare short-term predictions of traffic conditions to measured traffic conditions.
Modeling algorithms use standard traffic flow theory to model expected traffic conditions
on the basis of current traffic measurements. The advanced incident detection algorithms
include Fuzzy Sets, Neural Nets, Machine Vision, Automatic Vehicle Identification, and
others.

A recent review of the algorithms indicated that two of the Modified California

Algorithms (#7 and #8) and the McMaster Algorithms rated the highest on the basis of

32

reported performance, operational experience, and model complexity. When calibrated
properly, these algorithms can be expected to detect 70 to 85 percent of all incidents,
while incorrectly triggering a false alarm about 1 percent of the time or less (63). These

algorithms were also judged to be easy to understand and implement by the operator.

2.4.2 Incident Response and Clearance

Response involves the activation, coordination and management of appropriate
personnel, equipment communication links, and information media as soon as there is
reasonable certainty that an incident is present. Steps in the response process include:

o Verifying the existence and location of the incident.

0 Assessing the incident to determine the type of response needed to clear it.

o Initiating the appropriate response.

0 Removing the incident.

A quick and timely response by the necessary resources to clear the incident can
significantly reduce the incident duration and its impact on the traffic operations in the

region (64).

2.4.3 Electronic Incident Management and Control Strategies

The traffic management and control components of a freeway management
system are intended to help in incident management conditions. These components help
to warn motorists approaching the incident location about downstream traffic conditions,
advise about appropriate speeds, reduce traffic demand, and adjust control settings on

other roadways to accommodate increased traffic volumes due to diversion from the

33

freeway. The management and control components that can be used for this purpose
include ramp metering and information dissemination.

Ramp metering was implemented in 1960 and showed effectiveness in reducing
freeway congestion. They consist of a traffic signal to restrict motorists fiom entering the
freeway. The traffic signals are connected to a centralized computer that control their
timing based on metering control algorithms. When an incident occurs, ramp metering is
implemented in real-time upstream of the incidents to restrict motorists from entering the
freeway. Ramp metering typically operates during peak traffic periods, and could be
adjusted if an incident occurs during these periods or during off-peak periods in case of
high demand “‘5’.

Information dissemination components are usually activated to warn motorists
upstream of an incident about the traffic conditions ahead and inform them about which
travel lanes are closed and to recommend possible diversion routes around the incident.
Warning can occur through any of the different information dissemination technologies
available in the region such as CMS, HAR, commercial radio, traffic reports, and others.
Such information is coordinated and managed so that the information remains current
especially with respect to the affected location, expected duration, and its impact on
traffic conditions.

Diversion involves examining where and how much traffic should be diverted
when an incident or other blockage condition occurs on any section of the freeway at any
time of the day. Alternative route decisions involve determining not only where and how
much traffic should be diverted, but also when diverting traffic would produce positive

benefits. Since diverting traffic to alternate routes is often politically sensitive, how long

34

a freeway is to remain closed before an official detour route is established is often a
policy decision. For example, some areas divert traffic only when an incident is likely to
last more than one hour. Not all arterials near a freeway may be desirable alternative
routes. Examples that make an arterial not suitable as an alternative route include

schools, hospitals, and sensitive neighborhoods “’5’.

2.5 Summary

The literature review of the different topics related to the problem under
investigation showed the importance of the various ITS technologies in reducing traffic
congestion and softening the impact of traffic incidents on the quality of freeway
operations. Today, the various ITS components are implemented in almost all United
States urban areas that are suffering from congestion. As a result, ITS assessment studies
are extremely valuable to determine the magnitude of their benefits and cost and justify
the public firnds invested on such projects.

The MDOT ITS deployment was one of the first ITS plans of its kind. The
deployment offered a unique opportunity to perform a before and afier study to evaluate
the effectiveness of ITS. It is unique in the sense that the deployment is relatively
comprehensive, and the conditions are ideal for conducting a before and after evaluation
study.

The study author chose incident based congestion during off peak hours as one of
the topics to perform a before and after evaluation study. The study used average minute
speed as the only traffic variable to detect incident caused congestion. Chapter 3 covers

the development of the technique used to analyze the before and after data.

35

Chapter 3

Methodology

3.0 Introduction

The purpose of this chapter is to describe the development of the measures used to
conduct the comparative study on the effectiveness of the ITS deployment program in
Detroit. The measures are used to assess the incident-based congestion that occurs during
off-peak hours, and rely solely on the traffic loop detector data obtained from the MITSC.

In order to test the effectiveness of the ITS deployment, three measures were
established. The first measure quantifies individual station incident-based congestion.
The second is a measure of the aggregate impact of individual incidents across stations.
This procedure also measures the incidents’ queue length and categorizes them based on
this length. And finally, the third measure is an index to quantify the segment and
system-wide incident-based congestion.

The following sections of this chapter have been divided into the main subjects of
the methodology. The first topic is a description of the database variables and the
smoothing and reconstruction functions applied to these variables. The second topic
describes the establishment of the incident-based congestion speed threshold value. The
third topic covers the development of the algorithm to flag the start and end of the
incident-based congestion periods. The fourth part describes the algorithm established to
cluster station-based congestion measures into minute-station individual incidents. The

fifth and sixth topics describe the calculation of incident queue length and the procedure

36

for categorizing segment incidents. The calculation of the segment index is included in

section seven, and finally, a summary of the methodology is presented in section eight.

3.1 Description of the Database

The study utilized a traffic database obtained from the Michigan Intelligent
Transportation Systems Center (MITSC) in Detroit. The traffic database came from
inductive loop detectors imbedded at various locations in each freeway lane. The data
included minute by minute total lane volume, average speed and occupancy, an indication
of the number of lanes, and the number of lanes reporting data. The database was stored
in ASCII format files and recorded on 9-track 1600 CPI magnetic tapes. The files
consisted of 80 character records and the tape was written with 80 records per block.
Each file contained 74,902 records of 80 characters and 12 hours of average one-minute
data.

The first record on the tape was the title record, which showed the number of
stations and station minutes. Following the title record, the data was organized in sets of
720 station records. Each set represented a station and covered 720 minutes of traffic
data. Sets were arranged in an ascending order based on the station number. Table 3.1

shows the format of the records.

37

Table 3.1. Format of the traffic detector input file.

Column Contents
1 — 4 tatron num (range 1 to 400).
— 1 rn , year.
13 — 1 Four 't mrlrtary trme ).
type ( = = entrance

N 0 (one )-

o
T vo ume (sum 0 station
Average occupancy.
Average ).

 

The traffic variables such as volume, occupancy, and speed were measured for
each vehicle and computed for each lane. Volume was then summed for all lanes at a
station for each nrinute. The unit used for volume was vehicles per minute per station.
The station speed and occupancy were measured as the average one-minute speed and
occupancy for all lanes, and the units for speed and occupancy were miles per hour and
percent respectively. A smoothing function (equation 3.1) was then used to smooth the
station one-minute volume and the average one-minute speed and occupancy.
Speed=SPM+G * (STM—SPM) .................................................... 3.1
Where:
Speed = Smoothed speed for the current minute.
SPM = Smoothed speed from the previous minute.
G = Tunable variable currently set at 0.20.
STM = Raw speed measured for current minute.
To obtain the smoothed occupancy and volume, the speed variables replaced by the

corresponding volume and occupancy variables.

38

If the number of operational lanes was zero, then the value of volume,
occupancy, and speed was set at —1. If the number of lanes was greater than 0 and less
than the number of lanes, then the volume, occupancy, and speed were reconstructed

from the operational lanes. The formula used to reconstruct volume and occupancy was:

Number of lanes in ( Vol., or Occ.)

(VOL, 01' OCC.) RCCOI'ISU'UCICd = ................. 3.2
Number of operational loops

 

Table 3.2 shows a sample of the header, and functioning, reconstructed, and

malfunctioning records.

Table 3.2. Sample of a header record, and functioning, reconstructed, and
malfunctioning records.

 

 

 

Header VOS-MAIN-SMOOTH STA=285 =0720
record
Variable field number (1) (2) (3) (4) (5) (6) (7) (8) (9)
Functioning 57 090996 0002 1 3 3 10 0.9 66.0
record

 

Reconstructed record 59 011397 2323 1 3 2 22 2.2 64.3
from two stations.
Reconstructed record 78 011397 1611 l 3 1 80 12.9 41.5
from one station.
Malfunctioning 59 090996 1102 l 3 0 -l -1.0 -1.0
record

 

 

 

 

 

 

 

 

 

 

 

 

 

 

The variable field numbers represented the following variables:

(1) Station ID. (2) Date.

(3) Time. (4) Station type.

(5) Number of lanes. (6) Number of reporting lanes.

(7) Volume. (8) Occupancy. (9) Speed.

3.2 Establishment of the Incident-Based Congestion’s Speed Threshold Value
The primary traffic variable used in this study was the smoothed one-minute

average station speed; this variable is most frequently used in describing congestion and

39

in incident detection algorithms. To choose a speed threshold value (STHV) for
flagging incidents, the related literature was reviewed and is presented in the following
paragraphs.

The Highway Capacity Manual (8) Figures 3 and 4 of Chapter 4 show a gradual
decrease in speed as flow increases, with an increasing rate of change of speed as
capacity is approached. Speed at the border of Level of Service D and E that defines
capacity is around 35 mph. The recent Highway Capacity Manual (46) revised Chapter 3
by adding new speed versus flow curves. The new version indicates that the LOS E/F
breakpoint occurs at a significantly higher speed of 50 mph.

Thurgood ('3), in his study of developing a freeway congestion index, stated that
the onset of congestion on Utah freeways was based on traffic stream speeds falling
below a threshold speed of 40 mph. His reasoning was that 40 mph falls between the
new and old HCM values and the value of 40 mph is a strong indicator that flow is falling
into the LOS F, where forced operation is likely to occur. He also added that a speed
threshold value of 45 mph or even 50 mph would not likely have changed his results
significantly because, in most cases, once the speed fell below 50 mph, they also fell
below 40 mph. In contrast, Cottrell ('5), in his study of measuring the extent and duration
of freeway congestion in urbanized areas, chose LOS F and a speed threshold value of 35
mph as the congestion threshold value for two reasons. The first reason was based on the
General Accounting Office (GAO) study (9), which found that travel delay on the Urban
Interstate System was insensitive to changes in volume capacity ratio between 0.77 and
0.99. Secondly, the GAO also found that several metropolitan planning organizations and

local governments, as well as the California Department of Transportation, considered a

40

road to be congested when its volume capacity ratio was close to 1.0 with an average
travel speed near 35 mph.

This paragraph reviews the research regarding the choice of speed threshold value
for the McMaster Incident Detection algorithm. Forbes G. (67) used a conservative speed
of 70 km/hr (43.75 m/hr) to activate his congestion logic algorithm based on the three
previous studies. In the first study, Hurdle and Datta (68) identified an average speed of
80 krn/hr (50 m/hr) for capacity flows. In another study, Persuad and Hall (69) identified
the movement from uncongested to congested operations to be associated with a drop in
speed to a value of 70 km/hr. Third, Gall and Hall “4’ defined congestion data as having
occupancies that exceeded a maximum uncongested threshold value of 25% and a speed
threshold of 65 km/hr (40 mph).

Initially, three values of 40, 35, and 30 miles per hour were selected as the
threshold value to be used in this study, then the value of 35 mph was chosen as a

conservative value.

3.3 Incident-Based Congestion

The occurrence of traffic incidents on a freeway section may be reflected directly
in the traffic loop detector output at that section, depending on the incident occurrence
time and traffic demand. However, to quantify incident duration, it is necessary to know
the time all incidents occurred and were terminated. The following subsections describe
the criteria that were used to signal the start and end of incident-based congestion and the

algorithm developed to determine the start, end, and duration of an incident.

41

3.3.1 Criteria for the Start and End of Incident-Based Congestion

In order to decide when an incident occurred and was terminated, various criteria
were developed. One condition was established to define the start of incident-based
congestion, and a set of five conditions were used to determine the termination of an
incident. The start of incident-based congestion at any station and at any given time was
flagged if the speed of the present minute and the previous three nrinutes was below the
speed threshold value of 35 miles per hour (STHV) (provided that the station was
reporting valid data).

To flag the end of incident-based congestion at a given station, five conditions
were implemented, one of which was executed given that incident-based congestion had
started. The first two conditions were used for the case when the present time was not
noon or midnight. The first condition was for a station that was reporting valid data, and
the speed during the past five minutes excluding the present minute (J) was above STHV,
and the minute J-6 was below STHV. The second condition was for a station that had
been reporting valid data but was starting to report invalid data. This condition was met
when the station was reporting invalid data for the present minute and the past five
minutes and the speed at J-6 was valid and below STHV.

The last three conditions were developed for the case when the present time (J)
was noon or midnight. The third condition was met when the station was reporting valid
data with a speed below the STHV at the present minute and during the past 6 minutes.
The fourth condition was met when the station was reporting valid data and the speed
changed to above the STHV at any time during the past five-minute period. Finally, the

fifth condition was met when the speed was below the STHV and the station started to

42

report invalid data at any time during the previous five minutes. Figure 3.] illustrates
the condition for the onset of incident-based congestion and the five conditions to signal

the end of incident-based congestion.

3.3.2 Development of the Incident-Based Congestion Periods

This subsection describes the development of an algorithm to apply the criterion
established in section 3.3.1 to each of the station’s data. The algorithm started by
allocating memory for an array of 720 objects to store one set of station data, and each
object had member variables to store the values of the traffic detector variables. The 720
objects were used for all station sets. The algorithm also allocated memory for an array
of objects to store variables required for the functional and malfunctional detector
incident-based congestion periods. Other arrays of objects were also allocated to store
the periods of incident-based congestion that occurred during the pre-peak, post-peak,
and peak hours.

The algbrithm read the input file and once it reached a station that belonged to any
of the study segments, it recorded the station’s 720 minutes of information in the proper
objects allocated earlier in memory. Then it looped twice through the 720 objects. The
first loop involved the process of searching for periods of incident-based congestion, and
the second involved the search for periods of malfunctioning data. During the first loop,
the objects were checked if the condition for the start of congestion was satisfied; in
this case, it opened a functional object and stored the starting time of the incident. The
algorithm then continued the loop until one of the five conditions for the end of

congestion was satisfied, in which case it recorded the incident ending time and

43

Criterion used to define the onset of incident congestion:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

‘/Startmg time
E \\ STHV
ca
J—3 J-2 J-l J
(time now)
Time
Criteria used to define the end of congestion:
Condition 1:
Ending time
% 4_,_,.__._.
]/ STHV
/
i.
m
J-6 J-5 J-4 J-3 J-2 J-l J
(time now)
Time
Condition 2:
/ Ending time
'5 STHV
i “n a
m o
C.‘
0
.§
Detector
J-6 J-S J-4 J-3 J-2 J-l J Output = -1
Time

Figure 3.1. Criteria for the start and end of station incident-based congestion.

44

Condition 3: Ending time

 

'g STHV
r2' 5

 

 

 

 

 

 

 

 

 

11:54 11:55 11:56 11:57 11:58 1 1 :59 12:00 Time
or or or or or or or
23:54 23:55 23:56 23:57 23:58 23:59 24:00

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Condition 4: Ending time
w/
E // /' / //[/ f STHV
"i / / Q / a
O
/ / / G
\4 -4 g
11:54 11:55 11:56 11:57 11:58 11:59 12:00 .
or or or or or or or Trme
23:54 23:55 23:56 23:57 23:58 23:59 24:00
Condition 5: Ending time
'5
3 STHV
5' a
to
.§
[1254 11:55 11:56 11:57 11:58 11:59 [2:00
or or or or or or or De
23:54 23:55 23:56 23:57 23:58 23:59 24:00 tector
Output=-1
Time

Figure 3.1 (cont). Criteria for the start and end of station incident-based congestion.

45

calculated the incident duration. It continued looping through the remaining 720 objects
repeating the same checks looking for other incidents that might have occurred during the
remainder of the 12-hour period.

During the second loop, the algorithm looked for periods of invalid data. There
were no conditions established to start or end these periods. The algorithm went through
the station set of data looking for such periods. Once a station was detected reporting
invalid data, a malfunctioning object was opened and the starting time, ending time, and
duration of the failure period were recorded. Then it continued 100ping to the next period
following the same procedure until it finished looping through the 720 objects. The
algorithm then proceeded to the next relevant station and performed the same steps.

Afier the end of the input file was reached, the physical location for each station in
the firnctioning and malfunctioning objects was added to their objects. The periods of the
functioning and malflmctioning objects were then divided into pre-peak, post-peak, and
peak periods. The functional and malfunctional objects for each of the three periods were
combined and their values were stored in the objects previously allocated in memory.
Figures 3.2, and 3.3 depict a flow chart illustrating the algorithm’s steps and periods of

congestion for various stations versus time.

3.4 Clustering Station Incident-Based Congestion Periods
The impact of an incident can be detected at single or multiple stations, depending
on the incident severity and the level of traffic demand. The previous section described a

procedure to determine estimated individual station incident starting and ending times

46

Allocate memory for arrays of objects for:
0 Raw data objects
0 Functional and malfunctional detectors objects

0 Pre and post peak hour data structures objects

v

t J Read a new object of raw data. k fl
v

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Was the end of the file reached? 4
N v
N r Does the station belong to
e ‘L any of the study segments?
N Y Y
Y
¢ Speed 2 0 H Past minute speed < 0
y N l
H Past minute speed = -1
N v
Start a new malfunctioning Record the end time in
We“ and store the start the malfunctioning object —>
trme ‘ and calculate the duration.

 

 

 

 

% Speed 28THV Yfi

Is an incident object open? Is an incident Yobject open?

IN

 

 

T2

 

 

i:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

N r Is the condition of the startfl 13 one Of “16* conditions
‘ of congestion satisfied? 1 - 5 satisfied?
a J
jl Y t
f Open a new functioning fl Record the end time in the
‘— object and record the start functioning object and
time. J calculate duration.
‘ l
Add physical station location to the objects #—
Separate objects . .
into Combine the functional and Call the queue length
pre, peak, and post malfunctional objects. and duration subroutine.
period objects.

 

Figure 3.2. Flow chart for the station incident-based congestion.

47

3% oz .r. 8.55 5...: an v inn . .33: .8: .3; m3 .8 eoenoueoo .o 5:23 a... 2:2".

:6
IS
06
9’8
£8
L8
08
971.
21.
M
OL
979
29

- PIS-Ir .

r u
but 20058.91ng unfllvfldflgrgr figflhbccocr in 2% $48 in?

(II “013318

 

 

48

and duration. This section describes the criteria and procedure developed to cluster

related stations in the case of severe incidents where multiple stations are affected.

3.4.1 Criteria for Clustering Station Incident-based Congestion Periods

To determine the grouping of the incident-based congestion periods obtained from
individual stations, three conditions were developed. The three conditions are depicted in
figures 3.4, 3.5, and 3.6. Condition one was set when two incident-based congestion
periods ab and cd belonged to two adjacent stations. In this case, the two were grouped

into one incident if one of the following conditions was satisfied:

cSan ......................................................................... 33
csbsd ......................................................................... 3A
eSgSf ......................................................................... 35
eSth ......................................................................... 36

Where:
a, c, e, and g are the starting time of incident-based congestion at stations 1, 2, and

3 and b, d, f, and h are their corresponding ending times.

The second condition was established for a station that had two congestion
periods separated by a variable named endstart, which is the time between the end of
congestion period 1, and the start of congestion period 2. If the value of endstart is S 15
minutes then the second congestion period is added to the first incident. The third

condition dealt with the case where successive stations had malfunctioning detectors; in

49

this case a maximum of 3 successive malfunctioning stations that were enclosed in

functioning stations were allowed in each incident.

 

 

 

 

d3 g—h—
"‘ d r
.gzc e
:31 a b
(I)

’

 

Time

Figure 3.4. Clustering successive stations of incident-based congestion (Case 1).

Station I.D.

d e
2 k—1 endstart S 15 minutes..__p|
1

 

 

 

Time
Figure 3.5. Clustering periods of incident congestion belonged to the same

station (Case 2).

50

 

 

 

‘i

. Mam-4
? malfunctioning detectors

‘ \i
functioning detectors

Station LD.

 

r—NUAUIOk

 

 

 

 

>

Time

Figure 3.6. Clustering malfunctiong stations (Case 3).

3.4.2 Development of the Clustering Technique

This subsection describes the development of the procedure used to cluster related
station congestion periods by applying the criteria presented in section 3.4. 1. The
procedure was performed in three steps. The first step began by calculating the ntunber
of congestion periods in each segment, then the algorithm performed three imbedded
loops. The outer loop looped the number of segments, the middle loop looped the
number of incidents reserved once the program started executing, and the inner most loop
looped the number of periods in each segment. After starting the first pass in each of the
three loops, the algorithm read the first period in the segment; if the period belonged to a
malfunctioning period it was skipped and the second pass was performed. This process
continued until a functioning period was encountered; at that time the routine opened the
first incident object and stored the period’s information in its first array element of the

incident period objects. It then moved to the next period and determined whether the

51

period belonged to an adjacent downstream station or the same station (conditions
established in sec 3.4.1); if the conditions were satisfied, the period was clustered in the
subsequent element of the incident object. It continued looping through the segment
period’s objects and marked those that were clustered. The same procedure was
performed for all other segments. The unmarked periods were clustered during the
second and third passes. Figure 3.7 shows the flow chart of the program written to
accomplish the first step.

The following two steps involved clustering the periods of incident-based
congestion periods that were missed in the first pass. The algorithm was called after the
conditions on closing an incident were satisfied. The algorithm received the address in
memory of the incident objects and the incident-based congestion period objects as well
as the segment number and the starting and ending period numbers in the segment. The
algorithm performed two loops: the outer loop that looped the number of periods in the
segment and the inner loop that looped the number of periods in the incident. The outer
100p started by reading the first unmarked period in the segment and determined the
upstream station ID. It then looped through the incident periods to determine whether the
period fell upstream from the segments unmarked period; if this condition was satisfied,
the clustering conditions in section 3.4.1 were checked. If these conditions were satisfied,
the segment period was added to the incident and marked. The program then moved to
the next unmarked period and performed the same checks. It continued looping through
the remaining segment periods and performed the same checks. Figure 3.8 illustrates the

flow chart for the steps described above.

52

 

LQueue length and delay subroutinq

v

Gets the address in memory for the
congestion periods and incident objects.

9

l Calculate the # of periods of station
congestion for each segment.

 

 

 

 

 

;
v
Loop the number of segments = 12.
v

 

 

 

i Loop the number of incidents = 60.
[ Loop the # of periods in this segment. %

V
Calculate STDIF.

V

Is this the first period in the segment Y Move to next period
and is it malfunctioning? in the segment.
N * Y

L Is this a malf. object and was the ] #

 

 

 

 

 

 

 

 

 

 

 

 

 

previous statement executed?

Mark object
N ‘ from the segment. 4—
N +

[s the first object in the incident empty? Y

Store this period’s info in the
lfi 0 o a 0
Does this period belong to the first 013.19“ Offllls moldent.

same station of the previous

N
L period and endstart 515 minutes? (I) (I) is satisfied but Y
endstart > [5 minutes. fl
v i

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

a N
Add this period to the subsequent
periods of this incident.
J v
Increment T N Does this period belong to the
incident. adjacent downstream station?
Y Y I
[ Is condition 2 satisfied%N %

 

Figure 3.7. Flow chart for the first pass algorithm.

53

 

Get the incidents address in memory, period # in
incident to cluster, number of periods in segment,
and the segment number.

7,

Declare variables k , ka, station [D upstream (stdup). ]

i

W Loop the number of periods in segment. ]

i

initialize the variables k = ka = 0.

Y , ir

a Was this period marked? ]
I N

>Loop the number periods in the incident.

i

stdup = station ID upstream of this segment’s period

t

‘2 incident period ID - segment period ID = stdupfl

$Y

‘— On of equations 3.1 to 3.4 was satisfied? ]

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Y
Include segment period
in the incident array.

[ Mark the segment period. J

t

 

 

 

 

 

 

 

 

 

 

 

 

a N This was the last period in incident? ]
N f Y
1 This was the last period in segment?

 

 

 

iv

Return to the calling routine.

 

 

 

 

Figure 3.8. Flow chart for the subroutine second pass algorithm.

54

The last step had a structure almost identical to step two. In this step, the missed
periods in the segment were checked if they fell downstream from any of the incident
periods. Those periods were clustered if any of the conditions of section 3.4.] were
satisfied. The algorithm received the same addresses in memory as step two. Figure 3.9
shows the flow chart for the third part.

To test the performance of the clustering techniques, incidents that occurred at
different times and dates during the months of July and November of 1995 were collected
in files and tested. Figure 3.10 shows one of the incident files used for this purpose. The
incident time shown in the figure is not the actual time and day when the incidents
occurred.

Figure 3.10 illustrates ten different incidents of various lengths and
characteristics. They can be categorized into two groups; the first group required only
the first part of the clustering algorithm due to their simple pattern, and the second group
required the second or third or both parts to cluster. The first group is represented by
incidents l, 2,. 3, 4, 7, 8, and 10. The effect of incidents 1, 3, 7, and 10 was observed on
one station and each had one period of incident-based congestion except incident 3,
which had two periods that were 7 minutes apart. Incident 4 had three periods of
functioning stations. The last incidents of this category were 2 and 4, which had multiple
periods and included malfunctioning stations. The second group of incidents is
represented by incidents 5, 6, and 9. Incidents 6 and 9, each had one period, 6-1 and 9-1,
that were missed on the first pass; these periods were clustered during the second and
third pass algorithms respectively. Incident 5 had two periods, 5-1 and 5-2, that were

missed on the first pass and were clustered during the second and third pass algorithms.

55

 

Get the incidents address in memory, period # in
incident to cluster, number of periods in segment,
and the segment number.

i

Declare variables k , ka, station ID upstream (stdup). ]

l

H Loop the number of periods in segment.

Initialize the variables k = ka = 0.

Y 6

¢ Was this period marked?
I N

Loop the number periods in the incident.
v

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

stddn = station ID downstream of this segrnent’s period

 

 

 

v

N
H segment period 1D - incident period ID = stdup?
N i Y
F One of equations 3.1 to 3.4 was satisfied?
j v
Include segment period
in the incident array.

Mark the segment period.

6

b— This was the last period in incident?

iv

Q This was the last period in segment?

iv

Return to the calling routine.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 3.9. Flow chart for the subroutine third pass algorithm.

56

 

 

 

0
8

0. cozwuw

Time

 

 

Figure 3.10 Incident samples

57

3.5 Incident Queue Length

The incident queue length was calculated for each incident and consisted of two
parts. The first part was a random variable that had two parts. The first and second parts
of the random variable were located upstream (1..) and downstream (Id) of the incident.
Both parts of the random variable could take a minimum value of zero miles in the case
where the front and rear of an incident occurred exactly on stations 4 and 2 respectively
as shown in Figure 3.11. The maximum value might be equal to the difference between
the incident’s front station (station 4) and the station downstream (station 5) of the
incident for the downstream part of the random variable. The upstream part of the
random variable could take a maximum value equal to the difference between the
incident’s upstream station (station 2) and the station upstream of the incident (station 1).

Figure 3.11 illustrates the two parts of the random variable.

hUpstream Observed IDownstream
Flength length length

1a

 

Incident-based congestion

 

 

 

 

Station 1 Station 2 Station 3 Station 4 Station 5

Direction of traffic flow

 

*

Figure 3.11 Upstream and downstream parts of the random variable.

58

The second part of the incident queue length involved a numerical calculation
that was a result of the clustering technique developed in section 3.4. A routine was
written to locate the incident’s extreme upstream and downstream stations, and the queue
length was then determined by calculating the difference in miles between the upstream
and downstream locations. Figure 3.12 shows the flow chart of the subroutine written to

perform the queue length numerical calculation.

3.6 Categorizing Segment Incidents

Once the incident queue length was calculated, the segment was categorized
based on this length. The first incident category was for incidents that had a length of a
random variable L, and the subsequent categories were lengths incremented by 0.5 miles.
The shortest incident fell in category one with a queue length equal to a random variable
L, and the longest fell in category eight, which had a length of more than 3 miles. Then
the number of incidents within each category was calculated. Table 3.3 shows the
different categories of incident queue lengths and Figure 3.13 illustrates the flow chart

used to accomplish these steps.

Table 3.3. Categories of incident queue lengths.

 

Category number Incident queue length (miles)
L
(0.0 S Incident Queue Length 5 0.5) + L
(0.5 S Incident Queue Length 5 1.0) + L
(1.0 .<_ Incident Queue Length 5 1.5) + L
(1.5 s Incident Queue Length 5 2.0) + L
(2.0 S Incident Queue Length 5 2.5) + L
(2.5 s Incident Queue Length 5 3.0) + L
Queue Length 2 3.0

 

 

 

 

 

 

 

 

 

 

”QGUIbUNh-t

 

 

59

 

[ Get the incidents address in memory. J

Jr

[ Declare variables for the number of periods and ]

 

incident duration and initialize them to zero.

3

fl Loop the number of incidents = 60. 4 ‘
i

Is this the last incident?

v

Initialize the variables head = tail = temp = 0

 

 

 

 

 

 

 

 

 

 

 

 

 

 

[ Calculate the # of periods in this incidentj
N b
r Does this incident have more than one period?

 

 

 

 

 

I v
Incident queue length = 0 Head = location of first bar.

i

Loop the # of periods in this incident. fi—fi
T

[ Add bar duration to incident duration. ]

v

Temp = bar location.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

* T
Ltail s temp? N >
4 v

 

Tail =vtemp
i l
[ Is this the last bar in incident? N ,

f Y
Queue length = tail — head.
V

G 0 back to the Assign incident duration value. Was this the last N
calling routine. Y incident?

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

\

 

Figure 3.12. Flow chart for the queue length of incident-based congestion.

60

 

Sets the address in memory for incidents and segments. ]

 

 

[ Declare (temporary) tsmps and an array of 8

elements and initialize them to zero.

 

-—-fi Loop the number of segments = [2

Loop the number of incidents = 60 k

i

 

 

 

 

 

 

 

]

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

IN

Does this incident have 2.0 s ql s 2.5 ?

1

 

 

 

 

 

 

 

 

 

 

 

Y
Does this incident have a queue length (ql) = 0 ? H Increment ql[0]
I N
. . . Y
Does this moldent have 0.0 s ql s 0.5 ? 4 lncrementql[1]
I N
[ Does this incident have 0.5 s ql S [.0 ? —Yfl Increment ql[2]
I N
Y
Does this incident have [.0 s ql s 1.5 ? I—fl Increment ql[3]
I N
Y
Does this incident have 1.5 S ql s 2.0 ? —->I Increment ql[4]

 

 

 

 

Y

 

 

 

 

IN

 

 

 

 

 

 

 

 

 

 

{ Increment ql[5]

 

 

 

 

 

 

 

I Tl I TTTTT

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Y
1
Does this incident have 2.5 s ql s 3.0 ? —-fl Increment ql[6]
[ This incident has ql > 3.0 miles Increment ql[7]
9
Add incident duration to tsmps.
a . . i .
H Is this the last rncrdent in the segment?
I v
Assign the value of tsmps to segment duration. I—p tsmps = 0. ]

 

 

 

 

 

t

 

 

 

 

 

Y
Go back to the calling routine. k.— Was this the last incident? dfi

 

 

Figure 3.13. Flow chart for the queue length subroutine.

61

 

3.7 Incident-based Congestion Index

The final measure developed was the station, segment, and network incident-
based congestion index. This index is a measure of the time a station, segment, or the
network is congested due to incidents. The index is used to estimate the average daily
minutes of incident-based congestion by multiplying the index value times the number of
off-peak daily minutes. Equations 2, 3, and 4 define the station, segment, and network
incident-based congestion indices. The denominator in equation 1 represents the number
of the station off-peak minutes in the sample. This number was multiplied by the number
of stations in the segment or the network as a whole for equations 3 and 4 respectively.

Figure 3.14 illustrates the flow chart used to calculate the segment index.

in=n
ZID

STIBCI = _in_=_1____ ................................................................ 3.7
NOS'IM

sn=m in=n

Z 21D

scum]: m .................................................................. 3.8
NOSGS

sgi=12 srl=rn in=n

Z ZZID

 

= 89H sn=1 in=1 .......................................................... 3.9
NIBCI N ONSM
where;
STIBCI = Station incident-based congestion index
SGIBCI = Segment incident-based congestion index
NIBCI = Network incident-based congestion index
sgi = Segment number
sn = Station number
m = Maximum number of stations on the segment

62

 

Gets the segments address in memory

T

Store in an array the number of stations for each segment. ]

I

{Store in an array the number of minutes for the ]

 

 

 

 

 

AM or PM for the pre, peak, and post periods.

l

Declare a variable temp and initialize it to zero.

l

 

 

 

 

 

 

Loop the number of segments = 12 ]‘

f

(Calculate the segment minutes and assign the value to temp J

 

 

temp = smps / (# of stations per segment “' # of minutes)

i

[ Assign the value of temp to smps

;

Is this the last segment?

iv

Send the station minutes per segment and the
segment incident queue lengths to an output

file.
1

Return to the calling routine. ]

 

 

 

 

N

 

 

 

 

 

 

Figure 3.14. Flow chart for the subroutine the station minutes per segment.

63

 

in = Incident number

n = Maximum number of incidents that occurred on the station
ID = Incident duration

NOSTM = Length of observation period at a station (minutes)
NOSGSM = Length of observation period on a segment (minutes)
NONSM = Length of observation period on the network(minutes)

3.8 Summary

The methodology demonstrates the development of the different measures of
effectiveness to evaluate the ITS deployment program based on incident-based
congestion. The performance of the measures was tested using a sample file that
included incidents that occurred during the months of July and November of 1995. The
algorithm identified all of the incidents.

The measures of effectiveness can be used to compare non-recurring congestion
before and after ITS technologies are implemented. Implementing the ITS is intended to
reduce the impact of incidents by reducing the traffic demand through ramp metering and
diverting traffic to adjacent streets with available capacity by using CMS. This claim can
be verified by studying the change in the magnitude of the incident-based congestion
index and the change in incident queue length. Chapter 4 covers the analysis of the

traffic data using the methodology developed in this chapter.

64

Chapter 4

Analysis and Results

4.0 Introduction

The analysis of the traffic data gathered before the ITS deployment program
became firlly operational was conducted using the research methodology developed and
described in Chapter 3. The analysis was performed to accomplish four primary goals:
first, to estimate the station and segment monthly variation in the incident-based
congestion index; second, to estimate the monthly and segment variation in incident
frequency; third, to estimate the distribution of the monthly incident queue length and
frequency. And finally, to develop a plan to conduct the after analysis.

Chapter 4 is divided into sections that describe the steps followed to analyze the
traffic data. The first section describes the network study segments and the second
section presents the variation in traffic volume during off-peak hours. The third section
explains the method used to process a single traffic file and how the single processed
output files were integrated to yield monthly output files. The fourth and fifth sections
describe the station and segment monthly variation of the incident-based congestion
index and frequency, respectively. The sixth section presents the plan to conduct the

after analysis, and finally the chapter summary is included in section seven.

4.1 Description of the Study Segments
Two fi‘eeways (1-94 and M-10) were considered in this study. To help isolate

locations with high incident-based congestion and incident frequency, each direction of

65

the two freeways was divided into three segments of various lengths. The first segment
on each freeway is that portion of the freeway that would carry traffic destined for the
central city in the morning peak period. The second segment is that portion of each
freeway in the central city, and the third segment is the portion of the freeway that carries
the heavy outbound traffic in the evening peak period. Table 4.1 contains detailed
information concerning the study segments, including segment length and types of ITS

components and Figure 4.1 shows the location of the segments on the map of Detroit.

4.2 Processing the Traffic Data

The database included 180 files representing nine months of traffic data and each
file contained 12 hours of either AM or PM minute by minute traffic data. Each data set
consisted of ten consecutive working days. The database processing procedure was
performed in two steps and this section is divided into two subsections describing the
steps. The first subsection describes the output file of a single processed file and the
second presents the method used to integrate single output files to form monthly output

files. Table 4.2 shows the months and dates of each of the nine sets of traffic data.

66

 

 

 

 

 

 

 

 

 

 

 

 

 

 

wBBoE cox—83:2 - v m N o. KN 3-2 mm vNNAeN N—
oni—«3:2 E3425 .. .. v _ : MNN 2-2 mm ocNéVN 2
2.8335 .3 209805 .. - v - ON mad 2.: mm mVNéNN 2
20:520 cowgun .3 _ - v _ N and 3-2 m2 mNNich m
H.855 .m .35 98.5 _ - m .. S me 2-2 ml «8-3— m
coo—33:2 “Essex _ - m N A: cmN 3-2 ml STE; x.

$882 2.-— - - w - wN 3N flea—mm mENE 0

mm; film 935 - _ m - 2 SN vmmrmm _3-mN_ m
35 case mesa? - - N _ 2 and 42mm 272: 4
mean? .35 case - _ _ - 2 e2 33? 823 m
co; 3..— - N _ _ _ _ NwN “ea—-95 eqmw N
0:25 $8.82 ~ - o w aN v2. 37m? Nwrvm _
use 38 g EDD $208 mEU c.8883 Ame—=5

353%: cocaomoc no we 9:8 we we we Swab; .352m e 59:52

maxim mcmtfim “3852 33:52 c3832 conga—Z confisz “coEwom coufim EcEwum

 

 

 

 

 

 

 

 

 

 

 

 

downtown—u Eofiwom .3. 033.

67

.20on we 92: 2: :o 3560?. beam ._.v uSmE

 

 

11¢

dr/v-asr:-.lu.- 4:0:
49

 
  

a.“ n EoEmcm

  
 

  

.731» live \;

v EcEWom

  

 

 

 

 

e\...

    

ac... Jigs;

4

I

12...: .
urea-:27.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

II‘.

 

‘i‘ I //
e m EcEwom \
...L
r. "aqua"...
>:nI->.2—
u:—
1 9‘0:
m
a
\ I
x a W ..\ 010:2” —
t ‘ L I II.
D .930 anus _
'4
< c To, “NAN 004 so m J. / — oeoe
c Emeom in 55:55. 0 Z :58me I
i 2 EoEwom .
(M o “WHCDEMOM r 4 \_ u but.
6 L o .7 v r u .
in “gm . x M m 6. e... a
.. o t a
t! I
“D O— m m I
/ .nwuwuo >93. 090 .30: w .
/ no >5: >21! _ V . :2; u
i v r‘. -. K
L _m .to 83:0 55:.
l .313
. —m. m: a C...-
J . . .
o 9.0.. 3.: 2...)...» / ‘ o<oe
flag 5:.- Wfi _ = 230—0 l, E Q HEDEWDW cubs w m
r dual—£5020 ‘ — Z“.U—.Zu.l — a .4
_ ' 2.3:: .0: %/ _ No—

 

 

mtm; m.

 

 

 

68

Table 4.2 Months and dates of the traffic data sets.

Set M Dates 0 1 week 0 data Dates 0 week 0 data
to 1 1 l to l 1
12/9/96 12/16/96
1 3 tel 17 to]
to 1 to 9
to to 3 1

to to

to to

to to 7
to to 1
to to

 

4.2.1 Processing a Single Traffic Data File

The output of a single processed traffic data file was the backbone of this research
study. There are two major output files formatted to provide the basic measures that are
integrated later to yield the monthly measures. The first output file includes the
estimated incident-based congestion periods for each station and the second output file
includes the estimated incident frequency and queue length for each segment.

The first output file consists of three major parts, and each has four variables: the
duration of the pre-peak incident-based congestion and its associated traffic volume for
each station, and the total number of reported minutes and its associated volume during
the pre-peak period for each station. The second part includes the post-peak period
values of the same variables as in part one. The number of reported rrrinutes at each
station was used to estimate the value of the make-up factor for each station during the
pre-peak and post-peak periods. The make-up factors were then used to factor the entries
in parts one and two and summed to yield the entries of part three. Table 4.3 shows these
three major parts for segment 5 during the AM period on January 13, 1997. The PM files

had the same format as the AM files and the daily measures were obtained by adding the

69

AngoEwa 082.9 95.89 .38.

>lnmm

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Ea: use 3:89 .52 58.8 annoy.
$2253 c8898 33926205 c_ 8295 oE=_o> >lcoo
A .EEV cosmomcoo 389E365 .6 3:88 “Icoo
9 Sim gram
0 o o o o o o o o o o o 3.4
Er 2m 0 0 name mm? o o 9mm 0mm 0 o a?
E: 2m 0 o me: E o 0 SF 8m 0 o 8.
man: 3m 93 2. 8% mm? 32 m: mmow 0mm o o 22
wow? m5 NY: .8 mm: lmINF NC; mm mmwor omm o o v?
vwwom arm 0 o 3% mm? o o 33.. own 0 o mmw
vmmom m5 0 o 33 mm? o o Noor own 0 c we
SSN own 0 o 38 k? 0 o 88 own 0 o me
NVONN m5 o o 3.2: me o 0 SN? omm o o mN_.
mNNwF mwm o 0 9mm mg 0 o :2 own 0 0 mm?
83.. m E o 0 $3 ImNr o o mowlw 0mm 0 o mN_.
>Inwm “Imam >lcoo ulcoo >Iaom «lama >Icoo ulcoo >laom “Imam >lcoo “Icoo HIE
neon: come 9 88 5 come 2 098 E 88 2 85° 3
7 use

 

 

=2 .52: .o. 5.38:8 8392865 .o 3:82. 2: .o 2... 29.5 3 2.3

 

70

AM and PM files. Table 4.4 shows the duration of incident-based congestion for
segment 5 on 1/ 13/ 1997

The second output file, shown in Table 4.5, includes the frequency of the different
incident queue length categories for each segment for the AM period of 1/ 13/ 1997. The
station make-up factor obtained in the first output file was used to estimate the segment
make-up factor for each off-peak period. The segment make-up factor was then applied
to the observed incident frequency in Table 4.5 to yield the entries in table 4.6. The PM
files had the same format as the AM files and the daily incident frequencies were
obtained by adding the AM and PM files. Table 4.7 shows the incident frequencies for

1/13/1997.

4.2.2 Integration of Multiple Traffic Output Data Files

This subsection describes how the 20 output files that belonged to each set
(month) of traffic data were integrated to form an output file. The periods of incident-
based congestion were summed over the 20 files and the incident-based congestion index
was calculated for each station. Similarly, the incident frequency and incident queue
length categories were summed for each segment to yield the monthly segment incident
frequency with their corresponding queue length. Table 4.8 and Table 4.9 show the
monthly values of the station incident-based congestion index for segment 5 and the
segment incident frequency for the month of January 1997, respectively. The same
procedure was followed in processing the other eight months. The following sections
describe the utilization of the monthly output files to achieve the goals listed at the

beginning of this chapter.

71

36.0.53 3.5.2. 953%. .30... >13”.

2:5 use 8589 .32 3.8.8 .raom

36.0.53 cozmomcoo 683-5622. E 3202... oE:_o> >lcoo
NEE. cozmomcoo 339.520... .o 5.650 .lcoo

o. 5.65 gram

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

o o o o o o o o o o o o 2:
momma 9.9 o o 82m Rm o o 83— m3 o o m?
98v over 0 o 8on Rm o o «New. m3 0 o a?
ommom $3.. 2.3 9 thm Rm 0 c an»: m3 9m? 2. B.
039‘ 99 NF: 3 to; Rm o 0 wow? m5 NE 3” 32
2.8m 9.9 o o mmmNm Rm 0 0 35m 2.... o o m?
ENE 02: o 0 896 Km o o 50.8 m5 o o 5..
omwmm 96.. o o 85.... En o 0 BEN m5 0 0 amp
83m 03.. o o 9me Rm 0 o NBNN 3m o o NF
«NS». meow 03; 3 SEN Rm as: 3 2va 0:... o 0 cm.
.83 9.9 mme me 39% Rm 8N? we 8m? 9m 0 o 3
,>lamm .laom >Icoo alcoo >Inom. «lacy. >lcoo .lcoo >laom .131 .350 alcoo 63m
III [I [I ll

52.2: .8 88 2 SN. E 33 o. 88 E
9::

 

hw~n F2. 50* 203000600 UGOOD-«CO‘E—S *0 209230 Vs? 0—Dflh

72

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

o o o o o o o o N?

o o o o o o o o 3

o o o o o o o c or

o o o o o o o o m

o o o o o o o o m

o o o o o o o o N 2.5
o o o o o o o F m 3
o o o o o o N o m 86
o o o o o o o o v

o o o o o o o o m

o o o o o o o F N

o o o o o o o o r

o o o o o o o N N?

o o o o o o o o 2.

o o o o o o o c or

o o o o o o o o m

o o o o o o o o w

o o o o o o o N N ONE
0 o o o o o o o o 3
o o o o o o o o m 005
o o o o o o o w v

o o o o o o o o m

o o o o o o o o N

o o o o o o o o _.

 

son—Ea: GEE.

a...” 3 a.“ 3 n... 558m

as a o. o. a o. 2 8 ._

law as w... 3 m... a...
32...... 59.2 angler _

.J
'l-
_I
'0'
_l
.l.
.J
4.
.J
4.
.l
'l'
_l
'l-

 

 

°.
F

 

 

 

 

 

 

 

 

 

 

E< .33 5. .3. 59.0. 0:2... new ages—co...— EoEoc. .0 33:6 3: 6.9.5 mé 03a...

73

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ANNE 598. 23:01

 

 

 

 

o o o o o o o o o N,
o o o o o o o o o 3
o o o o o o o o o S
o o o o o o o o o m
o o o o o o o o o o
o o o o o o o o o N 86
o o o o o o o F 2.. o 2
o o o o o o N o o: m 36
o o o o o o o o o v
o o o o o o o o o N
o o o o o o o P m: N
o o o o o o o o o P
o o o o o o o N 8.2 N2
0 o o o o o o o o 3
o o o o o o o o o 2
o o o o o o o o o a
o o o o o o o o o N
o o o o o o o N 82 N 86
o o o o o o o o o m 2
o o o o o o o o o m 86
o c o o o o o P 8.. v
o o o o o o o o o N
o o o o o o o o o N
o o o o o o o o o P
.— + .— + ._ + ._ ... A + .— + .— + 38a...— 5325: 2::
3 3 9N 3 3 3 .5 SE?»
a.» A 2 2 2 2 2 2 4 2.2,.
L...N 9N 3 3 m... a...

E< .53 :w .8 598. 2.2:. can 3:259: 2520:. @2803 .0 55:0 0.: 29.5 mi. 233.

 

74

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

o o o o o o o N N.
o o o o o o o o :
o o o o o o o o 9
o o o o o o o o a
o o o o o o o o N
o o o o o o o N N
o o F o o o o N o
o o o o o o N o m
o o o o o o o N e
o o o o o o o o N
o o o o o o o 3 N
o o o o o o o o 3
4+ ..+ ._+ ._+ ..+ ._+ 4... .32.:
3 3 3 3 3 3 858
3 A 2 2 2 2 2 2 ..
MN 3 3 3 i3 3

 

 

Tom—5 :55. 2.2.01

 

33 _.=. :8 59:2 2.2.: 3:: 3:252. 3020:. :22on he 03:...

 

75

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

c0320. .2. 5:02.50 033.2020... 2.0: .0 2.2.0.3 0900
x09... 5.50980 083.2020... .ow.
$0.029. 0E:.o> 95.80. .20... >Iaom.

ES. 0:... 9.3.2.0. .22 .2022. .1001
30.0.53 5.7.09.8 0002.22.00.05 c. 00202.. 0E:_o> >Icoo
ASE. ceammmcoo ummmn..:m.u.0c. .0 5.6.30 .Icoo
o. 5.85 Edw
me 2:. mommmmv $39 539 v3 .08...
0.0 08.0 o o o o 3
ed oood 382. $53 o o m?
O... 80.0 ovmmom 883 o 0 mm?
v.m End 3:2. 52: 2mm mm n?
3. End mwmvom $83 8%.. 3 3
0.0 80.0 32% 883 o o m?
o.o 08.0 383 $83 o o :2
ed 80.0 58$ 88 o 0 m3
Wm mono Eommm 8mm 8mm mm mm?
m.v mmvd 805ml $33 mmmm 2. 3..
ma Emd 02.80 883 mwFN mm mm?
0200 .2: 5.0m 23. .350 3.8 .50
53 33:2. .8 5:03:00 0003.23.05 .0 532.5 Q: 03:...

 

 

 

76

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

o o o o o F F. «F «F
o o o o o F F o FF
o o o o o F o 2 2
o o o o o o o o m
o o o o o o F e o
o o o o o o o N F
o o F o o F o m o
o o o o o o o u m
o o o F o o o mF F.
o o o o o o o F m
o o o o o o N n N
o o o F F o F a F
._+ ._+ ._+ ._+ 4+ ._+ ._+ .383:
a... m." a.“ 3 3 o... 559m
S" A 2 2 8 8 2 2 4
MW 3 3 3 m... a...

 

 

“3:5. :55. 2.2.6

 

ham? Fran—CU... 505 sucw- 0:050 UCN >OCQSUGB uc¢fi_oc_ mg? 0.9“».

 

77

4.3 Incident-based Congestion Index

This section presents the incident-based congestion index obtained by processing
the nine sets of traffic data described in section 4.2. Tables 4.10 to 4.13 show the
estimated average values of the incident-based congestion index by month and season for
each station. Table 4.14 includes the values of the estimated incident-based congestion
index for the different months and segments. The values of the incident-based congestion
index included in these tables are utilized to determine the stations, months and segments
with high incident-induced congestion.

This section is divided into four subsections. The first subsection presents the
average monthly and seasonal station incident-based congestion index. The second
subsection presents the monthly variation in the incident-based congestion index by
segment. The third subsection includes the segment by segment variation in the incident-
based congestion index and the last subsection describes the system wide monthly

variation in the incident-based congestion index.

4.3.] Station incident-based Congestion Index
The average monthly station incident-based congestion index by season are
included in Tables 4.10 to 4.13. These Tables are plotted in Figures 4.2 to 4.5. and

analyzed in the following subsections.

4.3.1.1 Station incident-based Congestion Index on WB I-94
Figure 4.2 shows the seasonal variation in the incident-based congestion index for

segments 1, 2, and 3 on WB I-94. In the winter, segment 1 showed low values of

78

Table 4.10 Average monthly station incident-based congestion index

Station Station ‘ Station ‘ Station
ID IBCI ID IBCI ID IBCI ID IBCI

54 0.023 105 0.049 163 0.049 221 0.000
55 0.036 106 0.011 164 0.000 222 0.030
56 0.075 107 0.016 165 0.000 223 0.047
57 0.035 106 0.047 166 0.022 224 0.062
56 0.090 109 0.171 167 0.024 225 0.036
59 0.075 110 0.105 166 0.000 226 0.072
60 0.037 111 0.063 169 0.000 227 0.000
61 0.000 112 0.105 170 0.000 226 0.069
62 0.024 113 0.456 171 0.005 229 0.004
63 0.064 114 0.188 172 0.096 230 0.066
64 0.132 115 0.079 177 0.347 231 0.005
65 0.255 117 0.369 179 0.066 232 0.061
66 0.300 116 0.195 161 0.000 233 0.026
67 0.027 119 0.116 162 0.316 234 0.044
66 0.026 120 0.156 163 0.160 235 0.066
69 0.060 122 0.223 164 0.076 236 0.101
70 0.090 123 0.195 185 0.159 237 0.024
71 0.066 125 0.362 166 0.010 236 0.493
72 0.530 126 0.644 169 0.024 239 0.000
73 0.202 126 0.606 192 0.000 240 0.162
74 0.069 129 0.560 193 0.009 241 0.000
75 0.194 131 0.762 194 0.013 242 0.397
76 0.510 133 1.732 195 0.000 243 0.100
77 0.456 134 3.319 196 0.034 244 0.195
76 1.354 137 1.211 196 0.047 245 0.500
_79 0.412 136 0.749 199 0.001 246 0.063
60 0.295 139 0.450 200 0.003 246 0.070
61 0.747 141 0.066 201 0.113 250 0.031
62 0.007 142 0.576 202 0.000 251 0.143
63 0.066 143 0.465 204 0.174 253 0.036
64 0.369 144 0.406 205 0.000 254 0.035
65 0.672 146 0.439 206 0.000 255 0.054
66 0.933 147 1.126 207 0.000 256 0.006
69 0.615 146 1.161 206 0.031 257 0.000
90 0.665 149 1.106 209 0.049 259 0.019
92 1.116 150 0.504 210 0.005 260 0.040
93 0.602 151 0.290 211 0.046 263 0.345
94 0.266 152 0.262 212 0.104 264 0.222
95 0.136 153 0.414 213 0.127 266 0.036
96 0.066 155 0.605 214 0.071 267 0.000
96 0.035 156 0.699 215 0.071 266 0.000
99 0.219 156 0.103 216 0.059 269 0.047
100 0.065 159 0.055 217 0.070 270 0.163
101 0.014 160 0.012 216 0.080 272 0.312
102 0.273 161 0.000 219 0.063 273 0.609
104 0.001 162 0.000 220 0.065 27? 0.513

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

79

Table 4.11. Average winter station incident-based congestion index

 

Station I Station ' §tation Station ‘
ID IBCI ID IBCI ID IBCI ID IBCI

 

54 0.016 105 0.000 163 0.117 221 0.000
55 0.000 106 0.000 164 0.000 222 0.000
56 0.114 107 0.000 165 0.000 223 0.000
57 0.105 108 0.000 166 0.000 224 0.022
58 0.269 109 0.389 167 0.000 225 0.000
59 0.173 110 0.157 168 0.000 226 0.000
60 0.111 111 0.117 169 0.000 227 0.000
61 0.000 112 0.144 170 0.000 228 0.050
62 0.071 113 0.907 171 0.000 229 0.000
63 0.135 114 0.200 172 0.000 230 0.000
64 0.299 115 0.181 177 0.259 231 0.000
65 0.355 117 0.864 179 0.198 232 0.000
66 0.071 118 0.231 181 0.000 233 0.000
67 0.068 119 0.142 182 0.714 234 0.000
68 0.077 120 0.157 183 0.000 235 0.130
69 0.111 122 0.209 184 0.000 236 0.000
70 0.018 123 0.290 185 0.000 237 0.000
71 0.065 125 0.857 188 0.000 238 1.295
72 0.613 126 2.125 189 0.015 239 0.000
73 0.506 128 0.754 192 0.000 240 0.031
74 0.123 129 0.142 193 0.000 241 0.000
75 0.244 131 0.173 194 0.025 242 0.671
76 0.520 133 0.281 195 0.000 243 0.227
77 0.000 134 0.867 196 0.000 244 0.479
78 0.492 137 0.802 198 0.140 245 0.646

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

79 ‘ 0.827 138 0.278 199 0.003 246 0.028
80 0.296 139 0.192 200 0.009 248 0.031
81 0.243 141 0.000 201 0.003 250 0.080
82 0.000 142 0.724 202 0.000 251 0.077
63 0.135 143 0.469 204 0.313 253 0.000
84 0.453 144 0.586 205 0.000 254 0.000
85 1.217 146 0.458 206 0.000 255 0.000
86 1.225 147 0.516 207 0.000 256 0.000
89 0.740 148 0.891 208 0.000 257 0.000
90 1.347 149 0.904 209 0.000 259 0.000
92 1.764 150 0.849 210 0.000 260 0.058
93 1.267 151 0.503 211 0.043 263 0.454
94 0.240 152 0.451 212 0.099 264 0.213
95 0.055 153 0.927 213 0.108 266 0.031
96 0.043 155 1.431 214 0.000 267 0.000
98 0.037 156 1.564 215 0.000 268 0.000
99 0.000 158 0.000 216 0.090 269 0.028
100 0.000 159 0.000 217 0.210 270 0.086
101 0.012 160 0.000 218 0.241 272 0.644
102 0.037 161 0.000 219 0.188 273 1.346
104 0.000 162 0.000 220 0.255 2% 0.898

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

80

Table 4.12 Average spring station incident-based congestion index

Station Station Station Station
ID IBCI ID IBCI ID IBCI ID IBCI

 

 

54 0.000 105 0.148 163 0.000 221 0.000

 

55 0.000 106 0.034 164 0.000 222 0.089

 

56 0.000 107 0.049 165 0.000 223 0.141

 

57 0.000 108 0.022 166 0.040 224 0.166

 

58 0.000 109 0.124 167 0.071 225 0.114

 

59 0.000 110 0.127 168 0.000 226 0.000

 

60 0.000 111 0.071 169 0.000 227 0.000

 

61 0.000 112 0.170 170 0.000 228 0.158

 

62 0.000 113 0.417 171 0.016 229 0.012

 

63 0.015 114 0.362 172 0.288 230 0.258

 

64 0.022 115 0.056 177 0.000 231 0.015

 

65 0.176 117 0.240 179 0.000 232 0.184

 

66 0.454 118 0.320 181 0.000 233 0.028

 

67 0.012 119 0.099 182 0.000 234 0.000

 

68 0.000 120 0.133 183 0.000 235 0.000

 

69 0.000 122 0.278 184 0.025 236 0.133

 

70 0.000 123 0.204 185 0.158 237 0.071

 

71 0.000 125 0.052 188 0.003 238 0.185

 

72 0.723 126 0.102 189 0.000 239 0.000

 

73 0.102 128 0.405 192 0.000 240 0.235

 

74 0.083 129 0.338 193 0.026 241 0.000

 

75 0.120 131 0.607 194 0.015 242 0.343

 

76 0.006 133 1 .850 195 0.000 243 0.000

 

77 0.000 134 3.130 196 0.046 244 0.000

 

78 0.173 137 0.942 198 0.000 245 0.000

 

79 0.223 138 0.544 199 0.000 246 0.000

 

80 0.294 139 0.395 200 0.000 248 0.000

 

81 0.306 141 0.000 201 0.000 250 0.012

 

82 0.000 142 1.003 202 0.000 251 0.353

 

83 0.130 143 0.831 204 0.163 253 0.108

 

84 0.019 144 0.636 205 0.000 254 0.105

 

85 0.272 146 0.661 206 0.000 255 0.161

 

86 0.435 147 0.627 207 0.000 256 0.000

 

89 0.416 148 0.037 208 0.093 257 0.000

 

90 0.381 149 0.108 209 0.147 259 0.056

 

92 0.550 150 0.191 210 0.015 260 0.062

 

93 0.526 151 0.253 211 0.096 263 0.457—

 

94 0.121 152 0.309 212 0.214 264 0.285

 

95 0.080 153 0.314 213 0.272 266 0.000

 

96 0.090 155 0.385 214 0.056 267 0.000

 

98 0.000 156 0.527 215 0.095 268 0.000

 

99 0.068 158 0.310 216 0.086 269 0.000

 

100 0.068 159 0.124 217 0.000 270 0.000

 

101 0.031 160 0.000 218 0.000 272 0.000

 

102 0.205 161 0.000 219 0.000 273 0.043

 

 

104 0.003 162 0.000 220 0.000 274 0.559

 

 

 

 

 

 

 

 

 

81

Table 4.13 Average summer station incident-based congestion index

— 1 — ‘ l —
Station Station Station Station
ID IBCI ID IBCI ID IBCI ID IBCI

54 0.053 105 0.000 163 0.031 221 0.000

 

 

55 0.108 106 0.000 164 0.000 222 0.000

 

56 0.111 107 0.000 165 0.000 223 0.000

 

57 0.000 108 0.121 166 0.025 224 0.000

 

58 0.000 109 0.000 167 0.000 225 0.000

 

59 0.052 110 0.031 168 0.000 226 0.216

 

60 0.000 1 1 1 0.000 169 0.000 227 0.000

 

61 0.000 112 0.000 170 0.000 228 0.000

 

62 0.000 113 0.043 171 0.000 229 0.000

 

63 0.040 114 0.003 172 0.000 230 0.000

 

64 0.074 115 0.000 177 0.783 231 0.000

 

65 0.235 117 0.062 179 0.000 232 0.000

 

66 0.374 118 0.034 181 0.000 233 0.049

 

67 0.000 119 0.108 182 0.238 234 0.133

 

68 0.000 120 0.182 183 0.479 235 0.133

 

69 0.129 122 0.182 184 0.203 236 0.170

 

70 0.252 123 0.090 185 0.320 237 0.000

 

71 0.138 125 0.176 188 0.028 238 0.000

 

72 0.255 126 0.306 189 0.056 239 0.000

 

73 0.000 128 0.657 192 0.000 240 0.281

 

74 0.000 129 1 .262 193 0.000 241 0.000

 

75 0.219 131 1.506 194 0.000 242 0.176

 

76 1 .002 133 3.066 195 0.000 243 0.074

 

77 1.368 134 5.959 196 0.056 244 0.104

 

78 3.398 137 1 .887 198 0.000 245 0.853
79 0.186 138 1.425 199 0.000 246 0.163

 

 

80 0.294 139 0.764 200 0.000 248 0.178

 

81 1.691 141 0.199 201 0.337 250 0.000

 

82 0.020 142 0.000 202 0.000 251 0.000

 

83 0.000 143 0.096 204 0.046 253 0.000

 

84 0.636 144 0.000 205 0.000 254 0.000

 

85 1.126 146 0.197 206 0.000 255 0.000

 

86 1.140 147 2.237 207 0.000 256 0.025

 

89 0.690 148 2.617 208 0.000 257 0.000

 

90 0.866 149 2.312 209 0.000 259 0.000

 

92 1.041 150 0.473 210 0.000 260 0.000

 

93 0.614 151 0.114 211 0.000 263 0.123

 

94 0.502 152 0.025 212 0.000 264 0.169

 

95 0.280 153 0.000 213 0.000 266 0.083

 

96 0.126 155 0.000 214 0.157 267 0.000

 

98 0.068 156 0.006 215 0.118 268 0.000

 

99 0.588 158 0.000 216 0.000 269 0.115

 

100 0.127 159 0.040 217 0.000 270 0.402

 

101 0.000 160 0.037 218 0.000 272 0.293

 

102 0.578 161 0.000 219 0.000 273 0.439

 

 

 

 

 

 

104 0.000 162 0.000 220 0.000 274 0.083

 

 

 

 

 

82

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

SN... 8:. 8m... 8F... EN... 82. $2. 8:. £2. 6N»... 55... .93
mNN... 8F... 0N0... FNm... m8... 8N5 3F... 9:. 82. 22. NF
63... 8.... 82. 8o... 8.... F8... 8F... 23. $3. 8.... FF
NNF... So... 3F... 22. 98... 3F... o8... m8... SN... SN... 2
53... 68... 8o... 8.... 82. 8o... 82. :3... 8o... 58... a
58... SF... to... 1o... :2. 53... 9...... N8... .3... 68... m
2:. 2...... $2. 82. So... So... So... 2...... 8o... EN... F
N8... 23. m8... 8N... N8... tin... 2m... «Ne... 82. SN... o
F3... :3. ONFN NSN F3... F8... 33. 32. 3F... «FFF m
82. 82. m8... m2... 8N5 :8... EN... at... 22. F8... 6
«8... SF... 68... EN... Bo... Bo... Bod 93. F8... 82. m
F8... 5N”... NNX. EFF 68... F2... 8o... 23 32. N53 N
3N... F8... F8... IFS... 8:. N85 8°... 82. 82. EN... F
.558 .m>< B53. 3-..... 3.5% 8.6.... 5...? 3.5.2 3.8“. L652. 8.800 n .coemmml

 

3:333 6:: a5:o:. East... o5 .3.— xou:. 5.33:8 non—23:33:. 3.: snub

 

83

incident-based congestion. On this segment, station 79 had the highest index value of
0.827, which translates to 9.0 minutes of average daily incident-based congestion.
Segment 2 experienced very high index values. Station 92 had the highest index value of
1.764, which is equivalent to 19.1 minutes of average daily incident-based congestion.
Segment 3 did not show any noticeable congestion.

In the spring, there was an over all reduction in incident-based congestion.
Segment 1 had very low values of incident-based congestion compared to segments 2 and
3. Station 72 had the highest index value of 0.723. This value translates to 7.8 minutes
of average daily incident-based congestion. On segment 2, station 92 had the highest
index value of 0.550. This value corresponds to 5.9 minutes of average daily incident-
based congestion. On segment 3, station 102 had the highest index value of 0.205. This
value is equivalent to 2.2 minutes of average daily incident based congestion.

The pattern in the summer shows an increase in incident-based congestion on all
segments. Station 78 on segment 1 had the highest index value of 3.398. This value
translates to 36.7 minutes of average daily incident-based congestion. Segment 2 had
higher index values than the winter. On segment 2, station 86 had the highest index value
of 1.14. This value is equivalent to 12.3 minutes of average daily incident based
congestion. On segment 3, station 99 had the highest value of 0.588. This value
corresponds to 6.4 minutes of average daily incident-based congestion.

The average of the three seasons is plotted in Figure 4.2 (d). This figure shows
station 78 on segment 1 with an index values of 1.354. This value is equivalent to 16.6
minutes of daily incident-based congestion. On segment 2, station 92 had the highest

index value of 1.118, which translates to 12.1 minutes of daily incident-based congestion.

84

On segment 3, station 102 had the highest index value of 0.273, which corresponds to 2.9

minutes of daily incident-based congestion.

4.3.1.2 Station Incident-based Congestion Index on EB 1-94

The seasonal variation in the incident-based congestion index for segments 4, 5,
and 6 on EB I-94 is illustrated in Figure 4.3. The majority of the stations on this direction
of the freeway show incident-based congestion indices below 1.00. In the winter, station
113 on segment 4 demonstrated the highest index value of 0.907, which translates to 9.8
minutes of average daily incident-based congestion. On segment 5, station 126 had the
highest index value of 2.125, which corresponds to 23.0 minutes of average daily
incident-based congestion. Almost half the stations on segment 6 show no incident-based
congestion. On segment 6, station 156 had the highest index value of 1.564, which
translates to 16.9 minutes of average daily incident-based congestion.

In the spring, the incident-based congestion on all segments was slightly lower
than in the winter. On segment 4, Station 113 shows the highest index value among the
stations on this segment. Its index value was 0.417, which translates to 4.5 minutes of
average daily incident-based congestion. Segment 5 shows almost the same amount of
congestion compared to the winter, where station 134 had the highest value of 3.130.
This value corresponds to 33.8 minutes of average daily incident based congestion.
Segment 6 also had a slight decrease in the incident-based congestion index compared to
the winter. On this segment, station 142 had the highest index value of 1.003. This value

translates to 10.8 minutes of average daily incident based congestion.

85

a. Winter (WB 1-94)

——-——>_ ___K West,

 

 

 

b. Spring (WB l-94)

 

Incident-based congestion

c. Summer (WB l-94)

congestion index

 

lncident~beeed

(1. Average month (WB I-94)

 

incident—baud congestion index

Figure 4.2 Seasonal variation in the incident-based congestion index
on W8 l-94

86

The pattern in the summer exhibited a similar pattern as the spring, where
segment 4 experienced almost no incident-based congestion compared to the winter and
spring. Stations 120 and 122 on this segment show the highest index value among the
stations on this segment. Their index value was 0.182, which translates to 2.0 minutes of
average daily incident-based congestion. On segment 5, station 134 maintained its high
value among the rest of the segment’s stations. Its index value was 5.959. This value
translates to 64.4 minutes of average daily incident-based congestion. On segment 6,
stations 147, 148, and 149 had the highest index values of 2.237, 2.617, and 2.312 which
translates to 24.2, 28.3, and 25.0 minutes of average daily incident-based congestion
respectively. The rest of the stations on this segment show almost no incident-based
congestion.

Figure 4.3 (d) illustrates the season’s average in the incident-based congestion
index. On segment 4, station 113 showed a high index value of 0.456, which translates to
4.9 minutes of daily incident-based congestion. On segment 5, station 134 had the
highest index value of 3.319, which is 35.8 minutes of daily incident-based congestion.
On Segment 6, station 147 had the highest index value of 1.126, which is 12.2 minutes of

daily incident-based congestion.

4.3.1.3 Station Incident-based Congestion Index on NB M-l0

Figure 4.4 shows the seasonal variation in the incident-based congestion index for
segments 7, 8, and 9 on NB M-10. In the winter, most of the stations on segment 7 had
very low incident-based congestion. Station 182 had the highest index value of 0.714,

which translates to 7.7 minutes of average daily incident-based congestion. Segment 8

87

a. Winter (EB 1-94)

——-—-> East

incident-based congestion index

 

1). Spring (EB 1-94)

 

incident-based congestion index

c. Summer (EB l-M)

" 7.00
E...
5.00
g...
3.00

g 2.00
E 1.00
g 0.00

 

d. Average month (EB l-94)

 

Incident-based congestion index

Figure 4.3 Seasonal variation In the incident-based congestion index
on E8 1-94

88

experienced the lowest index values with station 198 showing the highest value among
the segment stations. This station had an index value of 0.14, which translates to 1.5
minutes of average daily incident-based congestion. The stations on segment 9 showed
higher index values compared to segments 7 and 8, and of those stations, station 220 had
the highest index value of 0.255, which corresponds to 2.8 minutes of average daily
incident-based congestion.

In the spring, segment 7 showed almost no incident-based congestion. Station 185
shows the highest index value of 0.158 that translates to 1.7 minutes of average daily
incident-based congestion. Segment 8 also showed no congestion except for station 204,
which had an index value of 0.163. This translates to 1.8 minutes of daily incident-based
congestion. Segment 9 showed higher value of the incident-based congestion; station 213
had the highest value of 0.272. This value translates to 2.9 minutes of average daily
incident based congestion.

The pattern in the smnmer exhibited slightly higher index values for some stations
compared to the winter and spring. Station 177 on segment 7 showed the highest index
value compared to the rest of the segment’s stations. Its index value was 0.783, which
translates to 8.5 minutes of average daily incident-based congestion. Segment 8 also
showed low index values. On segment 8, station 201 had the highest index value among
the rest of the segment’s stations. Its index value was 0.337, which translates to 3.6
minutes of average daily incident-based congestion. Segment 9 also showed very low
congestion. On this segment, station 214 had the highest index value of 0.157, which

translates to 1.7 minutes of average daily incident-based congestion.

89

Part “d” of Figure 4.4 illustrates the average of the three seasons. This figure
shows that on segment 7, station 177 had the highest index value of 0.347, which
translates to 3.7 minutes of daily incident-based congestion. On segment 8, station 204
had the highest index value of 0.174, which corresponds to 1.9 minutes of daily incident-
based congestion. On Segment 9, station 213 had the highest index value of 0.127, which

is equivalent to 1.4 minutes of daily incident-based congestion.

4.3.1.4 Station Incident-based Congestion Index on SB M-10

The seasonal variation in the incident-based congestion index for segments 10,
11, and 12 on SB M-10 is illustrated in Figure 4.5. In the winter, most of the stations on
segment 10 had very low incident-based congestion. Station 238 had the highest value of
1.295, which is equivalent to 14.0 minutes of average daily incident-based congestion.
Segment 11 experienced no incident-based congestion, while Segment 12 shows no
incident-based congestion except for a few stations. On this segment, station 273 had the
highest index value of 1.346, which is equal to 14.5 minutes of average daily incident-
based congestion.

In the spring, the incident-based congestion on most of the stations was below 0.4
with the exception of few stations. The station with the highest index value on segment
10 was station 242, which had an index value of 0.343. This corresponds to 3.7 minutes
of average daily incident-based congestion. Segment 11 also showed low incident-based
congestion where station 251 had an index value of 0.3 53, which translates to 3.8 minutes

of daily incident-based congestion. Segment 12 shows low incident-based congestion.

90

I. Winter (NB M—10)
1.00

-——> North

0.00
0.60
0.40

0.20

incident-based congestion index

 

0.00

1). Spring (NB M-10)
1.00
0.00
0.60
0.40

0.20

 

0.00

Incident-based congestion index

c. Summer (NB M-10)

 

1.00 7

0.80

 

0.60

 

0.40

 

 

I
°illll . L n

d. Average month (NB M-10)

 

Incident-based congestion index

§

 

 

0.00

 

 

 

 

0.40

 

'3

 

8

Incident-based congestion index
.°
8

Figure 4.4 Seasonal variation in the incident-based congestion index
on NB M10

91

The station with the highest index value on this segment was 274, with an index value of
0.559, which is equivalent to 6.0 minutes of average daily incident-based congestion.

The pattern in the summer showed lower index values compared to the winter and
slightly higher than the spring. Station 245 on segment 10 showed the highest index value
compared to the rest of the segment’s stations. Its index value was 0.853, which translates
to 9.2 minutes of average daily incident-based congestion. Segment 11 showed no
incident based congestion, and segment 12 showed very low incident-based congestion.
On this segment, station 273 had the highest index value of 0.439. This value is
equivalent to 4.7 minutes of average daily incident-based congestion.

Figure 4.5 “d” shows a plot of the average of the three seasons. On segment 10,
station 245 showed the highest index values of 0.50, which translates to 5.4 minutes of
daily incident-based congestion. On segment 11, station 251 had the highest index value
of 0.143, which corresponds to 1.5 minutes of daily incident-based congestion. On
Segment 12, station 273 had the highest index value of 0.609, which is equivalent to 6.6
minutes of daily incident-based congestion. The analysis shows that M-10 had

considerably lower incident-based congestion than I-94.

4.3.2 Segment Monthly Variation of the Incident-based Congestion Index

Figure 4.6 illustrates the monthly variation in the incident-based congestion index
on westbound I-94, which includes segments 1, 2, and 3. During the winter, segment 2
had higher values of incident-based congestion than segments 1 and 3, especially for the
month of December where the index value was strikingly high. Segment 3 shows the

lowest values for all months except for the months of March and June. During the spring,

92

a. Winter (88 M40)
———> South

 

1). Spring (SB M40)
1 .60

1 .40
1 .20
1 .00
0.80
0.60
0.40
0.20
0.00

Incident-based congestlon index

 

c. Summer (SB M40)

 

d. Average month (SB M40)
£1.60
F140
I:
81.20
D
§1.00
80.80

30.00

50.40

C

$0.20
0.00

 

Figure 4.5 Seasonal variation in the incident-based congestion index
on $8 M10

93

segment 2 had a higher index value than the other two segments except for the month of
March where segments 1 and 3 show higher index values. In the summer, the congestion
index was relatively high on segment 2 during the month of June. On the average,
segments 1, 2, and 3 had index values of 0.214, 0.561, and 0.073. These values translate
to 2.3, 6.1, and 0.8 minutes of incident-based congestion per day per station on segments
1, 2, and 3, respectively.

Figure 4.7 illustrates the monthly variation in the incident-based congestion index
on eastbound I-94, which includes segments 4, 5, and 6. Segment 5 shows a noticeably
higher index value than the other two segments. On the average, segments 4, 5, and 6
had index values of 0.188, 0.971, and 0.302. These values translate to 2.0, 10.5, and 3.3
minutes of daily incident-based congestion per station for segments 4, 5, and 6,
respectively.

Figure 4.8 illustrates the monthly variation of the incident-based congestion index
on northbound M-10, which includes segments 7, 8, and 9. These segments had
relatively low values. There is very little incident-based congestion on these segments
compared to I-94. On the average, segments 7, 8, and 9 had index values of 0.116, 0.039,
and 0.049. These values translate to 1.3, 0.4, and 0.5 minutes of incident-based
congestion per day per station for segments 7, 8, and 9, respectively.

Figure 4.9 illustrates the monthly variation of the incident-based congestion index
on southbound M-10, which includes segments 10, 11, and 12. On the average, segments
10, 11, and 12 had index values of 0.122, 0.045, and 0.225. These values translate to 1.3,

0.5, and 2.4 minutes of incident-based congestion per day per station for segments 10, 11,

94

 

*“ 1

 

 

 

E] Segment 1 E3 Segment 2 n Segment 3

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

l
; zs,,t ................................ . ...........

 

3.50

9|. 0
O O O O O O O
O lo q ll! 9 I!) O
M N N 1- 1- O O

xepul uogseBuoo peseq-auepgom

 

zs-fmv

96-390

Month

 

 

95

Figure 4.6 Monthly variation of the incident-based congestion index for segments 1, 2, and 3 on

WB l-94

 

:o o :5 .m .v 3:25: :8 x35 :oaeougo ceeenégEoE c5 .0 :caetg 25:05. E: 0.52".

Va; mm

 

 

 

      

 

 

    

 

 

 

 

 

 

  

 

 

 

 

 

 

 

 

 

 

55:
V
A
10
S
9 V P P: W V W I1. P G
n n a 0 a
m .4 4| u 0.. a. w a u ._.
_ .
MW 6 6 6 6 6 6 6 6 r0
1. I. I. I. I. I. I. I. I. 9
F - i i
l. é a
u 0 L. ”m u. x .0 u .l .l V x 0
.. .n H. I I H. .
x u. .. 9 .0... 3 .n v.
u. nu ”u l 0” Ir ”u I.
.n .u . 9 0 u.
u. H. 0 (.1. .u
.n .u .0 . 2 F.
”n ”u 6 8 Ir
n” nIt. " .Iv.
m. m it
.2 mm
L .u
r Wu H.
o z
9
5
z

 

 

 

 

 

 

m EoEmom a m EoEmom E v EoEmow a

 

 

cod
and
co. _.

- om...

ocfi
and
ac.”
on."

xepu! ucuseBuoo peseq-aueploul

 

 

96

 

 

 

E’J Segment 7 E31 Segment 8 0 Segment 9

 

 

 

 

 

 

 

 

 

 

 

 

d P———————- _.._-- -rfim-.-- m n_-

 

 

 

 

3 50
3 00
2.50
2 00
1 50

xepu! uogsefiuoo peseq-auepiom

 

Month

 

 

97

Figure 4.8 Monthly variation of the incident-based congestion index for segments 7, 8, and 9 on
NB M40

 

 

 

3 Segment 10 [:1 Segment 11 0 Segment 12

 

 

 

 

 

 

 

 

 

 

 

 

 

   
 

 

 

3 50
3 00
2.50
2 00
1 50

xepu! uonseBuoo peseq-zueplom

98

 

luawbas 'bAV

LG—bnv

Month

 

L...

Figure 4.9 Monthly variation of the incident-based congestion index for segments 10, 11, and 12 on
SB M40

 

and 12, respectively. These values are somewhat higher than the values for northbound

M-10, but still significantly less than the values for 1-94.

4.3.3 Segment Variation of the Incident-based Congestion Index

In order to isolate locations with high incident-based congestion, the nine sets of
data were used to estimate the incident-based congestion index for the different segments.
Figure 4.10 illustrates the variation in the incident-based congestion index among the
different segments.

On I-94, the eastbound shows about 1.6 times the westbound incident-based
congestion when aggregated by direction. The incident-based congestion index and
number of daily minutes of incident-based congestion per station were 0.262 and 2.8
minutes for westbound I-94 and 0.415 and 4.5 minutes for eastbound 1-94. Segments 2
and 5, which are located in the area that includes interchanges with I-96 and M-10
experienced higher levels of incident-based congestion than the other segments.

On M-lO, the southbound segments show a higher level of incident-based
congestion than the northbound segments, although both directions are relatively incident
free compared to I-94. The incident-based congestion index and number of daily minutes
of incident-based congestion per station were 0.063 and 0.7 minute for the northbound
segments and 0.127 and 1.4 minutes for the southbound segments. Segment 12 had the
highest index value on M-10.

The aggregate incident-based congestion index and number of daily minutes of
incident-based congestion per station were 0.34 and 3.7 minutes for I-94 and 0.95 and 1.0

minutes for M-10. The network wide incident-based congestion index and number of

99

 

.82. 5:005:00 0003.23.05 05 .0 531...: 30:58 0...: 0.3m.“—

._e:E:: EeEmem

iuawbas ’bAV
1 has

81 698
ll 598
01 693
6 has
8 693
was
9 has
9 has
17 598
9 has
a 693

- cod

 

- cud

 

cvd

 

- cod

 

- and

 

 

xepu! uonsefiuoo peseq-zuepgoul

 

 

 

 

100

daily minutes of incident—based congestion per station were 0.230 and 2.5 minutes
respectively. Table 4.15 shows the values of the incident-based congestion index and

number of daily minutes of congestion per station for the different directions of the

freeways.

Table 4.15 Incident-based congestion index and number of daily minutes of congestion
per station for the different directions of the fi'eeways.

 

 

 

 

 

 

 

 

Freeway incident-based Number of daily minutes of incident-
congestion index based congestion per station

WB 1-94 (Segments l, 2, and 3) 0.262 2.8
EB I-94 (Segments 4, 5, and 6) 0.415 4.5
NB M-10 (Segments 7, 8, and 9) 0.063 0.7
SB M-10(Segments 10, 11,and 12) 0.127 1.4
l-94 0.340 3.7
M-lO 0.095 1.0
l-94 and M-lO 0.230 2.5

 

 

 

 

 

4.3.4 System Monthly Variation of the Incident-based Congestion Index

In this subsection, the values of incident-based congestion for the different
months are compared. The average value of incident-based congestion for the whole
system is 0.230 as shown in Table 4.15. This value translates to 2.5 minutes of daily
incident-based congestion for any given station in the network. The month of January
1997 has the highest index value of 0.379, and this value translates to 4.1 minutes of
daily incident-based congestion at any given station in the network during that month.

The month of August 1997 shows the lowest index value of 0.136 and that is equivalent

101

to 1.5 minutes of daily incident-based congestion for any given station in the network.

Figure 4.11 illustrates the monthly variation of the incident-based congestion index.

4.4 Incident Frequency and Queue Length

This section presents the results of the incident frequency by length obtained by
processing the nine sets of traffic data described in section 4.2. It includes the results of
time-based variation in incident frequency, segment variation in incident frequency,
system-based variation in incident frequency, and incident queue length. Table 4.16
shows the incident frequency and queue length for the different months.

The distribution of incident frequency and queue length by month is illustrated in
Figures 4.12, 4.13, and 4.14. The month of December shows the highest frequencies and
April and August have the lowest. In all months the majority of incidents affected only
one station.

The frequency of incidents with various lengths that occurred on each of the
network segments is illustrated in Figure 4.15. Segments 5 and 6 had high rates of 9
incidents per mile per month, while segments 9 and 11 had low rates of 1 incident per
mile per month and the rest of the segments ranged from 3 to 7 incidents per mile per
month.

The variation in incident frequency per mile that occurred on the network for each
month is illustrated in Figure 4.16. This figure shows the network monthly variation of
incident frequency per mile. In the average month, there were 171 incidents of various
lengths on the network. During the winter and summer, the incident frequency tends to

be higher than in the spring.

102

 

 

.83. 3:000:00 3033:0205 00.3.5033 05 3 33...: 25:05. :2: 2.5.“.

 

qiuow '0AV
z6—bnv

[ES—Inf

 

IB—UW‘

 

zs-iow

5:02

[6—4dv

[6‘10W

[64193

 

 

 

 

I6”U0f‘

96—390

. 23.:
. cmod
. 03.6
. cmvd
. ccué

-- cmud
.- can...

 

omnd

 

 

 

 

03.0

 

 

 

cued

xepu! uogsefiuoo peseq-auepiotu

 

 

103

 

 

 

 

 

 

 

 

 

 

 

 

 

v o o P m S 2 mm lb: 582 .93
m o o N o m t m: 8 8-92
v o o o o N 5 mm 6 3......
o o o a c 9 8 8 m3 8.5...
v o o a o h 3 R x: 5%:
m o o o a v 2 S 8 3.a<
e o o o m w a 8 8 3.5:
v o o N v m cm 8 B 3.8“.
m o o a m m m 8 .3 3-5.
n o o o Lu 8 cu me «2 8.8g
2.5 ._+ ._+ ._+ .7, ._+ ._+ .7, 5.8:
.2. a.» 3 5 3 3 m...
>28.th :6 A 2 8 2 2 2 2 .—
236... MN 3. m... 3 m... a...
anticoagm .355 £98. 2.2:. can 35:68. «:36...

 

 

 

 

 

 

 

 

 

 

 

 

 

wficoE 290:6 m5 5. £062 930 new >833: #528. otv 2an

104

 

0 0'€<

.0 93:2: 05 .8 593. 2.2.3 3 5:259: «co—v.2: 3.3.5323 «5. 952....

kahun— uca $0.95.. 69000

 

0 0'9 - Q'Z

 

$0.25 59.0. 23:0

 

 

 

 

 

 

 

 

 

 

,0»

1 2:.

1 car

i ecu

 

 

 

 

nmunmm B hmémw D 8.02”. mm

 

 

omN

Aouenbeu zuopgaul

 

 

 

105

his: .23 .S.a< .53:

.0 «5:2: 05 3.. 598. 2.2:. 3 5:252. 2320:. 3.3.8826 2.6 0.59....

 

0 O'§<

32.8. 598. 2.25

0 0'9 — 92

u/

90 - 0'0

 

SK

  

 

  

at“ 0'1 - 9'0

  

 

 

 

er L\X\\\\D

 

 

 

 

 

 

 

I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I
I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I
I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I
u .

,ow

row

.00

-om

 

VOL

cow

 

 

 

3-32m. Egan. 3.52m.

 

 

 

our

Aauenbou wepgom

 

 

 

 

106

3.91 new .543... .325“.
*0 «5:05 05 3.— 59.2 25:3 3 3:250...— EIEoE ou_3.83m>m 3.6 952".

 

32.5 59.2 2.25

OZ — G'l

 

 

 

.0
CJ
_
.0

 

“N

\

 

 

 

0
0
_
.0
g
c._mmH
B
2

"mx

 

 

 

 

 

 

V‘TV'Y‘Y'

Y 7
.

 

 

 

\. 'wa “\

 

 

 

 

 

 

5-93m. 3-33 3-55

 

 

 

-cN

- co
- an
- 2:.
- cue
- :3.

cop
car

Aauenbou wopgom

 

 

107

 

 

 

 

22:50» .2. 0:8 .2. 3:33.: 2820:. E 2033:; 25:08 3223 23 2:9“.

.onEac «:oEuom

 

T

OQNCDIDV'MN

 

muow 10d emu 16d Aouonbu; wapgaul

 

 

O
F

 

 

 

 

108

 

 

 

 

 

 

[ES—W

 

[6—Unf‘

L6—K0w

 

 

LG-JDW

 

[641%

r -. t . e .V..-,. V
~ , w . 7.. "fi, ,vy ’ I Aq¢«_4.~a-~-o- "‘
' . o ‘ <4rxA$l~rA¢aVN'V,f¢‘rr'varhrlfl‘gijt.£t—k 'h! H
‘ .. e ' ~ ‘ . .. H ,_

q...~&~.---- '5 v -

 

 

96—39(1

 

 

 

emu 19d Aouenbeu zuepgeul

 

Month

 

 

109

Figure 4.16 System-wide monthly variation in incident frequency per mile

The incident frequency by length of queue for each month was also investigated.
Figure 4.17 shows the average monthly incident frequency and queue length. In the
average month, 115 incidents which affect one station, and one incident which has a
length between 2.0 and 2.5 miles plus a random length described in section 3 of chapter 3

occur over the network.

4.5 Analysis of Traffic Volume

The level of traffic demand has been reported to be one of the factors that
correlate to the duration of incident-based congestion. To investigate this correlation, the
average daily traffic volume was determined for each detector station. The number of
vehicles that passed each detector station at a speed below the threshold speed are then
extracted from these data. This section presents the steps followed to calculate and
analyze the relationship between the total traffic volume and the traffic volume during
congested time periods.

The off-peak trafiic volume was extracted for each station by adding the total
volume before and afier the peak hours. To account for the volume missed due to
malfunctioning detectors, the measured volume was factored using the same technique
described in section 4.2.1. The factored volume was then converted to vehicles per lane
per hour (vplph). The same procedure was applied to the congested volume, where the
volume during the periods of incident-based congestion was extracted, then divided by
the duration of incident-based congestion and the number of lanes to produce a measure
of vplph. This volume was also factored using the same method applied to the total

volume.

110

0'? <

0‘9 - G'Z

 

 

O

598. 28:6 3 3:259: «5205 25:9: ouuco>< 23¢ 2:2".

32:5 598. 28:0

 

 

 

.om

CV

.00

-ow

. cow

our

 

 

of.

auepgou! ;o .iequinN

 

 

111

The relationship between the incident-based congestion index and the average off-
peak volume was investigated for each station, direction of freeway and each freeway as
a whole. These relationships are illustrated in figures 4.18 to 4.20. The same procedure
was performed using volume data by month and summed by each of the twelve
previously defined segments. These relationships are depicted in Figures 4.21 and 4.22.
Finally, Figure 4.23 illustrates the relationship between the incident-based congestion
index and the nine month average daily off-peak volume for each of the stations involved
in this study. If the points with high index values are considered outliers and eliminated
from the graph, the slope of the line would have been close to zero indicating the absence
of a relationship between off-peak volume and the value of the congestion index.

The relationship between the average monthly incident-based congestion index
and the average hourly congestion volume is illustrated in Figure 4.24. This figure also
includes the hourly flow for the average month off-peak volume. The total volume was
low as was expected because only off-peak hours were considered. The volume during
the congested periods was below the capacity for an incident that was located on the
shoulder, with the three lanes open to traffic. For incidents where one lane was blocked,
the flow would reach the capacity of the opened two lanes. It could be inferred from this
graph that the majority of the incidents blocked one lane because there was a measurable
reduction in speed. There might have been a few incidents that blocked more than one
lane, causing the speed to be reduced even further. Neither the total volume nor the
congested volume reflects a strong linear relationship with the incident-based congestion

index as shown in the figure.

112

Station 85

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2.5
x 2 0 y I 0.0024x - 1.2933
§§ ' R’so.155 f
.3 E 1.5 ’e
'3. 8 o
o n 1.0 /
g c
0.0 - . , 9 , .
0 200 400 600 800 1000 1200
Average daily volume (vplph)
6 0 Station 137
5'0 y - 0.0011x + 0.2121
g 5 ' R’aomoe '
'6
0 = 4.0
n ..
.9 c
g .3 3.0
8 8
T: g: 2.0 .
E o
.. 1.0 M
o
0.0 . , - , ° , .
0 200 400 600 800 1000 1200 1400
Average daily volume (vplph)
20 Station 146
x y - -0.0005x + 0.9271 9
B 8 15 -
g 5 ' R’=0.008
.9 e
g g 1.0 c
a a
5 O ' . fl
” o 6o 0
0,0 , , . , , A—
0 200 400 600 800 1000 1200
Average daily volume (vplph)

Figure 4.18 Station incident-based congestion index versus average
monthly off-peak volume

113

WB l-94

 

0.50
y = 0.0009)! - 0.2837
0.40

O
A
R'aO.247 . . '
030 /

0:20 /

0.10
0.00

 

 

 

 

 

 

 

Incident-based congestion
Index

0 100 200 300 400 500 600 700 800
Volume (vplph)

EB l-94

0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00

 

Incident-based congestion
index

0100200300400500600700800900
Volume (vplph)

l-94
0.60

0.50
0.40
0.30
0.20
0.10
0.00

y - 0.0003x + 0.1648
. 0.0522

 

incident-booed congestion
index

0 100 200 300 400 500 600 700 800
Volume (vplph)

Figure 4.19 incident-based congestion index versus volume by
direction and freeway (I-94)

114

NB M-‘lO

 

 

 

 

 

 

 

 

 

C
° 0.14
E 0.12
5 0.10
3 g 0.08
3 5 0.06
§ 0.04
= 0.02
8 e
- 0.00
0
5 0 100 200 300 400 500 600 700
Volume (vplph)
SB M-10
C
O
0.30
3 o 25 y--0.0001x+0.1083
a .
C .
8 x 0.20 :- o 0032
g «8 0.15
C
3 - 0.10
‘g‘ 0.05
,3; 0.00
5 0 100 200 300 400 500 600
Volume (vplph)
c M-10
.9 0.20
i. 0 15 y - 0.0003x - 0.0504 e
§ x ‘ R’=0.2427 l
. 0.10 /
E / . O
.5 0.05 1: °
3
E‘ 0.00 - - , - . .
"' 0 100 200 300 400 500 000 700
Volume (vplph)

Figure 4.20 Incident-based congestion index versus volume by
direction and freeway (NI-10)

115

 

 

 

0.4

 

 

 

 

 

 

 

 

 

 

 

0
g 0.4 O
13 o 3 y=0.0003x+0.0781 °
§ ' R’=0.0459
O
2 0.3 4..
D
C /
8 0.2 I
'3 e e
E 0.1
2
E 0.1
0.0 . . , . . ,
0 100 200 300 400 500 600 700
Volume (vplph)

 

Figure 4.21 Incident-based congestion index vs average monthly off-peak
volume

 

 

,.____.____ ___r_._._.

 

 

 

 

 

 

 

 

 

1.2
g» y = 0.0012x - 0.4561
c 1.0 2 ,
'E R =0.312
O
a 0.0
0
D
C
3 0.6 .
g 0.4 /
§
2 0.2 0
.‘s’ . 3
O
0.0 , . - .
0 200 400 600 000 1000
Volume (vplph)

 

Figure 4.22 Incident-based congestion index vs average segment
off-peak volume

116

 

 

2.5.3 .3sz xuoofio

2.2. 00805 05 use .89.. eozuoucoo 033.2023. 05 .8253 2:05:23. and 2:9...—

 

Eniz 0523 Est equ><
com? 80.. com com oov com o

 

 

 

O md

 

o...

 

 

 

 

 

 

.83 u «m
o 28... - .825... u s

 

 

 

Wm

 

xepug uonsefiuoo peseq-zueplom

 

 

117

 

oE:_o> .83 ounce;
new .0823 US$9.00 ion... cosmoncoo commas—8205 05 52.33 35:23.3. 3.1 959".

 

 

 
 

 

 

 

 

5:0:
. . o
W 1 ,. 8.
11 W 1 8m
1 _. .1 M 8..” m
1 .. . 18v m
111 ,. 11 .. 1 00m N
1 1 .150 000 .m
1&1 1 19? Don M.
11111 11 1 1 .1 com M
1 11. 11 11m1111 1 .6 00m m.
1 1
1 n11.111m111mm1119111 11 11 10 11 1 m118§m
1 111111w118 1 1111111111 11111 111 1 11 com?
11 1 1 111111111 11111 1 1 11111 83
Dom—1

 

 

 

 

 

ink, .65 am: I ._o> .mcooy

118

 

4.6 Summary

This chapter demonstrated the application of the methodology developed and
presented in chapter 3 on the traffic database collected before the ITS deployment
program become fully operational. The chapter described the steps performed to yield
the basic results to conduct the analysis and presented the results.

The first part described the site location and the freeways involved in this study.
It also described how the freeways were divided into segments to isolate the locations and
months with high incident-based congestion and incident frequency, and how the data
were processed and integrated to yield monthly output files. The results of the analysis of
the incident-based congestion index, the incident frequency and the average monthly
incident frequency by queue length were then described. The data were then analyzed to
identify stations, segments and months with high traffic volume. The purpose of this
analysis was to investigate the relationship between traffic volume and the measures of

effectiveness values. Finally, a plan for conducting the after analysis was presented.

119

Chapter 5

Summary and Conclusions

This research study presented the development of measures of effectiveness to
evaluate the ITS deployment program in Detroit based on incident-related congestion
during off-peak hours. The measures of effectiveness were derived to identify and
quantify the frequency, duration, and location of incident-based congestion, and to
estimate delay and the characteristics of queues caused by incidents on various segments
of the freeway network. Three measures were established to meet these requirements; the
first measure quantifies individual station incident-based congestion; the second is a
measure of the aggregate impact of individual incidents across stations. This measure
also quantifies the incident frequency and queue length. Finally, the third measure is an
index that quantifies incident-based congestion at the segment level of aggregation.

The measures of effectiveness were utilized to analyze nine months of traffic data
on the SCANDI system before the ITS technologies were implemented. The morning and
evening rush hours were excluded from this study.

The station average monthly incident-based congestion index ranged from a low
of 0.00 to a high of 3.319 over the 184 stations included in the study. On westbound I-
94, station 78 on segment 1 had the highest incident-based congestion index of 1.354,
which is equivalent to 14.6 minutes of average daily incident-based congestion. On
eastbound I-94, station 134 located on segment 5 experienced the highest incident-based
congestion. The index value for the station was 3.319, which translates to 35.8 minutes

of average daily incident based congestion. This station had the highest index value

120

among all stations on the network. On northbound M-10, station 177 on segment 7
experienced the highest incident-based congestion of 0.347, which is equivalent to 3.7
minutes of average daily incident-based congestion. On southbound M-lO, station 273
on segment 12 experienced the highest incident-based congestion of 0.609, which
translates to 6.6 minutes of average daily incident-based congestion.

The segment average monthly incident-based congestion index ranged fiom 0.971
for segment 5 on eastbound 1-94 to 0.039 for segment 8 on northbound M-10. The
station average daily duration of incident-based congestion on segments 5 and 8 was 10.5
and 0.4 minutes respectively.

The variation in the average monthly incident-based congestion index ranged
from 0.379 for the month of January 1997 and 0.136 for the month of August 1997. The
station average daily duration of incident-based congestion in the months of January and
August 1997 was 4.1 and 1.5 minutes respectively.

The summer of 1997 demonstrated the highest incident-based congestion among
the seasons included in the data, and the spring showed the lowest index value. The
seasonal incident-based congestion index was 0.249, 0.162, and 0.281 for the winter,
spring, and summer of 1997, respectively, and their corresponding station average daily
duration of incident-based congestion was 2.7, 1.8, and 3.0 minutes respectively. On an
annual basis the average station index value was 0.23, which corresponds to 2.5 minutes
of average daily delay per station caused by incidents.

The incident-based congestion on I-94 was higher than M-lO. The index values
for 1-94 and M-10 were 0.340 and 0.095, which translate to 3.7 and 1.0 minutes of

average delay caused by incidents. The directional incident-based congestion index was

121

0.262 and 0.415 for westbound and eastbound I—94, respectively. These index values are
equivalent to 2.8 and 4.5 minutes of average daily incident-based congestion. Southbound
M-10 had about twice the value of incident-based congestion as northbound M-10. The
directional incident-based congestion index was 0.063 and 0.127 for these two directions.
These index values are equivalent to 0.7 and 1.4 minutes of average daily incident-based
congestion, respectively.

During the winter and summer, the incident frequency tended to be higher than in
the spring. In the winter, there were 635 incidents of various lengths on the network, of
which 423 incidents affected one station and 4 had a length between 2.0 and 2.5 miles
plus a random length. In the spring, there were 448 incidents of various lengths on the
network, of which 282 incidents affected one station and 2 had a length between 2.0 and
2.5 miles plus a random length. In the summer, there were 547 incidents of various
lengths on the network, of which 333 incidents affected one station and 4 had a length
between 2.0 and 2.5 miles plus a random length.

The segment average monthly incident frequency per mile ranged from 9
incidents for segment 6 on eastbound I-94 to 1 incident for segment 9 on northbound M-
10. The system-wide average monthly incident frequency was 4 incidents per mile.

The average monthly incident frequency ranged from 7 incidents per mile for the
month of December 1997 to 3 incidents per mile for the month of August 1997. In the
average month, there were 181 incidents that occurred system-wide, 115 affected only a
single station and one incident extended over 2.5 miles.

The relationship between the off-peak traffic volume and the incident-based

congestion index was investigated at the station, segment, direction of freeway, freeway,

122

and the system as a whole level. In all cases, the traffic volume analysis showed that there
is no strong linear relationship between the off-peak traffic volume and the incident-
based congestion index.

The number of sampling days required to perform the after analysis on the
SCANDI system and to perform the before and after analysis on the newly instrumented
freeways were estimated seasonally and annually. It is recommended that 25 days of data
be collected in the after period if the after analysis is performed seasonally and 40 days if
the evaluation is conducted over more than one season (see Appendix A).

The values of the various MOE for incident-based congestion will be determined
for the freeway system after the ATMS/ATIS system is fully operational, and the
differences will provide one measure of the effectiveness of the system. Other measures

will include changes in recurring congestion and traffic crashes.

123

Appendix A

Appendix A

Plan for Conducting the After Analysis

This research project was conducted to quantify the benefit of the ITS deployment
program in the Detroit metropolitan area by assessing changes in non-recurring
congestion during off-peak hours before and after the ITS deployment. The results of
this study utilized ninety days of traffic data representing the winter, spring and summer
of 1997. The number of sampling days required to perform the after analysis on the
SCAN DI system and the number of days required to perform the before and after analysis
on the newly instrumented freeways were estimated seasonally and annually. The
estimation methods are described in the following paragraphs.

To determine the sample size required to obtain a reliable estimate of the
congestion index, the following analysis was conducted: a) the daily incident-based
congestion index was calculated for each of the ninety days and ordered randomly; b) the
index was calculated for the first day selected from the random list; c) the index was then
calculated for the combination of the first and second days from the list, and this
procedure was continued through the ninety days. Figure A.1 shows the cumulative
index versus the number of days included in the sample. This figure indicates that the
index stabilized after about 40 days. This suggests that if days were sampled randomly
throughout this nine months period, 40 days would be required to estimate the system-
wide congestion index.

For the seasonal case, the ninety days were divided into three sets of thirty days to

represent the winter, spring, and summer. The same procedure used to estimate the

124

23333.32
mmommmowmnonmmoommommvovmmonmmowmvor m C

can 553533353 _..< 952“.

 

 

 

 

 

 

 

 

 

 

 

F . l7 F r _ P L _ _ if h h r p! 8.0 o
n
w
n
1-1111111111111111 1111111111111 186 w
9
A
9
1 1 1 11 111 11 11 1 11111 -11 11111 .26 w
1 m

u
1 1 1- 11111 111111111111111111111111111111.11 m; w w.
x m
a
11 omd P
O
O
. u
a 11 . -111 mg .m
nlav...
one u

{an 8

 

 

125

required sample was applied to each of the three sets. Figure A.2 illustrates the
cumulative incident-based congestion index versus the number of sampling days for the
winter, spring and summer. This figure shows that the index converged afier twenty five
days for the winter, spring, and summer.

Based on this analysis, it is recommended that 25 days of data should be collected
in the after period if the after analysis is performed seasonally and 40 days if the
evaluation is conducted over more than one season. Afier data is collected for the before
and after periods, the before and after analysis will be performed by comparing the means
of the incident-based congestion index before and after the ITS deployment. Since the
sample sizes are 25 or 40, (111,, m) both distributions can be assumed normally distributed.
The test will be conducted by calculating the before and after means (1.0,, u.) and standard
deviations (01,, 0,) for the incident-based congestion index. With these data, the 95%
confidence interval for the difference in population mean indices between the before and
after samples will be determined. Given a 5% level of significance, a test will be
constructed to determine if there is a significant difference between the population mean

indices. Since 1 - at = 0.95, oc/2 = 0.025 and 20,025 = 1.96. Thus a 95% confidence

interval for uh - pa is:

2 2
a 0'
#b—flaiZ0.025 4+4 .............................. A.1
"b "a

If the confidence interval includes zero, then the null hypothesis H0: uh = pa is
accepted at the level at = 0.05 and the alternative null hypothesis H1: uh at u, is rejected

which means that the deployment had no affect on the incident-based congestion index.

126

mm

get no 53:52

mm cm 9 or

p L L p

53 959.3» Econ—wow N.< 2:2".

 

,1 A

 

 

 

11 1 1111111|-l11 111l11 111.1111 11111111111111|11 no.0
1111! -111 2.0

. cod

 

 

2.0
.1 owd

 

 

 

’4 11 mNd

 

 

 

1 and

 

 

11111 1|111|111111111|11 94.0

.s.‘ 1 and

 

xepul
uogseBuoe peseq-iuepgouI engemuing

 

 

 

.mEEam LT mctam 1.1 .9531;

 

 

mvd

 

11L

 

127

The steps remaining to complete the comparison of the level of congestion before

and after the ATMS deployment include:

Collect 25 days of data on the SCANDI system after the ITS becomes fully
operational;

Using the program developed in this study, process the afier data to determine the
frequency, duration, and location of incident-based congestion;

Compare the before and after incident-based congestion index for the SCANDI
system in terms of the changes in the incident-based congestion index, incident
frequency per mile and average queue length;

Collect a set of 25 days of both before and after data on the newly instrumented
freeway system;

Analyze the data in the same manner carried out on the SCANDI system;

Compare the before and after incident-based congestion index for the newly
instrumented system in terms of the changes in the incident-based congestion index
and incident frequency per mile.

The before and after analysis may be performed in terms of observing the changes

in the incident—based congestion index, incident frequency per mile for each station,

segment, and freeway, and the system as a whole. Table A1 illustrates a suggested

summary of the before and after comparisons.

128

Table A] Summary of the before and after comparisons

 

Location Average index lb Average index 1,
(before) (afier)

 

Station 1
2

174

 

Segment 1
2
3

12

 

__Freeway
I-94
M- l 0

 

System

 

 

 

 

 

 

129

References

References

. http://www.itsa.org/wh_atits.html

. Lindley, A. Quantification of Urban Freeway Congestion and Analysis of Remedial
measures. Report FHWA/RD-87/052. FHWA, US. Department of Transportation,
1 986.

. Lindley A. Jeffrey. Urban Freeway Congestion: Quantification of the Problem and
Effectiveness of Potential Solutions. ITE Journal, Vol. 57, January 1987.

. Lindley, Jeffrey A. Urban Freeway Congestion Problems and Solutions: An Update.
ITE Journal, Vol. 59, December 1989.

. Turner S. M. Examination of Indicators of Congestion Level. Transportation
Research Record, 1326, 1992.

. ISTEA Implementation. ITE Journal, June 1992.

. Forbes, Gerald. Identifying Freeway Incidents Using One Detector Station. Masters
Thesis, McMaster University, August 1993.

. Special Report 209. Highway Capacity Manual. TRB, National Research Council,
Washington, DC, 1985.

. United States General Accounting Office. Traffic Congestion Trends, Measures, and
Effects, Report to the Chairman, Subcommittee on Transportation and Related
Agencies, Committee on Appropriations, United States Senates, November 1989,
GAO/PEMD-90-l .

10. The Status of the Nation’s Highways: Conditions and Performance. Report of the

Secretary of Transportation to the US. Congress. U. S. Government Printing Office,
June 1985.

11. Lomax, J. and Christiansen, D. L. Estimates of relative Mobility in Major Texas

Cities. Research Report 323- 1 F. Texas Transportation Institute, Texas State
Department of Highways and Public Transportation, 1982.

12. Hamm, R. A., Woods, D. L., Stover V. G., and Hawley, P. E. Congestion

Management Systems State-Of-The-Practice Review. FHWA/TX-94/563-1F.
August 1993.

13. Thurgood, Glen S. Development of a Freeway Congestion Index Using an

Instrumented Vehicle. Transportation Research Record 1494.

130

14. Gall, Anal. and Hall, Fred L. Distinguishing Between Incident Congestion and
Recurrent Congestion: A Proposed Logic. Transportation Research Record 1232.

15. Cottrell, W.D. Measurement of the Extent and Duration of Freeway Congestion in
Urbanized Area. In Compendium of Technical Papers, 61 st. Annual Meeting,
Institute of Transportation Engineers, Washington, D. C., 1991, pp. 427-432.

16. Traffic Management and Planning for Freeway Emergencies and Special Events.
Transportation Research Circular, Number 298, January 1986.

17. Palen, Joe. The need for Surveillance in Intelligent Transportation Systems.
Intellimotion, Vol. 6, No. 1, 1997.

18. Smith, Steven A. and Perez, Cesar. Evaluation of INFORM: Lessons Learned and
Application to Other Systems. Transportation Research Record 1360, 1992.

19. Dudek, C. L., Weaver, G. D., Hatcher, D. R., and Richards, S. H. Field Evaluation of
Messages for Real-Tirne Diversion of Freeway Traffic for Special Events.
Transportation Research Record 682, 1978.

20. Dabney, J. C., and Dudek, C. L. Driver Response to a Highway Advisory Radio
System in New Braunfels, Texas. Transportation Research Record 808, 1981.

21. Huchingson, D., McNees, R. W., and Dudek, C. L. Survey of Motorist Route
Selection Criteria. Transportation Research Record 643, 1977.

22. Sharazi, S. Anderson, and J. Stesney. Commuter’s Attitudes Toward Traffic
Information Systems and Route Diversion. Presented at the 67th Annual Meeting of
the Transportation Research Board, Washington D. C., 1988.

23. Haselkorrr, J. Spyridakis, L. Conquest and W. Barfield. Surveying Commuter
Behavior as a Basis for Designing Motorist Information System. First Vehicle
Navigation and Information Systems Conference (VNIS, 89). Toronto, Canada, 1989.

24. Dudek, L and Huchingson, R.D. Manual on Real Time Motorist Information
Displays. Report No. FHWA-IP-86-l6. Federal Highway Administration, US.
Department of Transportation, Washington D. C., 1986.

25. Judycki, D. C., and Robinson, J. 1991. Freeway Incident Management. Technical
Papers fi'om ITE’s 1990, 1989, and 1988 Conference, contained in Massachusetts
Incident Management Conference Proceeding, DOT-T-92-05, June6. 359-368.

26. Giuliano, Geneview. Incident Characteristics, Frequency, and Duration on a High
Volume Urban Freeway. Transportation Research, Part A, Vol. 23 A, No. 5, 1989,
pp. 387-396.

131

27.

28.

29.

30.

31.

32.

33.

34.

35.

36.

37.

38.

Urbanek, G. L., and Rogers. Alternative Surveillance Concepts and Methods for
Freeway Incident Management. Volume 1, Executive Summary. Report No. FHWA
RD-77-58, March 1978.

DeRose Jr. F. Analysis of Random Freeway Traffic Accidents and Vehicle
Disabilities. Highway Research Record, 59, 1964.

Juge J. and Thompson R. Service Patrol Pilot Study, Report No. 4 of Service for the
Standard Motorist Project. Los Angeles CA: California Department of
Transportation and California Highway Patrol 1978.

Federal Highway Administration. Traffic Control Systems Handbook. US.
Department of Transportation, Washington, DC, 1976.

Arceneaux, J. L. and Mikhalkin, B. Inventory of Freeway Geometric Bottleneck
Congestion, Final Report, Freeway Operations Department, District 7, California
Division of Highways, 1973.

Tignor, C. Traffic Management and Planning for Freeway Emergencies and Special
Events. Transportation Research Circular, No. 298, January 1986.

West, T. Proposed real-time surveillance and control system for Los Angeles. Paper
presented to the Highway Research Board Committee on Freeway Operations, Los
Angeles, California, 1969.

Schrank, D.L., Turner, S.M., and Lomax, T.J. Estimates of Urban Roadway
Congestion 1990. Report FHWA AfTX-90/1131-5. Texas Transportation Institute,
College Station, TX, March 1993.

Robinson, J. What’s New in Freeway incident Detection and Response. Technical
Papers fiom ITE Specialty Conferences 1990, 1989, and 1988. Institute of
Transportation Engineers, Washington , DC, pp244-257.

Goolsby, Merrell E., Smith Wilbur and Associates. Influence of Incidents on
Freeway Quality of Service. Transportation Research Record, Jan 1971.

Lari A. Christianson D., and Porter S. I-35 Incident Management and Impact of
Incidents on Freeway operations, Minnesota Department of Transportation, Office of
Traffic Engineering, Minneapolis, MN, 1982.

MacCalden M. S., Anderson R. W., and Morshed M. M. Feasibility Study of Closed-
Circuit Television for Traffic Surveillance, Report # DOT-HS-845016, National
Highway Traffic Safety Administration, Bus and Transportation Agency, Department
of Public Works, Division of Toll Crossing, Sacramento, 1973 (sited at 12).

132

39. Ullman, Gerald L. and Ogden, Michael A. Analysis of Major Freeway incidents in
Houston, Texas. Transportation Research Record, 1554. Sep 1996.

40. Reiss, R. A., and Dunn, W. M. Freeway Incident Management Handbook. Report
No. F HWA-SA-91-056. FHWA, US. Department of Transportation, Washington,
DC, July 1991.

41. Skabardonis, A., Petty, K., Noeimi, H., Rydzewski, D., and Varaiya, P.P. I-880 Field
Experiment: Data-Base Development and Incident Delay Estimation Procedures.
Transportation Research Record 1554. TRB, National Research Council, Washington
DC, 1996.

42. Ogden, M. A., Mounce, J. M., Middleton, D. R., and Ullman, B. L. Traffic
Management in Response to Major Freeway Incidents, Volume I. Report No.
FHWA/TX-94/1345-2F. Texas Transportation Institute, College Station, TX, August
1994.

43. Whitson, R. H., Buhr, J. H., Drew, D. R., and McCasland, W. R. Real-Time
Evaluation of Freeway Quality of Traffic Service. thway Research Record 289,
1969.

44. Hughes Aircraft and J HK & Associates. Direct Project, Incident Detection Issues.
Report for the Michigan Department of Transportation. MDOT-92-0413, Washingot,
D.C., August, 1992.

45. McShane, W. R., and R085, R. P. Traffic Engineering. Prentice Hall, Inc., Englewwd
Cliffs, N. J. 1990.

46. Transportation Research Board (TRB). Special Report 209. Highway Capacity
Manual. Nattional Research Council, Washington, D. C., 1994.

47. Morales, Juan M. Analytical Procedures for Estimating Freeway Traffic Congestion.
ITE Journal, Januray 1987, Vol. 50, No. 2.

48. Messer, C. J ., C. L. Dudek, and Friebele, J. D. Method for predicting Travel Time
and Other Operational Measures in Real Time During Freeway Incident Conditions
Highway Research Record, 461 , 1973.

49. Wirasinghe, S. Chandana. Determination of Traffic Delays from Shock-Wave
Analysis. Transportation Research, Vol. 12, 1978.

50. Chow, We-Min and Watson, Thomas J. A Study of Traffic Performance Models
Under an Incident Condition. Transportation Research Record, 567, 1976.

133

51. Wicks, D., and Lieberman, E. Development and Testing of INTRAS, a Microscopic
Freeway Simulation Model: Program Desigh, Parameter Calibration and Freeway
Dynamics Component Development. Report No. FHWA/RD-80/106, F HWA, US.
Department of Transportation, 1980.

52. Federal Highway Administration. FRESIM User’s Guide. Version 4.5, Washington
DC. 1994.

53. Alternative Surveillance Concepts and Methods for Freeway Incident Management.
F HWA-RD-77-58/63. Federal Highway Administration, Washington, DC, March
1 878.

54. Garib, A., Radwan, A. E., and AI-Deek, H. Estimating Magnitude and Duration of
Incident Delays. Journal of Transportation Engineering November/December 1997.

55. Greenshields, B. A Study of Traffic Capacity. HRB Proceedings, Vol. 14, 1934.

56. Cheu, R., Recker, W. and Ritchie, S. Calibration of INTRAS for Simulation of 30-
Sec Loop Detector Output. Transportation Research Board 72"d annual Meeting,
Washington DC, 1993.

57. Masters, Philip, Paesani, Gordon, and Coady, Ana. Congestion Management by
Dynamic Guidance on Toronto’s Dual Roadway Freeway.

58. Kyungsoo, Yoo, Sanghoon, Bae, Sigon, Kim and Keechoo, Choi. Analysis of the
Effectiveness of FTMS: A Korean Case. Proceedings of the 1997 Conference on
Traffic Congestion and Traffic Safety in the 21St Century, 1997.

59. Intelligent Transportation Systems Projects Book. US. Department of
Transportation, January 1998.

60. Carvel, J. D., Balke, K., Ulrnan J., Fitzpatrick, Nowlin, L., and Brehmer. Freeway
Management Handbook. Federal Highway Administration, Report No. F HWA-SA-
97-064, August 1997.

61. McLean, C.H. Cellular Phones—A Key Traffic Management Component.

Compenduim of Technical Papers, Institute of Transportation Engineers, Washington,
DC. 1991.

62. Wilshire, R., Black, R., Grochoske, R. and Higginbotham, J. Traffic control Systems
Handbook. Report No. FHWA-IP-85-11. FHWA, US. Department of
Transportation, Washington, D. G, April 1985.

63. Balke, K. N. An Evaluation of Existing Incident Detection Algorithms. Report No.

FHWA/TX-93/1232-20. Texas Transportation Institute, College Station, TX,
November 1993.

134

64. Wohlschlaeger, S. D. and K. N. Balke. Incident Response and Clearance in the State
of Texas: Case Studies of Four Motorist Assistance Patrols. Report No. FHWA/TX-
92/1232-15. Texas Transportation Institure, College Station, TX, November 1992.

65. Ritchie, S.G., and Prosser, N.A. Real-Time Expert System Approach to Freeway
Incident Management. Transportation Research Record 1320, 1991.

66. Dudek, C. L., and Ullman, G L. Freeway Corridor Management. NCHRP Synthesis
of Highway Practice No. 177. TRB, National Research Council, Washington, DC,
March 1992.

67. Forbes, Gerald. Identifying Incident Congestion. ITE Journal. June 1992.

68. Hurdle, V. F., and Datta, T. K. Speeds and Flows on an Urban Freeway: Some
Measurements and a Hypothesis. Transportation Research Record No. 905, 1983.

69. Persuad, B. N., and Hall, F. L. The McMaster Single Station Algorithm for Detection

of Freeway incidents. Traffic Research Group, Department of Civil Engineering and
Engineering Mechanics. McMaster University, Hamilton, Ontario, Canada, 1988.

135