Professional Documents
Culture Documents
165 PDF
165 PDF
A Monograph
1~1
SPECIAL REPORT 165
Transportation Research Board
National Research Council
Washington, D.C., 1975
..
Transportation Research Board Special Report 165 The project that is the subject of this report was
Price $20.00 approved by the Governing Board of the National Re-
Edited for TRB by H. P. Orland search Council, whose members are drawn from the
councils of the National Academy of Sciences, the Na-
Subject area tional Academy of Engineering, and the Institute of
54 traffic flow Medicine. The members of the committee responsible
for the report were chosen for their special competence
and with regard for appropriate balance.
This report has been reviewed by a group other than
the authors according to procedures approved by a
Transportation Research Board publications are Report Review Committee consisting of members of
available by ordering directly from the board. They the National Academy of Sciences, the National Acad-
may also be obtained on a regular basis through emy of Engineering, and the Institute of Medicine.
organizational or individual supporting membership in The views expressed in this report are those of the
the board; members or library subscribers are eligible authors and do not necessarily reflect the view of the
for substantial discounts. For further information, committee, the Transportation Research Board, the
write to the Transportation Research Board, National National Academy of Sciences, or the sponsors of the
Academy of Sciences, 2101 Constitution Avenue, N.W., project.
Washingt0n, D.C. 20418.
*On March 9, 1974, the Highway Research Board became the Transportation Research
Board to reflect the actual scope of its activities.
iii
CONTENTS
Chapter I-INTRODUCTION
1.1 Background
1.2 History
1.3 Background of Current Monograph 2
1.4 Context of Monograph 3
1.5 References 6
v
vi CONTENTS
197
ix
•
x
2 INTRODUCTION
Symposium Proceedings
Symposium on the Theory of Traffic Flow, Theory of Traffic Flow (ed. by R. Herman),
Detroit, Michigan, Dec. 7-8, 1959 Elsevier Publishing Co., 1961
Second International Symposium on the Proceedings of the Second International Sym-
Theory of Road "Traffic En- posh!m on the Theory of Traffic Flow (ed. by
~and,June25-27, 1963 J. Almond), OECD, Paris, 1965. Available
from OECD Publications Office, 17 50 Penn-
sylvania Avenue, N.W., Washington, D.C.
20006
Third International Symposium on the Theory Vehicular Traffic Science (ed. by L. C. Edie,
of Road Traffic Flow, New York, New York, R. Herman, R. Rothery), American Elsevier
June 1965 Publishing Co., 1967
Fourth International Symposium on the Beitrage zur Theorie des Verkehrsf/usses (ed.
Theory of Road Traffic Flow, Karlsruhe, Ger- by W. Leutzbach and P. Baron), Strassenbau
many, June 18-20, 1968 und Strassenverkehrstechnik, n. 86, Bundes-
anstalt fiir Strassenwesen, 5 Kain Raderthal,
Bruhler Strasse 324, West Germany
Fifth International Symposium on the Theory Traffic and Transportation (ed. by G. F.
of Traffic Flow and Transportation, Berkeley, Newell), American Elsevier Publishing Co.,
California, June 16-18, 1971 1972
Sixth International Symposium on Transporta- Transportation and Traffic Theory (ed. by
tion and Traffic Theory, Sydney, Australia, D. J. Buckley), American Elsevier Publishing
August26-28, 1974 Co., 1974
Symposium Proceedings
Symposium on the Theory of Traffic Flow, Theory of Traffic Flow (ed. by R. Herman),
Detroit, Michigan, Dec. 7-8, 1959 Elsevier Publishing Co., 1961
Second International Symposium on the Proceedings of the Second International Sym-
Theory of Road Traffic Flow, London, En- posium on the Theory of Traffic Flow (ed. by
gland, June 25-27, 1963 J. Almond), OECD, Paris, 1965. Available
from OECD Publications Office, 17 50 Penn-
sylvania Avenue, N.W., Washington, D.C.
20006
Third International Symposium on the Theory Vehicular Traffic Science (ed. by L. C. Edie,
of Road Traffic Flow, New York, New York, R. Herman, R. Rothery), American Elsevier
June 1965 Publishing Co., 1967
Fourth International Symposium on the Beitrage zur Theorie des Verkehrsflusses (ed.
Theory of Road Traffic Flow, Karlsruhe, Ger- by W. Leutzbach and P. Baron), Strassenbau
many, June 18-20, 1968 und Strassenverkehrstechnik, n. 86, Bundes-
anstalt fi.ir Strassenwesen, 5 Koln Raderthal,
Bruhler Strasse 324, West Germany
Fifth International Symposium on the Theory Traffic and Transportation (ed. by G. F.
of Traffic Flow and Transportation, Berkeley, Newell), American Elsevier Publishing Co.,
California, June 16-18, 19n 1972
Sixth International Symposium on Transporta- Trans port at ion and Traffic Theory (ed. by
tion and Traffic Theory, Sydney, Australia, D. J. Buckley), American Elsevier Publishing
August 26-28, 1974 Co., 1974
YS;'.ar First
Journal Publisher Published
Traffic Engineering and Control Printerhall Ltd., 29 Newman Street, 1958
London W1P3PE, England
Transportation Science Transportation Science Section, 1967
Operations Research Society of America,
428 East Preston Street, Baltimore,
Maryland 21202
Transportation Research Pergamon Press Inc., Maxwell House, 1967
Fairview Park, Elmsford, New York 10523
Transportation Planning and Gordon and Breach Science Publishers, 1972
Technology Inc., 440 Park A venue South, New York,
New York 10016
Transportation Elsevier Publishing Co., Journal Division, 1972
P.O. Box 211, Amsterdam, The Netherlands
dated and rewritten. The Highway Research TABLE 1.4 Principal Review Studies of
Board Department of Traffic and Operations Traffic Flow Theory Sponsored by
authorized the Committee to seek funding and Federal Highway Administration
an author to carry out the rewriting. As a result
of the Committee's action the Federal Highway Analytical Models of Unidirectional Multi-Lane
Administration (FHWA) agreed to provide Traffic Flow: A Survey of the Literature, Sys-
funding, and in June 1970 FHW A entered into a tem Development Corp. Technical Memo-
contract with the University of Minnesota randum TM(L)-379/001/02, 31January1969.
under which D. L. Gerlough and M. J. Huber NTIS No. PB 183075
would write a monograph on traffic flow
theory. It was agreed that partial funding Traffic Systems Reviews and Abstracts, issued
would be provided by the University of Min- monthly by Federal Highway Administration,
nesota. from September 1967 to 1971. Available from
NTIS
:,.
..
4 INTRODUCTION
biguous definitions of these characteristics in ful in describing the occurrence of the (time)
relationship to the methods of measurement, as intervals between events. Distributions are also
well as the appropriate methods of computing used, however, in describing such phenomena
averages, etc. Historically, the definitions of as speeds and gap acceptance. The uses of dis-
traffic characteristics were related to the meth- tributions in traffic are treated in Chapter 3.
ods of measurement. Unfortunately, methods
of averaging often used were influenced by a
lack of clear understanding of the processes Chapter 4: Traffic Stream Models
involved.
It requires little more than casual observa-
The earliest students of traffic behavior
tion to detect that as traffic flow or concentra-
had as the only instruments available to them
tion increases, there is a decrease in speed. In
stopwatches and manual counters. Thus flows
fact it is often possible to estimate one un-
past a point and headways were the important
known flow characteristic from another that is
measurements. These were supplemented by
known or easily measured. For example, the
measurements of speed by means of timing
ratio of flow to capacity may be estimated using
each car's transit across a "trap." As other
observed values of speed or concentration. In
methods of measurement have been developed,
Chapter 4 models that relate pairs of the basic
it has become evident that the numerical results
traffic flow characteristics (e.g., speed/flow,
depend on the method of measurement.
speed/ concentration, and flow/ concentration)
In Chapter 2 various methods of measure- are examined, as are models that consider travel
ment currently available to the traffic engineer time as one of the variables.
are considered, and the definitions of character-
istics are related to these methods of measure-
ment; methods of averaging are also considered.
Chapter 5: Driver Information
Processing Characteristics
Chapter 3: Statistical Distributions of In attempting to devise models to repre-
Traffic Characteristics sent traffic behavior we are, of course, indi-
rectly dealing with human behavior. Several
In designing new traffic facilities or new
traffic flow models to be discussed in later
control plans, it is necessary to predict the per- chapters contain parameters that are used to
formance of traffic with respect to some par- account for various characteristics of the driver
ticular characteristic (e.g., the frequency of in the driver-vehicle system. Some models
headways of a particular size, the number of deal with traffic as a deterministic phenomenon.
cars likely to arrive in an interval, and speeds even though the driver portion of the system, at
exceeding a certain value). It is often desirable least, is highly stochastic.
to be able to make a prediction with a mini- The whole field of human factors in traffic
mum amount of data available or assumed. could, of course, be the subject of a very
For instance, it may be necessary in designing lengthy treatise. Nevertheless, it is hoped that
a pedestrian control system to predict the fre- Chapter 5 will provide some insight into the
quency of headways of greater than 10 sec; in way drivers use the information they receive
designing a left-turn pocket it may be neces- and that this knowledge can then be of use in
sary to predict how many times per hour the various traffic flow models. Although a driver
number of cars arriving during one signal cycle is continuously making decisions, his control
will exceed four. Statistical distribution models actions are limited to control of heading (steer-
enable the traffic engineer to make these predic- ing) and control of acceleration.
tions with a minimal amount of information.
Statistical distributions are useful in de-
scribing a wide variety of phenomena where Chapter 6: Car Following and
there is a high clement of randomness. In Acceleration Noise
traffic the most important distributions are From the drivers' action interpretations of
counting distributions-those useful in describ- information received, given that actions are
ing the occurrence of things that can be limited to control of acceleration (braking and
counted-and interval distributions-those use- acceleration) and heading (steering), we now
CONTENT OF MONOGRAPH 5
consider the dynamics of a stream of traffic. unification of the continuum and car-following
These dynamics result when a series of drivers theories. Finally, a Boltzmann-like theory of
attempt to regulate their acceleration to ac- traffic is discussed in brief.
complish a smooth, safe trip.
In car-following analysis traffic is recog-
nized as being made up of discrete particles, Chapter 8: Queueing Models
whose interaction is examined. The techniques
A desirable goal for transportation engi-
of analysis are those of automatic control sys-
neers is to design and operate facilities that
tems. A principal effort of such studies has
minimize delay to the users. Delay resulting
been to try to understand the behavior of a
from congestion is a common phenomenon:
single-lane traffic stream by examining the man-
Vehicles wait in line for an opportunity to enter
ner in which individual vehicles follow one
a freeway with controlled access ramps; pedes-
another. Studies of this nature have been used
trians queue up on a crosswalk in anticipation
to improve flow through tunnels, explain the
of a gap in road traffic or at a turnstile in a
behavior of traffic at bottlenecks, and examine
transit station; left-turn' slots must be suffi-
control and communication techniques that
cientiy long to store the niaximum number of
will minimize the occurrence of rear-end chain
vehicles that can be expected to wait for a left-
collisions in dense traffic.
turn signal.
Acceleration noise is developed as a mea-
Interest may rest in the length of time a
sure of the quality of traffic flow.
user must wait, or the number of units waiting
in line, or the proportion of time that a facility
might be inactive (an empty parking stall, for
Chapter 7: Hydrodynamic and instance). Queueing models, which employ
Kinematic Models of Traffic methods of probability and statistics, provide a
Because traffic involves flows, concentra- means for predicting some of these delay char-
tions, speeds, etc., there is a natural tendency acteristics.
to attempt to describe traffic in terms of fluid The purpose of this chapter is to present
behavior. Although traffic was examined as the some of the results of studies of probability
interaction between particles in Chapter 6, in models of traffic delay. After an introduction to
Chapter 7 we apply to traffic those models that some elements of queueing or waiting-line
have been developed for fluids (i.e., continuum theory, examples concerned with delay prob-
models); by this we are implicitly saying that lems that occur when all users pass through a
we are more concerned with the over-all statis- single-channel control point, such as a left-turn
tical behavior of the traffic stream than with the slot or a single exit lane for a garage, are given.
interactions between particles. Because the Next, the analysis is extended to consider sev-
sample size for traffic includes only a few par- eral channels of service; for example, several
ticles relative to a true fluid, fluid models of parallel toll booths or the different stalls of a
traffic are applicable to the behavior of a parking facility. Also considered is the case of
stream rather than individual cars. a user who doesn't get served; for example, the
Despite some writers' postulated analogy person seeking a parking space who continues
between traffic and a real fluid, it is preferable to another destination when none is found.
to begin with fundamental observations and An analysis of delays at intersections is
postulates concerning traffic and, then, to iden- considered next, beginning with an analysis of
tify analogies with fluids as these analogies ap- unsignalized intersections. Queueing models
pear. Thus, we start by developing the con- for more complex intersection control, such as
tinuity equation for traffic, illustrating its anal- pedestrian control or traffic-signal control, are
ogy to the continuity equation for fluids. Next also considered in this section. Finally, queue-
to be developed is the concept of waves in ing theory is applied to delay on two-lane roads.
traffic, and examples of this application to prac- Except for the detailed development of the
tical problems are given. Such studies have formulas given in section 8.2, this chapter
been used to improve flow through tunnels and avoids detailed mathematical development, but
to explain behavior of traffic at bottlenecks. does present the theorists' assumptions and
Thereafter the model of choice points toward a some results of interest.
6 INTRODUCTION
7
• .
where N =the number of headways measured The average speed representing this travel time
and h=the mean headway. For some studies, would be
it may be useful to think of an instantaneous
flow associated with each vehicle. Thus, D D 1
Us=(- 1 N D 1 N 1 (2.6)
(2.3)
NLU. i:::l t N~li;
When using Eq. 2.3, care must be exercised in
computing an average flow. Eq. 2.2 shows that
For thethree cars this would be u=
1
1 1 1/3(1/20+1/ 40+1/ 60) = 32 ·7 mph
q=~=
h _1 ""<\:'h. (2.4) Thus, whereas the spot speed (i.e., the time mean
NL.., ' speed) is the arithmetic mean of the speeds ob-
served at a point, the speed that represents
Thus the average flow, when computed from average travel time is the harmonic mean of
individual flows associated with each vehicle, is speeds observed at a point. Furthermore, it
the harmonic mean of the individual flows. will be noted that the harmonic mean speed is
lower than the time mean speed. For reasons
2.2.2 Speed that are discussed in section 2.3.2, the harmonic
mean of speeds observed at a point is known as
The average speed is an important mea- the space mean speed.
sure of the traffic performance at a particular
point or along a particular route; in addition, it 2.2.2.3 Relationships of Mean Speeds.
is one of the fundamental characteristics of The relationship for computing time mean
traffic flow. There are two principal average speed from space mean speed was first recog-
speeds, the time mean speed and the space nized by Wardrop. 1 His relationship, derived
mean speed. 1 Unfortunately, in early traffic in Appendix A-1, is
literature there was confusion as to the use of (2.7)
these two averages. (As an example of the con-
fusion, the reader is invited to compare the dis- where er/ is the variance about the space mean
cussions in various editions of the Traffic Engi- speed.
neering Handbook). 8 - 10 In traffic engineering practice it is often
desirable to convert spot (time mean) speeds
2.2.2.1 Time Mean Speed (Spot Speed). to space mean speeds. For this purpose the
In the past, it has been common practice among following approximate relationship, also dis-
traffic eQgineers to report the "spot speed" for cussed in Appendix A-1, has been developed:*
a given location. This is computed as the arith-
metic mean of the observed speeds: (2.8)
where er/ is the variance about the time mean
speed.
For the speeds crossing line AA' in Figure
2.1 the arithmetic mean speed is 21.71 mph,
In theoretical discussions of traffic flow this while the harmonic mean speed is 1'6.18 mph.
value is referred to as the "time mean speed." 1 When the harmonic mean speed is estimated
from the arithmetic mean speed and the vari-
2.2.2.2 Space Mean Speed (Harmonic ance about it, the result is 15.58 mph. Tables
Mean Speed). Consider the situation where 2.1 and 2.2 illustrate the estimation of the har-
three cars, one at 20 mph, one at 40 mph, and monic mean (i.e., space mean) speed from
one at 60 mph, are traversing a length D. At
a point along D the spot speed would be re-
ported as (20 + 40 + 60) I 3 or 40 mph. Sup- * Haight and Mosher" have presented a
method for converting 1/, and <r.' to a, and <r.2 by
pose, however, that we are interested in average means of tables that assume that speeds follow the
speed as calculated from average travel time. Pearson type III distribution (see Appendix B-2).
MEASUREMENTS AT A POINT 9
1400
1200
1000
~
'+-
.s::
....Cl 800
c
Cl> Q
....J
A A'
600
400
200
2 4 6 8 10 12
B' C'
~ Time (sec)
Figure 2. 1 Vehicle trajectories: line AA' represents a fixed point in space; line BB' represents a fixed point in
time.
point measurements and the estimation of the other mean.) Because space mean speed is
arithmetic mean (i.e., time mean) speed from applied to the models of Chapter 4, the term
the harmonic mean. (Of course, the variance "speed" with no adjective implies space mean
about each mean is necessary in estimating the speed in all later sections of this monograph.
IO MEASUREMENT OF FLOW, SPEED, CONCENTRATION
2.2.3 Concentration from Point which may be derived as follows: Assume the
Measurements total stream is made up of substreams, with
each substream having its own (constant)
Although concentration (the number of
speed. Segregate flow into subflows according
vehicles per unit length) implies measurement to speed 1
along a distance, traffic engineers have tradi-
tionally estimated concentration from point
measurements, using the relationship
(2.9)
where k/ k is the fraction of the total density
in the stream having speed u;, and kt u;=q;
(from analysis of units). Therefore,
TABLE 2.1 Speed Measurements and
Averages a a=I.qi_q
"' -k--T
Eq. 2.9 is of fundamental importance and
Miles per Hour Frequency
is discussed further in Chapter 4.
57 2
58 7
59 1 2.2.4 Lane Occupancy
60 14
61 5 Although concentration or density is con-
62 25 sidered the fundamental characteristic to be
63 24 measured in freeway surveillance, density can-
64 32 not be measured directly by electronic means.
65 13 Thus, during the early days of freeway surveil-
66 22 lance various estimates of density were investi-
67 9 gated. This led to development of a measure
68 13
69 4
that is now called lane occupancy.
70 IO If one could at any instant measure the
71 0 lengths of all vehicles on a given roadway sec-
72 2 tion and then compute the ratio
73 1
__ sum of lengths of vehicles
Rl (2.10)
aFrom observations on Interstate 94 at Prior length of roadway section
Street Bridge, August 1970.
This ratio could then be divided by L,,,, the
average length of a vehicle expressed in miles,
to yield an estimate of the concentration or
TABLE 2.2 Harmonic Mean Derived density in vehicles per mile.
from Arithmetic Mean of Table 2.1 a For example, suppose vehicles having
lengths (in feet) 17, 13, 20, 40, 17 and 20 are
Arithmetic Harmonic distributed over a length of highway one lane
Mean Mean wide and 1,000 ft long. The ratio R is then
Direct 0.127. If the average length of a vehicle is
computation 64.1902 64.0368 taken as 21 ft (or 0.00398 mi), a computed
Estimate of arithmetic value of 31.91 vehicles/ mile is obtained as the
mean from harmonic concentration on this highway section.
mean 64.1925
Estimate of harmonic mean It is not feasible to use on-line methods to
from arithmetic mean 64.0353 measure the sum of the lengths of vehicles in a
Variance• about arithmetic mean =9.94723 given roadway section. It is possible, however,
Variance about harmonic mean =9.97089 to estimate this value by time measurements.
Several types of presence detector are
" See also Appendix B-8. available to the traffic engineer, including in-
• The variance computed here is the unbiased
estimate of the variance based on sample mea- duction loops, magnetometers, ultrasonic re-
surements; i.e., Bessel's correction has been in- flectors, and photo cells. A presence detector
cluded. has the characteristic that it remains in the "on"
MEASUREMENTS AT A POINT 11
or "closed" condition as long as the vehicle Vehicle in Earliest Position to Turn Detector "on"
remains within the zone of effectiveness of the
I
detector. This characteristic is the basis for
point measurements of lane occupancy.
It is not difficult to build an electronic in- Direction of Traffic Vehicle in Latest
Position to H~ld
strument to measure the ratio
R2=
sum of times vehicle detector is occupied
total time of observation period
I.to (2.11)
lfr;;QJ
I~'
dj--
_
L
Detector "on
Detector
T
Le
This ratio can then be used to estimate density
and speed. Figure 2.2 Effective length of vehicle at presence de-
Two general types of error are present- tector: t 0 == LelUt1, where f 0 is the occupied time, le is
instrumentation and estimation errors. In- the effective length of the vehicle, and Ud is the speed
strument errors depend on the design of the over the detector {time mean speed).
instrument and are not discussed here, except
to remark that time is often measured by count- Thus, 2:t 0 =5.85 and N= 13.
ing pulses from a generator and that the prin- 5
cipal errors are related to events whose begin- Then,</>= 6~5 (100) =9.75%.
ning or ending falls between pulses.
If the average effective length of a vehicle is
2.2.4.1 Estimation of Occupancy with 9.75 5,280 9 8 h" 1 I
taken as 26 ft, k= lOO · l 6 =I . ve 1c es
A
* Significance of asterisks is discussed follow- where Nt is the number of trucks and buses
ing Eq. 2.17. during the observation period and Ne is the
12 MEASUREMENT OF FLOW, SPEED, CONCENTRATION
number of cars during the observation period. car, and ui is the speed of the ith car (as be-
Extending this method to the estimation of fore).
concentration gives
(2.20) •
k- Nt+Ne ~--~ (2.16)
Nt Lt+Ne Le 100 Le 100
For k in vehicles/mile, Eq. 2.16 may be
restated or
length 20 ft, Nt=3; Ne= 10; k= . + . 2.3.2 Speed from Measurements Along
3 40 10 20
a Length
~·~g 5 ,280 = 20.9 vehicles/ mile.
When two (or more) aerial photos are
It should be noted that these estimates of taken in sequence with a short time interval
concentration and speed from presence de- between them, the situation represented in
tectors are biased. Mikhalkin et al.17 give meth- Figure 2.1 by lines BB' and CC', separated in
ods for obtaining an unbiased estimate. time by M, obtains. Although each vehicle
traverses a different distance, all are observed
for the same time, or
2.3 MEASUREMENTS ALONG A LENGTH
Either line BB' or line CC' in Figure 2.1 U·=2
I M
indicates the information one might obtain
from an aerial photograph (or photograph
from a tall building). From a photograph it is
possible to scale a distance and count the cars
in this distance. With two photos spaced a
short time apart (as the lines BB' and CC' in Because this average speed ii 8 is the mean taken
Fig. 2.1), it is possible to get speeds and flows. along a distance, Wardrop 1 gave it the name
(The aerial photo computations given here are "space mean speed." The relationship between
true for an infinitely long road; Breiman " this mean and its variance and the mean of
gives more accurate methods for finite lengths.) point measurements is given by Eq. 2.7.
9
6000
ii 8 = - =44 ft/sec=30 mph
136 .4
(2.26)
0.499
k=--=0.0113 veh/ft=59. 9 ve h /mile
.
where le= time observer is moving with stream 44
17
'·
Thus, from Eqs. 3 .4 and 3.5 it follows that For the case of fewer than x, the statement
P(O) = e- 111
becomes
x-1 ·
P(l) =!!!._P(O)
P( <x) = L m'.~-m (3.7)
i=O l.
1
For the case of more than x,
~
1 1
P(2) = P( 1)
x n1i e-m
P(>x)=I-L,--.1- (3.8)
m i=o l.
P(3) =)P(2), etc.
For the case of x or more,
Equations of the type illustrated by Eq. 3.5 are i m
known as "recursion formulas." P( zx) = 1- L m .~-
X-1
(3.9)
The theoretical number of observed inter- i=O l.
vals containing 0, 1, 2 cars, etc., is obtained by
multiplying the total number of intervals ob- or
served by, respectively, the probabilities P(O), 00 1nf e-m
P(l), P(2), etc. P(zx) = L,-.-,- (3.10)
i::::x l.
3.2.1.2 Combination of Poisson Popu·
lations. It is shown in Appendix B-1 that if For the case of at least x but not more than y,
several Poisson populations, having the param-
eters mA, mg ... mN, are combined, the result Y int e-m
P(xsisy) = L,--.,- (3.11)
is still Poisson distributed with the parameter i=x l.
N
3.2.1. 5 Limitations of the Poisson Distri-
tn= L,mi
i=A
bution. In section 3 .2.1 it was pointed out that
the Poisson distribution is appropriate for de-
3.2.1.3 Numerical Example of Poisson scribing discrete random events. When traffic
Distribution. The first published numerical is light and when there is no disturbing factor
examples of the use of the Poisson distribution such as a traffic signal, the behavior of traffic
in traffic were those of Adams 2 (see Table 3. l). may appear to be random, and the Poisson dis-
COUNTING DISTRIBUTIONS 19
tribution will give satisfactory results. How- TABLE 3.1 Poisson Arrival Frequencies,
ever, when traffic becomes congested or when Compared with Observed Arrival Counts
there is some cyclic disturbance in the arrival Measured on Vere Street, London
rate of traffic (such as produced by a traffic
signal), other elementary distributions give a No.
better description of traffic behavior (i.e., they Vehicles/
give a better fit between theoretical and ob- Sec-10 Observed Total Theoretical
served data). (See Appendix B-7 for a dis- Period" Frequency Vehicles• Frequency c
cussion of goodness of fit, including x2 and 0 94 0 97.0
Kolmogorov-Smirnoff (K-S) tests.) 1 63 63 59.9
It should be noted that for the Poisson 2 21 42 18.5
distribution the mean and variance are equal. 3 2 6 3.8
When the observed data present a ratio of >3 0 0 0.8
variance/ mean markedly different from 1.0, it 180
TOTAL 111 180.0
is an indication that the Poisson distribution is
not suitable. (See Appendix B-8 for a discus-
a Since there \Vere 111 vehicles in 180 10-sec
sion of the variance of observed data.) periods, the hourly volume was 222.
• Obtained by multiplying Col. 1 by Col. 2.
'(Theoretical frequency)=(total observed fre-
3.2.2 Binomial Distribution nrz e-m
q uency) ----;.y-,
For congested traffic (where the ratio of h (Total vehicles)
the observed variance/ mean is substantially w ere /11
= (Total observed frequency)
less than 1), * it has been found ·1 that the 111 ; x = va I ue m.
binomial distribution can be used to describe Col. 1.
180
the distribution of traffic arrivals. t
Stated in the form most useful for traffic
purposes, the binomial distribution is: 3.2.2.1 Numerical Example of Binomial
P(x) = C/' p·"(l- p) n-x, Distribution. Table 3.2 illustrates a binomial
distribution fitted to data for congested free-
X= 0, 1, 2 .. ., ll (3.12) way flow (where the variance/mean ratio
equals 0.535). For comparison a Poisson dis-
where p is the probability that one car arrives
tribution has also been fitted. It will be noted
and Ci." is the combinations of n things taken
that the binomial distribution produces a much
. ll !
x at a time= ( • better fit.
x! n-x).1
For the binomial distribution, m is the
mean, = n p, and s 2 is the variance, = n p( 1 - p). 3.2.3 Traffic Counts with High Variance
If p is the estimated value of binomial parame- (Negative Binomial Distribution)
ter p, used in fitting, and n
is the estimated
When traffic counts extend over both a
value of binomial parameter n, used in fitting,
peak period and an offpeak period, combining
these parameters may be estimated by the
relationships 6 : the results into one distribution results in a
high variance. A more common but less ob-
p= (m-s 2
)/m (3.13) vious situation occurs downstream from a traf-
fic signal: During the early portion of the
n= ml p=m 2 / (m-s 2 ) (3.14) signal cycle traffic flow is high (usually at satu-
ration level); during the later portion of the
where m and s 2 are computed from the ob- signal cycle there often will be very light traffic.
served data. Fitting may be accomplished by If the counting period corresponds to the green
direct computation or by the use of tables.' portion of the signal cycle, or to the complete
signal cycle, cyclic effects will be masked.
* In congested traffic the opportunity for free However, if the counting period is short (say
movement is decreased, resulting in decreased 10 sec), there will be periods of high flow and
variance. periods of low flow; there may even be periods
t It has been shown"" that as the binomial of intermediate flow. Thus, combining all
parameter n becomes very large and the parame-
ter p becomes very small but the product pn is a counting periods into one distribution will re-
constant, the Poisson distribution results . sult in a very high variance.
. ..
20 STATISTICAL DISTRIBUTIONS OF TRAFFIC CHARACTERISTICS
Table 3.3 illustrates the phenomenon where parameters p and k. If m is the mean of ob-
the counting period is synchronized with the served data and s 2 is the variance of observed
signal cycle; here, with 30-sec counting inter- data,
vals the observed data can fit a Poisson distri- •
bution acceptably (at the 5 percent confidence p= m/s 2 (3.16)
level with a chi-square test). When the same
data are analyzed in 10-sec intervals, the non- (3.17)
randomness appears, and a Poisson distribu-
tion will not fit the data; however, a negative and
binomial distribution may fit acceptably, as will
be shown.
q=(l-p) (3.18)
The negative binomial distribution (some- The various terms can then be obtained
times called the Pascal distribution) may be from tables 8 or by direct computation (e.g.,
stated 8 using a computer program). The recursion
equations for the negative binomial distribution
x=O, 1, 2, ...
are given by
(3.15)
3.2.3.2 Some Notes Concerning the TABLE 3.4 Comparison of Poisson and
Negative Binomial Distribution. Buckley 9 Negative Binomial Distributions for
points out that if a detector on a roadway Durfee Avenue, Northbound;
spans several lanes, the resulting count is 1 O·Sec Intervals
usually a negative binomial distribution. Ken-
dall and Stuart 10 stated that if the Poisson Theoretical Frequency
parameter is independently distributed with a Number Negative
type III distribution, the resulting distribution of Cars Poisson Binomial
is negative binomial. per Observed Distribu- Distribu-
Interval Frequency tion a tion a
0 139 129.6 140.4
3.2.4 Summary of Elementary Counting 1 128 132.4 122.0
Distributions 2 55 67.7 62.2
3 25 23.1 24.2
1. The Poisson distribution represents the 4 10 5.9 8.0
random occuHence of discrete events. 5 3 1.2 2.3
2. In counts of light traffic w·here the >5 0 O.i 0.9
observed data produce a ratio of variance/
mean of approximately 1.0, the Poisson distri- TOTAL 360 360.0 360.0
bution may be fitted to the observed data. m= 1.022
3. In counts of congested traffic where i'= 1.203
the observed data produce a ratio of variance/ i'lm= 1.177
mean of substantially less than 1.0, the bi-
nomial distribution may be fitted to the ob- ' The fit of the negative binomial distribution
served data. is acceptable by a x 2 test at the 5 percent sig-
nificance level; the fit of the Poisson distribution
4. In counts of traffic where there is a is not.
cyclic variation in the flow or where the mean
flow is changing during the counting period,
giving a ratio of variance/ mean substantially
greater than 1.0, the negative binomial distri-
bution can be fitted to the observed data.
If there is no vehicle arrival in a particu-
lar interval of length t, there will be a headway
3.3 INTERVAL DISTRIBUTION of at least t sec between the last previous
The previous analyses have developed the arrival and the next arrival. In other words,
probability of discrete events occurring within P(O) is also the probability of a headway
a specific time· interval. Another traffic char- equal to or greater than t sec. This may be
acteristic of great importance is the time be- expressed:
tween events; i.e., the time headways between P(h?.t) =e-Vt/3,aoo (3.19)
the arrival of vehicles. The class of distribu-
tions used for this purpose has been termed From this relationship it may be seen that
"interval distributions." (under conditions of random flow) the num-
ber of headways greater than any given value
will be distributed according to an exponential
3.3. l Negative Exponential Distribution curve. (Though correctly a negative exponen-
The elementary interval distribution is the tial, this is usually known simply as an expo-
negative exponential distribution, which may nential distribution.)
be derived as follows: In the above equation m or Vt/3,600 is
In Eq. 3.2 substitute A.=V/3,600 cars/sec, the mean of the arrival (counting) probability
where V is the hourly volume; thus, distribution. If we set m =ti T, T is the mean of
the interval (headway) probability distribution
P(x)= - -
Vt
( 3,600
)x -
e -rt/~.60,0 =3,600/V. Thus, the probability of a headway
x! equal to or greater than t may be written:
P(O) = e-rt/3,6oo P(h?.t) =e-t/T (3.20)
22 STATISTICAL DISTRIBUTIONS OF TRAFFIC CHARACTERISTICS
1.0 1.0 - - - - - - - -=-:.:;-::.-=-~-------
0.8 0.8
0..
>o.s
.~
0.2 0.2
2 4 6 8 10 2 4 6 8 10
Time, t
Time, t
Figure 3.1 Probability of headways equal to or greater Figure 3.2 Probability of headways less than t, with
than t, as treated by the exponential distribution, with T = 1 sec.
T= 1 sec.
A sketch representing the general (negative) These results are also depicted graphically in
exponential distribution of Eq. 3.20 is shown Figure 3.3. It may be seen that agreement
in Figure 3. I. between the curve and the data is reasonable
For some purposes it is more convenient (acceptable by x" test at the 0.05 level). It
to use the complementary relationship: should be noted that traffic was light at the
P ( h < t) = 1 - e- 1/ 7 ' ( 3 .21) time of these observations: Volume= (3,600/
1,753)214=439 vehicles/hr.
which is illustrated in Figure 3.2.
The exponential probability distribution is
fitted by computing the mean interval T and
3.3.3 Alternate Viewpoint
then using tables 72 to obtain values of e- 1/T
for various values of t. The negative exponential probability
The variance of the exponential distribu- models of Eqs. 3.20 and 3.21 can be viewed
tion is T2. (Appendix B-3 presents a deriva-
in an alternate way as follows: Consider the
tion.)
probability density of intervals (headways)
ft ft
ing 1,753 sec. Thus, T= 1,753/214=8.19 sec,
m=t/T=t/8.19=0.122t, P(h?.t) =e- 0 - 122 1, and 1
P(h< t) = p(h)dt= -e-t/Tdt= 1-e-t/T
H=expected number of headways?.t, or o o T
H=214e-o.1221
Table 3 .5 lists the results of the fitting.
ADV AN CED HEADWAY DISTRIBUTIONS 23
3.4 INADEQUACY OF THE TABLE 3.5 lntervehicle Headways,
EXPONENTIAL DISTRIBUTION FOR Arroyo Seco Freeway
INTERVALS
As discussed in sections 3.2 and 3.3, the Observed H, Expected
Headway, t Cumulative Number of
Poisson distribution for vehicle counts and the Frequency 2 t
(sec) Headways" 2 I
negative exponential distribution for time head-
ways are only applicable when traffic flows are 0 214 214.0
light; i.e., when there is no interaction between 1 185 191.3
vehicles, thus enabling them to move at ran- 2 171 171.0
3 149 153.0
dom. As traffic becomes heavier, vehicles are
4 136 136.7
restricted in their ability to pass at will, and 5 125 122.2
interaction between vehicles increases. (The 6 111 109.4
assumption of interaction between vehicles is 7 95 97.8
fundamental to car-following analysis, as dis- 8 84 87.3
cussed in Chapter 4.) Vehicles then tend to 9 72 78.1
operate in platoons where the minimum time iO 6i 69.8
headway is substantially greater than zero, 11 52 62.5
which results from vehicle finite length and 12 40 55.8
finite spacing between the rear of one vehicle 13 34 49.9
14 32 44.5
and the front of the following vehicle. The 15 29 39.8
exponential distribution, on the other hand, 16 26 35.7
describes probabilities of headways extending 17 19 31.9
down to very small values. (See Figure 3 .1 or 18 16 28.5
Figure 3.2.) When observations from several 19 14 25.5
parallel lanes are combined, it is possible to 20 11 22.7
have headways as low as zero. 21 10 20.3
Figures 3.1 and 3.3 illustrate the cumu- 22 9 18.2
23 8 16.3
lative probability of headways t or greater, and
24 8 14.6
Figure 3.2 illustrates the cumulative probabil- 25 7 13.1
ity of headways less than t. The phenomenon 26 7 11.6
of the scarcity of short headways can often be 27 6 10.5
better illustrated by the frequency density or 28 4 9.2
the probability density curve rather than the 29 3 8.3
cumulative curve; i.e., Eq. 3.22 or the equiva- 30 1 7.5
lent frequency curve obtained by multiplying 31 0 6.6
the values from Eq. 3.22 by the total number
"From negative exponential distribution.
of observations. Thus Figure 3.4 illustrates an
example of headway observations on the Holly-
wood Freeway for flows of 33, 34, and 35
vehicles/min. It will be noted from visual ex- quencies. For the purpose of this discussion
amination that the fit of the exponential density these have been termed "advanced headway
distribution is very poor. (Buckley 11 estimates distributions."
that "the probability of a worse fit is approxi-
mately 10-100 .") 3.5.1 Shifted Exponential Distribution
Section 3.4 emphasized one shortcoming
3.5 ADVANCED HEADWAY of the exponential distribution: It predicts too
DISTRIBUTIONS many short headways. One approach to treat-
Because of the poor agreement between ing this situation is to introduce a minimum
allowable headway; i.e., a region of the distri-
the frequencies of headways observed in prac-
bution in which headways are prohibited.
tice and the frequencies predicted by the nega- (Some writers 13 have maintained that a deter-
tive exponential distribution, as well as theo- ministic prohibited period (i.e., a deterministic
retical considerations precluding very short minimum headway) is philosophically unac-
headways, other distributions have been sought ceptable. They would rather have a period
as a means of improving the predicted fre- during which the probability of an arrival is
24 STATISTICAL DISTRIBUTIONS OF TRAFFIC CHARACTERISTICS
1.0
200
~
\
\
(> 0.8
160
\ -m
/ e , where m ; 0.122 t ....
....
....
Q)
....
Q)
"'....
Q) "'....
Q)
(!)
....
'f 0.6
(!)
....0
....0 120 \
6 ....
"'> \ >
"'
3: \6
"'3:
"O
"O
"' Q) \ "'
Q)
:c
I
..... \:::,
.....
0
0
.... \ >
.~
Q)
0.4
.0
80 ~ ii
E
z
::J
\ "'
.0
0....
a..
\
D. - Observed Frequencies
\ I - Theoretical Frequencies
0.2
40
5
Length of Headway, t (sec)
Figure 3.3 Plot of dota' in Table 3.5 for lone 1 from 2 to 2:30 PM, October 1950.
Note: Dashed curve applies only to probability scale.
very low but not zero. The authors wish to adjustments to maintain the total area of unity
point out, however, the use of a deterministic under the probability curve, from which the
"guarantee period" in reliability engineering. 14 ) following results 12 :
This may be visualized by taking the curve of P(h?_t)
Figure 3.2 and shifting it to the right by an
=e-<t-Tl/(T-Tl (3.22)
amount r equal to the minimum allowable (Appendix B-3 derives the parameters of the
headway. This is illustrated in Figure 3.5. In shifted exponential distribution.) Note that the
shifting, it is necessary to make appropriate probability density is
ADVANCED HEADWAY DISTRIBUTIONS 25
140
120
100
>
u 80
c:
m
::i
u
~ 60
LL
40 I I
20
0
~ 11111~. 2 3 4 5 6 7 8
.
9 10 11 12 1 3 14 15 16 1 7 18 19 20
1.0 ----------~=-=-==~=-...._..._ _ _ __
0.9
0.8
0.7
0.. 0.6
>
:!:
..c
cu
0.5 p (h < t) = 1 -e -(t - r)/(T - r)
..c
e
0..
0.4
0.3
0.2
0.1 Tl
0
0 2 3 4 5 6 7 8 9 10 11
Time, t
Figure 3.5 Shifted exponential distribution to represent the probability of headways
less than t with a prohibition of headways less than T. (Average of observed
headways is T.)
26 STATISTICAL DISTRIBUTIONS OF TRAFFIC CHARACTERISTICS
p(t) =0 for t<r and Buckley's data for the Hollywood Free-
way. Note that whereas the shifted exponen-
and tial fits data for low flows, it is not suitable for
1 high flows.
p(t) = - - e-<t-r>/tT-rJ for t~r.
T-r
3.5.2 Other Pearson Type Ill
Fitting a shifted exponential distribution Distributions
to field data requires estimation of the parame-
t
ters t and r, where is the mean of measure- Appendix B-2 discusses the Pearson type
ments from the origin, f is the shift of curve III family of distributions and points out that
with respect to the origin, and (T-7-) 2 is the both the negative exponential distribution and
variance about the origin. the shifted exponential distribution are special
An application of the shifted cumulative cases of this family. Other special cases have
exponential distribution is shown in Figure 3.6. been found to be useful for describing traffic
Figure 3.7 illustrates the relationship between headways. One such special case is the Erlang
a shifted exponential frequency density curve distribution.
400 ......
r = 1.0 sec
- -1.0
- -0.8
"'"'
<ll
....J
>
'°
;:
"O
,_ -0.6
'°<ll ....
J: .c
'° l5i
.... c
0 <ll
....0 )
0.. ....J
,_ -0.4 ~lij
.... ·- .c
<ll
..c ~ I-
z
E
:l
100 ,__ / '°
..c
0
.....
0..
,__ I ,_ -0.2
~~~....J.-J....L-L....__.__,__,__
L-.L..>-IJ
0 5 10 15 20 25 30
0
P(h>t)= L -T -~
k-l(kt)le-kt/1'
(3.23)
P(h~t) =[ +( ~ )+( ~y ~!]
1 e-kt/T
For k=2
+( ~ )3 ;! J
e-kt/T
260 ' T2
k=- (3.24)
s2
u.
120
observations on a freeway; more specifically,
Table 3.6 demonstrates the estimation of pa-
100 rameters from one-half of the data, and Table
3. 7 the testing of the fit to the other half of
the data by the Kolmogorov-Smirnoff (K-S)
80 test (see Appendix B-7). This test is an al-
ternative test for goodness of fit, especially
suited to small sample sizes. The fit shown in
60 Table 3.7 is acceptable at the 5 percent level.
40
20
0 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
'P(h ~ t) = [ 1 + 2 _;~ 4 Je- 2112 ·"''; fit is acceptable at the 5 percent level.
TABLE 3.8 Effect of Varying Erlang Parameter k for Theoretical Frequency of Headways
Greater Than ta
Observed
Odd Even
k=I k=2 k=3 k=4 k=5 Values Values
0 25.0 25.0 25.0 25.0 25.0 25 25
1 15.9 19.2 21.1 22.2 23.0 21 22
2 JO.I 11.5 12.2 12.7 13.1 11 13
3 6.4 6.1 5.7 5.2 4.8 7 6
4 4.1 3.1 2.3 1.7 1.3 4 3
5 2.6 1.5 0.9 0.5 0.3 2 2
6 1.6 0.7 0.3 0.1 1 0
7 1.0 0.3 0.1 0
8 0.7 0.1
9 0.4
10 0.3
"Data from Gerlough and Barnes' (as corrected by the authors Mar. 8, 1974).
where t=any time duration; •\=the minimum " Data from May. 23
"free" headway; y 1 =the average "free" head-
way; 8,=the minimum headway in the con-
strained headway distribution; y 2 =the average Buckley 11 has proposed a "semirandom"
headway in the constrained distribution; and model or a distribution representing two sub-
k =an index that indicates the degree of non- populations, one having type 1 headways and
randomness in the constrained headway distri- the other having type 2 headways. Type 1
bution. (See section 3.5.2.1.) headways result from a vehicle being placed
Table 3.10 lists the values for the various exactly at the rear limit of a zone of emptiness,
parameters as computed for data observed by where vehicles never enter; its length is mea-
investigators at Purdue University. Figure 3.11 sured in time units and is normally distributed.
plots the distributions from these parameters. Type 2 headways result when a vehicle occurs
Flow Rate
Moni- Com-
to red puted R' a_, 1'1 01 k a., 1'2 02
200
•
150
>
u
cCl)
::J
C"
Cl)
....
u.
Cl)
>
100
·p
.:!!
::J
E
.....
G
50
0 ..._~-"-~-'-2~~.__~~4~~-'--~~6,__~_._~-s~~~~--=-'10·
0
Time Spacings (sec)
Figure 3.9 Composite exponential arrivals: ' 0 curve A computed for constrained vehi-
cles; curve B, for unconstrained vehicles; curve C, for composite flow; curve D, from
field data (taken from Schuh!). Points in circles are results of computer generation.
at some position to the rear of the extreme rear results in a more complex computational pro-
limit of a zone of emptiness; i.e., vehicles with cedure. In some cases the intended use of the
type 2 headways are presumed to be not in- model can help in the selection procedure. For
fluenced by the front vehicle. instance, Newell 30 has shown that delays are
Using 609 headway observations from the relatively insensitive to the form of the distribu-
Hollywood Freeway at volumes of 33, 34, and tion of the arriving traffic. Thus, if the objec-
35 vehicles/min, Buckley plotted comparisons tive is simply the computation of delays, the
of data against several distributions. These simplest (i.e., the negative exponential) distri-
plots appear as Figures 3.4, 3.7, and 3.12. By bution should be used. If, however, the objec-
inspection, the semirandom distribution appears tive is the determination of gaps for, say, cross-
to provide the best fit. ing purposes, a more faithful distribution may
be needed.
3.7 SELECTION OF HEADWAY
DISTRIBUTION 3.8 ADVANCED COUNTING
DISTRIBUTIONS
As in many engineering selection proc-
esses, selection of a suitable headway distribu- In section 3.2 traffic counts (and head-
tion represents a compromise between eco- ways) for low traffic flows (i.e., where cars can
nomic considerations and faithfulness of the maneuver with relative ease) appear to be
model. Greater faithfulness is often obtained random. In such cases the Poisson distribution
by using a model with a greater number of can serve as a model of traffic counts. When
parameters; such a model, on the other hand, the flow becomes high, however, freedom to
32 STATISTICAL DISTRIBUTIONS OF TRAFFIC CHARACTERISTICS
maneuver is diminished and counts per unit changing average flow, the variance of the
time become more uniform. In this situation counting process becomes large. The simple
the binomial distribution provides a simple model for such situations is the negative bi-
means of modeling the counting process. On nomial distribution.
the other hand, when there is some disturbing In section 3.4 simple headway distribu-
factor, such as a traffic signal or a rapidly tions are not deemed adequate from a theo-
1.00
0.80
20SEC
19 SEC
0.60 18 SEC
17 SEC
16 SEC
> 15 SEC
.~
14 SEC
..0 13 SEC
"'
..0
0
....
12 SEC
0.. 11 SEC
10SEC
0.40 9 SEC
8 SEC
7 SEC
6 SEC
5 SEC
4 SEC
0.20
3 SEC
2 SEC
-;;
A.
.80
r Reported
Flow Rate
(veh/hr)
> 158
"'~
-0
.60
"'
Q)
.r.
;>.
.'::
.40
ii
"'
.0
0
ct
.20
0 0L_~~~~~~-15~~~~~~~J10-==============-1L5~~~~~--=~20
Time (sec)
Figure 3.11 Hyperlang headway distributions for Purdue research project data. 28
160
140
120
100
>
u
c:
Q)
::J
C' 80
....
Q)
u.
60
40
20
0
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Headway, Time Units of 10-4 hr (i.e., 0.36 sec)
Figure 3.12 Fit of Buckley's semirandom distribution to Hollywood Freeway data. 11 Bars are
observed data and curve is best fit for the semirandom distribution; sample size= 609.
•
retical standpoint, and in sections 3.5 and 3.6 k(a>+l J-1 )\i e-X
several advanced distributions of headways P(x)= L -.-,
I·
i=kx
x=O, 1, 2., ...
were discussed.
In section 3.1.1 it was pointed out that the or
negative exponential distribution for headways
can be easily derived from the Poisson distribu- ~ e-X ("A).rk+i-1
P(x)= L.., . x=O, 1, 2, ...
tion of counts. The derivation process could i=l (xk + 1- 1).1
have been reversed, taking the negative expo-
nential distribution of headways and deriving (3.28)
the Poisson distribution by the mathematical When k = 1, the simple Poisson distribution
operation known as convolution. This proce- results.
dure, even for the simple case of the negative
exponential headway distribution, is very te- When k=2
dious. For more complex headway distribu- P(O) = e-X+"Ae-X
tions the convolution process can be intractable
even though theoretically possible.
In summary, the following pertain: ( 1)
Although simple distributions will often work
for traffic data, they are not necessarily correct; "A4e-x "Ase-X
P(2) =~+-s-!, etc.
(2) for each headway distribution there is a
related counting distribution, even though such
When k=3
distributions cannot be stated explicitly.
By use of the approaches taken in de-
veloping advanced headway distributions we
can search for advanced counting distributions
directly (i.e., independent of the related head-
way distribution).
'l
50
20
10
5.0
Q)
CJ
c:
.!:!!
,_
<O
> 2.0
1.0
Mean
Figure 3.14 Nomograph for estimating parameter le for generalized Poisson distribution.33
Number of
Cars per Observed Total Cars Theoretical
Interval• Frequency Observed Probability c Frequency•
(x1) (f1) (f1x1) p(2x<) p(2x• + 1) P(x•) 64P(x1) /(x<) I F(xi)
'l
0 0.0000 0.0000 0.0000
1 0.0000+ 0.0000+ 0.0001 0.0
2 0.0005 0.0015 0.0020 0.1
0.0039 11.9
3 3 9 0.0084 0.0123 0.8
4 0 0 0.0162 0.0275 0.0437 2.8
5 8 40 0.0423 0.0590 0.1013 6.5
6 10 60 0.0755 0.0894 0.1649 10.6 9.4
7 11 77 0.0983 0.1010 0.1993 12.8 9.5
8 10 80 0.0974 0.0885 0.1859 11.9 8.4
9 11 99 0.0760 0.0619+ 0.1380 8.8 13.8
10 9 90 0.0480 0.0354 0.0834 5.3} 12.5
11 1 11 0.0691 10> 4.4
12 1 12
f ___
A e-A P( <
_x
)=~c.x+'(l- )x+1-i ;'Y[(x+l-i),y]
£_, , a a (x-i)!
/(LI, p) = ~"-'°
i
l=ll
'J...P d'J...
+a'r+ 1 (3.30)
e-A
where
where p = (x + 1) k- 1 and u =A.I (p + 1) 'I" a= proportion of restrained vehicles;
Appendix B-6 describes this procedure. y=µ.(1-a)[t-(x+l)r] for
For most traffic engineering applications µ. = mean flow rate for unrestrained ve-
the relatively modest improvement achieved hicles and
by use of noninteger values of k does not justify minimum headway; and
r=
the extra effort involved. 33
80
"'O "'
(I)
~"' 60
en fij
.... .t:.
0 1-
.... "'
~ ~ 40
u
.... ..J
(I)
0..
20
16 22 28 34 40 46 52 58 64 70 76 82
Speed (mph)
Figure 3.15 Cumulative (normal) distributions of speeds at four locations."
100
...
-.·.-.----:.~
g'] 80
:.= ~
~en
~ c:
I- ~ 60
·-
"'~ (!)
(.) c:
.... co
0 .t:. 40
.... I-
C: "'
~ ~
.... ..J
~ 1U 20
Speed
Mean Speed
Figure 3.16 Same data as Figure 3.15 but with each distribution normalized. 42
80
70
2
a.
E 60
"'O
(I) 50
~
en
40
1 2 5 10 20 30 40 50 60 70 80 90 95 98 99
Cumulative Percent
Figure 3.17 lognormal plot of freeway spot speeds.
GAP (AND LAG) ACCEPTANCE DISTRIBUTIONS 39
800
... \
80
/\
c.
co
l9
Cl 60
c
·;; Observed Erlang Observed
0
Ql k=7
'ii)
a: 40
...c:
Ql
...0
Ql
a. 20
0
0 2 4 6 8 0 2 4 6 8
15
10
9
8
:--~----~[ s - - - - - - -
uQl
V> 5
Ql
E 4
i=
3
•
2
1.0 10 50 90 95 99.9
Percent Acceptance
Figure 3.20 Lag and gap distribution for through movements in Lafayette and lndianapolis. 51
8.0
Regression Equations
7.0
YR= -2.41 + 8.55x
Y L = -2.63 + 8.55x
YT = -2.32 + 8.55x
6.0 /,
Q)
(.)
c /J
..."'a. ~
Q) ~
(.)
(.)
<x:
'+-
...
0
5.0 /,
_/,I
"
t(~";''""'"'
:.0
0.._
a..
4.0
- t : : . - Left Turns
f
/, ~ ...... ••• Through
Movements
3.0
. l'l
0.4
lfii
0.6 0.8 1.0 1.2
son (for low-density traffic), negative binomial ti on of order k (positive integer), the corre-
distribution (for varying flows), and the gen- sponding flow is itself a discrete event variable
eralized Poisson as well as the binomial (for consisting of every kth event of a Poisson
congested flow) . The second type are those series. This unique correspondence, although
that obey continuous distributions such as true, is difficult to obtain under real conditions
headways (exponential, shifted exponential, except for special cases; that is, to the discrete
composite exponential, Erlang, hyper-Erlang, flow process there corresponds a continuous
semirandom, and normal) and speeds (nor- headway process.
mal, lognormal, and gamma). The choice of Gap acceptances were found to obey either
distribution depends on how much complexity the Erlang or the lognormal distributions.
is desired as well as the behavior of the traffic. Probits were useful for expressing the normal
In several of the above cases it was shown scales.
that if the vehicle count (flow) obeys a given
discrete distribution the headway will obey a 3.12 REFERENCES
unique corresponding distribution. For ex-
ample, if flow is purely random (Poisson), 1. Kinzer, J.P., Application of the Theory of
headways are exponentially distributed. Simi- Probability to Problems of Highway
larly, if the headways have an Erlang distribu- Traffic. B.C.E. thesis, Polytechnic In-
42 STATISTICAL DISTRIBUTIONS OF TRAFFIC CHARACTERISTICS
stitute of Brooklyn (June l, 1933); tern Analysis for Engineers and Plan-
also in Proc. Inst. Traffic Eng., 5: 118- ners. McGraw-Hill, New York (1967),
124 (1934). (Abstr.) p. 506.
2. Adams, W. F., Road traffic considered 16. Burgin, T. A., and Wild, A. R., Stock •
as a random series. J. Inst. Civil Eng., control--experience and usable theory.
4: 121-130 (1936). Oper. Res. Q., 18(1):35-52 (1967).
3. Greenshields, B. D., Shapiro, D., and 17. Pearson, K., Tables of the Incomplete
Ericksen, E. L., Traffic Performance Gamma Function. Cambridge Univer-
at Urban Street Intersections. Tech. sity Press, Cambridge, England ( 1965).
Rep. No. 1, Yale Bureau of Highway 18. Whittlesey, J. R. B., Incomplete gamma
Traffic, New Haven, Conn. (1947). functions for evaluating Erlang process
4. Gerlough, D. L., and Barnes, F. C. The probabilities. Math. Comput., 17 ( 81):
Poisson and other probability distribu- 11-17 (1963).
tions in highway traffic. In Poisson and 19. Whittlesey, J. R. B., Normalized Incom-
Other Distributions in Traffic. Eno plete Gamma Function with Poisson
Foundation ( 1971). Term-Double Precision. SHARE Gen-
5. Gerlough, D. L., Use of Poisson distribu- eral Program Library UR GAM 2 (DO)
tion in highway traffic, pp. 1-58. In 3138, February 28, 1964.
Poisson and Traffic, Eno Foundation 20. Daou, A., On flow within platoons. Austr.
(1955). Road Res. 2 (7) :4-13 (1966).
6. Drew, D. R., Traffic Flow Theory and 21. Daou, A., The distribution of headways in
Control. McGraw-Hill, New York a platoon. Oper. Res., 12 (2) :360-361
(1968)' p. 130. (1964).
7. Harvard University Computation Labora- 22. Greenberg, I., The log-normal distribution
tory, Tables of the Cumulative Bino- of headways. Austr. Road Res., 2 (7):
mial Probability Distribution. Harvard 14-18(1966).
University Press, Cambridge, Mass. 23. May, A. D., Jr., Gap availability studies.
( 1955); Weintraub, S., Tables of the Highw. Res. Rec. 72 (1965), pp. 101-
Cumulative Binomial Probability Dis- 136.
tribution for Small Values of p. Col- 24. Tolle, J. E., The lognormal headway dis-
lier-Macmillan Ltd., London (1963). tribution model. Traffic Eng. Control,
13 (1):22-24 (1971).
8. Williamson, E., and Bretherton, M. H.,
25. Schuh!, A., The probability theory applied
Tables of the Negative Binomial Prob-
to distribution of vehicles on two-lane
ability Distribution. Wiley, New York
highways, pp. 59-75. In Poisson and
(1963)' pp. 9-11.
Traffic. Eno Foundation (1955).
9. Buckley, D. J., Road traffic counting dis- 26. Kell, J. H., Analyzing vehicular delay at
tributions. Transp. Res., 1 ( 2) : 105- intersections through simulation. Highw.
116 (1967). Res. Board Bull. 356 (1962), pp. 28-
10. Kendall, M. G., and Stuart, A., The Ad- 39.
vanced Theory of Statistics. Hafner 27. Grecco, W. L., and Sword, E. C., Predic-
Pub!. ( 1958), Vol. 1, p. 147. tion of parameters for Schuhl's head-
11. Buckley, D. J., Road traffic headway dis- way distribution. Traffic Eng., 38 (5):
tributions. Proc. Austr. Road Res. 36-38 (1968).
Board, 1962: 153-187. 28. Dawson, R. F., and Chimini, L. A., The
12. Gerlough, D. L., Traffic inputs for simu- hyperlang probability distribution-a
lation on a digital computer. Proc. generalized traffic headway model.
Highw. Res. Board, 38:480-492 Highw. Res. Rec. 230 (1968), pp. 1-14.
(1959). 29. Dawson, R. F., The hyperlang probability
13. Haight, F. A., The generalized Poisson distribution-a generalized traffic head-
distribution. Ann. Inst. Stat. Math. way model. Proceedings of Fourth In-
(Tokyo), 11:101-105 (1959). ternational Symposium on Theory of
14. Epstein, B., Estimation of the parameters Traffic Flow, held in Karlsruhe, Ger-
of two parameter exponential distri- many, June 1968. Published as Beitrage
butions from censored samples. Tech- zur Theorie des Verkehrsflusses, Stras-
nometrics, 2(3) :403-406 (1960). senbau und Strassenverkehrstechnik
15. Wohl, M., and Martin, B. V., Traffic Sys- (1969), Vol. 86, pp. 30-36.
REFERENCES 43
30. Newell, G. F., Statistical analysis of the 45. Taylor, W. C., Speed zoning-a theory
flow of highway traffic through a sig- and its proof. Proc. Inst. Traffic Eng.,
nalized intersection. Q. Appl. Math., 34: 11-23 (1964).
13 (4):353-369 (1956). 46. Haight, F. A., and Mosher, W. W., Jr.,
31. Jewell, W. S., The properties of recurrent- A practical method for improving the
event processes. Oper. Res., 8 ( 4) :446- accuracy of vehicular speed distribution
472(l960). measurements. Highw. Res. Board
32. Haight, F. A., Mathematica/ Theories of Bull. 341 (1962), pp. 92-116.
Traffic Flow. Academic Press, New 47. Raff, M. A., A Volume Warrant for
York (1963), pp. 97-98. Urban Stop Signs. Eno Foundation
33. Haight, F. A., Whisler, B. F., and Mosher, (1950).
W. W., Jr., New statistical method for 48. Blunden, W. R., Clissold, C. M., and
describing highway distribution of cars. Fisher, R. B., Distribution of accept-
Proc. Highw. Res. Board, 40:557-563 ance gaps for crossing and turning
(1961). maneuvers. Proc. Austr. Road Res.
34. Molina, E. C., Poisson's Exponential Bi- Board, 1: 188-205 (1962).
nomial Limit. Von Nostrand, Prince- 49. Bissell, H. H., Traffic Gap Acceptance
ton, New Jersey ( 1945). From a Stop Sign. Research Report
35. General Electric Co., Defense Systems submitted as partial fulfillment of the
Department, Tables of Individual and requirement for the degree of Master
Cumulative Terms of the Poisson Dis- of Engineering, University of Califor-
tribution. Von Nostrand, Princeton, nia, Berkeley (May 1960).
New Jersey (1962). 50. Robinson, C. C., Pedestrian interval ac-
36. Buckley, D. J., A semi-Poisson model of ceptance. Proc. Inst. Traffic Eng., 22:
traffic flow. Transp. Sci., 2 (2): 107- 144-150 (1951).
133 (1968). 51. Solberg, P., and Oppenlander, J. C., Lag
37. Serfling, R. J., Non-Poisson models for and gap acceptances at stop-controlled
traffic flow, Transp. Res., 3 (3) :299- intersections. Highw. Res. Rec. 118
306 (1969). (1966), pp. 48-67.
38. Oliver, R. M., and Thibault, B., A high- 52. Ashworth, R., The analysis and interpre-
flow traffic counting distribution. tation of gap acceptance data. Transp.
Highw. Res. Board Bull. 356 (1962), Sci., 4(3): 270-280 (1970).
pp. 15-27. 53. Tsongos, N. G., and Wiener, S., Com-
39. Underwood, R. T., Discussion of pak-Poy parison of day and night gap-acceptance
paper. Proc. Austr. Road Res. Board, probabilities. Public Roads, 35(7):
2:243 (1964). 157-165 (1969).
40. Pak-Poy, P. G., The use and limitation of 54. Wagner, F. A., Jr., An evaluation of fun-
the Poisson distribution in road traffic. damental driver decisions and reactions
Proc. Austr. Road Res. Board, 2: 223- at an intersection. Highw. Res. Rec.
247 (I 964). 118 (I 966), pp. 68-84.
41. Highway Research Board, Highway Ca- 55. Arley, N., and Buch, K. R., Introduction
pacity Manual. Washington, D.C. to the Theory of Probability and Sta-
(1950). tistics. Wiley, New York (1950), pp.
42. Leong, H. J. W., The distribution and 30-31.
trend of free speeds on two-lane rural 56. Feller, W., An Introduction to Probability
highways in New South Wales. Proc. and Its Applications, 1st ed. Wiley,
Austr. Road Res. Board, 4: 791-814 New York (1950), Vol. 1, pp. 364-367.
(1968). 57. Fry, T. C., Probability and Its Engineer-
43. Bleyl, R. L., Skewness of spot speed ing Uses. Van Nostrand, Princeton,
studies. Traffic Eng., 40(10): 36--37, New Jersey (1928), pp. 220-227.
• 40 (1970). 58. Ford, L. R., Differential Equations .
44. Berry, D. S., and Belmont, D. M., Distri- McGraw-Hill, New York (1933), p.
bution of vehicle speeds and travel 172.
times. Proc. Second Berkeley Symp. 59. Elderton, W. P., Frequency Curves and
Math. Stat. Probab. (1951), pp. 589- Correlation, 4th ed. Harren Press
602. (1953).
44 STATISTICAL DISTRIBUTIONS OF TRAFFIC CHARACTERISTICS
60. Kendall, M. G., and Stuart, A., The Ad- Newell, G. F., Stochastic properties of peak
vanced Theory of Statistics. Hafner short-time traffic counts. Transp. Sci., 1 (3):
Publ. (1958), Vol. 1, pp. 148-152. 167-183 (1967).
61. Aitchison, J., and Brown, J. A. C., The Oliver, R. M., A traffic counting distribution.
Lognormal Distribution. Cambridge Oper. Res., 9(6): 807-810 (1961).
University Press, Cambridge, England Oliver, R. M., Note on a traffic counting dis-
(1963). tribution. Oper. Res. Q., 13(2): 171-178
62. Hald, A., S,tatistical Theory with Engi- (1962).
neering Applications. Wiley, New York Shumway, R., and Gurland, J., A Fitting Pro-
(1952), pp. 160-174. cedure for Some Generalized Poisson Dis-
63. Kendall, M. G., and Stuart, A., The Ad- tributions. University of Wisconsin, Madison
vanced Theory of Statistics. Hafner, (Feb. 1960); NTIS No. PB 167 087.
Pub!. (1958), Vol. 1, pp. 168-173. Vaughan, R. J., The distribution of traffic vol-
64. Finney, D. J., Probit Analysis. Cambridge umes. Transp. Sci., 4(1):97-112 (1970).
Univ. Press (1947). Whittlesey, J. R. B., and Haight, F. A., Count-
65. Dixon, Wilfred J., and Massey, Frank J ., ing distributions for an Erlang process. Ann.
Jr., Introduction to Statistical Analysis. Stat. Math. (Tokyo), 13(2) (1961).
McGraw-Hill, New York (1957), pp.
226-227. Interval Distributions-Headways,
66. Yule, G. Udny, and Kendall, M. G., An Gaps, etc.
Introduction to the Theory of Statistics,
14th ed. Hafner Pub!. (1950), p. 475, Allan, R. R., Extension of the binomial model
477. of traffic to the continuous case. Proc.
67. Cramer, H., Mathematical Methods of Austr. Road Res. Board, 3: 276-316 (1966).
Statistics. Princeton University Press, Breiman, L., Gafarian, A. V., Lichtenstein, R.,
Princeton, New Jersey (1946), p. 435. and Murthy, V. K., An experimental analy-
68. Massey, F. J., Jr., The Kolmogorov- sis of single-lane headways in freely flowing
Smirnov test for goodness of fit. J. Am. traffic. Proceedings of Fourth International
Stat. Assoc., 46: 68-78 (1951). Symposium on Theory of Traffic Flow, held
69. Walker, H. M., and Lev, J., Statistical in Karlsruhe, Germany, June 1968. Pub-
Inference. Henry Holt and Co., New lished as Beitrage zia- Theorie des Verkehrs-
York (1953), p. 443. flusses, Strassenbau und Strassenverkehrs-
technik (1969), Vol. 86, pp. 22-29.
70. Neville, A. M., and Kennedy, J. B., Basic
Statistical Met hods for Engineers and Dunne, M. C., Rothery, R. W., and Potts,
Scientists. International Textbook Co. R. B., A discrete Markov model of vehicular
(1964),pp. 289-290. traffic. Transp. Sci., 2(3): 233-251 (1968).
71. Miller, A. M., An empirical model for Kell, J. H., A theory of traffic flow on urban
multilane road traffic. Transp. Sci., streets. Proc. West. Sec. Inst. Traffic Eng.,
4(2): 164-186 (1970). 1960: 66-70.
72. Handbook of Chemistry and Physics or Leutzbach, W., Distribution of time gaps be-
Standard Mathematical Tables. Chemi- tween successive vehicles. Int. Road Safety
Rubber Co., various editions; also, U.S. Traffic Rev., 5 (3): 31-36 (1957).
National Bureau of Standards, Tables of Lewis, R. M., A proposed headway distribution
the Exponential Function e, Applied for traffic simulation studies. Traffic Eng.,
Mathematics Series, No. 14, (June 29, 33(5): 16-19, 48 (1963).
1951). May, A. D., Jr., and Wagner, F. A., Jr., Head-
way characteristics and interrelationships of
the fundamental characteristics of traffic
3.13 RELATED LITERATURE flow. Proc. Highw. Res. Board, 39: 524-54 7
(1960).
Counting Distributions
McNeil, D. R., and Morgan, J. H. T., Esti-
Haight, F. A., Counting distributions for re- mating minimum gap acceptances for merg-
newal processes. Biometrika, 52(3,4): 395- ing motorists. Transp. Sci., 2(3): 265-277
403 (1965). (1968).
Haight, F. A., Handbook of the Poisson Dis- Miller, A. J., Nine estimates of gap-acceptance
tribution. Wiley, New York (1967). parameters. Presented at Fifth International
RELATED LITERATURE 45
Symposium on Theory of Traffic Flow and Lewis, B. J., Platoon movement of traffic from
Transportation, Berkeley, Calif. an isolated signalized intersection. Highw.
Mitchell, H. R., Freeway accidents and s!ow- Res. Board Bull. 178 (1958), pp. 1-11.
moving vehicles. Traffic Eng., 37 (2): 22-25 Nemeth, Z. A., and Vecellio, R. L., Investiga-
(1966). tion of the dynamics of platoon dispersion.
Moskowitz, K., Waiting for a gap in a traffic Paper presented at Annual Meeting of High-
stream. Proc. Highw. Res. Board, 33: 385- way Research Board, Washington, D.C.
394 (1954). (January 1970).
Oliver, R. M., Distribution of blocks and gaps Pacey, G. M., The Progress of a Bunch of
in a traffic stream. Oper. Res., 10(2): 197- Vehicles Released from a Traffic Signal.
217 (1962). Res. Note No. RN/2665/GMP. Road Re-
Raff, M. S., A Volume Warrant for .Urban search Laboratory (Jan. 1956), mimeo.
Stop Signs. Eno Foundation (1950). Underwood, R. T., Traffic flow and bunching.
Raff, M. S., The distribution of blocks in an Austr. Road Res., 1(6): 8-24 (1963).
uncongested stream of automobile traffic. Wright, C. C., Some characteristics of traffic
J. Am. Stat. Assoc., 46: 114-123 (1951). leaving a signalized intersection. Transp.
Schuh!, A., Distribution of gaps between suc- Sci., 4(4): 331-346 (1970).
cessive vehicles. (Letter to Editor) Int.
Road Safety Traffic Rev., 41 (1958).
Suzuki, T., Some results in road traffic distri- Speed Distributions, etc.
butions. Oper. Res. Soc. Jap., 9 (1): 16-25
(1966). Duncan, N. C., A Method of Estimating the
Weiss, G. H., The intersection delay problem Distribution of Speeds of Cars on a Motor-
with gap acceptance function depending on way. Transport and Road Research Labo-
speed and time. Transp. Res., 1(4) :367- ratory Rep. 598 (1973).
371 (1967). Lefeve, B. A., Speed habits observed on a rural
Widermuth, B., Average vehicle headways at highway. Proc. Highw. Res. Board, 33:
signalized intersections. Traffic Eng., 33 (2) : 409-428 (1954).
14-16 (1962). Mori, M., Takata, H., and Kisi, T., Funda-
mental considerations on speed distributions
of road traffic flow. Transp. Res., 2(1):
Platoon Phenomena 31-39 (1968).
Ogawa, T., Fisher, E. S., and Oppenlander,
Athol, P., Headway groupings. Highw. Res. J. C., Driver behavior study-influence of
Rec. 72 (1965), pp. 137-155. speed limits on spot speed characteristics in
Buckley, D. J., Tomlin, J. A., and Minson, a series of contiguous rural and urban
W. G., Delays to traffic platoons. Proc. areas. Highw. Res. Board Bull. 341 (1962),
Austr. Road Res. Board, 3: 364-376 (1966). pp. 18-29.
Edie, L. C., and Foote, R. S., Analysis of Oppenlander, J. C., Bunte, W. F., and Kadakia,
single lane traffic flow. Traffic Eng., 34( 4): P. L., Sample size requirements for vehicu-
21-27 (1963). lar speed studies. Highw. Res. Board Bull.
Gerlough, D. L., Some problems in intersec- 281 (1961), pp. 68-86.
tion traffic control, pp. 10-27. In R. Herman Oppenlander, J. E., A theory on vehicular
(ed.) Proceedings of [First] International speed regulation. Highw. Res. Board Bull.
Symposium on the Theory of Traffic Flow. 341 (1962)' pp. 77-91.
Elsevier, New York (1961). Pahl, J., The effect of discrete time measure-
Grace, M. J., and Potts, R. B., Diffusion of ment on speed data. Paper presented at
traffic platoons. Proc. Austr. Road Res. Annual Meeting of Highway Research
Board, 1: 260-267 (1962). Board, Washington, D.C. (January 1971).
Grace, M. J., and Potts, R. B., A theory of the Rowan, N. J., and Keese, C. 1., A study of
diffusion of traffic platoons. Oper. Res., factors influencing traffic speeds. Highw.
12(2): 255-275 (1964). Res. Board Bull. 341 (1962), pp. 30-76.
Herman, R., Potts, R. B., and Rothery, R. W., Schumate, R. P., and Crowther, R. F., Vari-
Behaviour of traffic leaving a signalized ability of fixed-point speed measurements.
intersection. Traffic Eng. Control, 5(9): Highw. Res. Board Bull. 281 (1961 ), pp.
539-533 (1964). . 87-96.
46 STATISTICAL DISTRIBUTIONS OF TRAFFIC CHARACTERISTICS
Wagner, F. A., Jr., and May, A. D., Jr., Vol- Dubey, S. D., On some permissible estimators
ume and speed characteristics at seven study of the location parameter of the Weibull and
locations. Highw. Res. Board Bull. 281 certain other distributions. Technometrics,
(1961), pp. 48-67. 9(2): 293-307 (1967).
Wortman, R. H., A multivariate analysis of Greenwood, J. A., and Durand, D., Aids for •
vehicular speeds on four-lane rural high- fitting the gamma distribution by maximum
ways. Highw. Res. Rec. 72 (1965), pp. likelihood. Technometrics, 2( 1): 55-65
1-18. ( 1960).
Harter, H. L., Maximum-likelihood estimation
of the parameters of a four-parameter gen-
Gap Acceptance
eralized gamma population from complete
Ashworth, R., and Green, B. D., Gap accept- and censored samples. Technometrics, 9 ( 1 ) :
ance at an uncontrolled inte.rsection. Traffic 159-165 (1967).
Eng. Control, 7: 676-678 (1966). Krishnaiah, P. R., and Rizvi, M. H., A note on
Ashworth, R., The capacity of priority-type moments of gamma order statistics. Tech-
intersections with a non-uniform distribution nometrics, 9(2): 315-318 (1967.
of critical acceptance gaps. Transp. Res., Sasieni, M., Yaspan, A., and Friedman, L.,
3(2): 273-278 (1969). Operations Research-Methods and Prob-
Bissell, H. H., Traffic Gap Acceptance from a lems. Wiley, New York (1959), pp. 32,
Stop Sign. Unpublished graduate research 56-57.
report. University of California, Berkeley
(1960).
Normal Distribution
Dart, 0. K., Jr., Left-turn characteristics at
signalized intersections on four-lane arterial Cohen, A. C., Jr., Estimation in mixtures of
streets. Highw. Res. Rec. 230 ( 1968), pp. two normal distributions. Technometrics,
45-59. 9(1): 15-28 (1967).
Drew, D. R., Gap acceptance characteristics
for ramp-freeway surveillance and control.
Highw. Res. Rec. 157 ( 1967), pp. 108-135. Lognormal Distribution
Hawkes, A. G., Gap-acceptance in road traffic. Joffe, A. D., and Sichel, H. S., A chart for
J. Appl. Probab., 5 ( 1): 84-92 ( 1968). sequentially testing observed arithmetic
means from lognormal populations against
a given standard. Technometrics, 10(3):
Exponential Distributions
605-612 (1968).
Rider, P. R., The method of moments applied
to a mixture of two exponential distributions.
Ann. Math. Stat., 32(1): 143-147 (1961); Distributions-General
reprinted by NTIS as AD 701 962. Cohen, A. C., Jr., Estimation of the Poisson
Tallis, G. M., and Light, R., The use of frac- parameter from truncated samples and from
tional moments for estimating the parameters censored samples. J. Am. Stat. Assoc., 158-
of a ·mixed exponential distribution. Tech- 168 (Mar. 1954).
nometrics, 10(1): 161-175 (1968). Falls, L. W., Estimation of parameters in
compound Weibull distributions. Techno-
metrics, 12(2): 399-407 (1970).
Gamma Distribution and Related
Haight, F. A., Index to the distributions of
Distributions
mathematical statistics. J. Res. Natl. Bur.
Bain, L. J., and Antle, C. E., Estimation of Stand. B. Math. Math. Phys., 65B(l): 23-60
parameters in the Weibull distribution. (1961).
Technometrics, 9(4): 621-627 (1967). Hammersley, J. M., and Morton, K. W., The
Choi, S. C., and Wette, R., Maximum likeli- estimation of location and scale parameters
hood estimation of the parameters of the from grouped data. Biometrika, 41: 296-
gamma distribution and their bias. Techno- 301 (1954).
metrics, 11 ( 4) : 683-690 ( 1969). Mann, N., Estimation of Location and Scale
Clarke, A. B., and Disney, R. L., Probability Parameters Under Various Models of Cen-
and Random Processes for Engineers and soring and Truncation, Rocketdyne Div.,
Scientists. Wiley, New York (1970), pp. North American Rockwell Corp. (February
186-189. 1970); NTIS No. AD 707 868.
PROBLEMS 47
49
50 TRAFFIC STREAM MODELS
Flow,q
Figure 4.1 A traffic stream model where u = q/k represents the surface of admissible traffic
stream models.
This model has shortcomings in that it does Munjal 5 •6 have proposed a family of models
not represent zero speed at high concentrations. of the form
In sections 4.3.2 and 4.3.2.1 it is shown that
( 4.4)
for Eq. 4.2 km= k/ e, whereas for Eq. 4.3
um=urle. where n is a real number greater than zero.
Three conditions of this model (n < 1, n = 1,
4.2.3 Generalized Single-Regime Speed- n > 1) are illustrated in Figure 4.6. It will be
Concentration Models seen that for n = 1 the relationship reduces to
Greenshields' model.
Recently, Pipes and Munjal 5 • 6 have de-
scribed a general family of speed-concentration 4.2.3.2 Drew Models. Drew 7 has pro-
models of which the linear model is a special posed a family of models of the form
case. Drew 7 has described a family of models
of which Greenberg's logarithmic model is a du- u
- k<n-1)/2
dk - m ( 4.5)
special case. Other families result from car-
following analyses.
where n is a real number. When n= -1, Eq.
4.2.3.1 Pipes-Munjal Models. Pipes and 4.5 can be solved to yield Greenberg's model.
SPEED-CONCENTRATION MODELS 51
qm
tan 8 = - = u
~ m /
... qm r-----:::;:;;:11,,..ic::::::------1- - - - - - - - - - - - - - - - 7 (
I /
--
..c.
..c.
Q,)
>
I
I /
//
3: I //
~
u.. I /
I //
kml -------7(
/I
I / I
I / I
I // I
! / I
I /
/ II
I / I
I /
I /
I
_ _ _ _ _ __J _ _ _ _ _ _
I //
--J..".:._ ____ - -
I I
I I
I I
I I
I I u
um
-------~------i----------:---------41
~
a. I -a
E I s
"C I ~
~
Q,)
Q,)
a. I
en I
0
0 0
Figure 4. 7 illustrates the results for three values normal curve as a model of speed-concentra-
of n ( - 1, 0, + 1). tion using the form
4.2.3.3 Car-Following Models. In Chap- u = llr e-1/2Hk/km i• ( 4.6)
ter 6 the methods of car-following analysis are
introduced, including parameters m and f. 4.2.4 Multiregime Speed-Concentration
Figure 4.8 shows how different speed-concen- Models
tration models can be obtained by manipulat-
ing these parameters. Figure 4.9 lists the more As noted in Section 4.2.2, Greenberg's
important models of this family. model is useful for high concentrations but not
for low concentrations; conversely, Under-
4.2.3.4 Bell-Shaped Curve Model. Drake wood's model is useful for low concentrations
et al. 9 have proposed use of the bell-shaped or but not for high concentrations. Additional
52 TRAFFIC STREAM MODELS
weight is added to this argument by noting finding that cars have a different speed and
that the high-congestion portion of the model acceleration after a slowdown caused by con-
usually represents cars accelerating from. a jam gestion.
condition and taking into account Forbes' 10- 12
4.2.4.1 Edie's Model. Edie 13 has de- •
scribed a model that is a composite of Eqs.
4.2 and 4.3, where Eq. 4.3 is at low concen-
40 trations and Eq. 4.2 at high concentrations.
When normalized speed is plotted against nor-
malized concentration, the two models become
tangent in the midrange of concentration
(Figure 4.10).
30
Inside Lane
.c R = 0.96
a.
E
20
"'O
~
a.
(/)
10
40
•
30
Outside Lane
R = 0.97
.c
a.
E
~ 20
Q)
a.
(/)
10
•
o~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
0 20 40 60 80 1-00 120
Density (veh/mile)
Figure 4.3 Study showing high correlation coefficient between field data and linear model for speed
vs. density. 2
SPEED-CONCENTRATION MODELS 53
50
40
..c
a.
..§. 30
i
·u 20
0
a;
>
10
~ S = 53.2e·D/67 Q.)
a.
Q.) (/)
a. r2 = 0.90
(/)
-0
c: Q.)
N
"'
Q.)
10 iii
:::? E
Q.)
(J
....
0
"'
a.
(/)
z 0.60
"'"'
~
5 c: 0.40
0
·;;;
4 c:
Q.)
-~ 0.20
0
3
.....
•
40 80 120 160 200
--
::J
::J
0
0.20 0.40 0.60 0.80 1.00
k/kj =Dimensionless (Normalized)
Density, D (veh/mi) Concentration
Figure 4.5 Underwood's• speed-concentrotion plot. for Figure 4.6 -Family of speed-concentration models pro-
Merritt Parkwoy. posed by Munjal and Pipes. 6
54 TRAFFIC STREAM MODELS
3.0
E
--
::J
::J
Linear Model (n = + 11
Parabolic Model (n=O)
.~
(,)
2.0 Exponential Model (n = -11
0
a;
>
~
(ij 1.0
...0
E
z
sumed that there is a fixed upper limit to and Keller 15 have discussed methods for fit-
speed. He combined this assumption with ting multiregime models.
Greenberg's model to produce the result shown
in Figure 4.12. 4.2.5 Summary
4.2.4.4 Fitting Multiregime Models. May Figure 4.13 summarizes several types of
speed-concentration models.
20
Density (veh/mile)
Figure 4.8 Example of car-following models 8 of speed-concentration, showing effett of vary-
ing parameter I.
FLOW-CONCENTRATION RELATIONSHIPS 55
I~ I= 0
m =o
-
s k kj
s = a constant
k.
I= 1 u= umln (+) -
1/2
I= 1.5 u=u (1-(~) J -
f k.
J
l (k/k )2
I= 3 u=u e-2 m
f
1.0
4.3 FLOW-CONCENTRATION
RELATIONSHIPS
0.8
Early studies of highway capacity followed
two principal approaches. Some investigators
examined speed-flow relationships at low con- '+-
0.6
centrations; others discussed headway phe-
nomena at high concentrations. Lighthill and
--
::i
: :i
k k
ki k1 ki
Uf u=umln(k/k)
k . k4 k
k2 k3 kj kj
V Underwood VI Edie
k k
km km ki
\
\
\ • Fixed speed, first series
... Fixed speed, second series (mixed traffic)
\•
40
.i:;
c.
E
"'
alCl>
c.
(/')
c:
"'
Cl>
:::'E
20
.......... .....,__ .
01--~~~-'--~-~--1.~~-~-L~~~~-'-~-----=="-"'--....._~-~--'
0 100 200 300
Concentration (veh/mile)
Figure 4.14 Speed-concentration curve for straight track at British Road Research
laboratory .16
2000
2
~
~
~
//
1500 if
!!"
~
J
,_ ...:-
--"'
.i:;
r
u 1000
.,._o
~"'
(;-
I
~ i~ ~IE
0
LL c: //
QJ-c:
Ef.r::r
.QI?
u. "'
>
.~I~
~f
ti l-0
Ill (\i
500 X 11
~~
'"I
~ O"E
_ Critica_!_ _ __j
km=80 veh/mile
0.4
1500
500
OL...-~-'-~--'-~-L~~'--~-'-~--'-~--'-~---''--~"--~~
0 50 100 150 200 250
k (cars/mile)
Figure 4.18 Discontinuous flow-concentration curve. 13
.....
.c
.__
.c
Cl>
>
Cl>
E
::J
0
>Cl>
c
"'
_J
Q,
Cl>
.....
.c 1000
I-
(Liq/ Lik) 1
Speed of Vehicles in
Bottleneck, u 2
Concentration, k
Figure 4.20 Flow-concentration curves for main highway and bottleneck."
most basic of such control systems limit the speed and concentration values, there will be
entrance rate when the freeway flow is near a second point of zero flow, corresponding to
capacity. Concentration has been found to be a zero speed (maximum concentration). Thus,
good estimator of the flow I capacity ratio. This regardless of the shape of the speed-concentra-
is partly because concentration increases in a tion curve, the speed-flow curve will have one
monotonic manner during periods of increasing point at the origin and one point on the speed
flow; that is, concentration does not decrease axis at the maximum value of speed. Between
until after the end of the peak period. To con- zero and maximum speeds, the diagram will
trol the facility represented by Figure 4.19, it form some type of loop toward maximum flow.
would be necessary to keep the combined con- If the speed-concentration curve is a straight
centration in the three lanes at or below about line, as suggested by Greenshields, the resulting
150 vehicles/ mile. At present, there is no good speed-flow curve is a parabola (Figure 4.2).
on-line instrument to measure freeway concen- Other shapes are associated with other speed-
tration (density) directly (although in certain concentration curves. (Creighton 20 has pre-
tunnels it is possible to provide direct concen- sented a qualitative description of the way in
tration measurement 43 ). Thus, lane occupancy which various portions of the speed-flow dia-
is used as an estimate of lane density or concen- gram come about.)
tration. References 37-42 provide advanced Some early investigators (e.g., Walker 26 )
methods of estimation. postulated a linear relationship between flow
and speed out to maximum flow, with a curvi-
linear segment between maximum flow and the
4.4 SPEED-FLOW MODELS
origin (Figure 4.21). An extreme case is the
As pointed out in section 4.2, once a model developed by the (British) Road Re-
speed-concentration model has been deter- search Laboratory (Figure 4.22). Here the
mined, a speed-flow model can be determined speed is taken as constant for a substantial
from it. In all realistic speed-concentration range of flow that finally breaks to a linear de-
models, the free-flow speed at zero concentra- crease of speed with increasing flow. Here
tion is the maximum attainable speed. (See road width was an important parameter.
Figure 4.13.) Thus, the highest point on the In the experimental speed-flow curves of
speed-flow curve will be the point at free-flow Figures 4.23, 4.24, 4.25 roadway radius of
speed and zero flow. Inasmuch as the flow curvature is shown to be an important param-
values are the products of the corresponding eter.
62 TRAFFIC STREAM MODELS
50
•
-a. 40
E
u
:Ero
.= 30
<i:
.....0
-0
Cl>
~ 20
(f)
Cl>
O'l
ro
....
Cl>
>
<(
10
Figure 4.21 Speed-flow model adopted by U.S. Bureau of Public Roads."' Relation between average speed of
traffic and traffic volume on a two-lane highway having a possible capacity of 2,000 vehicles per hour under fa-
vorable operating conditions.
The Highway Capacity Manual 41 defines proposed the use of speed-flow curves for
"level of service" and suggests its use in the establishing levels of service for traffic flow
design and evaluation of facilities. May 28 has (Figure 4.26).
30
.r:.
a.
E
-0
fil 20
a.
(/)
O'l
c:
c:
c:
:J
a: 10
c:
ro
Cl>
:"E Mean Road
Width (ft)
0 '--~~~~'---~~~~.1-~~~~-'-~~~~-'-~~~~--'-~~~~--'
0 400 800 1,200 1,600 2,000 2,400
60
' ..........
..........
" ..........
........
........
•
40
::cQ.
E
"O
Q.l
Q.l •
Q.
Cf)
c
"'"'
:2
20 •
Flow (veh/hr)
Figure 4.23 Experimenta I speed-flow relationship on test track with straight road-
way section.16
4.5 TRAVEL TIME RELATIONSHIPS where ti= total trip time for the ith car;
n; = the ith car to arrive on the on-
Although much attention has been given ramp queue;
by various investigators to speed-c~ncentra q 0 =the average arrival rate at the on-
tion, speed-flow, and flow-concentration rela- ramp queue;
tionships, relatively little attention has been q 1 =the average departure rate from
devoted to travel time-flow relationships. Roth- the on-ramp queue;
rock and Keefer 29 empirically demonstrated q 2 = the average departure rate from
the form of travel time-flow relationships (Fig- the off-ramp queue;
gure 4.27). Guerin 30 has proposed emp~rical
u = effective steady-state velocity of the
boundary curves representing travel tJme-
N cars on the freeway; and
volume (flow) and travel time-density (con-
centration), including percentage contours, as d= distance traveled on the freeway.
shown in Figure 4.28. Weinberg et al. 31 and Haase summarizes all trips for the facility and
Greenberg and Crowley 32 have also discussed produces curves such as Figure 4.29.
travel time relationships.
qi
1 J+_:!__+ni[-1 __
t;=n;[-l _ _
qo U
1
qz qi
J (4.11) order to minimize travel time. He has de-
veloped a model that includes the fraction of
64 TRAFFIC STREAM MODELS
30
:ca.
E 20
"O
...
Q)
Q)
a. 0
en
c
ca
Q)
10
~
30
.!:.
a. 0
E 20
"O
Q)
Q)
a.
en
c 10
ca
Q)
the central business district ( cso) area devoted A plot of this model is given in Figure 4.30.
to streets: Wardrop, 27 in his classic paper, postulated
two principles concerning travel time by regu-
T=t/2+ (7.409/ 10") A 1N[l -n/33tfA 1!2y:i lar users over several alternate routes. In such
( 4.12) situations regular users will distribute them-
selves over the various routes so that:
where T= average journey time measured
from the time the first vehicle (1) The journey times on all routes
enters the cso; actually used are equal, and less than
those which would be experienced by a
t = period over which entries to the single vehicle on any unused route.
cso are spread; (2) The average journey time is a
n = number of vehicles entering cso minimum.
during period t;
In transportation planning, the expression
A= area of cso in square feet; and
f =fraction of cso area devoted to
streets. ( 4.13)
TRAVEL TIME RELATIONSHIPS 65
1.0
0.9
0.8
-
I
I
I
0.7 I
:i I
---
:i
~- 0.6 ~/
I
\ I
Cl>
a.
(/)
0.5
*''I I
I
~ I
.,.ri. ----
----- -- --
N
co 0.4 "1 --
E
0
z 0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Normalized Volume, q/qm
Figure 4.26 Use of speed-flow (speed-volume) curve to establish levels of service. 28
4 4 4
6 - Min Intervals 12 - Min Intervals 18 - Min Intervals
~
Cl> 3 3 3
E
i= 2 2 2
a.
·;:
f-
Oo'---'---'40'---'---'8-0--1.~1~2-0_.
4 4 4
30-Mln~
24- Min Intervals 60 - Min Intervals
Cl> 3 : 3
E
i=
a.
·;:
f-
2
0..__._~_,_~..__..~_,_~..___.
2
_ _)
80 160 240 0 100 200 300 200 400 600
Volume Volume Volume
Figure 4.27 Form of empirical travel time-flow relationships for observation intervals of various durations.29
66 TRAFFIC STREAM MODELS
is sometimes used when estimating travel time relationship appears to be most fundamental
as a function of volume, 36 in that drivers appear to adjust their speeds ac-
cording to the concentration of the traffic
where T= the travel time for a particular fl.ow,
around them. The fl.ow-concentration relation-
•
q;
ship is generally the most useful because it uni-
qm =the maximum flow or capacity of
fies various theoretical ideas, and it provides
the facility; and
relationships for traffic control activities.
T min= the minimum travel time.
A variety of speed-concentration, ftow-
concentration, and speed-flow models has been
4.6 SUMMARY presented. The model(s) selected for a specific
Although the speed-concentration-flow re- application must depend on the particular situ-
lationship is three-dimensional in reality, it is ation and purpose of the analysis. In the ab-
often treated by using one or more of its two- sence of a strong reason to the contrary, the
dimensional orthographic projections. From a simplest appropriate model should be consid-
causation v.iewpoint, the speed-concentration ered. In Chapter 7 an example is presented of
traffic behavior at a signal in which solutions
are obtained using both linear and logarithmic
speed-concentration models. Which result is
better depends on local characteristics and
judgment.
4.7 REFERENCES
Ql
1. Greenshields, B. D., A study of traffic
E
i= capacity. Proc. Highw. Res. Board,
Q; 14:448-477 (1934).
> 2. Huber, M. J., Effect of temporary bridge
~
I- on parkway performance. Highw. Res.
Board Bull. 167 (1957), pp. 63-74.
3. Greenberg, H., An analysis of traffic flow.
Oper. Res., 7 (1):79-85 (1959).
4. Underwood, R. T., Speed, volume and
density relationships, pp. 141-187. In
Density Quality and Theory of Traffic Flow.
Bureau of Highway Traffic, Yale Uni-
versity, New Haven, Conn. (1961).
5. Pipes, L. A., Car following models and
the fundamental diagram of road traffic.
Transp. Res., 1 (1):21-29 (1967).
6. Munjal, P. K., and Pipes, L. A., Propaga-
tion of on-ramp density perturbations
Ql
on unidirectional and two- and three-
E lane freeways. Transp. Res., 5 (4) :241-
i= 255 (1971).
Q;
> 7. Drew, D. R., Deterministic aspects of free-
i="' way operations and control. Highw.
Res. Rec. 99 (1965), pp. 48-58.
8. May, A. D., Jr., and Keller, E. M., Non-
integer car-following models. Highw.
Res. Rec. 199 (1967), pp. 19-32.
9. Drake, J., Scofer, J., and May, A. D., Jr.,
A statistical analysis of speed-density
Volume hypotheses. Proceedings of Third Inter-
Figure 4.28 Guerin's travel time relationships. 30 Per- national Symposium on Theory of Traf-
centage indicated is percentage of observations of fic Flow (New York, June 1965). Pub-
travel time that are expected lo be below the curve. lished as Vehicular Traffic Science,
REFERENCES 67
Velocity, u (mph)
70 60 50 40 30 20 15
60~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
d = 16 mi
Max cars/min
14
12
40
10
c.E
I- 8
o.i
E
i=
6
20
40 80 120
Density, p (cars/mi)
5 10 15 20 25 28 29.4 28
American Elsevier, New York (1967), 11. Forbes, T. W., Human factor considera-
• pp. 112-117. tions in traffic flow theory. Highw. Res.
10. Forbes, T. W., Zagorski, H. J., Hols- Rec. 15 (1963), pp. 60-66.
houser, E. L., and Deterline, W. A., 12. Forbes, T. W., and Simpson, M. E.,
Measurement of driver reactions to tun- Driver-and-vehicle response in freeway
nel conditions. Proc. Highw. Res. deceleration waves. Transp. Sci., 2 ( 1) :
Board, 37:345-357 (1958). 77-104 (1968).
68 TRAFFIC STREAM MODELS
0.07
0.06
0.05
~,
Comparative Value
fA
~ r:: of-
II- n
0.04
0.03
0.02
111 I I
5678 9 10, 11 12 13 14 15 16
Speed (mph)
Figure 4.30 Plot of Smeed's 35 model reloting minimization of travel time to spread of period during
which travel takes place.
13. Edie, L. C., Car-following and steady-state traffic flow ~odels. Proceedings of
theory for noncongested traffic. Oper. Fourth International Symposium on
Res., 9 (1):66-76 (1961). Theory of Traffic Flow (Karlsruhe, Ger-
14. Dick, A. C., Speed/flow relationships many, June 1968). Published as Beitrage
within an urban area. Traffic Eng.
Control, 8 (6) :393-396 (1966). zur Theorie des Verkehrsflusses, Stras-
15. May, A. D., Jr., and Keller, H. E. M., senbau und Strassenverkehrstechnik
Evaluation of single and multi-regime (1969), vol. 86, pp. 37-48.
REFERENCES 69
16. Wardrop, J. G., Experimental speed/flow tion. Highw. Res. Board Bull. 156
relations in a single lane. Proceedings (1957), pp. 1-13.
of the Second International Symposium 30. Guerin, N. W., Travel time relationships,
on the Theory of Traffic Flow (London, pp. 69-103. In Quality and Theory of
1963). OECD (1965), pp. 104-119. Traffic Flow. Bureau of Highway Traf-
17. Lighthill, M. J., and Whitham, G. B., On fic, Yale University, New Haven, Conn.
·kinematic waves: II. A theory of traffic (1961).
flow on long crowded roads. Proc. R. 31. Weinberg, M. I., Deleys, N. J., and Schne-
Soc., Lond., Ser. A., 229 (1178): 317- berger, R. F., Surveillance and Ways
345 (1955); reprinted in An Introduc- and Means of Communicating with
tion to Traffic Flow Theory. Special Re- Drivers. NCHRP Report 28, Highway
port 79, Highway Research Board, Research Board, Washington, D.C.
Washington, D.C. (1964), pp. 7-35. (1966), p. 37.
18. Haight, F. A., The volume, density rela- 32. Greenberg, I. and Crowley, K. W., Hol-
tion in the theory of road traffic. Oper. land Tunnel study aids efficient increase
PM Q (A\ ·C:.7?_C:.71 (lQ/;0)
_.._......_ ..... ,:>,' V \ -T j • . J I ,,._, ..J I - '
\..I..-' ' - ' 'V / • of tube's use. Traffic Eng., 35(6) :20-
19. Haight, F. A., Mathematical Theories of 22, 38-41 (1965).
Traffic Flow. Academic Press, New 33. Haase, R. H., Decreasing Travel Time for
York (1963), pp. 67-95. Freeway Users. The Rand Corporation,
20. Edie, L. C., and Foote, R. S., Traffic flow Memorandum RM-3099-FF, October
in tunnels. Proc. Highw. Res. Board, 1962.
37:334-344 (1958). 34. Haase, R. H., Increasing freeway utiliza-
21. Edie, L. C., and Foote, R. S., Experiments tion for motorists. Paper presented at
on single-lane flow in tunnels. Proceed- 34th national meeting of Operations Re-
ings of [First] [International] Sympo- search Society of America, November
sium on the Theory of Traffic Flow 6-9, 1968, Philadelphia.
(Warren, Michigan, 1959). Elsevier, 35. Smeed, R. J., Some circumstances in
New York (1961), pp. 175-192. which vehicles will reach their destina-
22. Gazis, D. C., Herman, R., and Potts, R. B., tions earlier by starting later. Transp.
Car-following theory of steady-state Sci., 1 (4) :308-317 (1967).
traffic flow. Oper. Res., 7 (4) :499-505 36. Bureau of Public Roads, Federal Highway
(1959). Administration, U.S. D.0.T., Urban
23. Haynes, J. J., Some considerations of Transportation Planning, General In-
vehicular density on urban freeways. formation and Introduction to System
Highw. Res. Rec. 99 (1965), pp. 59-80. 360 (1970).
24. Drew, D. R., and Keese, C. J., Freeway 37. Gazis, D. C., and Knapp, C. H., On-line
level of service as influenced by volume estimation of traffic densities from
and capacity characteristics. Highw. time-series of flow and speed data.
Res. Rec. 99 (1965), pp. 1-39. Transp. Sci., 5 (3) :283-301 (1971).
25. Creighton, R. L., Deriving analytically the 38. Gazis, D. C., and Szeto, M. W., Estima-
relationships between speed and volume tion of traffic densities at the Lincoln
as observed on signalized roads in urban Tunnel from time series of flow and
areas. C.A.T.S. Res. News, 1 (11): speed data, pp. 151-165. In P. Lewis
6-11 (1957). (ed.) Stochastic Point Processes. John
26. Walker, W. P., Speed and travel time Wiley & Sons, New York (1972).
measurements in urban areas. Highw. 39. Szeto, M. W., and Gazis, D. C., Applica-
Res. Board Bull. 156 (1957), pp. tion of Kalman filtering to the surveil-
27-44. lance and control of traffic systems.
27. Wardrop, J. G., Some theoretical aspects Transp. Sci., 6 (4) :419-439 (1972).
of road traffic research. Proc. Inst. Civil 40. Gazis, D. C., and Foote, R. S., Surveil-
Eng., Part II, 1 (2):325-362 (1952). lance and control of tunnel traffic by an
28. May, A. D., Jr., Discussion to paper by on-line digital computer. Transp. Sci.,
Drew and Keese. Highw. Res. Rec. 99 3 (3) :255-275 (1969).
(1965), pp. 39-43. 41. Gazis, D. C., and Szeto, M. W., On the
29. Rothrock, C. A., and Keefer, L. A., design of density measuring systems for
Measurement of urban traffic conges- roadways. Paper presented at Highway
70 TRAFFIC STREAM MODELS
Research Board Annual Meeting, Wash- Franklin, R. E., On the flow-concentration re-
ington, D.C. (January 1974). lationship for traffic. Proceedings of the Sec-
42. Nahi, N. E., and Trivedi, A. N., Recursive ond International Symposium on the Theory
estimation of traffic variables: section of Traffic Flow (London, 1963). OECD •
density and average speed. Transp. Sci., (1965), pp. 120-128.
7 (3) :269-286 (1973). Keefer, L. E., Speed-density study. Chicago
43. Foote, R. S., Instrumentation development Area Transportation Study. C.A.T.S. Res.
and contributions from traffic flow News, 1 (13):6-10 (1957).
theory. Highw. Res. Rec. 368 (1971), Leutzbach, W., Testing the applicability of the
pp. 1-8. theory of continuity on traffic flow at bottle-
44. Highway Research Board, Highway Ca- necks. Proceedings of Third International
pacity Manual. Special Report 87. Symposium on Theory of Traffic Flow (New
Washington, D.C. (1965), pp. 7-8, York, June 1965). Published as Vehicular
78-87. Traffic Science, American Elsevier, New
York (1967), pp. 1-13.
4.8 RELATED LITERATURE May, A. D., Jr., and Wagner, F. A., Jr., Head-
way characteristics and interrelationships of
Campbell, E. W., Speed-volume relationship fundamental characteristics of traffic flow.
determined by probability of signal failure. Proc. Highw. Res. Board, 39:542-547
Chicago Area Transportation Study. C.A. (1960).
T.S. Res. News, 1 (16): 12-16 (1957). Walker, W. P., Speed and travel time measure-
Carstens, R. L., and Ring, S. L., Pedestrian ment in urban areas. Highw. Res. Board
capacities of shelter entrances. Traffic Eng., Bu!. 156 (1956), pp. 27-44.
41 (3): 38-43 (1970). Welding, P. I., Time series analysis applied to
Edie, L. C., Foote, R. S., Herman, R., and traffic flow. Proceedings of the Second Inter-
Rothery, R., Analysis of single lane traffic national Symposium on the Theory of Traffic
flow. Traffic Eng., 33 (5) :21-27 (1963). Flow (London, 1963). OECD (1965), pp.
Forbes, T. W., Speed, headway, and volume re- 60-72.
lationships on a freeway. Proc. Inst. Traffic Underwood, R. T., Some aspects of the theory
Eng., 22: 103-126. of traffic flow. Austr. Road Res., 1 (2):
Forbes, T. W., Mullin, J. J., and Simpson, M. 35. (1962).
E., Interchange spacing and driver behavior Underwood, R. T., Traffic flow models. Traffic
effects on freeway operation. Proceedings of Eng. Control, 5 (12) : 699-701 (1964).
Third International Symposium on Theory
of Traffic Flow (New York, June 1965).
4.9 PROBLEMS
Published as Vehicular Traffic Science.
American Elsevier, New York (1967) , pp. 1. For the speed-concentration model of
97-108. Figure 4.12, compute the coordinates and plot
Foster, J., An investigation of the hydrody- curves of speed-flow and flow-concentration
namic model for traffic flow with particular on rectilinear graph paper.
reference to the effect of various speed- 2. From the flow-concentration curve of
density relationships. Proc. Austr. Road Res. Figure 4.19, use radius vectors to obtain speeds
Board, 1 :229-257 (1962). and plot a speed-flow curve.
Chapter 5
DRIVER INFORMATION PROCESSING
CHARACTERISTICS
5.1 INTRODUCTION The tracking or steering subtasks can be
described in terms of a servo system (Figure
Model representation of traffic behavior 5.1) and several authors have proposed various
indirectly deals with human behavior. Several models to describe the driver's actions as part of
traffic flow models to be discussed in later chap- the servo loop.~-n To accomplish his steering
ters contain parameters used to account for function the driver attempts:
various characteristics of the driver in the
driver-vehicle system. Some models deal with 1. To select a reference (point, line,
traffic as a deterministic phenomenon even angle, curve, etc.) from which to determine
though the driver portion of the system, at least, the vehicle misalignment or deviation (i.e.,
is highly stochastic. error).
Although the whole field of human factors 2. To detect such errors.
in traffic could, of course, be the subject of a 3. To establish an error criterion.
very lengthy treatise, such as that by Forbes, 10 4. To respond to the detected error (via
it is hoped that this chapter will pnwide some the steering wheel) in such a manner as to
insight into the way drivers receive and use in- maintain the vehicle within the established
formation and that this knowledge can then be criterion limits throughout the duration of the
applicable in various traffic flow models. It steering task.
should be borne in mind that driver actions are
highly stochastic, but that it is often possible to The acceleration control subtask consists
represent stochastic data either by expected of detecting differences in velocity and/ or spac-
values or by worst cases. ing and taking actions that will prevent unsafe
The discussion starts with a brief examina- conditions and fulfill both the driver's goal of
tion of the driving task and its information re- proceeding at a particular speed and such other
quirements. Then the ways in which a driver goals as he may have.
receives and processes information are ex-
amined. Finally, several miscellaneous items of
driver behavior are considered. 5.2.1 Driver Information Needs
The information needs of the driver vary
5.2 NATURE OF THE DRIVING TASK with the portion of the trip and the immediate
maneuver to be accomplished. The trip may
The principal goal of the driver is to guide start with a planning phase in which macro-
his vehicle from origin to destination in a safe information provided by maps, weather reports,
manner. He may have some additional goals and road conditions is required. Once the driv-
such as arriving at his destination at the earliest ing has started, however, information comes
possible time; he may also have certain goals from observations of the roadway, other traffic,
concerning the environment through which he signs and signals, and instruments. Generally,
passes during his trip. At various points in cues to the driver may be visual, auditory,
the trip there may be special subtasks (e.g., tactile, kinesthetic, and even olfactory, but by
parking). 1 far the most important sense is visual. This is
The task of accomplishing the driver's followed by hearing, the ability to sense ac-
goals may be broken down into categories of celerations, and the ability to sense vibrations.
action: perception, judgment, decision, and Some characteristics of these senses are given
control. 2 • 41 The driver's control actions are in Table 5.1. (At one time auditory cues were
limited to control of acceleration (braking and an important source of information to the
accelerating) and control of heading (steering driver. However, with the present trend to
or tracking). make cars very quiet and to drive with the
71
72 DRIVER INFORMATION PROCESSING CHARACTERISTICS
Attitudes,
Personalities,
Motivation, etc.
MONITOR I
Processes ar
Sets Criteri;
Evaluates p,
Initiates Dr
Road
Data Driver Data Data DECISION-MAKING
Display Selective -
Sensor {Action Selection)
Kinematics Properties Attention
and including including
Geometry Sensing Perception
Other Thresholds
Vehicles Thresholds
---- -- -, I
L..--
Figure 5.1 Generalized block diagram of the car-driver-roadway system (adapted from Rockwell 6).
windows closed and air conditioning on (and 5.3 HUMAN RESPONSE TO STIMULI
perhaps a radio or tape player operating),
The human response to various stimuli
many auditory cues cannot be heard.)
varies with the type of stimulus and the level
The behavior of the driver can be likened
of the stimulus (intensity and/ or frequency)
to the observer who must scan several displays.
with respect to the related threshold value.
The driver, in addition to watching the road-
Furthermore, because each stimulus results in
way, must watch the car ahead and the cars
the transmission of certain information to the
in adjacent lanes; he must scan one or more
brain, response may be moderated by the simul-
rearview mirrors and from time to time must
taneous receipt of other information. As in any
check the instrument panel. At times it is neces-
instrument system, the human system contains
sary to actively scan the roadside area for di- an inherent transmission delay between the
rectional signs and other information such as stimulus and the response. Also, a finite time
landmarks. In many situations the driver must is required to process information. The proc-
perform a filtering function to extract the in- essing time is added to the transmission time to
formation needed from the surrounding fea- create a total reaction time to the various
tures (visual noise). Thus, it is important to stimuli.*
study the ways the driver processes and re-
sponds to the information at hand. ''' Private communication from S. F. Hulbert.
HUMAN RESPONSE TO STIMULI 73
Stresses Experience,
Skill
; FUNCTION
nterprets Data
lrmance
·Actions
Adaptation
lcoMMANDSETl
I Driver Actions Steering
I Command VEHICLE
Error DRIVER Brakes
Desired Vehicle I DYNAMICS CONTROL
DYNAMICS
Motions I + Throttle
Wind
VEHICLE
DISTURBANCE
Road Irregularities DYNAMICS
.thetic Cues
5.3.1 Simple Response Models cent of the usual range of sensory perception.
Based on Weber's law, Fechner suggested
The ability of an observer to detect a stim-
a generalization:
ulus is often stated by giving the threshold for
the specified stimulus." The (absolute) thresh- Magnitude of sensation
old may be defined as that amount of energy = k log (magnitude of stimulus)
that will result in the correct identification of
the signal 50 percent of the time. More recent A more accurate function was developed by
research has used the "detection-effectiveness Stevens.-'"· ""
parameter." Another (relative) measure of ..y=k(rp-rpo)"
stimulus is the "just noticeable difference"
(JND). This has led to Weber's law, which may where '¥=sensation magnitude; ¢=stimulus
be stated t magnitude; k =a constant; and n =an exponent
for a given sense area (e.g., vision, hearing).
JND = D.s Is= constant
• where s is the previous stimulus energy value 5.3.2 Driver as Sampled-Data System
and D.s is the change in stimulus energy level.
Experiments have shown that this relationship In observing any object to extract informa-
is a good approximation over about 99.9 per- tion, it is necessary for a person to fixate his
view on one portion at a time. Thus, normal
t Private communication from. T. W. Forbes. viewing is a process of sampling data. Further-
74 DRIVER INFORMATION PROCESSING CHARACTERISTICS
TABLE 5.1 Human Senses of Principal Importance to Drivers a
Hearing Ear Some amplitude and IX 10-" Approx. At 2000 Hz With pure tones
frequency varia- ergs/cm 2 1014 times Approx 325 about 3-5
tions of pressure threshold discernible identifiable
of surrounding levels levels
media (e.g., air)
Interrupted white
noise
more, because the driving task requires observ- rectly and process all information presented to
ing not only the roadway but also other targets them. Drivers, however, are limited in their
and displays, the sampling nature is increased. capacity for processing information. If the rate
Figure 5.2 shows histograms of fixations in of information presentation exceeds this capac-
normal traffic and while passing. Bekey 13 has ity, the excess will not get processed and may
investigated some of the phenomena associated even confuse the driver. Thus, the driving
with humans as data samplers. Some experi- environment should not present information at
ments to determine the characteristics of this a rate that exceeds the information processing
sampling process have been conducted by capacity of drivers.
Senders et al. M-lG; these results are discussed in The driver is considered to have a single
Section 5.3.4. information channel of fixed capacity in which
tasks are linearly additive. 19 (Several studies
5.3.3 Driver Information Processing undertaken to measure the capacity of this
channel are discussed in the following para-
Tasks, such as driving, can be thought of graphs.) Interestingly, the level to which a
as responses to information-containing stimuli. driver is stimulated can have a beneficial effect
It is convenient to quantify this information by up to a certain point, but thereafter further
the number of "bits" it contains; i.e., the num- stimulation can degrade his performance. 20
ber of mutually exclusive decisions that must
be made to correctly execute the task. 17 •46 5.3.4.1 Short-Viewing-Interval Experi-
Driving involves stimuli that are continuously ments. Experiments reported by Szafran 21
changing; hence, it presents the driver with a using short viewing intervals have demonstrated
flow of information in bits per second. If this the limits of the human information processing
rate is slow enough, drivers can respond cor- ability. As the amount of information in-
HUMAN RESPONSE TO STIMULI 75
Maximum lnfor-
Lowest Highest Relative Absolute mation Rate
300 nm 1500 nm Approx 128 discernible hues 12-13 hues 4.6 X JOU bits/sec
at moderate intensities
creases, or when there is divided attention or dents but a high number of violations had the
unwanted or excessive information (visual highest information processing capacity. Ferg-
noise), the additional information may be only enson 23 offers the following interpretation:
marginally processed. The results are illustrated Possibly those individuals with many
in Figure 5.3. violations are involved in more critical
5.3.3.2 Measuring the Spare Mental situations due to their driving habits; but
they still avoid accidents because their
Capacity of Drivers. Brown and Poulton 22 information processing capability is not
have described experiments by which they overloaded (they have large channel ca-
sought to measure the spare mental capacity of pacity.
drivers under certain tasks. Although they
demonstrated a method, they did not obtain a
specific rate in bits per second. TABLE 5.2 Average Information Processing
Ability of Subjects With and Without
5.3.3.3 Drivers' Channel Capacity Re- Violations and With and Without
lated to Accident Experience. Groups of driv- Accidents"
ers having high and low accident incidences
and high and low violation experiences were Bits per Second
tested to determine, by measuring decision time, Zero High
their abilities to process information (Table Violations Violations
5.2). It will be noted that the subjects with Zero accidents 26.09 38.67
high accident rates had a very low information High accidents 21.13 15.69
processing capacity. One unexpected result of
the experiments is that drivers with zero acci- "From Fergenson.'"
76 DRIVER INFORMATION PROCESSING CHARACTERISTICS
30 30 (b)
(a)
25 25
20 > 20
>
u u
cQ) cQ)
::J 15 ::J 15
O" rr
Q)
....
Q)
I-
LL LL
10 10
5 5
0 0
0 2 3 0 2 3
Time (sec)
Figure 5.2 Frequency distributions of eye fixation duration (a) during passing and (b) during normal driving
(from Whalen et al. as given by Briggs 18 ).
t = T,,, at which time vision is restored ity (or the occlusion time) so that his uncer-
and the cycle repeats. 1 " tainty U(Tll) at the encl of occlusion is less
The following assumptions and derivations than some critical value Uc.
of a driver-information processing model are The uncertainty is made up of two ele-
made: ments: ( 1) uncertainty about the road and
(2) uncertainty about the vehicle position and
l. The driving situation is in the steady orientation of the vehicle. With the introduc-
state. The vehicle proceeds at a constant ve- tion of uncertainty, the model of driver behav-
locity ( V miles/sec); the timing of looks is ior becomes
periodic such that vision is allowed for T1 sec
and occluded for Tll sec. U(T, 1 ) =H·D[l -e-·11·;1J+i/F1Tt1]
2. The road has a constant information +K,, V"(Tt1)"i 2 '.5oU., (5.1)
density of H bits/ mile.
3. The information density of the driver's where K,, includes the power density spectrum
stored image is H e-"·1 n bits/ mile, where D is related to drivers' uncertainty about vehicle
a weighting factor in miles and x is the distance position, as well as other scaling factors; other
from the start of occlusion. variables arc as previously defined.
4. The period of view is sufficient for the In the first experiment, on an unopened
driver to absorb all the information available. section of 1-495, drivers were given views of
The amount of information stored at t=O is the roadway of 0.25-, 0.50-, or J .0-sec dura-
HD bits. tion. Between views the occlusion time was
5. Information is forgotten at the rate of varied from 1.0 to 9.0 sec, and drivers were
/,.(1)/ F and becomes obsolete at the rate of expected to adjust speed to reflect the informa-
I,.(t)BI D bits/sec. The amount of informa- tion received.
tion in storage t sec after the onset of oc- When Eq. 5.1 was calibrated against ob-
clusion is H D e-<V/D+i/Fit bits, where /,. is served driving data on the highway, the follow-
the bits of information stored, F is the time ing ranges of results were obtained from five
constant (sec), and Vis speed (miles/ sec). different drivers (viewing time=0.50 sec):
The driver is assumed to adjust his veloc- H =information bits per mile 12.0-34.0
Ir Max
....Q)
c .:::!
0 ....
·.;:; 0
"' Q)
E r.
.........
0 >
c .0
- -0
-0 Q)
"'0 ....0
0:: ci)
•
Tl o Time Td
~ '----------r---------~
Vision Occlusion
0 0 V) 0
-~ stant viewing and occlusion time or more fre-
.._ quent viewing for constant speed and viewing
0
'+-- N time.
0
~ 0
M
0
0
Verification of adequacy of the informa-
2Q) 0 0 tion driving model suggests that trained drivers
E 4-<
0 might be used to identify and quantify exces-
~ N c: sively demanding road configurations and that
ro
a.. 00
0
0 .g this technique might be used on "calibrated"
0 ·u;
Q)
-0
0
"'
0 0 0
p.
<U
roadways to make a preliminary classification
of drivers in terms of skill level.
~ "'
0
-5
0 0 ..c:
N 0 u
0 0 V)
0\
:a 5.3.5 Driver Perception-Response Time
2:
N
!
4-<
~ One of the important measures of driver
0 c:
0 00 0 r-- 0
0
response to information received is his reaction
0 0 0 c: ·u;
ci ci \0 0 .9 :::>
time. The best known of these responses is
u
<U
uu brake reaction time.
N
0
"'
'O
0
4-<
In early experiments brake reaction times
<U
0
N
0
~
c: 0 were measured in the laboratory, and the re-
<U
0 0 p. sults listed values that are now considered
0
c: relatively short. Recent experiments in real
:::>
N
0 c: traffic have produced brake reaction times that
~
0
0 "'
c: may be considered more reliable. 21 In these
0 0 0
N experiments the time measured represented the
N
sum of the perception time (time to perceive
Q 0 N
0
0
the need for braking) and the time to move the
::::
u
~
0
~
0
0
0
~
0
foot from the accelerator to the brake pedal.
Some drivers were tested under conditions of
N both surprise and limited anticipation. A large
0
0 0 group was tested under limited anticipation
0 0
0 alone (Figure 5.5). To correct for surprise,
results in Figure 5.5 should be multiplied by
1.35. For surprise situations, then, the median
is 0.9 sec, with 10 percent of the reactions at
1.5 sec or longer.
VELOCITY DIFFERENCES IN CAR FOLLOWING 79
100
90
80
70
'+-
60
>
(.)
c
"':::l 50
O"
~
u...
40
30
20
10
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1
Time (sec)
Figure 5.5 Histogram of brake reaction times of 321 drivers. The sharp peak at the left is the distribu-
tion of the reaction times of the measuring process.:.:•
5.4 DETECTING VELOCITY DIFFERENCES Hoffmann "' has proposed a model for latency
IN CAR FOLLOWING as follows:
One of the most important driver tasks
from the viewpoint of traffic flow theory is
detection of differences of velocity between the where
lead car and that which the driver operates.
Michaels "" has given one approach to quanti- t= latency time:
fying this task; his work is discussed later. \a\= magnitude of acceleration 111 ft/ sec";
Another approach to the problem is the ability and
of the driver to detect acceleration. Under R=vehicle separation in ft.
general conditions, the human thresholds for The minus sign represents acceleration: the
sensing acceleration are s linear-horizontal plus sign, deceleration.
(12-20 cm/ sec 2 ), linear-vertical ( 4-12 cm/ Field observations have led Lee and
sec 2 ), and angular (0.2 degrees/sec 2 ). Jones 0 !• to make the following statement:
For car following at night, Todosiev and
Fenton 26 found the velocity threshold gen- The mean time lag for the queue-
erally to be smaller than the corresponding day forming condition is less than that for the
threshold. queue-releasing condition. In the queue-
forming condition the driver of the fol-
An important aspect of velocity detection lowing car must make quick decisions to
is the reaction or latency time. Braunstein and avoid collision as the following car gets
Laughery 2 ' have measured response latencies nearer to the vehicle ahead. In the queue-
of drivers in detecting accelerations and de- releasing condition there is less urgency
celerations of the lead car (Figure 5.6). for decisive action.
80 DRIVER INFORMATION PROCESSING CHARACTERISTICS
5.4.1 Estimated Driving Speed ingly impeded by congestion, and flow departs
progressively from the predicted lines.
Estimating the speed actually being driven
Forbes' predictions of flow based on fixed
is a problem related to velocity differences in
reaction time headways are shown in the nega-
car following. Experiments have shown that a
tively sloped portions of the flow-concentration
driver's estimate of his speed depends on the
curves. As concentration increases, drivers
speed at which he has been traveling pre-
move more slowly to preserve reaction time
viously. Denton 30 has stated, "Drivers under-
headway. (Haight, Bisbee, and Wojcik 1 9 have
estimate their speed when decelerating and
shown that the driver cannot reduce time
overestimate it when accelerating." Beers and
headway below man-machine reaction time if
Hulbert 4 " have found that accuracy of speed
the risk of collision with the car ahead is to
estimation increases with age, but that older
be avoided. The derivation assumes that the
drivers tend to underestimate speed while
driver's braking is as effective as that of the
younger drivers overestimate.
driver in front.) For example, if concentration
increases from 120 cars to 240 cars per lane
5.4.2 Behavior in Platoons ··· per mile, distance headway decreases from 44
Forbes :n has derived the flow-concentra- to 22 ft, and clearance between 1 7-ft cars
de~reases from 27 to 5 ft. In this example, the
tion curve from driver characteristics. The
curves on the left of Figure 5.7 would result if dnver must reduce speed to less than a fifth to
drivers maintained the same speed as concen- equalize his time heading at the two concen-
tration increases. At low concentration where tration levels.
cars do not interfere with each other, flow Flow data from highway sites are also
would increase linearly with concentration. If shown in Figure 5.8. Curves a to d show data
there is ?ne car per mile moving at 25 mph, obtained at four freeway stations. The Edsel
the flow 1s 25 cars per hour. If concentration Ford Expressway (curve c) is fitted by a time
is doubled to two cars per mile, flow is also headway between 1.0 and 1.5 sec; the Lincoln
doubled (in this case to 50 cars per hour). It Tunnel (curve b) and the Penn-Lincoln Park-
may be seen from Figure 5.8 that Forbes' pre- way (curve d) are predicted by headways be-
dictions are validated by actual traffic flow data tween 1.5 and 2.0 sec. The traffic flow taken
from tunnels and expressways. However, as at another station on the Edsel Ford Express-
concentration increases, the driver is increas- way and shown in curve a did not reach a
slowdown or stoppage, although flow volume
•:• Material in section 5.4.2 is quoted from reached as high as 2,000 vehicles per hour. It
Gordon " with appropriate renumbering of refer-· may be seen that the congested-flow data can
ence citations and figures. be fairly well fitted by straight lines represent-
ing the assumption that drivers' average time
heading is constant and independent of traffic
density.
7 DECELERATION ACCELERATION Forbes also showed that drivers reacted to
sudden braking of the car ahead by dropping
u
' -........<so
-...:;;: Ft
.........
back to increase their time headways. If the
leader of a three-car queue reduced speed by
21 about 10 mph and after several seconds re-
"<V '-..
-----
6
> ~
-2so tr turned to cruising speed, the driver immediately
u behind dropped back to increase his time head-
c
Q.l
+" way. The effect at various road sites differing
"'
_J in horizontal and vertical curvature is shown
c 5 in Table 5 .4. It may be seen that the rear
"'
Q.l
driver increased time headway 0.8 sec or more
~
after the deceleration maneuver. This change
in time headway is observed after deceleration
4
of actual traffic and affects flow rate.
(-) 0.8 (-) 1.4 1.7 2.2 Forbes' experimental work under dense
Rate of Change (ft/sec2) traffic is described in Forbes and Simpson.'"
Figure 5.6 Effects of rate of change and initial distance 5.4.2. l Platoon Behavior as a Weber-
on response latency.:r. Fechner Relationship. Daou 3 " suggests that
VELOCITY DIFFERENCES IN CAR FOLLOWING 81
.._
.r:: 3
---
.r::
Q)
>
0
0
~
---
O"
2
L = 18
OL-~~~4L0~~~--'80~~~~1~20~~~-1L60~~~-2~0-0~~~2-4~0~~~-2~8~0--~~3~20
k (veh/mile)
Figure 5.7 Relation of volume, speed, and concentration from driver response time as a limit (right} and scat-
tered free flow (left). 31
.._
.r::
---
.r::
Q)
4
>
..,,
cQ)
'.2'.°
::J
O" 3
UJ
Q)
E
::J
0
> 2
Q)
c
ro
....J
.._
0
0
0
~
---
O"
to be made. 0 10 20 30 40 50 60
Speed (ft/sec)
Figure 5.9 Plot of Daou's relationship for platoon be-
5.7 REFERENCES huvior.3:!
1. Allen, T. M., Lunenfeld, H., and Alexan-
der, G. J., Drivers information needs.
Highw. Res. Rec. 366 (1971), pp. 102-
115. S. N. Alexander (ed.) System Engi-
2. Cumming, R. W., The analysis of skills neering Handbook. McGraw-Hill, New
in driving. Austr. Road Res., 1 (9): York (1965).
4-14 (1964). 12. Fogel, L. J., Biotechnology: Concepts and
3. Wohl, J. G., Man-machine steering dy- Applications. Prentice-Hall, Englewood
namics. Human Fact., 3 ( 6) : 222-228 Cliffs, N.J. (1963), pp. 200-203.
(1961 ). 13. Bekey, G. A., The human operator as a
4. McRuer, D., and Weir, D. H., Theory of sampled-data system. IRE Trans. Hu-
manual vehicular control. Ergonomics, man Fact. Electron., HFE-3(2): 43-51
12(4): 599-633 (1969). (1962).
5. Rashevsky, N., Mathematical biology of 14. Senders, J. W., Kristofferson, A. B., Levi-
automobile driving. Bull. Math. Bio- son, W., Dietrich, C. W., and Ward,
phys., 29(1): 181-186 (1967). J. L., An Investigation of Automobile
Driver Information Processing. Bolt,
6. Rockwell, T., Skills, judgment and infor-
Beranek, and Newman Inc., Report No.
mation acquisition in driving, pp. 133-
13335, April 26, 1966. Submitted to
164. In T. W. Forbes (ed.) Human
U.S. Bureau of Public Roads under
Factors in Highway Traffic Research.
contract CPR-11-0956. NTIS No. PB
Wiley, New York (1971).
170879.
7. Mowbray, G. H., and Gebhard, J. W., 15. Senders, J. W., and Ward, J. L., Addi-
Man's senses as information<Jl channels, tional Studies on Driver Information
pp. 115-149. In H. W. Sinaiko (ed.)
Processing. Bolt, Beranek, and New-
Selected Papers on Human Factors in man Inc., Final Report to Federal
the Design and Use of Control Systems. Highway Administration on contract
Dover, New York (1961). CPR-11-5096, 15 April 1969. NTIS
8. Fogel, L. J., Biotechnology: Concepts No. PB184499.
and Applications. Prentice-Hall, Engle- 16. Senders, J. W., Kristofferson, A. B., Levi-
wood Cliffs, N .J. (1963), p. 151. son, W. H., Dietrich, C. W., and Ward,
9. Jacobson, H., The informational capacity J. L., The attentional demand of auto-
of the human eye. Science, 113: 292- mobile driving. Highw. Res. Board Rec.
195 (1967), pp. 15-33.
293 (1951).
17. Shannon, C. E., and Weaver, W., The
10. Jacobson, H., The informational capacity
• Mathematical Theory of Communica-
of the human ear. Science, 112: 143- tion. University of Illinois Press, Chi-
144 (1950). cago (1949).
11. Pew, R. W., Human information-process- 18. Briggs, G. E., Driving as a skilled per-
ing concepts for system engineers. In formance. Proceedings of the Second
R. S. Machol, W. P. Tanner, Jr., and Annual Automobile Insurance Industry
84 DRIVER INFORMATION PROCESSING CHARACTERISTICS
Traffic Safety Research Symposium. 33. Rockwell, T., Skills, judgment, and infor-
(1968),pp. 123-143. mation acquisition in driving, pp. 133-
19. Senders, J. W., The estimation of operator 164. In T. W. Forbes (ed.) Human
workload in complex systems, pp. 207-· Factors in Traffic Safety Research.
216. In K. B. DeGreene (ed.) Systems Wiley, New York (1972). •
Psychology. McGraw-Hill, New York 34. Farber, E., Silver, C. A., and Landis, D.,
(1970). Knowledge of closing rate versus on-
20. Welford, A. T., Arousal, channel-capacity coming-car speed as determiners of
and decision. Nature, 194 ( 4826) : driver passing behavior. Highw. Res.
365-366 (1962). Rec. 247 (1968), pp. 1-6.
21. Szafran, J., Limitations and reliability of 35. Gordon, D. A., and Mast, T. M., Drivers'
the human operator of control systems decisions m overtaking and passing.
to process information. Aerosp. Med., Highw. Res. Rec. 247 (1968), pp.
239-242 (Mar. 1966). 42-50.
22. Brown, I. D., and Poulton, E. C., Mea- 36. Farber, E., and Silver, C. A., Behavior of
suring the spare "mental capacity" of drivers performing a flying pass. Highw.
car drivers by a subsidiary task. Ergo- Res. Rec. 247 (1968), pp. 51-56.
nomics, 4 (1):35-40 (1961). 37. Silver, C. A., and Farber, E., Driver judg-
23. Fergenson, P. E., A Study of the Relation- ment in overtaking situations. Highw.
ship Between Driver Decision Time and Res. Rec. 247 (1968), pp. 57-62.
Driving Accident and Violation Record. 38. Jones, H. R., and Heimstra, N. W., Ability
Davidson Laboratory, Stevens Institute of drivers to make critical passing judg-
of Technology, Report SIT-DL-69- ments. Highw. Res. Rec. 122 (1966),
1329 (1968). NTIS No. PB188159. pp. 89-92.
24. Johansson, G., and Rumar, H., Drivers' 39. Forbes, T. W., Visibility and legibility of
brake reaction :imes. Human Fact., 13 highway signs, pp. 95-109. In T. W.
(1) :23-27 (1971). Forbes (ed.) Human Factors in Traffic
25. Michaels, R. M., Perceptual factors m Safety Research. Wiley, New York
car following. Proceedings of the Sec- (1972).
ond International Symposium on the 40. Forbes, T. W. (ed.), Human Factors in
Theory of Traffic Flow (London, June Traffic Safety Research. Wiley, New
1963). Published by OECD (1965), York (1972).
pp. 44-59. 41. Forbes, T. W., and Katz, M. S., Summary
26. Todosiev, E. P., and Fenton, R. E., Veloc- of Human Engineering Research Data
ity thresholds in car-following at night. and Principles Related to Highway De-
Highw. Res. Rec. 122 (1966), pp. 94- sign and Traffic Engineering Problems.
104. American Institute for Research,
27. Braunstein, M. L., and Laughery, K. R., (1957), mimeo.
Detection of vehicle velocity changes 42. Beers, J., and Hulbert, S. F., Judgment of
during expressway driving. Human Vehicle Speeds and Traffic Patterns.
Fact., 6 ( 4): 327-331 (1964). UCLA Engineering Report 7281, June
28. Hoffmann, E. R., Note on detection of 1972, PB21 8925.
vehicle velocity changes. Human fact., 43. Stevens, S. S., The surprising simplicity of
8(1):139-141 ( 1966). sensory metrics. Am. Psycho!., 27:
29. Lee, J., and Jones, J. G., Traffic dynamics: (1962).
visual angle car following models. 44. Von Fieadt, G., The World of Perception,
Traffic Eng. Control, 9(7) : 348-350
Dorsey Press (1966), pp. 169-170.
(1967).
45. Forbes, T. W., and Simpson, M. E., Driver
30. Denton, G. G., A subjective scale of speed
when driving a motor vehicle. Ergo- and vehicle response in freeway deceler-
nomics, 9 (3) : 203-210 (1966). ation waves. Transp. Sci. 2 ( 1): 77-104
31. Forbes, T. W., Human factor considera- (1968).
tions in traffic flow theory. Highw. Res. 46. Attneave, F., Applications of Information
Rec. 15 (1963), pp. 60-66. Theory to Psychology. Holt, Rine-
32. Daou, A., On flow within platoons. Austr. hart and Winston, New York ( 1959),
Road Res., 2 (7) :4-13 (1966). pp. 1-12.
RELATED LITERATURE 85
47. Forbes, T. W., Zagorski, H. J., Hols- King, G. F., and Lunenfeld, H., Development
houser, E. L., and Deterline, W. A., of Information Requirements and Transmis-
Measurement of driver reactions to tun- sion Techniques for Highway Users. Na-
nel conditions. Proc. Highw. Res. tional Cooperative Highway Research Pro-
Board, 37:345-357 (1958). gram Report 123. Highway Research Board,
48. Gordon, D. A., Psychological Contribu- Washington, D.C. (1971 ), p. 239.
tions to Traffic Flow Theory. Draft re- Kolers, P. A., Aspects of Motion Perception.
port to FHWA staff (June 1973). Pergamon Press, New York ( 1972).
49. Haight, F. A., Bisbee, E. F., and Wojcik, Lazarus, R. S., Deese, J ., and Osler, S. F., The
C., Some mathematical aspects of the effects of psychological stress upon perform-
problem of merging. Highw. Res. Board ance. Psycho!. Bull., 49 (4):293-317
Bull. 356 (1962), pp. 1-14. (1952).
Llewellyn, K. R., Visual guidance of locomo-
5.8 RELATED LITERATURE tion. J. Exp. Psycho!., 91 (2) :245-261
(1971).
Baker, C. A., and Steedman, W. C., Perceived Rashevsky, N., A note on the mathematical bi-
movement in depth as a function of lu- ology of automobile driving. Bull. Math.
minance and velocity. Human Fact., 3: 166- Biophys., 29 (1):187-188 (1967).
173 (1961).
Rockwell, T. H., and Snider, J. S., An Investi-
Brown, I. D., Human factors in the control of gation of Variability in Driving Performance
road vehicles. Electron. Power, 14:275-279 on the Highway, Report RF-1450 17-31.
(1968). Systems Research Group, Ohio State Univer-
Bryant, J. F. M., Progress of research in human sity, Columbus (1965).
factors. Aus tr. Road Res., 3 (1 ) : 3 8-4 7 Senders, J. W., Elkind, J. I., Grignetti, M. C.,
(1967). and Smallwood, R., An Investigation of the
Duckstein, L. A., Unwin, E. A., and Boyd, E. Visual Sampling Behaviour of Human Ob-
T., Variable perception time in car following servers. Bolt, Beranek and Newman Inc.,
and its effect on model stability. IEEE Report NASA Cr-434. Submitted to Na-
Trans. Man-Mach. Sys., MMS-11 (3):149- tional Aviation and Space Administration
156 (1970). under contract NAS 1-3860, April 1966.
Forbes, T. W., Factors in highway sign visibil- NITS No. N66-22910.
ity. Traffic Eng., 39 (12) :20-27 (1969). Senders, J. W., Human performance. Int. Sci.
Good, M. C., Rolls, K. J., and Joubert, P. N., Technol., No. 55: 58-60, 62, 64, 66-68, 89
Driver-vehicle behaviour in free-path turns. (1966).
Transp. Res., 3 (1):43-68 (1969). Snider, J. N., The capability of automobile driv-
Gordon, D. A., Experimental isolation of the ers to sense vehicle velocity. Highw. Res.
driver's visual input. Highw. Res. Rec. 122 Rec. 159 (1967), pp. 25-35.
(1966), pp. 19-34. Steedman, W. C., and Baker, C. A., Perceived
Gordon, D. A., The driver in single lane traffic. movement in depth as a stimulus size.
Highw. Res. Rec. 349 (1971), pp. 31-40. Human Fact., 4: 349-354 ( 1962).
Gordon, D. A., Traffic interactions and delays Weir, D. H., and McRuer, D. T., A theory for
in freeway traffic. Highw. Res. Rec. 336 driver steering control of motor vehicles.
(1970)' pp. 76-91. Highw. Res. Rec. 247 (1968), pp. 7-28.
Gordon, D. A., Experimental isolation of the Olson, P. L. The driver's reference point as a
driver's visual input, Human Fact., 8 ( 1): function of vehicle type, direction and radius
129-137 (1966). of turn. Human fact., 6(4) :319-325
Haight, F. A., A mathematical model of driver (1964).
alertness. Ergonomics, 15 ( 4): 367-378 Wierwille, W. W., Gagne, G. A., and Knight,
(1972). J. R., An experimental study of human op-
Hoffman, E. R., The interaction between the erator models and closed-loop analysis meth-
driver, vehicle and road. Austr. Road Res. ods for high-speed automobile driving. IEEE
3 (2) :4-26 (1967). Trans. Human Fact. Electron., HFE-8 (3):
Hulbert, S. F., and Burg, A., Human factors in 187-201 (1967).
transportation systems, pp. 471-509. In K. Wooler, J., The measurement of driver per-
B. DeGreene (ed.) Systems Psychology. Mc- formance. Ergonomics, 15 ( 1 ) : 81-87
Graw-Hill, New York, (1970). ( 1972).
86 DRIVER INFORMATION PROCESSING CHARACTERISTICS
87
.. ..
f-olll------s (t)-----~
-Xn+1 (t)
- - - - - X n (tl-----~
f-olll-----d1------+-------~d2------11...+-o,._
Distance Traveled Stopping Distance for
During Time T n+1
dx(I) . x+
11 1 (1+l)=l.O[.X 11 (1)-i 11 +1 (t)] (6.9)
u(I) = d t =x(I) ( 6.4a)
A direct analytical solution of Eq. 6.9 is
and the acceleration as cumbersome; it is easier to find a numerical
solution. As an. example, consider the case
d 2x where the second vehicle attempts to follow a
a(I) =--=x(I) ( 6.4b)
d1 2 first vehicle that has instantly attained a veloc-
ity of 30.0 ft/ sec from a stopped position as
and substituting in Eq. 6.3 the appropriate developed below. For this solution the posi-
relations for d 1 , d 2 , and cf;:,,, gives tion of the first vehicle is advanced 30 ft
s(I) = x,,(1)-x,,+ 1 (1) during each 1-sec time slice. At each time t
=Tx 11 + 1 (1) separated by an increment of time 6.1, the
+ [x2,,+r ( t + T) ]/ [2x,,+ 1 (I+ T)] acceleration of the second vehicle is calculated
+L-[.F 11 (1)]/[2.X 11 (1)] (6.5) ( Eq. 6 .9) . All distances measured are from
the stop line at location 0 ft.
If the stopping distances and velocities of Assuming that during each time slice M
the two vehicles are assumed equal, so that ( 1.0 sec in this example) the acceleration is
d 2 =d 3 , the spacing becomes uniform and equal to the average of the accel-
x 11 (1)-x 11 +1 (1) =Tx 11 + 1 (t+T) +L (6.6) erations calculated at the start and end of each
time slice, the equations of velocity u 2 (t) and
which is the distance traveled by the following position x 2 (t) are given by
vehicle during reaction time (T), d 1 , plus the
separation between the front bumpers at rest, i2(1) = .<' 2 (1-D.t)
L. It will be observed that this spacing is less +1/i(x 2 (t-M) +x 2 (1)]6.1 (6.10)
than that assumed in Eq. 6.2. and
Differentiating with respect to ( t),
X 2 (t) = X 2 (t-D.t) -f-X 2 (t-M)D.t
.X,,(1)-.i,,+ 1 (1)=T[x,,+ 1 C1+T)J (6.7) +1/i[i2(t-6.I) +x 2 (t)]M 2 /2
so that the acceleration of the n +1st vehicle =x 2 (t-M)
x+ (1+T) =T- 1 [.X"(t)-.X 11 + 1 (1)] (6.8) where M is the time increment between suc-
11 1
cessive calculations (sec).
which is of the form given previously, A numerical solution of these three for-
Response (I+ T) =Sensitivity x Stimulus ( t). mulas is given in Table 6.1, where the relation-
The response of the n + 1st driver, which ship is solved at 1.0-sec increments. It is
takes place at time (t+ T), is to accelerate evident that vehicle 2 quickly reaches the speed
(decelerate) by an amount proportional to the of the lead vehicle and settles down to follow
positive (negative) difference in the relative at a distance of about 55.0 ft from it, with only
velocity of the nth and n +1st driver and the minor corrections in velocity and spacing after
measure of sensitivity is given by T-1 (sec). 7 or 8 sec have elapsed. The spacing of 55.0 ft
How well does a simple car-following is the same as that given in Eq. 6.6 when the
model of the type given in Eq. 6.8 describe reaction time is assumed to be 1.0 sec and the
driver behavior? Assume two vehicles waiting stopped spacing L between vehicles is 25 ft.
at a traffic signal, with the front bumper of the So far concern has been with the behavior
second vehicle positioned 25 ft from the front of only the first and second cars waiting in a
bumper of the lead vehicle. The reaction time queue at a stop signal. What of a vehicle that
( T) of the drivers is taken to be 1.0 sec and is not immediately behind the lead vehicle but
the sensitivity ( T-1 ) is 1.0. At time 0.0, shortly is the fourth or fifth vehicle in line at the
after the signal changes to green, the first moment the lead vehicle moved at 30 ft/ sec?
vehicle immediately (a physically impossible The behavior of a series of vehicles could be
90 CAR FOLLOWING AND ACCELERATION NOISE
TABLE 6.1 Car-Following Calculations
extended by extending the type of arithmetic results are presented at 1.0-sec intervals, and
calculations given in Table 6.1, but the calcu- the reader will observe that the location of the
lations are lengthy and numerical accuracy second vehicle x 2 is not exactly that calculated
dictates that the calculations be done at 0.1-sec by using 1.0-sec intervals as in Table 6.1.
intervals (see Fox and Lehman H for a discus- Although the simple rule for car following
sion of solving the car-following problem with developed in Eq. 6.9 gave a reasonable pattern
an electronic computer). for the motion of the second vehicle relative
An alternate method for calculating the to the first, it will be observed that the third
behavior of a platoon of vehicles for the car- and fourth vehicles will be involved in a rear-
following situation is to apply the Laplace end collision about 90 ft and 7.0 sec from the
transform to the solution of the problem. The starting point when the spacing (front bumper
particular problem postulated here, the behav- to front bumper) is reduced to less than
ior of a platoon of vehicles starting from rest 18.0 ft. It is evident that if all drivers were to
when the velocity of the lead vehicle instanta- follow the postulated behavior there would be
neously becomes u 0 at time t ::>: 0, has been a rash of rearend collisions at signalized inter-
solved by Kometani and Saski. 7 Further dis- sections. The amplitude of the response to the
cussion of the application of the Laplace trans- instant change in velocity of vehicle 1 becomes
form will be found in references 9, 15, 16. greater for each successive vehicle and the sys-
A numerical solution of the behavior of tem is said to be unstable.
the first five vehicles in a platoon, based on
the equations of Kometani and Sasaki 7 is given
6.3 TRAFFIC STABILITY
in Table 6.2. The assumptions are the same as
for the two-car situation, with T= 1.0 sec and The question of stability in a platoon of
vehicles starting from a queue at rest with vehicles is important in reviewing different
25-ft spacing between vehicles, except these patterns of car-following behavior. If driver
calculations are done at 0.1-sec intervals. The behavior is to be modified, or the mechanical
TRAFFIC ST ABILITY 91
"Front ends of vehicles are separated by less than 18.0 ft (collision imminent).
70 26
24
60
22
50 20
70 18
16
60
-;:;- u
..... Q.)
14
Cl 50 .!:!.
c: Q.) 12
u E
"'a.
(/)
90 f- 10
....
8
u"' 80
6
4
2
0
98765432
-2'---'-~'---'-~-L--->~...L....--lL..--'----'
-320-280-240-200-160-120-80-40 0
10 20 30 40 Car Positions (ft)
Time (sec)
Figure 6.4 Asymptotic instability of a line of
Figure 6.3 Car spacings of a line of cars with constant control for
nine cars under the influence of velocity control
varying values of C ( =
etT) and for a condition where the first car (Ct= 0.8, T = 2.0) relative lo a phantom lead
decelerated and then accelerated back to its initial velocity, with an
car moving with constant speed at an initial spac-
initial spacing of 70 ft between vehicles. 10 ing of 40 ft. Vehicle 1 decelerates and then ac-
celerates back to its original velocity. In the
process its spacing from the phantom vehicle is
increased but is stable. Successive vehicles make
corrections at a later time and with increasing
proposed that the sens1t1v1ty be inversely pro- amplitude of oscillation until a collision occurs
portional to the spacing so that between vehicles 7 and 8.10
the sensitivity and the units of a. 0 are distance/ ( n + 1st) vehicle, giving an expression for
time. the velocity of that vehicle, which in turn is the
A number of car-following experiments steady-state velocity of the traffic stream. The
were performed in the Holland and Lincoln resulting equation of the velocity can then
Tunnels in New York and at the General be solved for known boundary conditions, de- •
Motors test track. Two vehicles were con- termining the constant of integration and then
nected by a length of wire on a reel, permitting substituting in terms of the appropriate quanti-
direct measurement of spacing and relative ties, q and k, used in the definition of flow
velocity simultaneously with measurements of (q=uk) (Chapter 2).
the acceleration and velocity of the following Consider the application of the procedure
vehicle. Different drivers were given instruc- to the simple linear car-following model given
tions to follow the lead vehicle as they would in Eq. 6.12:
under normal driving conditions. A summary
of the values of a. 0 and T found in the car- 1. Express the acceleration for the
following experiments is given in Table 6.4. 12 (n+ 1st) vehicle,
As will be seen in section 6.5, a. 0 is representa-
tive of the speed associated with maximum
traffic volumes. 2. Integrate the expression to obtain the
The values given are the averages for all velocity of the n + 1st vehicle (the velocity of
of the drivers tested. In the General Motors the traffic stream)
experiment, for example, the value of T ranged
from 1.0 to 2.2 sec. Xn+i (t + T) =a[x 11 (t) -Xn+i ( t)] + C 0
stream data (see Figures 4.15 and 4.17), indi- been verified by experimental data (Figure
cating that the car-following model of Eq. 6.12 4.15).
is not a realistic model, at least at low densities.
The same procedure may now be applied 6.6 GENERAL EXPRESSION FOR
to Eq. 6.13. CAR-FOLLOWING MODELS
.. (
I . x""' t+
T)- x11 (t)-i,,. 1(t)
-ao - - - - - -
Continuing examination of the relationship.
x11(t) -x11+1(t) between microscopic and macroscopic models
led to a generalized form of the car-following
2. Integration (this is of the form equation:
f d'lll)
.. ( ) x"'11+1Ct+T)
C • results m u=a0En,[x 11 -x 11 +1]+C,, X11+1 t+T =ao [x"(t)-x11+1Ct)]I
= a 0 fu[s] + C 0 = a 0 fu[l/ k] + C 0 • [.:i:"(t) -i,,+ 1 (t)] ( 6.18)
3. Since for u= 0, k=ki
0= a 0 fu[l/ kj]+C 0 , where l and m are constants.
C 0 = - a 0 fu[l I kj]. First proposed by Gazis, Herman, and
4. U= a 0 fu[l/ k]-a 0 fu[l/ kj] Rothery, 13 the general expression has been
= a 0 Ei,[l/k-En, 1/ki] further examined by May and Keller. 18
= a 0 Ei,[k/ k] (6.16) Integration of the generalized equation
5. q= u k (Eq. 6.18) by Gazis et al. 13 has given
=a 0 kfu(k/k) (6.17)
(6.19)
6. To determine the proportionality con- where u is the steady-state speed of a stream of
stant a 0 , refer to Figure 4.15 and Section 4.3 .2 traffic, s is the constant average spacing, and a 0
to determine the known physical conditions. and C 0 are appropriate constants consistent
Observe that slope dq/ dk = 0 at maximum with physical restrictions. Further,
volume qrn. Differentiation of Eq. 6.17 yields:
f,,,(u) = ul-111 for (m =I= 1)
dq!dk= a 0 [k(kl ki) (-k/ k 2 ) +En,(k/k)]=O and
= a 0 [ - l +En,(k/ k)]=O f,,,(u)=fuu for (m=l) (6.20)
= a 0 Ei,[k/ ke] = 0
fi(s)=s'-' for (I =I= 1 )
where e =base of natural logarithms.
and
Assuming that a 0 =I= 0 and defining k 111 as the f 1(s)=Ei,s for (l= 1) ( 6.21)
density at maximum flow qm yields En,(k/ krne)
The value of C 0 is related either to free
= 0 so that k/ kme= 1 and km= k/ e. speed uf (the velocity of a single vehicle whose
Defining u 111 as the velocity at qm, qm speed is not influenced by interaction with
=umkm=umk/e. other vehicles) or velocity at jam spacing,
Substituting um and km in Eq. 6.16, gives Sj= 1/ ki.
um= a 0 En,[k/ km]= a 0 En,[kje/ kj] = a 0 fu e, so The values of C 0 are shown to be
that
for m > 1, l =I= 1 or m = 1, l > 1
(6.22)
Eq. 6.16 was first proposed by Green- and
berg,17 who developed the relationship from the
equation describing unidimensional flow in C 0 =a 0 / 1 (si) for all other combina-
fluids and that denoting conservation of matter. tions of m and l (except m= 1, l> 1
The relationship was confirmed from experi- not bounded) (6.23)
mental observations of flow, density, and veloc-
ity. The Greenberg equation of state (section The traffic stream model equations result-
7.4.1), using the application of a fluid flow ing from the application of Eqs. 6.19-6.23 are
analogy, is based on a macroscopic approach shown in Table 6.5 and Figure 6.5. These
and is mathematically equivalent to Eq. 6.13, steady-state solutions were originally derived
which is based on the principles of car follow- independently of car-following assumptions.
ing, a microscopic approach. This model has However, as shown previously, they have a
96 CAR FOLLOWING AND ACCELERATION NOISE
TABLE 6.5 Steady-State Flow Equations (q-k)(l,
m=l
2 q= a.ke'kfko! Edie 21
direct correspondence with car-following to a comparison with the Greenbt;rg fluid flow
models. A generalized discussion of the steady- analogy data.
state flow for various values of m and l is given Greenshields' 20 analysis of traffic flow,
in Gazis et al., 13 but only those combinations which corresponds to the case m=O, 1=2, was
presented in Table 6.5 have been verified by first developed from photographic observations
observations of vehicle flow on roadways. of traffic flow made in 1934. Values for flow
May and Keller 18 have also examined the rate and mean velocity were calculated for
case for noninteger values of m and /, propos- 100-vehicle groupings; density was calculated
ing a model with m = 0.8, I= 2.8 when fitted to from the flow rate and velocity information. A
data on the Eisenhower Expressway in Chicago. simple straight-line fit between velocity and
The following steady-state flow equation density was deduced from a plot of the data.
results: Although the steady-state flow equation
resulting from this analysis does not fit observed
(6.24)
data as well as some other models listed here, it
Values for m and l have evolved as various does have the advantage of being amenable to
investigators attempted to fit observed data to calculation and manipulation.
proposed models of driver and/ or stream flow An independent verification of the Green-
behavior. The case m = 0, I= 0 evolved from shields model is reported by Pipes and
the "simple" linear car-following situation de- Wojcik "" in which perceptual factors are re-
veloped in Eq. 6.8. The same model had been lated to the car-following model. (A similar
developed by Pipes 1 and analyzed for stability derivation is shown in Fox and Lehman.'" In
by Herman et al. 10 • 1 e Figure 6.6 let 8 be the visual angle subtended
Further examination of experimental data by the lead vehicle. Taking w as the width of
based on car-following experiments led to the vehicle (and neglecting the distance L, the
hypothesis that a driver does not have a con- length of the lead vehicle),
stant reaction to a stimulus by the lead vehicle w
but that the reaction will vary inversely as the 8=- (6.25)
s
distance between the subject vehicles.ll This
is the case for m = 0, I= 1, which led directly Differentiating with respect to t,
GENERAL CAR-FOLLOWING MODELS 97
0 2 3
LL
LL
Figure 6.6
-1~ --s=xn-x(n+1)_J..,_L--=:f
Conditions of the Pipes and Wojcik 23
1
veriflcation of the Greenshields traffic flow model.
If it is assumed that the acceleration for which implies acceleration if e is negative and
the following vehicle is proportional to the deceleration if iJ is positive. Because s=x,.-
driver's perception of the rate of change of the x,,+1 and dsldt=x,.-x,,+ 1 , substituting Eq. 6.26
visual angle, e, in Eq. 6.27 yields
98 CAR FOLLOWING AND ACCELERATION NOISE
passenger cars are greater than the acceleration vehicle that has been following a lead vehicle
capabilities and that in congested traffic greater at a constant velocity and at a desired spacing.
action is taken when the vehicle ( s) in front is The lead vehicle accelerates but the following
(are) decelerating rather than accelerating vehicle is in no hurry to close the spacing, be-
relative to the subject vehicle. Herman and ing satisfied with the longer spacing until some
Rothery 25 examined this hypothesis by weight- further change takes place in the velocity (and
ing the negative relative speeds by values in the therefore spacing) of the lead vehicle. When.
range 1 to 1.25. The linear model (Eq. 6.12) the lead car decelerates, the following car al-
was modified so that lows some of the excess gap to dissipate before
reacting to the deceleration.
for relative In Figure 6.7 it is assumed that there are
velocity two velocity-headway relationships of the form
positive
(6.31a) Xn+ 1 (t+T) =G 11 + 1 [x 11 (1)-x 11 + 1 (t)) (6.32)
town traffic, and to examine safety in car +k[x 11 (1) -X 11 + 1 (1)] ( 6.33)
following.
where IX and k are sensitivity constants for
relative speed and relative spacing, respectively.
6.8.1 Applications to Driver-Aided In a comprehensive study and review of
Car Following the "car-following" problem as applied to im-
proving flow (safety and volume) Rockwell and
One method of improving lane capacity Treiterer 33 suggest a control system in which
would be to give a driver more information, the acceleration of the following vehicles is
permitting him to decrease his reaction time given by:
and therefore follow the lead vehicle at a lesser
headway. Fenton and Montano 31 report the .x 11 +1 (1+T)=1X[x 11 ( 1) -.x 11 + 1 (r) J+ Kx 11 (t)
results of an experiment in which additional ( 6.34)
information about the lead car was given to a
driver by means of a tactile device built into The relationship suggests that the accelera-
the single control stick used for steering, ac- tion, after some lag ( T), is a function of differ-
celeration, and deceleration (as opposed to a ence in velocity plus a term indicating that the
steering wheel, accelerator pedal, and brake following car exactly duplicates the acceleration
pedal as normally used). The tactile device of the lead car after a lag of T sec. (This latter
was a finger flush with the control stick at the component is termed "acceleration control" by
correct spacing, recessed when headways were Rockwell and Treiterer.) The parameter K
too small and protruding when headways were varies from 0 to 1.0, representing the propor-
EXTENSIONS OF CAR-FOLLOWING MODELS 101
tion of the control that is acceleration control. 6.8.3 Effect of Small Cars on
For K = 1.0 (total acceleration control) each Downtown Traffic
vehicle would exactly duplicate the acceleration
Two major advantages are usually claimed
of the preceding vehicle after a lag of T sec,
for small cars: they occupy less space when
a system that cannot be controlled by drivers
parked and they will reduce congestion. Mc-
without the aid of supplementary devices.
Clenahan and Simkowitz :ic, have estimated the
anticipated reduction in congestion by com-
6.8.2 Single-Lane Bus Flow puter simulation of a single lane of traffic
An application of car-following theory to consisting of a number of cars with identical
an analysis of single-lane flow of heavy vehicles acceleration and performance characteristics,
(large buses were used in the experiment) is but with varying lengths.
reported by Rothery et al. 34 Car-following The simulation modeled the behavior of
experiments, similar to those involving pairs of a file of cars down one lane of a street with
automobiles, were conducted with pairs of arbitrarily fixed-cycle traffic signals. Driver
behavior was based on a car-following model
buses directly connected by the apparatus used
to measure the required parameters of car- plus a model of a lead-car driver's behavior.
The model was designed to reflect driver
following models.
Three versions of the general expression reaction to the traffic signal indication if it took
for car following (Eq. 6.18) were analyzed. precedence over the stimulus received from
the lead car. When the lead· car provided the
The three models evaluated were for the fol-
stimulus, the driver behavior was based on the
lowing cases:
model given in Eq. 6.13 (the reciprocal spac-
ing model). The results of the simulation are
Model m synopsized in Table 6.6. They indicate that
linear 0 0 the flow increased by 70 percent and the
reciprocal spacing 1 0 velocity by 57 percent if all small cars ( 10 ft
reciprocal spacing-speed 2 1 long) are substituted for all long cars (20 ft
long), and congestion is such that there is a
Of the three models tested, the reciprocal queue of 15 cars at each light.
spacing model provided the best fit to the data. The validity of the simulation model was
For the 22 drivers involved, the value of lag verified by data collected during a Friday
rush-hour period on a three-block section of
time ( T) ranged from 0.4 to 1.6 sec (the higher
Walnut Street in downtown Philadelphia. Re-
value occurred during a test conducted in the sults of the field experiment corresponded
rain). These values can be compared with closely with the simulation results.
those found ·in Table 6.4, where the average
value of T is about 1.4 sec. The value of a 0 for
the reciprocal spacing model was approximately 6.8.4 Safety in Car Following
36 mph for expressway facilities, substantially A further example of simulating car-
greater than the values of about 20 mph found following behavior on a digital computer is
for passenger vehicles in the vehicular tunnels given by Fox and Lehman. 11 The purpose of
as reported in Table 6.4. the study was to investigate those driver and
An analysis of stability for the bus-follow- vehicle characteristics that are most important
ing experiment showed that all of the data in eliminating rearend collisions. The car-
points were in the region of asymptotic stability, following model used is that given by Eq. 6.18
whereas in previous automobile experiments with m= l, 1=2. Refinements included the
only about one-half of the data points fell in option of including the next-ahead vehicle as a
this region. stimulus, a driver sensitivity factor that reflected
Applicability of the car-following model whether the driver was accelerating or decel-
to steady-state flow was tested by observations erating, and a distance "threshold" beyond
of platoons ranging from two to ten buses. The which the driver did not adjust his velocity as
platoon experiments verified the predictions of dictated by the car-following equation.
the bus-following model; i.e., maximum flows The implications drawn from the simulation
at a rate of 1,450 buses/ hr at a constant speed studies were concerned with the reduction in
of 33 mph were measured. accidents that might follow by sharpening
102 CAR FOLLOWING AND ACCELERATION NOISE
TABLE 6.6 Flow and Velocity as a Function of Traffic Concentration
and Vehicle Size a
drivers' perceptions and minimizing response A "smooth" trip will have minor deviations,
lag. The simulation demonstrated that the a "rough" trip greater deviations from the
quantities most efficient for system behavior mean acceleration.
are (1) driver reaction time, (2) desired The concept of acceleration noise devel-
spacing, and (3) threshold boundary for rela- oped as a result of car-following studies 9 • 10
tive velocity perception. Specific suggestions and was further analyzed by Jones and Potts 36
for improving or minimizing the effect of these who suggested that the parameter might be used
three quantities are given in Fox and Lehman, to give partial answers to such questions as,
Chapter 4. 14 "How much safer and more economical is a
four-lane dual highway than a twisty two-lane
6.9 ACCELERATION NOISE road?"; or "Are teenage drivers more reckless
than other drivers?"; or "How much congestion
It is reasonable to assume that a driver is produced by increasing traffic volume and
will attempt to maintain a uniform velocity the general side activity generated by a shop-
when he is traveling along a roadway. Even ping area?"
at low volumes on a limited-access roadway,
however, he will fluctuate from his desired
velocity. In the presence of high volumes, 6.9. l Calculation of Acceleration Noise
where his velocity and acceleration are a func-
tion of the car-following laws, or in the urban Eqs. 6.35 and 6.36 do not lend themselves
situation, where traffic controls dictate his to ease of calculation of data collected in field
velocity, there will be greater fluctuations about studies. An equation for acceleration noise,
his desired velocity. which is adaptable to reduction and analysis
A measure of the fluctuations of a driver of data, is
is given by the standard deviation <T of the
acceleration about the mean acceleration and is (6.37)
defined as the acceleration noise. The mathe-
matical definition of this quantity, assuming
mean acceleration to be zero, is where T is the time in motion for the trip
segment, 6.u is taken as a constant increment
<T=
{T }% . (6.35)
of velocity change (mph), ti.ti is the time in-
{(l!T) }o [a(t)]2dt
terval (sec) for a change in velocity of mag-
nitude nit!..u (n is integer), and V 0 and VT
where a(t) is the acceleration (positive or represent the velocity (mph) at the start and
negative) at time t and T is the total time in end of the trip segment. K represents the
motion. An alternative form, in which accel- number of segments of uniform acceleration,
eration is sampled at successive time intervals
(ti.t) becomes
<T=[(l!T):2'.[a(t)]2 .6.1]% ( 6.36)
ACCELERATION NOISE 103
where M; and ni are measured along a trace highway and then entering an urban area at
of velocity versus time, with measurements be- about 5.0 min from the start of the record is
ing recorded for every change in acceleration. shown in Figure 6.8. The initial velocity V 0
For a long trip, or a trip where VT and V 0 is 54 mph and at 30 sec the driver begins to
are nearly identical, the second term of the decelerate for a curve with an advisory speed
equation can be neglected. The derivation of limit of 45 mph, gradually accelerating again
this equation is given by Drew, Dudek, and to 60 mph at 96 sec. He continues at this
Keese. 37 velocity until 132 sec, at which time he is·
Data may be collected by connecting a forced to adjust his speed to a slower-moving
recording pen directly to a vehicle speedometer vehicle. He continues with minor velocity
cable or to a fifth wheel and tracing the velocity fluctuations until 246 sec and then decelerates
directly on a moving strip of paper, where the to adjust to urban traffic and a speed-zone
distance the paper moves is proportional to restriction. For the second 5-min interval the
time. A hypothetical trace of this type is shown driver is subject to a rapid series of accelera-
in Figure 6.8. Data may also be collected by tions and decelerations, including a complete
recording the total distance traveled (ft) at stop for a traffic signal 450 sec after the start
fixed intervals (about 1 sec) and then cal- of the record. The acceleration noise for the
culating velocity and acceleration from these first 5 min of the record is determined as in
distance-time measuremer;ts. The distance may Table 6.7.
be recorded on film, advancing at one frame In the example it is assumed that ~u=2.0
per second, or on tape, where the distance is mph and the acceleration noise is calculated for
printed or punched at the proper time interval.
the first 5-min interval as shown in Figure 6.8. It
Except for punch tape that may be read directly
into a computer, the data reduction is lengthy is convenient to use a table of values of 11 2/ M
and tedious. (such as Table 6.8) in order to calculate the
A trace of a velocity-time graph for a value of n 2 1M progressively on a desk calcula-
hypothetical vehicle proceeding on a rural tor. Substituting the appropriate values in
60
I
50 I
I I
I I
I I
I I I
I I
40 I
I I I
I
..r=
0.
I I
I
E I I I
I I I
> 30 I I I
-~ I I
(..)
0 I
I
I I I
Q)
I I I
> I I
I
I
20 I I I I
I I I I
I I I I
I I I I
At 1 I ~t3 At
5 At7 I IAt91 At 11 1
10
1At2: At 4 At
6 I At8 j At
10
I
I I
I I I I
I I I I
I
00 2 3 4 5 6 7 8 9 10
Elapsed Time (min)
Value of n'l!it
lit n= I n=2 11=3 n=4 n=5 11=6 n=7 n=8
1 1.00 4.00 9.00 16.00 25.00 36.00 49.00 64.00
2 0.50 2.00 4.50 8.00 12.50 18.00 24.50 32.00
3 0.33 1.33 3.00 5.33 8.33 12.00 16.33 21.33
4 0.25 1.00 2.25 4.00 6.25 9.00 12.25 16.00
5 0.20 0.80 1.80 3.20 5.00 7.20 9.80 12.80
6 0.17 0.67 1.50 2.67 4.17 6.00 8.17 10.67
7 0.14 0.57 1.29 2.29 3.57 5.14 7.00 9.15
8 0.13 0.50 1.13 2.00 3. 13 4.50 6.13 8.00
9 0.11 0.44 1.00 1.78 2.78 4.00 5.44 7.11
10 0.10 0.40 0.90 1.50 2.50 3.60 4.90 6.40
11 0.09 0.36 0.82 1.45 2.27 3.27 4.45 5.82
12 0.08 0.33 0.75 1.33 2.08 3.00 4.08 5.33
13 0.08 0.31 0.69 1.23 1.92 2.77 3.77 4.92
14 0.07 0.29 0.64 1.14 1.79 2.57 3.50 4.57
15 0.07 0.27 0.60 1.07 1.67 2.40 3.27 4.27
16 0.06 0.25 0.56 1.00 1.56 2.25 3.06 4.00
17 0.06 0.24 0.53 0.94 1.47 2.12 2.88 3.76
18 0.06 0.22 0.50 0.89 1.39 2.00 2.72 3.56
19 0.05 0.21 0.47 0.84 1.32 1.89 2.58 3.37
20 0.05 0.20 0.45 0.80 1.25 1.80 2.45 3.20
ACCELERATION NOISE 105
6.9.2 Acceleration Noise Related to culations of a- was the total trip time, including
Roadway and Traffic stop time.
Underwood 30 reported results of observa-
Acceleration noise is mainly influenced by tions made in Australia that suggest that G
three factors-the driver, the road, and traffic seems to be a better measure of traffic conges-
conditions. An aggressive driver, with frequent tion than er, but contradicts item 7 of the
and relatively large speed changes, will have conclusion of Jones and Potts 36 by questioning
greater "noise" than a passive driver. A narrow, whether the velocity gradient is any better than
winding road or a signalized urban street will total travel time as a measure of congestion.
show greater and more frequent changes in
velocity than will occur on a multilane freeway.
Finally, a driver in congested traffic will gen- 6.9.3 Acceleration Noise of a Vehicle in
erate more acceleration noise than that obtained Traffic
at low traffic volumes, as demonstrated by the
The acceleration noise of an isolated
example shown in Figure 6.8.
vehicle was discussed in Section 6.9.1. In
Jones and Potts 36 measured acceleration
Sections 6.1-6. 7, several sim pie car-following
noise over different roads, varying traffic con-
laws for traffic were exhibited. Clearly, the
ditions, and different drivers. They reported
total acceleration noise of a vehicle in traffic is
the following conclusions:
a superposition of its natural noise (the ac-
1. For two roads through hilly coun- celeration noise at very low traffic volume)
try, rY is much greater for a narrow 2- and its response to that of its predecessors
lane road than for a 4-lane dual highway. through the law of following. The total accel-
2. For a road in hilly country, rY is eration noise of vehicles at different locations
greater for a downhill journey than for in a platoon has been measured by Herman
an uphill one. and Rothery 10 (Figure 6.9). It is noted that
3. For two drivers driving different
speeds below the design speed of a high- traffic has broadened the acceleration distribu-
way, rY is much the same. tion function so that the acceleration noise far
4. If one or both drivers exceed the down the platoon is about three times that of
design speed, rY is greater for the faster the lead car, which is effectively moving freely
driver. on the road. Figure 6.9 also shows that if the
5. Increasing traffic volume increases rY. traffic stream is proceeding in a stable manner,
6. Increasing traffic congestion pro- where stability means the damping out of small
duced by parking cars, stopping busses,
cross traffic, crossing pedestrians, etc., in-
creases rY.
7. The value of rY may be a better mea-
sure of traffic congestion than travel times N
and stopped times. u
8. High values of rY indicate a po-
tentially dangerous situation. --.....
N
~
+ -'
(")
0.05
cessor. Montroll 41 has shown that in smoothly makes its appearance. (See section 6.3.2.)
moving traffic separation distance varies only The closer the traffic reaches the limit of
slightly from the equilibrium distance s; hence, stability (2CT-7 1), the larger is the traffic
Eq. 6.13 can be linearized so that addition broadening of the acceleration noise.
of the natural acceleration /3( t) gives If Eq. 6.40 is substituted in Eq. 6.41, the
average spacing may be expressed as
.x,,+,(t+ n =C[.:i:,,(1)-.X,,+11J+f3(1) (6.39)
( 6.42)
This equation was checked with the Hol-
land Tunnel observations of Herman, Potts,
and Rothery. The traffic broadening of the
acceleration noise dispersions a/ cr0 in the tun-
nel varied from about 1.50 to 1.75, depending
on the density during the experiment. The
average time lag of 1.5 sec, which was ob-
served in car-following experiments, was sub-
stituted in Eq. 6.42, as was the observed ratio
er/ cr0 • The computed values of s were then
converted into appropriate densities (s = 1I k),
which were compared with the observed den-
sities made at the same time as er/ cr0 was
determined. These calculated values generally
did not deviate from the measured ones by
more than 10 or 15 percent.
-2 -I 0 I 2 6.10 REFERENCES
1. Highway Capacity Manual. U.S. Govern-
Acceleration in Units of 0.059 (-1.6 ft/sec2) ment Printing Office, Washington, D.C.
Figure 6.10 Acceleration distribution functions for a
(1950), pp. 2-3.
driver (A) moving with a traffic stream at approximately 2. Reuschel, A., Vehicle movements in a
35 mph and (BJ affempting to drive 5 to 10 mph faster platoon. Oesterreichishes Ing.-Arch., 4:
than the stream average. 10 193-215 (1950) (In German).
REFERENCES 107
3. Reuschel, A., Vehicle movements m a International Symposium on the Theory
platoon with uniform acceleration or of Road Traffic Flow (London, 1963).
deceleration of the lead vehicle. Z. OCED (1965), pp. 12-26.
Oesterreichischen Ing.-Archit.-Ver., 9 5: 17. Greenberg, H., An analysis of traffic flow.
59-62, 73-77 (1950) (In German). Oper. Res., 7 (1) :79-85 (1959).
4. Pipes, L. A., An operational analysis of 18. May, A. D., Jr., and Keller, H. E. M.,
traffic dynamics. J. Appl. Phys., 24 Non-integer car-following models.
(3): 274-281 (1953). Highw. Res. Rec. 199 (1967), pp.
5. Kometani, E., and Saski, T., On the 19-32.
stability of traffic flow (report-I). Oper. 19. Drew, D. R., Deterministic aspects of
Res. Soc. Jap., 2 (I): 11-26 (1958). freeway operations and control. Highw.
6. Kometani, E., and Sasaki, T., Dynamic Res. Rec. 99 (1965), pp. 48-58.
behavior of traffic with a nonlinear 20. Greenshields, B. D., A study in highway
spacing-speed relationship. Proceedings, capacity. Proc. Highw. Res. Board, 14:
Symposium on the Theory of Traffic 448-477 (1934).
Flow. Elsevier, Amsterdam ( 1961), 21. Edie, L. C., Car-following and steady-
pp. 105-119. state theory for noncongested traffic.
7. Kometani, E., and Sasaki, T., A safety Oper. Res., 9 (1):66-76 (1961).
index for traffic with linear spacing. 22. Drake, J. S., Schoefer, J. L., and May,
Oper. Res., 7 (6): 704-720 (1959). A. D., Jr., A statistical analysis of speed
8. Kometani, E., and Sasaki, T., Car follow- density hypotheses. Highw. Res. Rec.
ing theory and stability limit of traffic 154 (1967), pp. 53-87.
volume. Oper. Res. Soc. Jap., 3 ( 4) : 23. Pipes, L. A., and Wojcik, C. K., A con-
176-190 (1961). tribution to theory of traffic flow. SAE
9. Chandler, R. E., Herman, R., and Mon- Conference Proceedings-Analysis and
troll, E. W., Traffic dynamics: studies Control of Traffic Flow Symposium
in car following. Oper. Res., 6 (2): (Detroit, 1968), pp. 53-60.
165-184 (1958). 24. Michaels, R. M., Perceptual factors in
10. Herman, R., Montroll, E. W., Potts, R. car following. Proceedings, Second In-
B., and Rothery, R. W., Traffic dy- ternational Symposium on the Theory
namics: analysis of stability m car- of Road Traffic Flow (London, 1963).
following. Oper. Res., 7 (1) : 86-106 OECD (1965), pp. 44-59.
(1959). 25. Herman, R., and Rothery, R. W., Car
11. Gazis, D. C., Herman, R., and Potts, R. following and steady state flow. Pro-
.B., Car-following theory of steady state ceedings, Second International Sym-
flow. Oper. Res., 7 ( 4) :499-505 posium on the Theory of Road Traffic
(1959). Flow (London, 1963). OCED (1965),
12. Herman, R., and Potts, R. B., Single-lane pp. 1-11.
traffic theory and experiment. Proceed- 26. Forbes, T. W., Zagorski, H.J., Holshouser,
ings, Symposium on the Theory of E. L., and Deterline, W. A., Measure-
Traffic Flow. Elsevier, ·Amsterdam ment of driver reactions to tunnel con-
(1961), pp. 120-146. ditions. Proc. Highw. Res. Board, 37:
13. Gazis, D. C., Herman, R., and Rothery, 345-357 (1958).
R. W., Nonlinear follow-the-leader 27. Forbes, T. W., Human factor considera-
models of traffic flow. Oper. Res., 9 tions in traffic flow theory. Highw. Res.
(4) :545-567 (1961). Rec. 15 (1963), pp. 60-66.
14. Fox, P., and Lehman, F. G., Safety in 28. Forbes, T. W., and Simpson, M. E.,
Car-Following-A Computer Simula- Driver and vehicle response in freeway
tion. Newark College of Engineering, deceleration waves. Transp. Sci., 2 ( 1) :
New Jersey (1967), 173 pp. 77-104 (1968).
15. Chow, T., Operational analysis of a traffic 29. Newell, G. F., Theories of instability in
dynamics problem. Oper. Res. 6 (6): dense highway traffic. J. Oper. Res. Soc.
827-834 (1958). Jap., 5:9-54 (1963).
16. Oliver, R. M., Catoire, B., and Skeates, 30. Newell, G. F., Instability in dense highway
R. S., On a mathematical function of traffic, a review. Proceedings, Second
traffic flow theory. Proceedings, Second International Symposium on the Theory
108 CAR FOLLOWING AND ACCELERATION NOISE
111
112 HYDRODYNAMIC AND KINEMATIC TRAFFIC MODELS
(7.1)
or
Eq. 7.1 will be recognized as the continuity (u 1 -uw)k 1 =(u 2 -uw)k 2 (7.2)
expression for a fluid.
Eq. 7.2 is a restatement of the conserva-
tion of matter applied to the vehicles that cross
7.3 WAVES IN TRAFFIC the line and may be written in the form
To anyone slightly familiar with the be-
havior of fluids, especially with shock waves in u 2 k 2 -u 1 k1 = Uw(k 2 - k,) (7.3)
fluids, the behavior of traffic at a bottleneck
If the rate of traffic flow in region A is
appears to be acting in a shock wave-like q,, and the rate of traffic flow in region B is q 2 ,
manner. (In the late 1950s wave phenomena
in traffic even became the subject of advertise- q 1 =k 1 u 1
ments.1) The existence and behavior of
traffic shock waves has been demonstrated by and
Lighthill and Whitham. 2 But the use of traffic
wave analysis is not limited to shock waves.
Lighthill and Whitham have also demonstrated On insertion of these values, Eq. 7.3 may be
that several traffic problems can be analyzed manipulated to
by assuming a system of traffic waves. In the
following sections a variety of analytic tech-
niques is used to demonstrate and analyze If the rates of flow and the concentrations
traffic waves. are nearly equal,
(qz-q,) =!iq,
7 .3.1 Fundamental of Traffic
Shock Wave Motion ':' and Eq. 7.4 becomes
du ou ou
-=-+-u (7.7)
df ot ox
du 1s
H ere dt · t h e acce1erat1on
· o f an o b server
ou . h
moving with t h e tra ffi c stream, an d at 1s t e
Figure 7.3 Small discontinuity in concentration. 3
acceleration of the traffic stream as viewed by
an observer at a fixed point at the side of the
road.
If it is assumed that u is a function of k,
u=u(k) (7.8)
then~=~~l
of dk of Figure 7.4 Shock wave caused by stopping.'
(7.9)
OU du ok
ax--<lkax-
where
(7.13)
Substituting Eq. 7.15 in Eq. 7.16 gives
Eq. 7.1 may now be rewritten using Eq. 7.13 as
(~) ok
2
du= -k (7.17)
ok oq ok dt dk ox
-=--=-u - (7.14)
Of OX w ox
Eq. 7.17 expresses the acceleration of an ob-
Noting that q=ku, u,,. may be restated''' server moving with traffic as a function of the
. d"
density gra Jent
ok
ax·
du
Regardless of the sign of dk' the square
Substituting Eq. 7.14 in Eq. 7.10 gives
makes the value of the parentheses positive.t
Thus, from Eq. 7.17, when an observer mov- 7.3.3.1 Numerical Example of Shock
ing with the traffic fluid moves into a less Wave Analysis. The following numerical ex-
dense region his acceleration is positive; when ample of shock wave analysis was suggested
he moves into a more dense region his accel- by L. C. Edie.* Consider traffic flowing at
eration is negative. 1,000 vehicles/ hr with a concentration of 20
•
The acceleration of traffic as seen by a vehicles/ mile and a speed of 50 mph, as rep-
fixed observer can be restated, using Eq. 7.14: resented by point 1 in Figure 7.6. A truck
with a speed of 12 mph (as represented by
OU du ok
ai=<lkai=
[
-uw
du
dk ox
J ok (7.18) the slope of the radius vector 0-2 in Figure
7.6) enters the traffic stream and travels for
2 miles. Because it is impossible to pass, cars
where the quantity in brackets can take on immediately behind the truck are forced to
positive, negative, or zero values. match his speed, so that a platoon forms with
a platoon concentration of 100 vehicles/ mile,
a space mean speed of 12 mph, and a platoon
7 .3.3 An Application flow of 1,200 vehicles/hr, as represented by
Before considering an application, let us point 2 of Figure 7.6. The rear of the platoon
summarize: (i.e., the point at which the free-flowing traffic
behind the platoon catches up with the cars in
u =speed of traffic stream (i.e., speed the platoon) moves with a speed represented
of an observer moving with the by the slope of chord 1-2 in Figure 7.6, so
traffic stream) ;
q2-q1 1,200-1,000
du/ dt =acceleration of traffic stream (i.e., Uw=~= _ =2.5 mph.
2 1 100 20
acceleration of an observer mov-
ing with the traffic stream); Thus, the rear of the platoon (shock wave of
ou!ot= acceleration of the traffic stream increased concentration) is moving forward
as seen by a fixed observer; with a speed of 2.5 mph with respect to the
roadway. But the front of the platoon (the
u"' = speed of propagation of small truck) is moving forward at a speed of 12 mph.
disturbance in concentration and Therefore, the length of the platoon is growing
du at a rate of (12.0-2.5) =9.5 mph. The truck
flow=u+k d{; and
requires ~ hr for a 2-mile trip so by the time
ou/ox= speed gradient along the roadway. the truck turns off, the length of the platoon is
(~) (9.5) = 1.58 miles. At 100 cars/mile, the
du!dt= [ -uw
du J ok
dk ox +u ox
ou (7.19)
* Private communication.
2000
platoon will contain 158 cars. Note that while 7 .3.4.1 The Case of Nearly Equal Con·
the platoon is growing, although the rear moves centrations. If the normalized concentrations
forward at a speed of 2.5 mph, the rear is mov- YJ 1 and Y/z on both sides of the boundary line S
ing backward at a speed of 9.5 mph with re- are nearly equal, the situation shown in Fig-
spect to the cars in the platoon. ure 7.3 exists. The normalized concentration
After the truck turns off, flow increases to to the left of S is YJ, whereas the normalized
optimum flow (capacity) of the facility. This concentration to the right of S 1s (71+71 0 ),
condition is represented by point 3 of Figure 7 .6, where YJ + Y/o :':'. 1. In this case, let
with a flow of 1,500 vehicles/hr, a concentra-
tion of 50 vehicles/ mile, and a space mean (7.24)
speed of 30 mph. The front of the platoon and
moves with a speed represented by the slope [I - (7J 1 +YJJ]=[I - (2YJ+YJ 0 )]=[1-271]
of the chord 2-3 in Figure 7.6, or (7.25)
=
= q 3 -q 2 1,500-1,200 = _ 6 _0 m h.
in which Y/o is neglected. If Eq. 7.25 is substi-
u k 3 -k 2 50-100 P
With the rear of the platoon moving for- tuted in Eq. 7.23, the wave of discontinuity is
ward at 2.5 mph and the front of the platoon propagated with the following velocity:
moving to the rear at 6.0 mph, the platoon llw = uf[l - 271]
(originally 1.58 miles long) will dissipate in
1.58/(2.5+6.0)=0.174 hr (=10.44 min). This is the equation for the propagation of
Thus, for approximately 10 min after the truck shock waves obtained by Lighthill and
has turned off there is still a platoon of queued Whitham 1 by a more elaborate analysis.
vehicles. 7.3.4.2 Stopping Waves. Consider a line
of traffic moving with a normalized concentra-
7 .3.4 Shock Wave Behavior for tion 7J 1 and a mean vehicle velocity of
Specific u-k Relationship
(7.26)
So far, the analysis has not considered any
At a pos1tton x =x 0 on the highway, a
specific relation between the mean velocities
traffic signal causes the traffic to halt, and the
11 1 and 11 2 and the concentrations k 1 and k 2 •
stream immediately assumes a saturated nor-
If we now assume Greenshields' model,
malized concentration of 71 2 = 1, as shown in
ui=ur(l-k;I ki) (4.1) Figure 7.4. To the left of the line S, the traffic
is still moving with the mean velocity given by
If we further let Eq. 7.26 at the original concentration of 71 1.
Under these conditions the shock wave velocity
(7.20)
is obtained by substituting 71 1 =71 1 and 71 2 = 1 in
we can now write Eq. 7.23 to give
where u, is the free-flow speed of the traffic which indicates that the shock wave of stop-
stream and 711 and 712 are the normalized con- ping travels backward with a velocity of u 171 1 .
centrations on both sides of the boundary line If the signal at x=x 0 turns red at t=O, then
S. Substituting these values in Eq. 7.4 gives a in time t later, a line of cars of length u 171 1t
wave speed of will be stopped behind x 0 •
[k1 ur0 -711)]- [k2ur(l - 712)] 7.3.4.3 Starting Waves. To discuss the
(7.22)
Uw= (k1 -k2) nature of the shock wave produced by the
starting of a line of vehicles, assume that at
The relationships for 71 1 and 71 2 from Eq. t = 0 a line of vehicles has accumulated behind
7.20 may be used to simplify Eq. 7.22 with a signal located at x=x 0 • Because this line of
the result vehicles is standing still, it has a saturated con-
(7.23)
centration of 71 1 =1 (Figure 7.5). Assume that
at t = 0 the signal at x = x 0 turns green and
which gives the velocity of the line S in terms permits vehicles to move forward with a veloc-
of the normalized concentrations on either side ity of u 2 • Because u 2 =u 1 (1-71 2 ) there exists a
of the moving discontinuity. concentration of
116 HYDRODYNAMIC AND KINEMATIC TRAFFIC MODELS
(7.28) side of the figure is a flow-concentration (q-k)
curve; the right-hand side, a time-space dia-
Therefore, a starting shock wave forms as soon gram. On the q-k curve, point A represents a
as the line of vehicles begins to move. The situation where traffic is flowing near capacity
velocity of this shock wave is obtained by and the speed is reduced to a value well below
substituting 71 1 =1and71 2 =71 2 in Eq. 7.23; thus, free-flow speed. Point B represents a situation
where traffic flows at a somewhat higher speed
Uw =uf[l- (1 +112)] = - uf112= - (uf-U2)
because of the lower density. Tangents at
(7.29)
points A and B represent the wave velocities
Therefore, the shock wave of starting travels for these two situations. Now, if the faster flow
backward from x 0 with a velocity of (uf-u 2). of point B occurs later in time than that of
Because the starting velocity is small, it is seen point A, the waves of point B will eventually
that the shock wave of starting travels back- catch up with those of point A. This is shown
ward with a velocity essentially equal to - uf. in the time-space diagram of Figure 7.7. The
Another way of deducing a wave motion intersection of these two sets of waves has a
is as follows: If q=q(k,x), then slope equal to the chord connecting the two
points on the q-k curve, and this intersection
oq oq ak represents the path of the shock wave.* Note
OX ak ax that the velocity of the shock wave is often
negative with respect to the roadway and is
If k is rel.atively constant with small variations always negative with respect to the traffic.
about a mean value, then At this point it is necessary to clarify that
the waves on the time-space diagram in this
analysis are not trajectories of vehicles but lines
of constant flow and thus lines of constant
speed. The vehicles have a greater velocity than
when substituted in Eq. 7.1 gives the waves, because the speed of the vehicle
ok ok stream is represented by the radius vector,
--+u
at w -ax= 0 whereas the velocity of the waves is represented
by the tangent. Thus the reader may wish to
This differential equation has the solution regard these waves as imaginary but useful as
k=B(x-uwt), where() is an arbitrary function. an analysis tool.
Thus small changes in concentration will be
7.3.5.l Progress of a Traffic Hump. An
propagated in the direction of traffic flow with
example of the application of such analysis
velocity u"°. (See, for instance, Pipes 0 .)
techniques is the progress of a traffic "hump"
as discussed by Lighthill and Whitham. 1 A
hump is a parcel of increased density, such as
7 .3.5 Wave Flow Traffic Analysis might occur on a freeway, flowing at a constant
Thus far the discussion has centered level along the main stream, when there is a
around shock waves, which are observable in short-term influx of substantial proportions at
the field. However, analysis of several traffic one on-ramp.
situations can be performed by use of a postu- Figure 7.8 portrays the traffic waves asso-
lated system of traffic waves that are not ob- ciated with the formation of a hump. The
servable in the field, as first pointed out by speed of the front of the hump can be stated
Lighthill and Whitham. 1 (In a sense these immediately from Figure 7.8 as the velocity of
waves are analogous to radio waves that can-
not be seen and are thus difficult to compre- * The slope of the chord equals the mean
hend.) velocity of the two waves for parabolic q-k
The flow-concentration curve is an essen- curves. This is also approximately true for non-
tial part of any traffic wave analysis. In Figure parabolic curves, except those containing a vertical
4.15 it was pointed out that for any point on tangent, because the following series expression
the flow-concentration curve, the radius vector applies:
represents the traffic speed ii 8 and the tangent q(ko)-q(k1) 1
k,-k, =1·(q'(k1)+q'(k,))
represents the w.ave velocity u". Figure 7. 7
demonstrates the use of traffic waves to predict (k..-k, )'
the occurrence of a shock wave. The left-hand - 24 [q"'(k,) +q'"(k,)J+.
WA YES IN TRAFFIC 117
q x
'-~~~~~~~~~~---' k
Figure 7.7 Analysis of traffic shock wave using flow-concentration curve and
time-space diagram. The wave speeds in the space diagram are drawn parallel
to the respective tangents of the flow-concentration diagram. 1
Feed Point
the earliest wave representing the increased finite. This situation might be associated with
flow. The construction of the path of the shock the passage of a hump as previously discussed
wave is carried out as discussed in Figure 7.7 or attributable to the buildup of traffic during
and is shown in Figure 7.9. the peak period, which is followed by dissipa-
tion.
7 .3.5.2 Behavior of Traffic at Bottle· Figures 7.12 and 7.13 illustrate this analy-
necks. The study of traffic behavior at bottle- sis. First, no wave carrying a flow greater than
necks represents an important application of the capacity of the bottleneck can pass the
traffic wave analysis. A bottleneck is here de- bottleneck zone. (Here, it is necesary to re-
fined as a stretch of roadway where the capacity emphasize that waves are being discussed and
is less than that of the roadway sections up- not the actual traffic stream. Given time, the
stream and downstream from it. Figure 7 .10 actual traffic stream will eventually pass
depicts the flow-concentration curves at vari- through the bottleneck.) Consider the arrival of
ous points within the bottleneck. Note the hori- waves as shown in Figure 7.13. At the left-
zontal line at the left of the diagram. This line hand side waves of low flow (volume) are
indicates how the speed of the stream suddenly seen; hence, high-velocity waves are arriving
drops as the bottleneck is reached. Figure 7.11 at the bottleneck area. As they arrive, there is
indicates the passage of the traffic waves a jump from one flow-concentration q-k curve
through the bottleneck area. to another of lesser capacity (maximum flow)
When the approach flow attains the point (Figure 7 .12) , resulting in a decrease in
where it exceeds the capacity of the bottleneck, velocity. As the arriving flow increases, arriv-
the duration of such a condition would be ing wave velocities are less. Eventually, the
• .. ..
F=-k(:~r
in a negative wave velocity, the wave leaving
the bottleneck in a rearward direction with the (7.30)
velocity indicated by point C. As waves start
toward the rear they meet arriving waves; the where F may be regarded as a transfer func-
intersection of these two sets of waves forms a tion. This allows Eq. 7 .17 to be rewritten
Flow-Concentration
Curve on Main Road
Flow-Concentration
Curves Inside Bottleneck
k
Figure 7.10 Changes in flow-concentration curve at a bottleneck.'
Distance
Main Road
Bottleneck
Main Road
Time
Figure 7.11 Passage of waves through a bottleneck whose capacity exceeds the arrival
flow rate.'
(7.33)
~=F~ (7.31)
F=-c 2 /k
dt ox Equating the values of F from Eqs. 7.33 and
Up to now the form of the u-k relationship has 7.30 gives
not been specified. Specific forms can be in-
vestigated, as well as the resulting flow-concen-
tration models. Postulating a form for F is
-k(:~r =- ~ (7.34)
..
120 HYDRODYNAMIC AND KINEMATIC TRAFFIC MODELS
Velocity with Which Wave Enters Bottleneck
q
B
•
Velocity with
Which Wave Leaves Bottleneck
k
Figure 7.12 Flow-concentration curve for reflection of waves from a bottleneck as shown in
Figure 7.13. 1
Main Road
Bottleneck
Main Road
Figure 7.13 Formation of shock wave in front of a hump as it enters a bottleneck of inadequate capacity. 1
the negative sign must be used to obtain a posi- ogy) approaches to the theory of traffic flow.
tive u/ c: Care must be exercised, however, as to the
extent to which one carries this reasoning.
k
u= -c&-=c[&ki-&k] (7.37)
kj 7.4.2 Richards' Equation of State
This is the same model as Eq. 4.2 when um is Richards ' has implied a state equation of
substituted for c. In Section 4.3.2 the maxi- the form
mum flow for this model was shown to be
qm=umk/e. (7.38)
Note that the model of Eq. 7.37 is the where A is a constant having the dimensions of
same as that obtained in the car-following speed-concentration. 5 Equating the expressions
theory where l=l and m=O (See Table 6.5). for F in Eqs. 7.38 and 7.30 and performing
Because of this equivalence, there is a tendency algebraic manipulations, one obtains
to look on this as a unifying of the microscopic
(car-following) and macroscopic (fluid anal- du/dk= -A (7 .39)
QUANTITATIVE ANALYSIS (SIGNALIZED INTERSECTION) ·121
Solving the differential Eq. 7.39 with the stant the signal turns green. Because the re-
boundary conditions: sponse of a traffic system is slow, traffic does
not start immediately on receiving the green
indication, and some cars may still be required
to come to a complete stop even after the green
has been displayed. Point B represents the
yields maximum distance from the stopline for which
(7.40) complete stoppage is required. X A represents
the distance of point A from the stopline, and
Note that this is equivalent to assuming that Xu represents the distance of point B from the
A= url ki as a traffic fluid state. Eq. 7.40, of stopline.
course, is the same as Greenshields' model of
Eq. 4.1.
XA =tr tan a
(7.47)
This gives two values of arrival density-
one for flows less than maximum on the low
density side and one for flows less than maxi-
mum on the high density side. The lower
~~
value is accepted and the higher is discarded.
Maximum flow (q",) will occur at =0: Thus,
kA= 0.5kj(l-\fl-p)
X tr(Pqm) F ( trqm)
A= kj-0.5kj(l-(1-p)'l2] := A ---r;
(7.48)
where
Velocity of waves representing jam density is p p
expressed FA= 1-0.5[1- ( l-p)l/2] 0.5 +0.5(1-p)'/2
(7.49)
~~I
2
,c=l'J = uf- ~:ki_ = - uf (7.46)
--(trqm)F
qA=Pqm
where p is a decimal fraction (i.e., 0::::; p::::; 1 ) , IS II
(7.50)
tance will the queue of cars occupy at the time Figure 7.16 Sketch of angles for computation of queue
length at a signal.
the signal turns green? Assuming that Green-
shields' flow-concentration model has been ac-
trqm O.211 ( ,30 )(1,800)
cepted, XA =FA~=
3 600 200 7 .6.1 Pacey's Diffusion Study
=0.015825 miles=83.6 ft.
Pacey 12 has postulated that the speeds of
vehicles in a platoon are normally distributed,
7.6 PLATOON DIFFUSION from which he deduced the distribution of
travel times. The mathematical derivation is
One common objective in timing traffic given in Appendix F-1. The result of Pacey's
signals is to synchronize them so that a platoon method is shown in Figure 7.19.
of cars being released from one signal arrives at
the next signal at such time that it can pass
through this second signal without interruption. 7 .6.2 Diffusion Model of Grace and
The usual practice is to assume that all cars in Potts
the platoon move with the average speed of By first changing to new variables and then
traffic. This simple approach breaks down be- changing back Grace and Potts 13 have shown
cause platoons do not remain in a compact state that Pacey's model corresponds to a unidimen-
but tend to diffuse as they move away from sional fluid diffusion equation
the point of their formation. For example,
Graham and Chenu 10 have experimentally
found the percentage of the original number of
cars in the platoon that remain in the platoon at TABLE 7.1 Values of FA and Fn
various distances along a highway from the
point of origin (Table 7 .2). When platoons are Greenshields' Greenberg's
Flow-Concen- Flow-Concen-
represented by frequency distributions of per- tration Model tration Model
cent of total platoon flow versus arrival time,
the result is as shown in Figure 7 .17. It will be p FA Fn FA Fn
noted that as distance increases, the peak of the 0.1 0.103 0.105 0.101 0.105
distribution becomes lower, with an increased 0.2 0.211 0.223 0.204 0.220
tendency for the distribution to "tail" out to the 0.3 0.327 0.356 0.310 0.350
0.4 0.451 0.508 0.421 0.498
• rear. Part of the problem lies in the fact that 0.5 0.585 0.686 0.536 0.668
platoon leaders from one signal cycle to another 0.6 0.735 0.901 0.662 0.874
do not travel at the same speed. Figure 7.18 0.7 0.904 1.169 0.798 1.130
illustrates the fact that arrival times of platoon 0.8 1.105 1.527 0.954 1.469
leaders become more dispersed as the distance 0.9 1.368 2.079 1.146 1.983
1.0 2.000 4.000 1.582 3.787
from the point of platoon formation increases.
124 HYDRODYNAMIC AND KINEMATIC TRAFFIC MODELS
01d OT= a.2 ( 0 2 K! ox 2 ) (7.53) a= s/il=diffusion constant;
r= 1/2 t 2 ;
which is equivalent to K= (x/u) -t;
(oklot) +u(oklox) = u a.2(0
2 2
k!ox 2
) (7.54) x= distance from stopline at original sig-
nal; and
where t= time since display of green signal.
u =mean speed of vehicle speed distribu- More importantly, Grace and Potts have
tion; ( 1 ) pointed out that the establishment of the
s= dispersion of speed distribution; ratio a computed from stream measurements
as a parameter in the kinematic diffusion theory
serves to link it to stream flow and (2) de-
scribed a suitable way of timing traffic signals
by representing platoons as trapezoidal pulses.
TABLE 7.2 Percentage of Vehicles
Remaining in a Highway Traffic Platoon a
7.6.3 Platoons as Trapezoidal Pulses
Vehicles When the green indication of a traffic sig-
Distance Remaining in nal is displayed, the flow does not immediately
(miles) Platoon(%) jump to its maximum value but builds up over
0.25 91 a period of several seconds. The solid line in
0.50 85 Figure 7.20 is a plot of flow versus time for
0.75 80 the flow leaving a traffic signal (assuming there
1 77 is sufficient demand to assure flow throughout
the entire green interval). The other lines in
•From Graham and Chenu.10 Figure 7.20 indicate the (time) shape of the
4.0
"'
C1l
.1?
.c
C1l
>
3.0
....0ctJ
I-
....c
C1l
u
....
C1l
a.. 2.0
>
u
c
C1l
::i
O'
....
C1l
u..
-g 1.0
.c
....
0
0
E
(/)
--
10 20 30 40 50 60 70 80 90 100 110
Seconds After Green
Figure 7.17 Frequency distributions of vehicle orrival at various stations." Station l, 0.03 miles from stop·
line of signal; station 2, 0.21 miles; station 3, 0.34 miles; station 4, 0.50 miles; station 5, 0.65 miles.
PLATOON DIFFUSION 125
20 Station 1 2 3 4 5
Distance (miles) 0.03 0.21 0.34 0.50 0.65
16
"'
(l)
:"Q
.c
(l) 12
>
-ro....
0
I- 2
....c::
(l)
u 8
....
(l)
Cl..
60 70
1.2
- - - - - Predicted by Diffusion Theory
- - - - Observed
1.0
-0.8
~ VI
.......
.c(l)
2. 0.6
:;:
.2
u..
0.4
0.2
0
10 20 30 40 50 60 70 80 90 100 110 120
Time After Green (sec)
Figure 7.19 Result of Pacey's diffusion prediction.
126 HYDRODYNAMIC AND KINEMATIC TRAFFIC MODELS
flow pulse after it has been diffused by passage derivation, the reader is referred to Grace and
down the roadway by distances of 0.25 and Potts. 13
0.50 mile. Definitions
For purposes of analysis, consider the
K = maximum density (e.g., in cars/
trapezoidal pulse indicated by the dotted lines ft) of the platoon at t=O;
of Figure 7.21. It is most convenient to treat Q= Ka maximum flow (e.g., in cars/
the front and the rear of the pulse separately. sec) of platoon at x=O;
Only the significant steps for the case where a= initial length (in ft) of a pla-
b 1 =b 2 (Figure 7.21) are given. For a detailed toon;
--
0
t:T
~-
1.0
0.8
.2
LL 0.6
"O
Q) 0.4
N
I
I t=O \ K
t = t'
\
-a 0
Distance, x
Figure 7.21 The density k (x, t) of a platoon sketched (not to scale) as a function
of distance x down the highway for t = 0 and a later time t = t'. The solid lines
indicate an initial rectangular pulse, showing how it spreads as the platoon moves
down the highway. The dashed lines indicate an initial trapezoidal pulse. 13
PLATOON DIFFUSION 127
bpb 2 = length (in ft) of front (rear) of tension, the ratios of N rl A and N / Q are
platoon in which the density given in Table 7.3 (for a diffusion constant
builds to (falls from) its maxi- a.=0.15).
mum value (Actually, only the
Computational Procedure
case where b, = b 2 = b has been
treated); To apply these results to allow for platoon
erf(z) =---=
2
v"'
1:
o
exp(-x 2 )dx
diffusion in the design of the coordination of
two successive traffic lights, the following pro-
cedure is suggested:
For the Front of the Platoon 1. Under conditions of the maximum
traffic flow to be coped with, the initial flow for
Nr= V2 K(rtt-x-112b) + (112b )K u a. 12 2 2
all lanes combined is determined from a time
[(x 2 + 1!2)erf(z) + (ll7r)Z exp(-z 2 )J:,z, arrival study, from which Q (the maximum
(7.55) flow in cars/sec) and b,u and b 2 u (sec) (the
where Nr= number of cars that are stopped build-up and fall-off times) are estimated.
at the front of a platoon with an 2. The distribution of car speeds is deter-
offset xi u sec from the previous mined and the mean speed ii and dispersion sa-
signal and a preset (early start) (both in ft sec-1 ) are estimated, from which the
of the green (xi u- t) sec; diffusion constant a.= a-I u is calculated.
z, = (alu-t)la.tVf.; and 3. From the distance x (ft between the
Z 3 = [(x/ u- t) + blii]/ a.tVf.. traffic lights), the offset time xi ii (sec) is
calculated.
For the Rear of the Platoon
4. If Nr and Nr-the expected number of
Nr= 112 K(x-iit-112b) + ( l/2b )Kii 2 a. 2 t 2 cars to be stopped, respectively, at the front
[(z 2 + 112) erf(z) + (lly;)exp(-z 2 )Jz1z, and rear of the platoons-are decided upon,
(7.56) then Nrf Q and N/ Qare calculable.
5. The appropriate values of the preset
where Nr= number of cars that are stopped time x (sec) and the extension time -x(sec)
at the rear of a platoon when the are then read from Table 7.3 where x=
offset is as above and the exten- (xlu-t).
sion (late termination) of the
green is - (xi ii- t); and Numerical Example
z,= (x!u-blii-t)!a.tVL
Consider the following data: Q=two cars/
For various values of offset and preset or ex- sec (for a multilane facility), x=2,640 ft (the
•The odd lines give the values (sec) of the ratio of the number of cars stopped at the front of the
platoon to the maximum initial flow for the specified values of offset and preset. The even lines give
the values (sec) of the ratio of the number of cars stopped at the rear of the platoon to the maximum
initial flow for the specified values of offset and extension.The diffusion constant a.=0.15. 13
128 HYDRODYNAMIC AND KINEMATIC TRAFFIC MODELS
distance between signals), a= 30 mph= 44 ft/ Laboratories, and by the System Development
sec (the average speed), cx.=s/u=0.15 (the Corporation. Until now, no data have been
diffusion constant, b 1 / a= 4 sec (the time to collected under conditions prescribed by the
build up to full-flow condition), and b 2 /u=4 theory; however, recent analyses based on
sec (the time to decay from full-flow condi- available data demonstrate the attractiveness of
tion). It is a matter of engineering judgment the approach. Prigogine and Herman 17 dis-
to select the number of cars to be stopped at cuss the development of the theory, and a brief
the front and rear of the platoon. If Nt=2 and review of assumptions, results, and some nu-
Nr=2, the offset as computed from the data is merical applications is given in the following.
60 sec.
The ratio N ti Q = 1.0. Referring to Table
7.3, line three lists values of NfQ for offsets of 7 .7 .1 Velocity Distribution Function
60 sec. A ratio of 1.0 falls between a preset of An essential part of the theory is the in-
4 and 5 sec; a preset of 5 sec is selected. troduction of the "velocity distribution func-
The ratio N rQ = 1.0. Referring to Table tion" f (x,u,t), which is analogous to the
7.3, line four lists values of Nr/ Q for offsets probability of finding a particle in a gas at x
of 60 sec. Because the ratio 1.0 falls between and t with a specific momentum. In absence of
7 and 8 sec, an extension of 8 sec is selected. any interactions, this distribution must satisfy
Suppose now the normal duration of the the continuity equation for a given region*:
green interval at the second intersection is 40
sec. The green interval as designed here would (7.57)
start 60- 5 = 55 sec after the start of the green
at the first intersection in order to account for
forward diffusion of platoons. Ordinarily, the
green at the second intersection would terminate
7 .7 .2 Kinetic Equation
at 60 + 40 = 100 sec after the start of the green Actm\lly, three different main features of
at the first intersection. As designed to account multilane traffic are recognized and treated as
for backward platoon diffusion, it will ter- separate processes in the theory. They are the
minate 100 + 8 = 108 sec after the start of the relaxation process, or the speeding-up process,
green at the first intersection. that expresses the attempts of drivers to achieve
their own desired speeds; the interaction proc-
ess, or the slowing-down process, that arises
7 .6.4 Other Platoon Studies and in the conflict between a faster driver and a
Comments slower driver; and the adjustment process that
reduces the variance around the mean speed. 18
Edie et al. 20 have performed experimental
These processes are then expressed in a kinetic
studies to test the kinematic model of Grace
equation:
and Potts. They state: "The kinematic model
well describes the spreading of platoons in
medium traffic without interference."
Hillier and Rothery 21 have demonstrated
dr =ir + u( ~~) =(~~)relaxation
that consideration of platoon phenomena can
improve signal timing in the delay-difference +( ir\nteraction +(~~)adjustment (7 ·53 )
of offset approach. As written, this equation describes the time
Nemeth and Vecellio 15 have concluded evolution of the speed distribution, f[u,k(x,t) ,t],
that Pacey's model is valid. of cars on a homogeneous highway at location
x and time t, where k(x,t) is the concentration
of cars at that point. ·
7.7 BOLTZMANN-LIKE BEHAVIOR
OF TRAFFIC 7.7.2.1 Relaxation. The distribution of
desired speeds is defined as / 0 (x,u,1) and / 0
The fact that traffic under low densities (x,u,t) dxdu is the number of cars at time t in
is essentially the flow of individual cars, whereas
at high densities flow is by platoons, suggests
" Prigogine, Herman, et al. considered the
an analogy between traffic and gases. The early distribution of desired speeds to be independent of
work on this approach was performed by density (concentration). More recent work by
Prigogine. 16 Continuing work has been carried Andrews''·"" discusses the case of this distribution
on by him, by the General Motors Research being a function of concentration.
BOLTZMANN-LIKE BEHAVIOR OF TRAFFIC 129
the interval dx with desired speeds between u where
and u +du. When the distribution f is different
from f 0 it will relax toward f 0 with a time
constant T. The relaxation process is described
by
k
(of)
-=71
kj
- =-U-fo)IT (7.59)
ot relaxation Thus, for very small values of 71, q is linear
The desired speed distribution can exist only with increase in concentration (see, for ex-
at low concentrations. ample, Figure 7.22); the deviations in linearity
are determined by the dispersion of the desired
7.7.2.2 Interaction. The interaction proc- speed distribution function.
ess describes the effects of various cars on each To summarize, as k ~ 0, f ~ f 0 ; i.e., for
other. The probability of passing (P) is re- a very small density, or light traffic, the actual
lated to the interaction phenomenon, which is speed distribution is in fact the desired distribu-
described by tion of each individual,
( ~1t ). .
1nteractlon
= (1-P)k(u-u)f (7.60) f= fo
l+Tk(l-P)(u-a)
(7.62)
u 0
= 100 ft/sec
u0 = 100 ft/sec
•
20 25
20
0Q)
--....
VJ
!::..
15
--;r
C"
10
Free-Speed
Distribution
No. Function /o iio
Exponential cl~) Ur
-;;( exp c-u) Ut
kin/ut[exp(- u/ur)]
(fut)·'exp(fur)-1 {"' dx x- 1exp(- x)
A+'Yu } (fur)-1
2 Modified
exponential ( 4kinu) ex
ur p
(=-")
ur
Ut
(4kinu/ur 2 )[exp( - 2u/ur)]
A+ 'YU 2(fur)·1 { 1 - 2(fut)-1[ exp ( r:,) J
! r,,,
00
dxx- 1exp(-x)}
2< J-1
SUMMARY 131
Indeed, every driver experiences that at 25
some moment, with increasing traffic u0 = 80 ft/sec
concentration, his speed drops abruptly,
and he is trapped in a collective flow to
which he himself contributes. 20
The results of the kinetic theory are
nevertheless impressive, as the fl.ow-concen-
tration curves in Figures 7.22, 7.23, and 7.24 15
indicate. The material presented here is in-
tended to give the reader the basic assumptions
and conceptual formulation of the kinetic
theory of traffic flow, but the interested reader
may refer to the referenced literature for more
details.
- - - - - f11
----f12 r-,
-------- f13
I I
0.06
- - - - - - f14
I I
I I
:::J
I
;j:°
r:'
-1 I L-1
0
·.;;
u
r:
l_J
I I
:::J 0.04
u.
>
u
r:
I r----
•I I
I
I
I
<I>
:::J
er
I
L----1 I
.... :·-,• : I
I
<I>
.--J I
u.
L.--{
r---- I I
F-·--::c_-_-__.~!~ L_l
0.02 I
i I
t=-
I=:-..:..:!
- - __ J
1
I l
I
r- r·--:. . ___ _ I
L--
L.-+----
ol:===::!::=:=L_____JL.__....c=.~::.:::i'=..:=-:J~--1-----1:'-~-~-~-~-1-.-__J;;:;;:;::;:s::t.:...::..=.=:1
0 10 20 30 40 50 60 70
Speed, u (ft/sec)
Figure 7.25 The frequency function, f(u), of the observed speed distributions from the Port Authority of New
York and New Jersey. The concentrations corresponding to distributions fn, f12, f,,, and fu are 35.0, 62.0, 74.3,
and 93.0 veh/mile, respectively.
0.08
-------- f21
f22 n
. L.1
I .
0.06 - - - · - f23
., _r--1
::J
;:;:- rJ I
c· fl . _j '
'L--1_ __
0
·.;:::
u r-~
. I1 r-"'
' .. _"t_
..,
c
::J
u.. 0.04 ! u r- _,i
I
i'
I ''
I
~
> L--
u
c
Q)
I ''
~
::J
C'
--,I
,_
I
Q)
u..
~
I
~
I
0.02
r i
!
L.L. ----1
• 40 50 70
Speed, u (mph)
Figure 7.26 The frequency function, f(u), of the observed speed distributions deduced from the Bureau of
Public Roads data in Pipes. 3 The concentrations corresponding to f21, f~2, and f,, are 32.7, 49.5, and 88.4 veh/
mile, respectively.
134 HYDRODYNAMIC AND KINEMATIC TRAFFIC MODELS
Highway Research Board Special Re- Amsterdam, New York (1961 ), pp.
port 79 (1964), pp. 7-35. 158-164.
3. Pipes, L. A., Hydrodynamic approaches- 17. Prigogine, I., and Herman R., Kinetic
Part I, pp. 3-5. In D. L. Gerlough and Theory of Vehicular Traffic. American
D. G. Capelle (ed) An Introduction to Elsevier Publishing Co., New York
Traffic Flow Theory. Special Report 79, (1971), pp. 17-54.
Highway Research Board, Washington, 18. Herman, R., Lam, T., and Prigogine, I.,
D.C. (1964). Kinetic theory of vehicular traffic: com-
4. Pipes, L. A., Vehicle accelerations in the parison with data. Transp. Sci., 6 ( 4) :
hydrodynamic theory of traffic flow. 440-452 ( 1972).
Transp. Res., 3 (2) :229-234 (1969). 19. Payne, H. J., Models of freeway traffic
5. Pipes, L. A., Topics in the hydrodynamic and control. In G. A. Bekey (ed.)
theory of traffic flow. Transp. Res., 2 Mathematical Models of Public Sys-
(2): 143-149 (1968). tems. Simulation Counc. Proc. Ser., 1
6. Greenberg, H., An analysis of traffic flow. (1):51-61 (1971).
Oper. Res., 7 (1) :79-85 (1959). 20. Edie, L. C., Foote, R. S., Herman, R., and
7. Richards, P. I., Shock waves on the high- Rothery, R., Analysis of single lane
way. Oper. Res., 4 (1) :42-51 (1956). traffic flow. Traffic Eng., 33 (4):21-27
(1963).
8. Pipes, L. A., Car following models and
the fundamental diagram of road traffic. 21. Hillier, J. A., and Rothery, R., The syn-
Transp. Res., 1 (1):21-29 (1967). chronization of traffic signals for mini-
mum delay. Transp. Sci., 1 (2) :81-94
9. Rorbech, J., Determining the length of (1967).
the approach lanes required at signal-
22. Andrews, F. C., A statistical theory of
controlled intersections on through
highways-an application of the shock traffic flow on highways. III. Distribu-
wave theory of Lighthill and Whitham. tions of desired speeds. Transp. Res., 7
(3) :223-232 (1973).
Transp. Res., 2 (3) :283-291 (1968).
10. Graham, E. F., and Chenu, D. C., A study 23. Andrews, F. C., A statistical theory of
traffic flow on highways. IV. Semi-
of unrestricted platoon movement of
traffic. Traffic Eng., 32 (7): 11-13 empirical steady state theory. Transp.
(1962). Res., 7 (3) :233-241 (1973).
11. Lewis, B. J., Platoon movement of traffic 24. Anderson, R. L., Herman, R., and Prigo-
from an isolated signalized intersection. gine, I., On the statistical distribution of
Highw. Res. Board Bull. 178 (1958), traffic flow. Oper. Res., 10 (2) :180-
196 (1962).
pp. 1-11.
12. Pacey, G. M., The Progress of a Bunch of
Vehicles Released f ram a Traffic Signal. 7 .10 RELATED LITERATURE
Research Note No. RN/2665/GMP.
Road Research Laboratory, London Hydrodynamic and Wave Models
(1956), mimeo.
Bick, J. H., and Newell, G. F., A continuum
13. Grace, M. J., and Potts, R. B., A theory model for two-directional traffic flow. Q.
of the diffusion of traffic platoons. Oper. Appl. Math., 18 (2): 191-204 (1960).
Res., 12 (2):255-275 (1964). De, S. C., Kinematic wave theory of bottle-
14. Herman, R., Potts, R. B., and Rothery, necks of varying capacity. Proc. Camb.
R. W., Behaviour of traffic leaving a Philos. Soc., 52 (Part 3):564-572 (1956).
signalized intersection. Traffic Eng. Foster, J., An investigation of the hydrody-
Control, 5 (9) :529-533 (1964). namic model for traffic flow with particular
15. Nemeth, Z. A., and Vecellio, R. L., In- reference to the effect of various speed-
vestigation of the dynamics of platoon density relationships. Proc. Austr. Road Res.
dispersion. Highw. Res. Rec. 334 Board, 1: 229-257 (1962).
(1970), pp. 23-33. Franklin, R. E., The structure of a traffic shock
16. Prigogine, I., A Boltzmann-like approach wave. Civil Eng. Public Works Rev., 56
to the statistical theory of traffic flow. (662):1186-1188 (1961).
Proceedings of the [First) [International) Franklin, R. E., On the flow-concentration
Symposium on the Theory of Traffic relationship for traffic. Proceedings of Sec-
Flow (Warren, Mich., 19 59). Elsevier, ond International Symposium on Theory of
PROBLEMS 135
Traffic Flow (London, 1963). O.E.C.D. mental validation of two Boltzmann-type
(1965), pp. 120-128. statistical models for multi-lane traffic flow.
Lighthill, M. J., Dynamic instability of trans- Transp. Res., 5 (3):211-224 (1971).
port systems: the hydrodynamic analogy. Herman, R., and Lam, T., On the mean speed
Oper. Res. Q., 8 (3):109-114 (1957). in the "Boltzmann-like" traffic theory: ana-
Oliver, R. M., Fluid Flow Models of Traffic . lytical derivation. Transp. Sci., 5 (3): 314-
Streams. Institute of Transportation and 327 (1971).
Traffic Engineering, University of California, Herman, R., and Lam, T., On the mean speed
Berkeley (1962), 52 pp. mimeo. in the "Boltzmann-like" traffic theory: a
Pipes, L. A., Wave theories of traffic flow. J. numerical method. Transp. Sci., 5 ( 4) : 418-
Franklin Inst., 280 ( 1 ) : 23-41 ( 1965). 429 (1971).
Wojcik, C. K., and Pipes, L. A., A study in
wave theories of traffic flow. Paper pre-
sented at the National Meeting of the Opera- 7.11 PROBLEMS
tions Research Society of America, held in 1. Given a single-lane roadway as shown
Santa Monica, California, May 1966. in the sketch and the three conditions at loca-
Platoon Phenomena
Athol, P., Headway groupings. Highw. Res.
Rec. 72 (1965), pp. 137-155.
Buckley, D. J., Tomlin, J. A., and Minson,
W. G., Delays to traffic platoons. Proc.
A I B I c
I ,
Austr. Road Res. Board, 3:364-376 (1966). ,---1/4 Mile~
Daou, A., The distribution of headways in a
platoon. Oper. Res., 12 (2) :360-361 School Zone,
(1964). Operates Only for 15 Min
Daou, A., On flow within platoons. Austr.
Road Res., 2 (7) :4-13 (1966).
Gerlough, D. L., Some problems in intersection tions A, B, and C as: qA=l,250 vehicles/hr,
traffic control, pp. 10-27. In R. Herman uA=50 mph; qn=l,000 vehicles/hr, un=20
(ed.), Proceedings of [First] International mph; qc=l,200 vehicles/hr, u<'=40 mph. De-
Symposium on the Theory of Traffic Flow. termine (a) uw(AB) (speed of shock wave
Elsevier, New York (1961). AB), (b) llw(BC) (speed of shock wave BC),
Grace, M. J., and Potts, R. B., Diffusion of ( c) length of queue at end of 15-min period,
traffic platoons. Proc. Austr. Road Res. ( d) time to dissipate queue after the end of
Board, 1 :260-267 (1962). period, and ( e) distance from beginning of
Kleinman, S., A mathematical model for pla- school zone to point of dissipation.
toon formation. Transp. Sci., 6 (2) :202-204 2. A line of traffic is moving with a free-
(1972). flow speed, uf, of 35 mph and a concentration
Newell, G. F., Platoon formation in tunnel of 40 vehicles/ mile. This traffic stream is
traffic. Oper. Res., 7 (5) :589-598 (1959). stopped for 30 sec at a signal. (a) What would
Seddon, P. A., The prediction of platoon dis- be the velocity and direction of the wave of
persion in the combination methods of link- stopping? (b) What would be the length of
ing traffic signals. Transp. Res., 6 (2): the line of cars stopped for the signal? ( c) How
125-130 (1972). many cars would be stopped for the signal?
(Jam concentration, kj = 200 vehicles/ mile.)
3. Consider a one-lane approach to a traf-
Boltzmann-Like Traffic Models fic signal with an approach volume of 400
Andrews, F. C., A statistical theory of traffic vehicles/hr, a capacity volume of 1,600 ve-
flow on highways. I. Steady flow in low- hicles/hr, a red signal time of 25 sec, and a
density limit. Transp. Res., 4 ( 4): 359-366 jam density of 200 vehicles/ mile. Based on the
(1971). information provided in Section 7.5 and assum-
Andrews, F. C., A statistical theory of traffic ing Greenshield's flow-concentration model,
flow on highways. II. Three-car interac- what distance will the queue of cars occupy
tions and the onset of queueing. Transp. at the time the signal turns green? What is the
Res., 4 (4) :367-377 (1970). distance from the stopline for which complete
Gafarian, A. V., and Munjal, P. K., An experi- stoppage is required?
• •
Chapter 8
QUEUEING MODELS*
8.1 INTRODUCTION control point, such as a left-turn slot or a single
exit lane for a garage.
A desirable goal for transportation engi- In section 8.3 the analysis is extended to
neers is to design and operate facilities that consider several channels of service; for ex-
minimize delay to the users. Delay resulting ample, several parallel toll booths or the dif-
from congestion is a common phenomenon ferent stalls of a parking facility. In this sec-
associated with many types of transportation tion the case of a user who does not get served
problems. Vehicles wait in line on access is also considered; for example, the person
ramps for an opportunity to enter a freeway; seeking a parking space but who then continues
pedestrians queue up on a crosswalk in antici- on to another destination when none is found.
pation of a gap in road traffic or at a turnstile in Much urban traffic engineering is related
a transit station; left-turn slots must be suffi- to the operation of urban intersections. An
ciently long to store the maximum number of understanding of delay at these intersections is
vehicles that can be expected to wait for a necessary to obtain the greatest efficiency from
left-turn signal. existing and planned transportation systems.
How long a user must wait, or what is The analysis of delays at intersections is con-
the number of units waiting in line, or the sidered in section 8.5, beginning with an
proportion. of time that a facility might be in- analysis of unsignalized intersections. Queue-
active (an empty parking stall, for instance)? ing models for more complex intersection con-
Queueing models, employing the methods of trol, such as pedestrian control or traffic-signal
probability and statistics, provide a means by control, are also considered in this section.
which it is possible to predict some of these A final application of queueing theory, the
delay characteristics. treatment of delay on roadways, is included in
Queueing theory was first developed early section 8.6. Except for the detailed develop-
in the twentieth century to deal with problems ment of the formulas given in section 8.2, this
of telephone switching. Following World War chapter avoids detailed mathematical develop-
II queueing was accepted for use in a wide ment, but does present the theorists' assump-
range of situations. Adam 2 considered the tions and some results of interest. The model
problem of pedestrian delay at an unsignalized for service through a single channel is devel-
intersection in 1936. Tanner 3 expanded on the oped in detail because it demonstrates the
pedestrian problem in 1951, and in 1954 Edie 4 relationship between probability theory and the
evaluated delays at toll booths by applying behavior of waiting lines. Readers interested in
queueing models to an analysis of their opera- further theoretical development of queueing
tion. In the same year Moskowitz 5 reported models should consult textbooks such as
on an empirical study of vehicles waiting for a Haight, 6 Prabhu, 7 Cox and Smith, 8 or Newell. 9
gap in traffic.
The purpose of this chapter is to present
some of the results of studies of probability 8.2 FUNDAMENTALS OF QUEUEING
models of traffic delay. Section 8.2 introduces THEORY
some elements of queueing or waiting-line Queueing theory draws heavily on prob-
theory. The examples used in section 8.2 are ability theory. To mathematically predict the
concerned with delay problems that occur characteristics of a queueing system, it is
when all users pass through a single-channel necessary to specify the following system char-
•
acteristics and parameters:
* In preparing this chapter, the authors have
drawn freely on the material of Cleveland and A. Arrival pattern characteristics: ( 1)
Capelle.1 References to that material are acknowl- average rate of arrival and (2) statistical dis-
edged at this time. Specific references are noted tribution of time between arrivals;
only for the illustrations used in this text. B. Service facility characteristics: (1)
137
138 QUEUEING MODELS
service time average rates and distribution and actly n items (this includes all items being
(2) number of customers that can be served served as well as those waiting to be served).
simultaneously, or number of channels avail- The value of n may be either 0 or some positive
able; integer.
C. Queue discipline characteristics, such If the average arrival rate is called A, the •
as the means by which the next customer to be average interval between arrivals is 1I A. If
served is selected; for example, "first come the service rate of the system is µ, the average
first served," or "most profitable customer service time is 1 Iµ. The ratio p =A/µ, some-
first." times called the traffic intensity or utilization
To facilitate reference to these charac- factor, determines the nature of the various
teristics, a short notation in the form a/b/ c has states. If p < 1 (that is, A<µ) and a sufficiently
come into use. In this notation a letter denoting long time elapses, each state will be recurrent.
the type of arrival pattern is substituted for a; This means that there is a finite probability of
a letter denoting the type of service is sub- the queue being in any state n. If, on the other
stituted for b; a number designating the number hand, p?_ 1, every state is transient and the
of service channels is substituted for c. Sym-
queue length (the number in the system) will
bols in the first two places are as follows:
become longer and longer without limit. A
M= exponentially distributed (i.e., ran- fundamental theorem states that the queue will
dom) interarrival or service time; be in equilibrium only if p < 1.
D= deterministic or constant interarrival An understanding of the characteristics of
or service time; queueing systems can be obtained from simple
G = general distribution of service times;
cases. Consider the case of a single-channel
GI= general distribution of interarrival
times; queueing system with a mean random Poisson
Ek= Erlang distribution of interarrival or arrival rate of A customers per unit of time and
service times with Erlang parameter where service times are independent and dis-
k. tributed exponentially with a mean rate µ. Let
P,, ( t) be the probability that the queueing
Thus, MIG! 1 designates a queue with random
system has n items at time t. Consider the
arrivals, general service distribution, and one
server. situation at time t+M where Mis so short that
In some discussions it may be desirable to only one customer can enter or leave th~
indicate queue length limitations and queue system during this time.
discipline. For such purposes the notation, Thus, for the period M, the following
M/M/1: (LI Disc) is used, where Lis replaced probabilities can be stated:
by the maximum allowable length and Disc is
replaced by a symbol for the appropriate queue Mt= probability that one unit enters
discipline. The following are common dis- the system;
ciplines: 1-Mt= probability that no unit enters the
system;
FIFO= first in-first out (i.e., service in µD.t = probability that one unit leaves
order of arrival) ; the system;
SIRO = service in random order; 1 - µM = probability that no unit leaves the
system.
LIFO= last in-first out.
Thus, Ek/D/2 ( oo /FIFO) denotes a system There are three ways in which the system
with Erlangian arrivals, constant service, two can reach state n at time (t+M) (when n > 0):
service channels, infinite queue length (i.e., 1. The system was in state n at t and no
no limitation on queue length), and first come-
customers arrived or departed in M. (The
first served discipline.
probability of simultaneous arrival and depar-
ture in M is considered to be zero.)
8.2.1. System State for M/M/l 2. The system was in state n - 1 at t and
The fundamental quantities characterizing one customer arrived in M.
a waiting line are the states of the system. The 3. The system was in state n+ 1 at t
system is said to be in state n if it contains ex- and one customer departed in M.
FUNDAMENTALS OF QUEUEING THEORY 139
The probability of the system being in The resulting equations are of the form
state n at (t+M) is
for n > 0
P,.(t+tlt) = P,,(t) [( 1-Mt) (1-µ6.t)]+Pn_,(t)
[(AM) (I - µ6.t)] + P,.+i (1)[ (1- Mt) and
(µM)] (for n2 I) (8.1)
for n=O
Expanding and collecting terms, (8.5)
The variance of the number in the system the average time an arrival spends in the sys-
is tem. The expected time to wait before service
(that is, the time spent waiting in a queue) is
Var (n)='.I[n-E(n)]2P,,=(l_:' )2 (8.14) E(w) =E(v)-E(s) (8.16)
.~ p
where E(s) is the average service time (1/µ.);
This relationship is plotted in Figure 8.2. thus, Eq. 8.16 may be rewritten as
The derivation of this expression may be found
in a standard text on queueing theory. 1 A
E(w)=~--11µ.= (8.17)
µ.-A. µ.(µ.-A)
8.2.3 Delay Time in the ·System The average number of customers waiting
(M/M/l) to be served (the average queue length), E(m),
is the product of the average waiting time,
Consider the total time a customer spends
E(w), multiplied by the arrival rate, A.:
in the system ( v) to be made up of two
components: a time to wait before service, w
(queueing time) plus a time in service, s (ser-
vice time). The average number in the system,
25
20
20
E E
....<ll 16 ....<ll
"'>
U) Ji"' 15
j
I
c 12
.... ....
<ll
<ll
.0 .0
E E 10
z
:J
8
:J
z I
....0
<ll
O>
~ ~ 5 J
c
v /
<ll
> 4
<( ·~
Constant
>"' ~
v
0
0 0.1 0.2 0.3 0.4 o.s 0.6 0.7 0.8 0.9
Figure 8.1 Average number in system as a function of Figure 8.2 Variance of number in system as a function
traffic intensity.' of traffic intensity.'
THE CASE OF MULTIPLE CHANNELS 141
E(m) = [ p,(p,-!t)
It JIt= 1'2
p,(p,-1')
( 8.18)
If the garage operator wanted to be certain
with 0.9 5 probability that departing vehicles
would not interfere with other operations, he
Eq. 8.18 considers the average queue would need to provide space for about three
length over all time, including the periods when vehicles, one in service, two in queue. Similarly,
the queue is empty. Of interest is the average if he wished to be certain at the 0.99 prob-
queue length, given that the queue length is ability level, he would have to provide space
greater than zero. This is defined as for 5 or 6 vehicles, one in servict;, the others
· average queue length in queue.
E( m\m > 0) = ----''--~---
prob. of nonempty queue (c) The average number in the system (Eq.
E(m) 8.13) is E(n) = 120/ (240-120) = 1.
(8.19) ( d) The average number waiting in a queue
P(m>O)
(Eq. 8.18) is E(rn)=(120x120)/
A zero queue length will occur if the [240(120)] =0.5.
system is in state zero or state one so that by ( e) The average length of a nonempty queue
Eqs. 8.11 and 8.12 the probability of a non- (Eq. 8.20) is E(m\m>O) =240/120=2.
empty queue is (f) The average time in the system (Eq. 8.15)
is E(v) = 1/120 hr=0.5 min.
P(m>O) =1-(P 0 +Pi)
(g) The average time waiting in a queue (Eq.
=l-[(1-p)+p(l-p)] 8.17) is E(w)=120/[240(120)]=1/240
2
hr=0.25 min.
=l-l+p-p+p 2 =p 2 = ~2 (8.20)
Substituting Eqs. 8.18 and 8.20 in Eq. 8.3 THE CASE OF MULTIPLE
8.19 gives CHANNELS WITH EXPONENTIAL
ARRIVALS AND EXPONENTIAL SERVICE
E(m\rn>O)=(µ(;~!t) )(~:)=(fL~A) TIMES (M/M/N)
( 8.21) A parking lot (or the face of a block with
on-street parking) may be considered as an
example of a system with parallel service chan-
8.2.4 Example of Application of nels where the N parking slots represent the
Queueing Formulas (M/M/1) service channels. An arriving vehicle will
occupy an empty slot if one is available; if not,
The exit from a parking garage is through
it joins the waiting queue. The arrivals into the
a single gate where a variable fee is collected
system are assumed to be random with rate It
and change is made for drivers. Vehicles
and the service time per service channel ( dura-
arrive at the gate at random at a rate, It, of
tion of parking) is also random with mean 1 Iµ.
120 vehicles/ hr. The time to collect fees is
Again, p is defined as It/ p,. Further, pl N is
exponentially distributed, with a mean duration
defined as the utilization factor for the entire
(1 Iµ) of 15 sec. What are the characteristics
facility, representing the mean proportion of
of the operation when 1'=120 arrivals/hr, µ=4
busy channels (full parking spaces). For the
services/min=240 services/hr, and p=!tl fL=
multiple-channel case the value of p may be
120/240=0.5
greater than one but the following formulas
(a) The probability of an idle booth (Eq. apply only for the case where the utilization
8.12) is 1-0.5=0.5. factor pl N < 1.
( b) The probability that n vehicles will be in
the system (Eq. 8.11) is
8.3.l Synopsis of Equations for
Px=n PX5,» Queues with Multiple Channels
Po= 0.5 0.5 Probability of n units in system
• 0.25 0.75
P1=
0.125 0.875 p"
P2= Pn= - 1 P0 for nsN (8.22)
Pa= 0.0625 0.9375 n.
P.= 0.03125 0.96875
Ps= 0.015625 0.984375
r.
8.3.4 The System with Loss If Eq. 8.36 is multiplied by e-P/ e-P, the expres-
sion becomes
A more realistic type of operation for a
.Y-1
parking lot occurs when vehicles unable to park ~e-Pp"
go away instead of waiting in line; that is, L..- n!
E(n) =p _11=__
0_ _ (8.37)
P,,=0 for n>N
±e-Pp"
For this model the probability that n 11=0 n!
vehicles will be parked is
in which case it is possible to use tabulations of
pn/11!
P,,=-,,-,-- for n=O,l,2, ... ,N the Poisson distribution function to get the
desired answer.
LP;;;i A more elegant discussion of the relation-
i=O ( 8.34) ship between the Poisson distribution and the
The probability of an empty lot is queueing formulas developed here is given by
1 Kometani and Kato 10 and by Haight and
Po=-_,-.-- (8.35) Jacobson. n
LPi/i!
i:::O 8.3.5 Example of a Queueing System
and the probability that a car cannot park is Operating with Loss
the probability that there are N slots occupied, Assume the same data given in exampfo
so that 8.3 .2, except that vehicles do not wait for an
pX/X! empty space. Again, N=5, A.=4 arrivals/hr,
x (8.36) u=2 service/hr, p=A.lu=412=2, and utiliza-
Px= L_i)li! tion factor= pl N = 215 = 0.4.
i=O
The probability of an empty slot (Eq. 8.35) is
Eq. 8.36 is called Erlang's Loss Formula, L;y(p),
the probability that an incoming unit is "lost" Po=------ -----
to the system. 2
3
2' 25 2• 2
1 +1+2 +31+41+5f
Finally, the average number of vehicles in
the parking lot may be developed as follows: =0.137614
N P, = (2/ 1 )P 0 =0.275228
E(n)= L,n P,, P~= (2/2)P 1 =0.275228
ti=O P 3 = (2/3)P~=0.183485
P 4 = (2/ 4 )P 3 =0.091743
P,-,= (2!5)P,=0.036697 (probabil-
ity that a
car cannot
park)
The average number of vehicles parked
. 0.947 9
(Eq. 8.37) IS E(n) =2 _ =1. 3.
0 983
Raff 15 and Oliver 16 have considered the prob- vehicle. If h is greater than the critical head-
lem of defining the intervals between vehicles way T, the waiting driver or pedestrian will
that might be considered as acceptable to a cross, otherwise he will wait.
vehicle crossing or merging with a stream of Intervals 2, 6, 7, and 9 are all greater than
traffic. T, but it is evident that only a portion of these
Consider a stream of traffic as shown in four intervals is available for crossing. That
Figure 8.3 where the time of arrival of main- time during which no crossing is possible has
street vehicles is shown on a time scale (events been defined by Raff as a "block"; conversely,
2 through 9) . the rest of the time is defined as "antiblocks."
Intervals 2 through 9 are time intervals Oliver defined any time interval ( h > T)
(h) between the arrival of main-street vehicles as a gap and the remaining intervals as non-
at the projected path of the crossing side-street gaps, as shown in Figure 8.3. The length of a
vehicle. The first interval (a lag) is defined by gap is seen to be the length of a block less the
the time from the arrival of the side-street critical headway (for h > T). Time intervals
vehicle at the point of crossing (event number 2, 6, 7, and 9 are each a gap; time intervals 3,
1 ) to the passage of the next main-street 4, and 5 are grouped into a single nongap; and
Later Earlier
10 9 8 7 6 5 4 3 2
Events
• '
9 8 7 6 5 4 3 2 1 Interval
Number
-.-T--!
I
L
I
I
I
I
I
I
- Blocks
Antiblocks
I
- - Nongaps
Gaps
Start of
() () () () 0 Gaps
I I I
Figure 8.3
J--1ntergap Headway, 1---f
intervals 1 and 8 are also nongaps. It is of vehicles over an elapsed time t, during which
interest to note that the number of blocks, the number of events is the accumulated vol-
antiblocks, and gaps are all equal. The number ume, qt. Further, the mean headway (1 I q) is
of vehicles in an intergap headway is here defined as T.
defined as the number occurring just after the Each time interval ( h > r) is the beginning
vehicle (or event) defining the start of a gap of an antiblock and therefore also marks the
but including the last vehicle that defines the end of a block, so that in the elapsed time t
end of the nongap. For example, the first the number of time intervals
intergap headway in Figure 8.3 contains four
vehicles (numbers 3, 4, 5, and 6); the last (h>r) =number of antiblocks
intergap headway shown contains one vehicle =number of blocks
(number 10). If event 3 were to represent the =number of gaps
=qt e-qr (8.42)
arrival of a pedestrian, he would be delayed
until events 4, 5, and 6 (passage of main-street the time spent in antiblocks is
vehicles) had taken place.
=te-qr ( 8.43)
and the time spent in blocks is
8.5.2 Unsignalized Control-Pedestrian
Delay t- te-qr= t( 1 -e-qr) (8.43a)
Delays are discussed with the assumption The total time spent in gaps (Figure 8.3)
that the delayed person is in a position to accept is the sum of antiblock time + r x (number of
or reject a gap without the added delay of antiblocks)
waiting for the person in front to enter or cross
= te-qr + qte-qr r
the traffic stream. A group of pedestrians
waiting to enter a crosswalk is an example of = (t+qtr)e-qr (8.44)
this case. and the proportion of time spent in gaps (Eq.
Pedestrian delay at an unsignalized inter- 8.44/ t)
section was first treated by Adams 2 in 1936 in
one of the earliest theoretical traffic papers. = (1 + qr) e-qr (8.45)
He assumed that pedestrian and vehicle arrivals which was first proposed by Adams in his
are random and made field observations that 1936 study of delay to pedestrian traffic.
generally justified the assumption. If it is as- The average time duration (sec) of all
sumed that the main-street flow is q (vehicles/ gaps= (total gap time, Eq. 8.44)/ (number of
sec) and that an interval r (the critical gap in gaps, Eq. 8.42)
seconds) is required between successive arrivals
on the main street for a pedestrian to cross (t+qtr)e-qr
safely, several delay relationships can be de- T+r (8.46)
qt e-qr
rived.
By the equations of Chapter 3, the proba- which is equation III of Adams.
bility that a pedestrian will pass without delay The average time duration for all intervals
is (h<r) =average length of nongaps
0
anti blocks. ....0
In considering the problem of delays,
>
reference is again made to Figure 8.3, recalling .':::'.
that the beginning of a gap can be defined by :0
co
the arrival of a side-street vehicle or pedestrian, .0 0.2
0
as well as by the passage of a main-street ct
vehicle. A pedestrian may arrive in two posi-
tions: (a) as the event defining a gap (no
0
delay) or (b) during the nongap interval. In 0 2 3 4 !!
the latter instance, the arrival must wait for the
remaining vehicles within the gap to pass before Vehicles per Minimum Acceptable Gap, qr
he can cross the stream. Figure 8.4 Probability of pedestrian delay.'
The average number of vehicles between
the start of gaps (see Figure 8.3 for relation-
ship between start of gaps and the start of
blocks)
1
Ed(t) = ( - _
qe-qr q
_l__r)x ( 1 - 1e-qr )
volume qt 1 T
(8.49) -------
qe-qr
( 8.52)
number of blocks - qte-qr - e-qr 1 -e-qr
A delayed vehicle or pedestrian has to wait which is also plotted in Figure 8.5.
for one less vehicle than is given in Eq. 8.49, or Underwood, 17 in applying the formulas for
pedestrian delay to determine warrants for in-
1-e-ar stallation of pedestrian crossings, considered
e-qr
( 8.50)
three levels of pedestrian treatment: (a) no
treatment zone; (b) a "walking legs" sign zone
From this, it is noted that the expected (pedestrian sign and crosswalk markings in
delay, E(t), is found by multiplying the U.S. practice); and (c) a traffic control signal
average number of waits (Eq. 8.50) by the zone. His proposed method requires that three
average length of gap (h<r), Eq. 8.47; that is, values be established: (a) minimum vehicular
volume warrant; (b) minimum pedestrian
E(t) = ( l :-:~qr) X ( T- l ~e:~ar) volume warrant; and ( c) maximum pedestrian
volume warrant.
1 If the volume is less than that required by
=---T-·r
qe-qr
(8.51) warrants (a) or ( b), no treatment is required.
In the case of warrant (a), the delay to
Eq. 8.51 is plotted in Figure 8.5 with delay in pedestrians would be acceptable; in the case of
terms of the minimum crossing gap required. warrant ( b), because there are so few pedes-
For example, if there is one vehicle per mini- trians delay would be acceptable. The "walk-
mum gap and the minimum required gap is 5 ing legs" sign zone (includes painted cross-
sec (equivalent flow= 720 vehicles/ hr), the walk) is used if warrants (a) and (b) are
delay will be (0.7 x 5) = 3.5 sec. Similarly, for exceeded. If warrant (c) is also exceeded, a
• a 10-sec value of r (equivalent flow=360 traffic control signal is justified.
vehicles/hr) the expected delay will be (0.7 The minimum vehicular volume warrant is
x 10) =7.0 sec. determined from the application of Eq. 8.52.
The amount of delay for all delayed which is plotted in Figure 8.6 for values of r
pedestrians is equal to the average delay I pro- at 9, 12, 15, and 18 sec. Line AA represents
portion delayed: a point such that below the line delays increase
148 QUEUEING MODELS
200
100
/
70
/
40 /
/
20
Average Deloy for /
u0)
10
Those Delayed
7 ~
/
VI
'1--o
7/
>
"'
7
.J r
v
Qi 4
Cl
0)
/.
E' 2
0)
>
<{
I
///
0.7
/"
/ j K_Averoge Deloy for All Pedestrians
0.4
0.2 I
0.1
0
I 2 4 5 6 7
(.)
0)
gradually; above it, increase at an accelerating
Vl 140 rate. In each case the value V x r is approxi-
\J mately 6,000, where V =hourly vehicular
0)
> volume ( =3,600 q) and r is the intervehicular
"'
0) 120 spacing (sec) required by pedestrians. To
Cl
....
0) attain the minimum,
<{
0 100 V= 6,000 (8.53)
.c:
T
~
an approximate formula for r as given by
Underwood is
RxS W
r=---1--+2
30 ' v
where
R = perception time (sec; 2 or 3 sec);
S= speed limit (mph);
W = width of roadway (ft); and
v =crossing speed (ft/ sec); about 4
ft/ sec.
200 400 600 800 1000 The minimum pedestrian volume is taken
to be an average of one pedestrian per anti-
Volume (veh/hr)
block. The number of pedestrians delayed at
Figure B.6 Pedestrian delay vs. vehicular volume. 17 an average of one delayed pedestrian per anti-
DELAYS AT INTERSECTIONS 149
block is Ve-fJT (from Eq. 8.42) and the propor- the same minimum acceptable gap and a pedes-
tion of pedestrians delayed is ( 1 - e-rjT) ; thus, trian crossing velocity of 4 ft/ sec, a signal
the total number of pedestrians who may cross will be required for pedestrian volumes over
the traffic stream such that, on the average, not 360/hr if the vehicular volume exceeds 600
more than one pedestrian is waiting at any vehicles/ hr.
particular time is In 1951, Tanner 18 published the results of
a comprehensive study of pedestrian crossing
Ve-QT delays. His work is an extension of the delay
Pernin)= ( 1-e-(jT ) pedestrians/hr (8.54)
relationships developed by Adams. 2 He as-
sumed random arrivals of both main-street
which is the minimum pedestrian warrant. vehicles and crossing pedestrians and presented
For example, at V=720 vehicles/hr three proofs of the crossing delay distribution.
(q=0.2 vehicles/sec) and r=9 sec, Pernin) (Details of these proofs are beyond the scope
720xe-i.s 720x0.165 d . /h of this presentation but may be found in the
(1-e-LS) 0. 835 -1 42 pe estnans r.
original paper.) Tanner also developed a
Finally, Underwood reasoned that moto·r- method of considering varying values of gap
ists who were required to give way to pedes- acceptance for different pedestrians and gave
trians in the crosswalks should have at least some attention to the problem of groups of
60 percent of the available time. If less than pedestrians crossing the street.
60 percent of the time is available, traffic Of particular interest to those designing
signals are warranted to assure adequate time pedestrian controls are the distribution of the
to the motorists. size of pedestrian groups crossing together and
The proportion of nondelayed vehicular the distribution of the number of pedestrians
traffic (from Eq. 8.40) is e-Py/ 3 .soo where P is waiting at a random time.
the hourly pedestrian volume and g is the time The average size of a group crossing
interval between pedestrians in the crosswalk together is
such that there is a minimum safe distance
between pedestrians. pePT +qe-qT
E(n )=----~ (8.56)
Underwood takes the minimum safe dis- c (p+q)eCP-(j!T
tance between pedestrians as equal to the width
of a vehicle plus 6-ft clearance on either side, in which pis pedestrian flow and q is vehicular
or, for an 8-ft vehicle, a spacing of 20 ft. The flow. Figure 8.8 shows this relationship. The
time interval between pedestrians is given by average number waiting to cross is
g = 201 v, where v is the crossing speed of pedes-
trians (ft/ sec) , so that (8.57)
Probability (no delay) =e-20Pja,soov
=e-P/1sov (8.55) which is plotted in Figure 8.9.
As an example, consider a crossing with a
For 60 percent probability e-P/ 180 "=0.60 vehicular volume of 720 vehicles/hr and a
P/180v=0.51 and P = 90v. pedestrian volume of 360 persons/hr. The
If v=4 ft/sec, a signal is warranted when minimum acceptable gap, r, is constant at 10
pedestrian volume exceeds 360 pedestrians/hr. sec. For this example, q=0.2, p=O.l.
Underwood prepared a family of curves The probability that a pedestrian will be
based on Eqs. 8.53, 8.54, and 8.55, as shown in delayed (Eq. 8.41) is Pa=l-e- 2 =0.865. The
Figure 8. 7. Areas to the left of and below the mean delay for all pedestrians (Eq. 8.51) is
dashed curve associated with each minimum
gap represent the domain of combined pedes- 1 1
trian and vehicular volumes for which no E(t)= . e_2 - . -10=21.95 sec; the mean
02 02
treatment is required. To the right of and
above the same curves "walking legs" sign delay per delayed pedestrian (Eq. 8.52) is
zones are required. Traffic control signals are 1 10
Ea(t) = . e_2 - 0. =25.38 sec; the average
required when the pedestrian volume exceeds 02 865
360 persons per hour; for example, if the number of pedestrians crossing together (Eq.
pedestrian volume is 100/hr and the minimum 56 . E( )- 0.1e1+0.2e-2 -2 71 e-
acceptable gap is 10 sec, no control is required 8. ) is n'" - ern.1-0.2,10(0.1 +0.2) . p
for vehicular volumes less than 800/hr. For destrians; the average number waiting to cross at
500
Traffic Control •
Signal Zone
400
Q)
a.. 300 c
0
ai N
E ...c
.2 Q)
>
0 ...E
Sl
c ~
.E"' 0
V'>
Q}
z Walking Legs
-0 Sign Zone
Q}
a..
>
-;:::
::J
0
I 200
10 ..
11"
100 12 ..
13"
14"
15"
16"
17"
18"
o....._~~--'-~~~-'-~~~-'-~~~_._~~~<--~~--'
0 200 400 600 800 1000 1200
Vehicular Volume (veh/hr)
40 Cl
c 50
'+J
'(ij
35 :l:
....
Cl) V'>
.c c 40
....
...."'
·.::
Cl)
Cl
30
0 V'>
I- cipr = 2.0 "C
Cl)
Cl Cl)
c 25 0.. 30
·v;
V'>
.....
e
u 20
0
....
.... 13
Cl) E 20
.0 :::l
E 15 z
:::l Cl)
z Cl
c
10 -1-- "'....
Cl)
10
"'
Cl)
<(
>
~
5 o~llliiiii~;;;;;;;;;=t::::::::I~~L____J
0 I 2 3 4 5
Vehicles per Minimum Acceptable Gap, qr
o~~~.._~~.._~~...1-~~..L~___i
0 2 3 4 5
Figure 8.9 Average number of pedestrians waiting to
Vehicles per Minimum Acceptable Gap, qr cross street .1
30
E(t) = eqr_ 1-r+ .!.[eqr_ l-qr+
q b
(-q-)
q+b
2
-0
X (l+qr+br)(l-e-qr)
J
*'.,,
Ill
"C
+e-qrC!b +qr) (8.59)
c: "C
~ c: 20 in which r is the mm1mum acceptable gap, q
~~ is the main-street flow, and b is the parameter
(1)~
gi 0 of the shifted exponential gap acceptance dis-
~z tribution, which equals 1/(T-r). Eq. 8.59
a: .i:::.
..... may be compared with Eq. 8.51, the delay
.i:::. ·-
..... 3: with a constant gap acceptance value. The
-~ > probability, F(t), of accepting a gap of t sec
Ill
>- in the main-street flow is
a;O 10
Ill Cl>
Cl F(t) =0 (tSr)
Cl>
Cl
Ill F(t) = 1-exp[-b(t-r)] (t:'.::r) (8.60)
Q;
> The upper curve of Figure 8.11 presents a
< graph of this relationship for Herman and
Weiss's constants, r=3.3 sec and b=2.7 sec-1 .
2 4 6 The lower curve shows the results of assuming
that all drivers have an acceptable gap of 3.3
Vehicles per Minimum Acceptable Gap, qr
sec.
Figure 8.10 Pedestrian delays with and without refuge Weiss and Maradudin also considered the
island.' yield sign problem, which differs from the
pedestrian delay problem in acceptable gap
between moving and stopped vehicles. If a
moving vehicle requires a gap of r 1 and a
variance of delay increases in proportion to stopped vehicle requires a gap of r 2 (r 1 Sr 2 ),
the third or higher power of the critical gap. the mean delay is
Weiss and Maradudin 21 and Herman and
Weiss 22 developed several generalizations of
the crossing delay problem studied by earlier
investigators. The approach is based on re-
newal theory. A renewal process in time is the
occurrence of random spacings from a known As an example, assume r 2 =3.3 sec and
gap distributiol).. With their technique, it is r 1 =2.0 sec. These values substituted into Eq.
possible to deal with a general independent 8.61 yield a plot, as shown in Figure 8 .12, that
distribution of main-street gaps and a general compares the average side-street vehicle delay
gap acceptance distribution. This makes it (at a yield sign) with a stop sign situation where
possible to consider the "yield sign" delay all side-street drivers are required to stop and
problem where the side-street vehicle has a wait for a main-street gap of 3.3 sec.
different critical lag, depending on whether it Weiss 2 " further demonstrated that the
is moving or stopped. It is also possible to delay to a single vehicle crossing or merging
develop delay functions for the impatient driver with a traffic stream is practically independent
whose probability of accepting a given gap in of the velocity distribution of the mainstream
the main street increases with the passage of flow, for flow rates less than 1,600 vehicles/hr.
main-street vehicles. A different approach to the distribution of
Weiss and Maradudin expressed delay gaps in the main-street flow is given by Miller, 24
characteristics for several gap and gap accept- who postulated that bunches of vehicles (non-
ance distributions. Herman and Weiss fitted gaps in Figure 8.3) are randomly distributed on
shifted exponential constants experimentally. a highway. Letting the flow of queues (bunches)
For Poisson main-street traffic with mean be q/unit time and defining >< as the parameter
headway T and shifted exponential gap accept- of the exponential distribution of intervals be-
ance, the mean delay to side-street traffic as tween queues, ><e-1. 1, Miller relates q to >< as
developed by Herman and Weiss 22 is 1/ q = 1/1' + t, where tis the average length of a
DELAYS AT INTERSECTIONS 153
0Q)
~
4
>-
.2
Q)
Cl
Q)
1;71
0
....
Q) 2
~
queue (Eq. 8.47 if traffic flow is random). For Miller made a limited comparison of the
example, if the average length of a queue is average side-street delay and frequency of
10 sec (T=lO) and A.=0.025, llq=l/0.025+ undelayed crossings predicted by the random
10=50 and q=0.02 queues/sec=72 queues/hr. bunches model with those produced by the
One can arrive at an intuitive feeling for A. random vehicle model. He found little differ-
by observing that in one hour there will be 72 ence in average waiting time for crossing
queues of 10-sec average duration T, or a total vehicles. The random bunches model predicted
of 720 sec of queues (nongap time). The re- the opportunities for immediate crossing better
maining gap time (3,600- 720=2,880 sec) is
distributed among the 72 gaps (intervals be-
tween queues) so that the average gap length
=2,880172=40 sec(l/A.). This is equivalent
to a flow rate A. of 0.025 units/ sec.
Miller derived an expression for the mean
waiting time for a pedestrian or side-street
vehicle as
E(t) =>-(T+-r) 2 /2, (8.62)
where -r is the minimum acceptable gap and
A and tare defined as previously. For example,
if it is assumed that a pedestrian needs a time
gap of at least 10 sec (-r= 10), that A. is
equivalent to 90 events/hr (A.=l/40), and that
it takes on the average 10 sec (t = 10) for a
queue to pass, E(t)=l/2xl/40 (10+10) 2 =
5 sec.
The probability that a side-street vehicle
can cross immediately is given by
P 0 = (1-qf)e-h ( 8.63) 400 800 1200 l600
,. ,. ,.
154 QUEUEING MODELS
than did the random vehicles model. Figure 2T and 3T, two vehicles enter, etc. The number
8.13 gives the observed values for immediate of vehicles N that can enter from the side street,
crossing opportunities and the values predicted developed by Major and Buckley, 26 is found as
by the two theoretical models for 14 sets of follows:
observations, each set representing one hour •
of data collection. Number
of Vehicles
Size of Entering Number of Headways
8.5.3 Unsignalized Control-Vehicular Headway Headway This Size/Unit Time
Delay
<T 0 q(l-e-qT)
Because the equations in section 8.5.2 T-2T 1 q(e-qT -e-2q7)
make no allowance for queueing on the minor 2T-3T 2 q ( e-2q7 _ e-3qT)
street, it is necessary to introduce a delay that 3T-4T 3 q(e-3q7_e-4qT)
reflects the time required for a second-in-line etc.
vehicle to get into position to accept or reject a
gap. The case of a queue of vehicles waiting
on a side road or entrance ramp before merging The number of vehicles that will enter or
or crossing traffic on a main highway has been cross the main-street flow per unit time
discussed by Evans, Herman, and Weiss, 25 (capacity of cross flow) is
Major and Buckley, 26 and Ashworth, 27 among
others. N =q( e-qr -e-2qr) + 2q(e-2qr -e-3qr)
Consider a single inexhaustible queue of
+3q( e-3qr -e-4qr) + .. .
= qe-qr + qe-2qr + qe-3qr + .. .
vehicles on the side street under the following
conditions: when a main-street highway is less from which
than T, no vehicle enters; when a main-street
headway is between T and 2T, one vehicle qe-qr
N= -- (8.64)
enters; when a main-street headway is between 1-e-qr
60 ,___
Observed
,___
.... _
> ,___
~
..__
50
c::===:
"'
-
Q.)
F:=~
Cl
z
0 - -
---
-
==-=-=
- ---- ~
Random Vehicles Model
..c
..... 40
·~ ---- ,,,.,-- Random Bunches Model
V>
Ol
.!::::
30
-- -
-- - ===- -
--
-
--
V>
V>
0 ---
u ---
'+-
0
-=-
.... 20
Q.)
.c
E
:::l
.---
z
10 -
0 2 6
3 4 5 7 8 9 10 11 12 13 14
Observation Set Number
Figure B.13 Comparison of undelayed crossing opportunities.' Each set represents 1 hr of observations. (Adapted
from Miller, 21 Table 3.)
DELAYS AT INTERSECTIONS 155
The more realistic assumption can be LI v. Suppose a merge occurs at some time
made that the headway for following vehicles t < TL; if the driver had not been merging, the
is /3" so that two vehicles require a headway of distance traveled would have been Vt instead
T + /3 2 and three vehicles require T + 2/3 2 , etc. of vt actually traveled. This delay is defined as
Provided /3 2 ~T, Eq. 8.64 becomes
qe-ar D = ( 1- ~} = f3t
N=---o- (8.65)
1-e-q/3,
where /3= 1-v/ V.
If it is assumed that only one vehicle enters If the time to merge is greater then TL
during each antiblock, the capacity of the cross (say t=TL+ti> where t 1 >0) the delay is
street (from Eq. 8.4) is equal to qe-ar vehicles/ defined by
unit time.
Ashworth 27 modified the approach used
by Major and Buckley by assuming that the
critical gap of the driver is r 1 , whereas that of
the second driver is r 2 sec. Further, he assumed The total delay to a vehicle in the accelera-
a move-up time equal to a constant /3 2 sec, tion lane can be written as the sum of two
during which time-following the departure of random variables, Dill+ D., where Dill is the
the vehicle at the head of the queue-the sec- time spent traveling on the lane and D. is
ond vehicle moves into the front position but is the delay while stopped. The formula for the
unable to take advantage of any suitable gap expected delay (developed in detail by Blumen-
offered. With these assumptions the author feld and Weiss 28 ) is awkward, but the authors
gives the average waiting delay at the head of have found numerical solutions to relate the
the queue for those vehicles actually delayed: probability of reaching the end of the ramp
without merging (Figure 8.14) and the ex-
pected delay to merging vehicles (Figure 8.15)
( 8.66)
as a function of the length of the acceleration
lane.
where
P= 1 +q(T1 -/32)-eq<r,-r,-/32l(l +qrz) 8.5.4 Signalized Intersections
q( 1-eq<r,-r,-{32 \ )
There are several models that may be used
and T=llq. in the investigation of queues and delays at
signalized intersections. In section 8.5 .4.1 a
The average waiting delay for all vehicles continuum or fluid model is considered in
(provided r 2 > r 1 -/3 2 ) is which various measures of queue length and
delay are developed. In section 8.5.4.2 various
E(t) =r 1 - r2 -/32+ T(eq 7 2-eq<r,-/32J) (8.67) probability models are compared and different
assumptions implicit to several types of models
Eqs. 8.67 and 8.66 may be compared with are noted.
Eqs. 8.51 and 8.52, respectively, where the last In estimating delay at intersections, the
two equations consider delay for the pedestrian traffic is considered as consisting of identical
case. passenger car units (PCU). A truck, for ex-
The influence of an acceleration lane on ample, may be considered as 1.5 or 2 Pcu
the merging problem is treated by Blumenfeld
and a turning vehicle may be assigned some
and Weiss. 28 In their model, vehicles on the
acceleration lane of length L are assumed to value depending on the type of maneuver that
travel at a constant velocity v and vehicles on is made.
the main road travel at a constant velocity The following notation, after Allsop, 29 is
V( >v). The model further assumes that the used. Let
merging driver continues to move along the
acceleration lane at a velocity v until he either c= the cycle time (sec);
finds a suitable gap or continues to the end of g= the effective green time (sec);
the lane, at which point the velocity goes r= the effective red time (sec);
instantaneously to 0. The time TL is the time q= the average arrival rate of traffic on
for the vehicle to reach the end of the lane= the approach ( PCUI sec) ;
156 QUEUEING MODELS
variance of number of PCU arriving the degree of saturation of the approach, and
in one signal cycle y is called the flow ratio of the approach.
1
= mean number of PCU arriving in' The effective green time is the portion of
one signal cycle the cycle time during which Pcus are assumed
to pass the signal at a constant rate s, providing
s= the saturation flow on the approach vehicles are waiting on the approach. Green-
(Pcu/sec); shields et al., 30 for example, reported that the
d= the average delay to PCU on the total time for a queue of n stopped vehicles to
approach (sec); pass a signal can be given by
Q 0 = the overflow (PCU);
>-.= g/ c (i.e., the proportion of the cycle Total time=14.2+2.l(n-5) sec for n"2.5
that is effectively green; Had all of the vehicles departed at the satura-
y= q/s (i.e., the ratio of average arrival tion rate s (1I2.1), the first five vehicles would
rate to saturation flow); and have required 10.5 sec; that is, the effective
x= qc/ gs (i.e., the ratio of average num- green is -the signal green time less 3. 7 sec. In
ber of arrivals/ cycle to the maximum most studies it is assumed that a waiting queue
number of departures/ cycle). of vehicles will take advantage of the yellow
Thus, r + g = c and >-.x = y. The ratio x is called clearance interval, although the effective green
time may be adjusted to reflect particular
operating conditions.
The meaning of arrival time and departure
time for a pcu on an approach can be demon-
strated by reference to Figure 8.16, in which
0.5 distance-time curves are plotted for each of
four vehicles. AB represents the passage of an
undelayed vehicle, where the line PQ represents
Volume q in Right-Hand the stopline at which the first vehicle waits
Freeway Lane when there is a queue. CDEF represents the
trajectory of the first vehicle that is delayed
0.4
by a signal. The straight portions of CD and
q = 1200 cars/hr EF are parallel to AB and projected to meet PQ
q = 600 cars/hr
3
E
Cl..
~2
>
co q = 1200 cars/hr
Q)
0
q = 1000 cars/hr
"O
Q)
0.1 tl
Q)
c.
x q = 600 cars/hr
UJ
O'---~--'-~~'--~-'-~---'~~~~---'~~
200 400 600 0 400 800 1200
Length, L (ft) Length, L (ft)
Figure 8.14 Probability of reaching the end of the
acceleration lane without merging as a function of Figure 8.15 Expected delay as a function of accelera-
length. 28 tion lane length. 28
DELAYS AT INTERSECTIONS 157
at X and Y so that the length XY is the delay 8.5.4.1 Continuum Model for Pretimed
to the first vehicle. Similarly, X'Y' and X"Y" Signal. A representation of a continuum
represent the delays for the next two vehicles. model proposed by May 11 is given in Figure
XX' and X'X" represent the arrival headways. 8.17. The vertical axis represents the cumula-
In a number of the studies it has been tive arrivals qt and the horizontal axis the
convenient to assume that events such as the time t.
arrival of a vehicle may occur only at certain Case I represents the behavior when the
instants that are equally and closely spaced in capacity of the green interval exceeds the ar-
time. It is convenient to choose the interval rival during the green+ red time. Case II is
in seconds between two events such that concerned with the instance when the discharge
:':.t = 1Is and to define time in multiples of t:.t. during the green phase is equal to the arrivals
If the time origin t = 0, events occur at time during the green+ red period. In Figure 8.17
t=nM (n=l,2,3, ... ). the vertical distance ca represents the number
Let c=Ct:.t; r=Rt:.t; and g=Gt:.t, where of vehicles that have accumulated since the
C, R, and G are integers, and let a.= average signal entered the red phase. The horizontal
number of vehicles in time t:.t=qM=y. These distance ab represents the total time from ar-
definitions are useful when using a binomial rival to departure for any given vehicle.
model of vehicular arrivals as used in section May developed the following measures of
8.5.4.2. queue behavior:
Position of t-:------+--------r-::::;111-_."-771-+.~+:'7"::--------Q
Stop line
Time - -
Figure 8.16 Diagram to illustrate the definitions of supposed arrival and departure times.2°
158 QUEUEING MODELS
co"'
.~
.....
.;c
/\
....>
·o
"'
a.
"'
(.)
Q)
"'
"'
(.)
Time
Red Green
....>
·o
"'a.
"'
(.)
Q)
"'"'
(.)
Time
2(~=y)] ;c = 2c(;~y)
1. Time after start of green that queue is
dissipated (t 0 ); (8.74)
d= [
2. Proportion of cycle with queue (Pq);
3. Proportion of vehicles stopped (P8 ) ; 8. The maximum individual vehicular
4. Maximum number of vehicles in queue delay will be seen from Figure 8.16 to be
(Qm); (8.75)
5. Average number of vehicles in queue
(Q); If the departures sc are less than the arrivals
6. Total vehicle-hours of delay per cycle qc, the queue grows with each successive cycle
(D); and the foregoing formulas are no longer ap-
7. Average individual vehicle delay ( d) ; plicable.
8. Maximum individual vehicular delay Consider the following example (after
(dm). May) of the behavior of a queue at a signalized
intersection. Assume that the green phase g is
The formulas for these two cases are developed 40 sec, the red phase r is 20 sec, the discharge
from simple geometric relationships: rate s is 1,200 vehicles/hr, and the input rate
q is in one case 600 vehicles/ hr and in another
1. For any given cycle it is evident that
case 800 vehicles/ hr. The results are given in
at time t 0 after the start of green, the arrivals
Table 8.2.
equal the discharge:
q(r+t 0 ) =st0 8.5.4.2 Probability Models for Pretimed
Signals. This section owes much to the paper
letting y=qls, by Allsop 29 in which he critically reviews the
various theoretical models of delay at fixed-
t 0 =yrl(l-y) (8.68) time signals. The reader should refer to this
2. Proportion of cycle with queue is equal paper and those of the individual authors cited
to queue time/ cycle length for a more detailed analysis than that presented
here.
(8.69) Several models have been used to describe
the arrival of vehicles at intersections. The
3. Proportion of vehicles stopped is equal simplest involves regular arrivals, as discussed
to vehicles stopped/ total vehicles per cycle in section 8.5.4.1. Winsten and co-workers 32
were the first to use the binomial model in
P8 q(r+t0 ) / q(r+g) =!0 / (ye) (8.70) analyzing delays at pretimed signals.
4. The maximum vehicles in queue will An approach has binomial arrivals if for
be seen by inspection to be the height of the some fixed M and a,, P ( 1 PCU arrives at time
triangle at r units after start of red: nM) =a and P (no PCU arrives at time nM) =
1 - a for each n, independent of any other
Qm=qr (8.71) instant, and no arrivals can occur at other
times. The average arrival rate is a,/ flt; if AN
5. The average number of vehicles in the
queue, over the total length of cycle ( c) is
is the number of PCU in a period containing N the average delay per PCU will tend to d =r 2 /
instants of nilt, then AN has a binomial distribu- [2c(l-y)], the result given in Eq. 8.74, based
tion. (See section 3.2.2.) on an assumed continuum model.
For this distribution the ratio of variance Wins ten et al. 32 have demonstrated that
to mean (I), equal to 1- a, is less than 1; for for a traffic signal with binomial arrivals the
urban roads, however, the v;ilue of I is often average delay to a Pcu passing through the
observed to exceed 1, as reported by Miller. 33 approach is
The Poisson distribution (section 3 .2.1)
has been used by Adams, 2 Webster, 34 and
Wardrop. 35 The value of I will equal 1 and the
d=-R-[E(Qo)
(1-a) a
+ R+l
2
JM (8.78)
model is most appropriate when an approach is Although the Winsten group was not able to
lightly loaded in relation to the capacity of the develop the probability distribution of the
approach. overflow (the number of vehicles failing to
Newell 3 6 used a model in which the arrival clear the intersection during a given cycle),
headways were assumed to have a shifted ex- Dunne 41 and Potts 12 have developed the prob-
ponential distribution (section 3.5.1). This ability-generating function for the total delay
assumption imposes a minimum headway of in a cycle when binomial arrivals are assumed
1/ s. Other models of arrivals cited by Allsop and the green time is long enough for the queue
include those of Darroch, 37 Kleinecke, 38 Mc- to clear.
Neil, 39 and Miller. 40 Newell 43 was able to develop an estimate
Models for the departure of PCU from a of the mean value of the overflow E(Q 0 ) ,
queue are simpler than the arrival models. Most which, as a-gl c (i.e., the ratio x approaches
models assume departures at equal time inter- 1), can be approximated by
vals 1 / s providing a queue exists and the first
departure is at the start of the effective green Rg
E(Qo) = 2 c ( g-ac ) ( 8.79)
time. For the discrete time assumption, M is
taken as 1/ s. The first Pcu departs at the start
of the effective green time nM, and one PCU One of the better-known formuliis for
departs at each succeeding nM until the queue delay is that developed by Webster 34 using data
clears or the green time ends. Other departure resulting from computer simulation of inter-
models have been proposed, but variations in section operations:
assumptions for departure models do not have
the same impact on delay as the variations in d=c(l-A.)2+ x2 -0.65(~)1hxc2+s;_i
2(1-A.x) 2q(l-x) q2
arrival models.
Allsop demonstrates that for an approach ( 8.80)
with regular arrivals at intervals 1/ q (arrivals Because c(l-A.) =r and A.x=y, the first term
would be plotted as a step function rather than is the same as that obtained by assuming
as a straight line in Figure 8.17), the average continuum flow (Eq. 8.74). Allsop 29 shows
delay I PCU (sec) is that the second term is obtained by assuming
that a queue with constant service 1/ A.s is
interposed between the signal and the arriving
traffic. The mean waiting time in the interposed
(8.76) queue is x 2 /2q(l-x).
The third term, developed by Webster
from regression analysis of data generated by
where e is shown by Allsop to have the range simulated signal behavior, is a correction term
1 2 representing from 5 to 15 percent of the total
_ x 1/ <0< x 11z' and c, q, r, s, and y are
3 3 2 3 3 mean delay. From this, Allsop suggests that
as previously defined. the average delay may be taken as
The first term of Eq. 8.76 is the same as
that developed by Wardrop 35 : d-
~
_9_[ c(l-A.)2
10 2 ( 1 -
Ax)
..L
1
x2
_2_q_(_l___x_)
J (8.81)
+
Ix
2
2q(l -x)
J (8.86)
When x < 1I2, the middle term in the brackets
Because Webster's full expression has been sub-
is replaced by zero. Miller found that his equa- ject to the most extensive field testing, it has
tion and Webster's (Eq. 8.80) gave similar been taken as the standard for the numerical
results with I near to 1, and that Eq. 8.84 gave comparisons made by Hutchinson.
better results when I was appreciably greater Curves of Webster's average delay (Eq.
than 1. 8.80) versus degree of saturation x and propor-
Newell ·10 considered the continuum model tion of cycle effectively green are shown in
developed in section 8.5 .4.1 and the further Figures 8.19 and 8.20, respectively.
delay caused by the overflow Q0 • His work The influer:ce of the parameter I (vari-
leads to the expression for delay ance/ mean) for the several expressions for
delay is presented in Figure 8.21.
c(l-A.) 2 IH(µ,)x _L _J(l-A.) Hutchinson also calculated the percentage
d- + _ _ __
differences between various estimates of aver-
- 2(1-A.x) 2q(l-x) ' 2s(l-A.x) 2
age delay and the delay given by Webster's full
(8.85) expression as a function of the flow ratio y.
where H (µ,) is a function of the spare capacity Figure 8.22 shows the results obtained when
s=0.5, c=90, and A.=0.5. Similar results are
of the approach and
presented by Hutchinson for different combi-
µ,= (sg-qc)! (lsg)l/z nations of s, c, and A.. The heavy pair of curves
= (1-x) I (sg/ 1)1/2 (8.86) that nearly envelope the other curves are the
limiting values of the percentage differences
(See Figure 8.18.) that result as y ~ 0 (x ~ 0) and as y ~A.
(x~ 1).
Hutchinson 46 made numerical compari-
sons of several of the delay expressions that had The percentage differences related to the
proportion of cycle effectively green A., as
been discussed by Allsop, 29 as well as. analyzed
calculated by Hutchinson, are shown in Figure
• the following expressions: (a) Webster's full 8.23. As a result of his analysis, the author
expression (Eq. 8.80), (b) Miller's first ex- concluded that the results are sufficiently simi-
pression (Eq. 8.84), (c) Miller's second expres- lar that convenience and simplicity of calcula-
sion (Eq. 8.83), (d) Newell's expression (Eq. tion may dictate the choice of model. For
8. 85), ( e) a modification of the simplified example, if one were required to differentiate
••
0
g 60 ! 70
·~ c:
Q)
.... 0
'iii
a.
x VI
Q)
w 50 0. 60
:; x
u. w
:;
e
.vi
40 u. 50
VI .vi
.D ....
Q)
s: ....
Q)
VI
.D
Q)
B 30 s: 40
Cl
.£: ....0
"E Cl
c:
0
8 20 "E 30
<( 0
u
> u
~ <(
Q)
Cl 10 ~ 20
Q)
Cl
a;
Cl
"'....
Q) Q)
> Cl
<( o~~~-'-~~~~~~-'-~~--'~~~~
0 0.1 0.2 0.3 0.4 0.5
"'....
Q)
>
10
<(
Flow Ratio, y
36
32
28
24
>
"'
Q)
0 20
Q)
Cl
..."'
Q)
>
<( 16
"O ----
f::::::.:::::::::::::::::::::=::::~:;;;:;;;:.::::.:~::i~::~::~;;;;,,,"""""'"':::::
E - ---...r=---=-== ==-===-------
+-" ---- -- -------------
Q)
·~
w 12
0
0 0.5 1.0 1.5
Value of I
•
50 Webster's simplified expression
with modification
incorporating the ratio I
Miller's first expression
A
E
Miller's second expression /I
.....e
>
"'
a;
40
-
Newell's expression
Largest of changes in Webster's
full expression resulting from
changes in parameters
I/ I
I
Cl described in text I
Q)
I
c:n
a;"' 30
>
<(
0 c:
: .Q
c: "'
.Q :G 20
"'"' ....c.
Q) x
O..w
X-
W :5
.... LI.
-s -.. .
Q) "'
-
o~
0
~3
c:
-2"'
10
Q)
.....
.....
~
Q)
Cl
-20
Degree of Saturation, x
12 20 50 00
:i 40
LL
"'
'._
~
"'
.0
OJ
3: 20
E
.....e
>
"'
OJ
Cl
OJ
Cl
~
OJ
>
<(
._ c
0 0
'+- ·v; -20
"'c "'OJ
.Q a.
~~
._
0.
x
w -40
._
OJ
..c
.....
0
.....0
OJ
u -60
c
._
OJ
.....
OJ
~
Cl
OJ
Cl
~ -80
c
OJ
~
OJ
a..
-100
0.4 0.5 0.6 0.7 0.8 0.9 1.0
Proportion of the Cycle Effectively Green, ;\
}
0 0 0 00 nitely long trip. He was able to express the
average speed E(u) in terms of the average
waiting time behind all vehicles, E(tw), which
Speed = u Flow Rate= q 1 •
Velocity= u Concurring included zero waits. The expression
1
Minimum Spacing = s1 Flow
E(u) = uu1 2 +q1(U-U1) (u1-s1q1)u1ECtw)
U1 2 +qi (u- U1) (u1 -s1q1) E(tw)
Figure 8.24 Two-lane roadway, showing the corre-
sponding assumed terms in the Tanner model. 1 (8.89)
was developed; thus the problem became one
a distance of at least d,,=d+nS 1 (u+u 2 ) / of solving for E( tw). Algebra involved in the
(u-u,) in the q" stream, the vehicle passes computation of E(tw) is formidable:
without decelerating. [d is defined as the least
acceptable clear distance between the u vehicle
and the passed traffic as the u vehicle clears the
bunch in passing. It can be expressed by d =
A,u!(u-u 1 ), with A 1 being some distance
between 50 and 100 ft.]
3. If the required distanced,, is not avail-
able, the vehicle decelerates instantaneously to
u 1 , follows as closely as possible behind the
vehicle ahead, waits for a clear distance of at
least D,,_=d,,+t(u 1 +u 2 ) in the q 2 stream, waits
a further time t, accelerates instantly to u, and
passes. The additional time t required for the in which g=q 1 1u 1 , G=q 2 /u 2 , r=S1 g, R=S 2 G,
overtaking vehicle to remain in the q 1 stream c=q 1 (u-u 1 )/u 1 (u+u 2 ), K=root between 0
because of having slowed down is used to com- and 1 of K=exp[R(K-1-c/G)], and N=the
pensate for the assumed instantaneous accelera- smaller real root of N = exp[n (N - 1 +GI c)]
(which exists only when r exp (1-n+nG/
c):; 1).
Limited solutions for K and N have been
included in Figures 8.25 and 8.26, respectively.
Substitution of the values of E(tw) in Eq.
8.89 gives an expression for the average speed
E(u) in terms of the desired speed u, the
velocity of the q 1 stream u1, and the flow rate
q 1 of the stream. Limited solutions of this
equation were made by Tanner using specific
K values of the various parameters. Figure 8.27
shows the effect of traffic flow when q 1 =q 2 for
various values of u and u1= 30 mph. This
model indicates that for a total flow (q 1+q 2 )
of more than 800 vehicles/hr a vehicle will
have to assume very nearly the velocity of the
q 1 stream, regardless of its own desired velocity.
The effect of varying proportions of q 1
0.1 0.2 0.3 0.4 0.!5
and q 2 on the average speed E(u) is shown in
R Figure 8.28, which shows that in this model
the average speed E(u) is least when one-half
Figure 8.25 Relationship between K and the parame- to three-fourths of the total traffic is traveling
ters R and C/G. 1 in the opposite direction of flow q 2 (one-half
QUEUEING MODELS FOR ROADWAYS 169
6 being applicable at low volumes and the value
50
of three-quarters as one approaches higher
volumes).
5 It is worthwhile to point out that the delay
implied by E(u) is the only delay involved; all
10 other vehicles are, by the assumptions, not
4
delayed. The u vehicle would ultimately pass
all q, vehicles, and no passing would occur
among q, or q" vehicles.
N Morse and Yaffe 02 have developed a
3
queueing model for the two-way two-lane situa-
tion at low volumes. The model assumes a
sequence of free or lead cars driving at various
2 speeds, some of which are followed by a queue
of trapped cars that are traveling at the same
speed as the lead car. No assumption is made
of the desired speed of the trapped vehicle, but
some are able to escape the queue by passing
and becoming lead cars. The authors develop
a formula for the length of queue trapped be-
hind a lead vehicle with given velocity v, the
mean velocity of the stream, and the mean time
for a faster vehicle to pass a slower vehicle.
Figure 8.26 Relationship between N and the parame-
ters rand G/C. 1
Tables that aid in the solution of the resulting
equations are included.
Queueing models on multilane roadways
differ from the two-lane situation by virtue of
Total Volume
I
! u = 30 mph
1
80
--·--+-1 I
.c 70
a. II I. Ii
s
3 60 -J__ j_-+ . -1
~l . t---~-
LU
.3
UJ
50
1 I I
I
40 -""'1"-<~--··--+-·-
1
II
I
30
0 180 360 540 720 900 30o'-~~2~0,--~._,4~0~~-60~~~~80~~~IOO
q1 = q2 (veh/hr)
q (as Percent of Total Vall
Figure 8.27 Effect of traffic when equally divided be-
2
tween directions for various values of u (two-way Figure 8.28 Effect of varying proportion of opposing
traffic).1 traffic for various levels of total traffic.1
170 QUEUEING MODELS
the fact that a vehicle is free to use al).y of the 3. Tanner, J. C., The delay to pedestrians
two or more lanes available in his direction of crossing a road. Biometrika, 38 (Parts
travel and is restricted from passing only dur- 3 and4):383-392 (1951).
ing the period when another vehicle is in the 4. Edie, L. C., Traffic delays at toll booths.
adjacent passing lane. Miller 53 proposes an Oper. Res., 2 (2): 107-138 (1954).
empirical model, which is an extension of the 5. Moskowitz, K. "Waiting for a gap in a
"product density" method of analysis. Miller traffic stream. Proc. Highw. Res. Board,
develops and criticizes the product density ap- 33:385-395 (1954)".
proach and then develops a technique that for 6. Haight, F. A., Mathematical Models for
want of a better term, he calls the "termination Traffic Flow. Academic Press, New
rate method," where this latter method is analo- York (1963).
gous to the statistical model of mortality rates 7. Prabhu, N. U., Queues and Inventories.
for humans. Miller provides some verification Wiley, New York (1965).
of his model with data collected on a multilane 8. Cox, D. R., and Smith, W. L., Queues.
freeway. Chapman and Hall, London ( 1971) .
Schach 5 ·1 gives an example of the Markov 9. Newell, G. F., Applications of Queueing
process as it might be applied to a model of Theory. Chapman and Hall, London
multilane flow. Some quantities that he is able (1971).
to calculate are the average use of each lane, 10. Kometani, E., and Kato, A., On the theo-
the average speed of cars in the system, and the retical capacity of an off-street parking
expected number of lane changes per unit time. space. Memoirs, Faculty of Engineer-
Input parameters are N, the density of vehicles ing, Kyoto University, 18 (4) :315-328
on a roadway, intensities of weaving from lane (1956).
1 to lane 2 or lane 2 to lane 1 (A, µ,), and a 11. Haight, F. A., and Jacobson, A. S., Some
parameter r that the author sets equal to 2. mathematical aspects of the parking
Holland 55 considers the behavior of three problem. Proc. Highw. Res. Board, 41:
lanes of traffic, all moving in the same direction. 363-374 (1962).
He allows the special behavior of each lane-
12. Haight, F. A., Overflow at a traffic light.
traffic in the right lane is free to exit or move to
Biometrika, 46:420-424 (1949).
the middle lane; traffic in the middle lane may
move to either of the two adjacent lanes; traffic. 13. Tanner, J. C., A simplified model for de-
in the left lane may enter the middle lane only. lays in overtaking on a two-lane road.
The solution technique is iterative (a flow chart J. R. Stat. Soc., Ser. B, 20 (2) :408-414
of the computational process is included), and (1958).
his results are plotted against previously pub- 14. Tanner, J. C., Delays on a two-lane road.
lished data. J. R. Stat. Soc., Ser. B, 23 (1) :38-63
Drew 5 " and Worrall, Bullen, and Gur 59 (1961 ).
have used the Markov process to describe the 15. Raff, M.A., A Volume Warrant for Urban
lane-changing process on a multilane freeway. Stop Signs. Eno Foundation for High-
The interested reader will find information on way Traffic Control, Saugatuck, Con-
application of this process to traffic analysis in necticut (1950), pp. 62-75.
these two references. 16. Oliver, R. M., Distribution of gaps and
blocks in a traffic stream. Oper. Res.,
10(2):197-217 (1962).
17. Underwood, R. T., Tentative warrants for
8.7 REFERENCES the installation of pedestrian crossings.
I. Cleveland, D. E., and Capelle, D. G., Res. Memo. No. 13. Country Roads
Queueing theory approaches, pp. 49-98. Board, Victoria (1957), p. 18.
In D. L. Gerlough and D. G. Capelle I 8. Tanner, J. C., The delay to pedestrians
crossing a road. Biometrika, 38:383-
(ed.) An Introduction to Traffic Flow
392 (1951 ).
Theory. Special Report 79. Highway
19. Mayne, A. J., Some further results in the
Research Board, Washington, D.C. theory of pedestrians and road traffic.
(1964). Biometrika, 41 :375-389 (1954).
2. Adams, W. F., Road traffic considered as 20. Jewell, W. S., Waiting for a Gap in Traffic.
a random series.: J. Inst. Civil Eng., Res. Rep. No. 6. Operations Research
4: 121-130 (1936). Center, University of California (1961).
REFERENCES 171
21. Weiss, G. H., and Maradudin, A. A., Some 37. Darroch, J. N., On the traffic light queue.
problems in traffic delay. Oper. Res., Ann. Math. Stat., 35:38-88 (1964).
10:1, 74-104 (1962). 38. Kleinecke, D. C., Discrete-time queues at
22. Herman, R., and Weiss, G., Comments on a periodic traffic light. Oper. Res., 12:
the highway-crossing problem. Oper. 809-814 (1964) .
Res., 9: 828-840 (1961 ) . 39. McNeil, D. R., A solution to the fixed-
23. Weiss, G. H., The intersection delay prob- cycle traffic light problem for compound
lem with gap acceptance function de- Poisson arrivals. J. Appl. Probab., 5:
pending on speed and time. Transp. 624-635 (1968).
Res., 1 (4) :367-371 (1967). 40. Miller, A. J., Settings for fixed-cycle traffic
24. Miller, A. J., A queueing model for road signals. Oper. Res. Q., 14:373-386
traffic flow. J. R. Stat. Soc., Ser. B, ( 1963).
No. 23 (1961), pp. 64-75. 41. Dunne, M. C., Traffic delay at a signalized
25. Evans, D. H., Herman, R., and Weiss, intersection with binomial arrivals.
G. H., The highway merging and queue- Transp. Sci., 1:24-31 (1967).
ing problem. Oper. Res., 12 ( 6): 832- 42. Potts, R. B., Traffic delay at a signalized
857 (1964). intersection with binomial arrivals.
26. Major, N. G., and Buckley, D. J., Entry to Transp.Sci., 1:126-128 (1967).
a traffic stream. Proc. Austr. Road Res. 43. Newell, G. F., Queues for a fixed-cycle
Board (1962), pp. 206-228. traffic light. Ann. Math. St~t., 31: 589-
27. Ashworth, R., The capacity of priority- 597 (1960).
type intersections with a non-uniform 44. Miller, A. J., Some operating characteris-
distribution of critical acceptance gaps. tics of fixed time signals with random
Transp. Res., 3 (2) :273-278 (1969). arrivals. Institute of Highway and Traf-
28. Blumenfeld, D. E., and Weiss, G. H., fic Research, University of New South
Merging from an acceleration lane. Wales (1969).
Transp. Sci., 5 (2): 161-168 (1971).
45. Newell, G. F., Approximation methods for
29. Allsop, R. E., Delay at a fixed time traffic
queues with applications to the fixed-
signal. I: Theoretical analysis. Transp.
cycle traffic light. SIAM Rev., 7 (2):
Sci., 6 (3) :260-285 (1972).
223-240 (1965).
30. Greenshields, B. D., Schapiro, D., and
Erickson, E. L., Traffic performance at 46. Hutchinson, T. P., Delay at a fixed time
urban intersections. Bureau of Highway signal. II: Numerical comparisons of
Traffic, Tech. Rep. No. 1. Yale Uni- some theoretical expressions. Transp.
versity Press, New Haven, Conn. (1947). Sci., 6:286-305 (1972).
31. May, A. D., Jr., Traffic flow theory-the 47. Little, J. G., Jr., Queueing of side street
traffic engineer's challenge. Proc. Inst. traffic at a priority type vehicle-actuated
Traffic Eng. (1965), pp. 290-303. signal. Transp. Res., 5:295-300 (1971).
32. Beckmann, M., McGuire, C. B., and Win- 48. Newell, G. F., Properties of vehicle actu-
sten, C. B., Studies in the Economics of ated signals. II. One-way streets.
Transportation. Yale University Press, Transp. Sci., 3:30-52 (1969).
New Haven, Conn. (1956). 49. Newell, G. F., and Osuna, E. E., Proper-
33. Miller, A. J., Settings for fixed-cycle traffic ties of vehicle actuated signals: II.
signals. Proc. Austr. Road Res. Board, Two-way streets. Transp. Sci., 3:99-125
2:342-365 (1964). (1969).
34. Webster, F. V., Traffic Signal Settings. 50. May, A. D., Jr., and Keller, H. E. M.,
Road Research Tech. Paper No. 39. A deterministic queueing model. Transp.
Res., 1:117-128 (1967).
Great Britain Road Research Labora-
51. McNeil, D. R., Growth and dissipation of
tory, London (1958).
a traffic jam. Transp. Res., 3:115-121
35. Wardrop, J. G., Some theoretical aspects (1969).
of road traffic research. Proc. Inst. Civ. 52. Morse, P. M., and Yaffe, H. J., A queue-
Eng. 1 :325-362 (1952). ing model for car passing. Transp. Sci.,
36. Newell, G. F., Statistical analysis of the 5:48-63 (1971).
flow of highway traffic through a sig- 53. Miller, A. J., An empirical model for
nalized intersection. Q. Appl. Math., multilane road traffic. Transp. Sci., 4:
13:353-369 (1956). 164-186 (1970). .
..
54. Schach, S., Markov models for multi-lane Ashworth, R., The capacity of priority-type
freeway traffic. Transp. Res., 4:259- intersections with non-uniform distribution
266 (1970). of critical acceptance gaps. Transp. Res., 3:
55. Holland, H. J., A stochastic model for 273-278 (1969).
multilane traffic flow. Transp. Sci., l:
•
Blumenfeld, D. E., and Weiss, G. H., On the
184-205 ( 1967).
robustness of certain assumptions in the
56. Wagner, F. A., Ger\ough, D. L., and
merging delay problem. Transp. Res., 4:
Barnes, F. C., Improved criteria for
125-140 (1970).
traffic signal systems on urban arterials.
National Cooperative Highway Research Blumenfeld, D. E., and Weiss, G. H., On queue
Program Rep. 73. Highway Research splitting to reduce waiting times. Transp.
Board, Washington, D.C. (1969), 55 pp. Res., 4: 141-144 (1970).
57. Wagner, F. A., Barnes, F. C., and Ger- Blunden, W. R., and Pretty, R. L., On the
lough, D. L., Improved criteria for theory of deterministic cyclic traffic flows in
traffic signal systems in urban networks. networks, pp. 287-299. In Vehicular Traffic
National Cooperative Highway Research Science. Proceedings Third International
Program Rep. 124. Highway Research Symposium on the Theory of Traffic Flow,
Board, Washington, D.C. (1971), 86 pp. New York, 1967.
58. Drew, D. R., Traffic Flow Theory and Buckley, D. J., Leong, Y. F., Ong, C. K., Some
Control. McGraw-Hill, New York
results for a deterministic model of an ele-
(1968).
ment of a traffic signal system. Strassenbau
59. Worrall, R. D., Bullen, A. G. R., and
Gur, Y., An elementary stochastic und Strassenverkehrstechnik 1969 (IVth In-
model of lane-changing on a multilane ternational Symposium on the Theory of
highway. Highw. Res. Record 308 Traffic Flow). Karlsruhe, 3: 364-376 (1968).
( 1970), pp. 1-12. Darroch, J. N., Newell, F. F., and Morris,
R. W. J., Queues for a vehicle-actuated traffic
light. Oper. Res., 12:882-895 (1964).
8.8 RELATED LITERATURE DeSmit, J. H. A., The transient behaviour of
the queue at a fixed cycle traffic light. Transp.
Allan, R. R., On the highway crossing problem. Res., 5:1-14 (1971).
Proc. Austr. Road Res. Board, 4: 529-548 Dunne, M. C., Traffic delay at a signalized in-
(1968). tersection with binomial arrivals. Transp.
Allan, R. R., Queueing on the minor road. Sci., 1:24-31 (1967).
Proc. Austr. Road Res. Board, 4: 549-566 Dunne, M. C., and Buckley, D. M., Delays and
(1968).
capacities at unsignalized intersections. Proc.
Allsop, R. E., An analysis of delays to vehicle Austr. Road Res. Board, 6:345-362 (1972).
platoons at traffic signals, pp. 98-104. In
Dunne, M. C., and Potts, R. B., Algorithms for
Strassenbau und Strassenverkehrstechnik
I 969 (IV th International Symposium on the traffic control. Oper. Res., 12: 870-881
Theory of Traffic Flow). Karlsruhe (1968). (1964).
Andrews, F. C., A statistical theory of traffic Dunne, M. C., and Potts, R. B., Control of
flow on highways. I. Steady-state flow in traffic at an intersection. Proc. Austr. Road
low density limit. II. Three-car interactions Res. Board, 2: 265-273 (1964).
and the onset of queueing. Transp. Res., 4: Fry, A. T., and Buckley, D. J., Delay and inter-
359-377 (1970). ference in combined lanes at signalized inter-
Andrews, F. C., A statistical theory of traffic sections. Proc. Austr. Road Res. Board, 5:
flow on highways. III. Distributions of de- 15-39 (1970).
sired speeds. IV. Semi-empirical steady state Gaver, D. P., Jr., Time-dependent delays at
theory. Transp. Res., 7:223-242. traffic merges. Oper. Res., 14: 812-821
Ashton, N. R., Buckley, D. J., and Miller, A. J., (1966).
Some aspects of capacity and queueing in Gazis, D. C., Newell, G. F., Warren, P., and
the vicinity of slow vehicles on a rural two- Weiss, G. H., The delay problem for crossing
lane road. Proc. Austr. Road Res. Board, an n-lane highway, pp. 267-279. In V ehic-
4:595-612 (1968). ular Traffic Science. Proceedings of the
PROBLEMS 173
Third International Symposium on the 8. 9 PROBLEMS
Theory of Traffic Flow, New York, 1967.
I. A parking authority has found that for
Hauer, E., and Templeton, J. G. C., Queueing
every day an automatic ticket dispenser is out
in lanes. Transp. Res. 6:247-259 (1972).
of service $50 is spent to assign operators to
Iyer, K. S. S., and Maheswaran Nambudripad,
that station. Ticket dispensers fail at an aver-
D. B., On a stochastic study of highway traf- age rate (A.) of three/ day. The ·authority has
fic. Strassenbau und Strassenverkehrstechnik the option of using repair system A that costs
1969 (IVth International Symposium on the
$40/ day and is capable of repairing equipment
Theory of Traffic Flow). Karlsruhe (1968), at a rate (µ) of four/day. A more expensive
pp. 67-71.
system, B, capable of repairs at a rate (µ) of
Kleinecke, D. C., Discrete time queues at a six/ day, will cost $60/ day. The breakdowns
periodic traffic light. Oper. Res., 12: 809- occur at random (Poisson), and the length of
814 (1964 ). time to repair a dispenser is exponentially dis-
Leong, H. J. W., Some aspects of urban inter- tributed. Which of the two systems, A or B,
section capacity. Proc. Austr. Road Res. will minimize the total daily cost of repairs plus
Board, 2:305-338 (1964). breakdowns? (Answer: A=$190, B=$110).
Martin-Lof, A., Computation of an optimal 2. Vehicles arrive at a toll plaza at ran-
control for a signalized traffic intersection. dom at a rate of 10/ min. Calculate the queue
Transp. Sci., 1: 1-5 (1967). characteristics for each of the following ar-
Miller, A. J., Analysis of bunching in rural rangements: (a) a single toll booth where the
two-lane traffic. Oper. Res., 11 :236-247 service is random at an average rate of 20
( 1963). vehicles/min; (b) two toll booths where at each
Miller, A. J., Queueing in rural traffic, pp. 122- the service is random at an average rate of
13 7. In V ehicu/ar Traffic Science. Proceed- 10 vehicles/ min.
ings of the Third International Symposium 3. Pedestrians require a minimum head-
on the Theory of Traffic Flow, New York, way T of 6.0 sec to cross a roadway. Using the
1967. equations of section 8.5.2, find the mean delay
Pearce, C. E., Queueing systems with transport per pedestrian and the mean delay per delayed
service processes. Transp. Sci., 1: 218-223 pedestrian for flow rates varying from 100 to
(1967). 1,200 vehicles/hr (by 100-vehicle/ hr incre-
Pretty, R. L., and Blunden, W. R., On the com- ments). Plot the results.
puter simulation of a single channel queueing 4. The cycle length c at an intersection is
facility for a wide range of arrival and de- 90 sec. Effective green time g is 60 sec. Ve-
parture distribl1tions. Proc. Austr. Road Res. hicles arrive at a rate of 720 vehicles/hr and
Board, 2:248-260 (1964). the saturation flow s is 1I3 vehicle/ sec. (All
values are passenger car units.) The ratio of
Rosenshine, M., Queues with state-dependent
variance:mean, 1, is 1.10. Find the mean delay
service times. Transp. Res., 1 :97-104
by each of the methods discussed in section
(1967).
8.5.4.
Shaw, L., On traffic jam queues. Transp. Res., 5. The peak-hour demand volume on a
4:281-292 (1970). freeway is 4,500 vehicles/hr and the capacity
Weiss, G. H., The intersection delay problem for uninterrupted flow is 5,700 vehicles/hr. In
with correlated gap acceptance. Oper. Res., the event of a vehicle breakdown on the free-
14:614-618 (1966). way, the capacity is reduced to 4,200 vehicles/
Weiss, G. H., Pedestrian queueing at an n-lane hr. It requires 15 min to clear the breakdown.
intersection, pp. 280-286. In Vehicular Use the technique discussed by May (section
Traffic Science. Proceedings Third Interna- 8.6.1) to determine the queue and delay char-
tional Symposium on the Theory of Traffic acteristics. What happens if the breakdown can
Flow, New York, 1967. be reduced to 10 min?
Chapter 9
SIMULATION OF TRAFFIC FLOW
9.1 INTRODUCTION is the case with traffic flow simulations. Al-
though some early investi.gators examined the
As any discussion of theory proceeds, one
possible use of analog computers for traffic
usually has a growing desire to conduct experi-
flow simulation, and a few recent investigators
mentation. In some cases there is a desire to
have proposed applications of hybrid comput-
test a particular model; in others, a desire to
ers for traffic flow simulation (see, for instance,
evaluate a parameter or constant. In still other
Green et al. 43 ), the important traffic flow simu-
situations there is a desire to investigate situa-
lations remain those performed using a digital
tions that have not yet yielded to theoretical
computer.
treatment. At times there is a need for experi-
mentation simply to gain enough knowledge of The reasons for simulation include (but
a system to begin modeling. The conduct of need not be limited to) :
traffic experiments on operating facilities has 1. The need to test the behavior of a new
many difficulties. The experimenter must find system or operating procedure prior to its
a suitable site, prepare suitable instrumenta- actual construction:
tion, and then wait for the appropriate traffic
condition to occur. If the condition lasts only (a) The construction of the new system
a short time, tests may have to be conducted on may be very expensive and/ or time
several days or weeks; as a result, it may not consuming.
be possible to repeat a problem in the field. ( b) Experimentation with the real system
Some traffic situations may not occur at all on may entail considerable risk (such
an operating facility. Some experimental runs as traffic accidents).
may imply conditions that would be hazardous. 2. The need to test alternate systems
Some experiments might require the construc- under identical conditions. (For instance, it is
tion of expensive facilities. Test tracks, such never possible to exactly reproduce a specific
as that maintained by the (British) Transport traffic condition in the field; in simulation it is
and Road Research Laboratory in Crowthorne, quite routine to submit the same traffic condi-
England, enable execution of a certain range tions to several alternative systems.)
of tests neither requiring elaborate construction
nor entailing hazard. But even experiments on Any simulation may be divided into the
a test track can be quite expensive, requiring following steps:
the provision of vehicles, drivers, etc.
Since the development of the high-speed 1. Formulation of a model.
automatic computer, there has been a growing 2. Reduction of the model to a language
tendency to use digital computer simulation as acceptable by the computer.
a method of conducting a variety of experi- 3. Program checkout and internal verifica-
ments, especially those concerning systems tion of the model.
having important stochastic features. 4. Experiment planning and design:
(a) Design of an experiment that will
yield the desired information. 1
9.2 NATURE OF SIMULATION (b) Determination of how each of the
test runs will be executed.
Over the past 20 years various definitions
5. Performance of experiment(s):
and interpretations have been given to the
(a) Model validation.
term "simulation." Generally, present-day
(b) Simulation of new conditions.
usage refers to an experiment performed on an
6. Interpretation of results.
artificial model of a real system. The pre-
ponderance of current simulations makes use of Each of these steps is discussed in the follow-
a digital computer to implement a model. Such ing paragraphs.
175
176 SIMULATION OF TRAFFIC FLOW
9.3 HISTORICAL NOTE ON TRAFFIC not exhibit any nonrandomness. There are
FLOW SIMULATION many algorithms that can be used to generate
pseudorandom numbers," but the one that re-
Although suggestions for the simulation of mains the most common by virtue of its relia-
traffic flow took place as early as 1949, * it was bility and ease of implementation is the simple
not until several years later that there were multiplicative procedure. 10 This method may
published papers discussing possible tech- be described as follows:
niques,2·'1 and it was about 1954 before traffic
simulations were actually run on computers in
the United States;•-G, t Many studies were con-
cerned with how to conduct traffic simulation. where
By about 1960 it became generally accepted
that traffic simulation was possible and feasible, R,,,= the mth random number;
and efforts were directed to the development, p= the multiplier;
validation, and use of large-scale simulation n = number of digits in a normal word
programs. The work described by Wagner on the particular computer used;
et al.' is typical of these efforts. b = number base of computer;
Mod b"= instruction to use only the low-
order or less significant half of the
9.4 GENERATION OF RANDOM INPUTS full (2n-digit) product (the re-
mainder after dividing the product
One of the most important features of by b" the maximum integral num-
simulating traffic is the ability to generate ran- ber of times) ; and
dom events. Such generation takes place in two R 0 =any odd number selected as a start-
steps: First, a random number following a ing number.
uniform (rectangular) distribution is generated.
Second, this random number is treated as a This multiplicative procedure is a special
probability to substitute into an appropriate case of a general "congruential" algorithm,
distribution function in order to solve for the which may be stated 11 :
associated event.
Rm=[pR,,,_ 1 +c] Mod b11
P(h<t)=l-e-t/T
Variable x
which simplifies to Figure 9.1 Use of cumulative probability distribution
to convert random numbers from uniform distribution
P=l-e-t/T
to desired distribution.1°
t= -Tf-n(l-P) (9.2)
For the shifted negative exponential the 2. Take the sum of these fractions, giving
relationship is: the first normal variate.
3. Repeat using 12 different random frac-
t=(T-T)[-En.(1-P)]+r (9.3)
tions, giving the second normal variate.
If, however, a counting distribution such The results will be variates from a distribution
as the Poisson distribution is required, a step- having a mean of 6 and a variance of 1. (This
by-step solution must be used. This procedure result may be explained by the central limit
for the Poisson distribution is illustrated by theorem.)
Figure 9 .2 and described as follows: First, a
random fraction, R, is generated as previously
described. The cumulative Poisson distribution 9.5 MODEL FORMULATION
is then formed, term by term, using Eq. 3.6. At The formulation of a simulation model
each step the cumulation is compared with R. starts with the definition or selection of the
When the first value P(x) satisfying the rela- following items:
tionship P(x) ?.R is found, the corresponding
value of x is taken as the random variate (num- 1. The traffic situation to be simulated.
ber of arrivals). A flow diagram for accom- 2. The "figure of merit" or "measure of
plishing this process on a computer is given by effectiveness" to be used.
Gerlough. 10 3. The degree of complexity to be in-
An alternate approach is to generate and cluded.
store in advance a table with probability as an 4. The traffic generation (arrival) model
argument and time as a function. This may to be used.
be done by assuming various values of t and 5. Model for processing traffic through the
computing P. When in operation, random simulated situation.
fractions are used to search the P values of the 6. Computer language to be used (see
table. When a match is found, the correspond- section 9 .6).
ing t-value is used as the arrival headway. 7. Computer to be used (which is usually
For the production of random variates the one most readily available).
from a normal distribution the following special
At first it may seem inappropriate to in-
procedure may be used:
clude items 6 and 7 under model generation.
1. Draw 12 random fractions from a uni- However, in many cases these items influence
form distribution. the selection of other items, especially items 3
and 5.
Rather than attempt to describe model
formulation in general terms, the procedures
are illustrated by selected examples.
~I 0.4
~
·--·
I
I
.. -1
green intervals for that approach during the
same period.")
a.. 0.3 I
I
I
Once it has been decided that a study of
0.2
0.1
0
·--
I
__ JI
I
I
Value of x
Corresponding
to R
load factor as it relates to capacity should be
conducted by simulation, the following steps
of model formulation are implemented:
(j) At the end of an experimental run 6. The program will be written in BASIC.
the various figures of merit are com- 7. The simulation will be run on a com-
puted and printed out. puter having an on-line time-sharing system
using a remote terminal.
The flow chart for this program is shown
Start in Figure 9.4. The program listing is given in
Appendix D-2. This program is relatively sim-
ple in a number of ways; principally, (a) no
opposing or cross traffic is considered, (b) time
is advanced in uniform intervals (periodic
Initialize scan), and ( c) only one type of random event
(arrival time) is generat(!d. The program does,
however, illustrate the s'teps involved in con-
structing a simulation.
Generate Arrival
Store if Queue Present
9.5.2 Simulation Example: Simple Four-
Way Intersection
It is now possible to progress to the simu-
Process Departure lation of a more complex situation; namely, the
four-way intersection shown in Figure 9.5, as
described by Worrall.1' Several types of event
are generated randomly. The intersection con-
sists of two one-way streets, each with a single
Compile Statistics lane. The east-west street has priority over the
north-south street. Low flows are assumed,
and the arrival headways are assumed to follow
the negative exponential distribution of Eq.
3.8. The value T in the equation is taken as
3,600/ V, where V is the arriving volume in
Is Run Complete?
vehicles/ hr. Each approach is assigned (as
input data) probabilities that vehicles will turn
No
or continue straight ahead. A gap acceptance
distribution is provided, in tabular form, and
Increment Clock the same distribution is used for all gaps. (See
Table 9.2.)
Figure 9.5 illustrates the decisions by
north-south (N-S) vehicles in accepting or
rejecting gaps between east-west (E-W) ve-
Check Signal Status hicles. The simulation model may be described
Update if Ready with the aid of the flow diagram of Figure 9.6.
The input data consist of volume levels on
the two approaches and a gap acceptance table.
The input operation is indicated by the first box
Yes Is New Arrival Needed? at the top of Figure 9.6. This includes the
initial settings of all counters, etc. The simula-
tion loop starts with the second box, "Generate
Next N-S Arrival Gap." This arrival gap is
generated by first generating a random fraction
Compute Results and Print and then substituting this fraction in Eq. 9.3.
The actual arrival time of the new arrival is
obtained by adding the generated time headway
to the arrival time of the previous N-S vehicle.
Stop
At this point it is appropriate to test
Figure 9 .4 Flow chart for simulation example. whether the undelayed arrival time of the new
MODEL FORMULATION 181
N-S vehicle exceeds the maximum duration of the vehicle will arrive at the stop!ine. This
the experiment. If the limit has been reached, calculation starts by determining whether the
the program jumps to the calculation and print- previous N-S vehicle has crossed the intersec-
ing of output results; if not, the program con- tion: if it has not, the new vehicle joins the
tinues. queue; if it has, the new effective arrival time is
Next the effective arrival time of the new equal to the new actual arrival time.
N-S vehicle is computed. This is the time that Queues may be handled in several ways.
N N
_J__~
N.1
_J f71
N.1
-~~,~~~~~~~~~
J___, E.2
-c=J E.1 E w-.- -L__j E.1
I
E -c=J
It!
Stage 1
s
Stage 2
N N
J
w-c:=J E.1
~
-~-....__,
1
-CJ
N.2
N._1- - -
E.2 ~E
J_ ~_
W-c:=J E.2
I
I
I
N.2
E.3
-c E
~N.l
I
I
•
s:l s
17
s I
Figure 9.5 Simulated intersection shows schematically the various conditions: Stage 1, vehicles N. l and E. l ap·
proach intersection; Stage 2, vehicle N. l reaches intersection and examines E·W stream for acceptable gap, but the
available gap between E. l and E.2 is not acceptable, so vehicle N. l is delayed; Stage 3, vehicle N. l continues to
examine E-W stream for acceptable gap and vehicle N.2 arrives at intersection and forms a queue; Stage 4, gap
between vehicles E.2 and E.3 is acceptable to vehicle N. l, so vehicle N. l moves off across intersection, and vehi·
cle N.2 commences looking for an acceptable gap. However, the lag between N. l moving off and E.3 arriving is not
· acceptable to N.2, so vehicle N.2 is delayed further.
Set Input Data
Volume Levels, Turning Mvts., Gap Acceptance, etc.
Increase
N-S Count by
One
-----No - - - - - -
Yes en
......
Has Previous N-S Vehicle Crossed the Intersection? ~
c:::
t'"'
Yes No ~
...... t
Compute Queue Length* 0
z
0
"I1
Effective N-S Arrival Time Effective N-S Arrival Time= (Departure ,...,
=Actual N-S Arrival Time Time of Previous N-S Vehicle+ Starting Delay) :;tl
>
"I1
"I1
......
(')
"I1
t'"'
0
Select Appropriate Acceptable Gap Length in E-W Stream
~
t
Is Effective N-S Arrival Time Later Than
Last E-W Arrival Time?
Yes No
• No
Effective N-S Arrival Time =
Arrival Time of Last E-W Vehicle
Departure Time of N-S Vehicle=
(Effective N-S Arrival Time+
Starting Delay)
• Calculate Average Queue Length and Max. Queue Length for N-S Vehicles
Print Out Results (Volumes, Turning Mvts., Delays, Queue Lengths, etc.)
co
w
Figure 9.6
B
Flow chart for intersection simulation.17
184 SIMULATION OF TRAFFIC FLOW
TABLE 9.2 Gap Acceptance Distribution whether the effective arrival time of the (first-
for Example of Section 9.4.2 in-queue) N-S vehicle is later (greater) than
the arrival time of the last E-W vehicle gener-
Percent ated. If it is, it is necessary to generate a new
Gap (sec) Acceptance E-W vehicle arrival gap and arrival time. The
0-2 0 method used is similar to that used for N-S
2-3 9 vehicles. If the new E-W arrival time exceeds
3-4 18 the time limit for the experime,nt, the run is
4-5 22 terminated (after calculating and printing the
5-6 40 output results); otherwise, the E-W traffic
6-7 59 count is increased, and the available gap in the
7-8 61 E-W traffic is examined for acceptance by the
8-9 80 first-in-queue N-S car. The available gap is
9-10 90 the arrival time of the last E-W car minus the
10-11 92 effective arrival time of the N-S car. After this
11-12 93
12-13 94 gap is computed it is compared with the previ-
13-14 95 ously computed minimum acceptable gap.
14-15 96 If the gap is not acceptable, the effective
15-16 97 arrival time of the first-in-queue N-S vehicle is
16-17 98 reset to the arrival time of the last E-W vehicle
17-18 99 and a new E-W vehicle is generated. If th~
>18 100 arrival time of this new E-W vehicle does not
exceed the time duration of the experiment,
the E-W traffic count is increased by one, and
the acceptability of the gap in front of this
new E-W car is tested. If the available gap
One effective way is to establish within the (discussed in the previous paragraph) is found
computer an array three columns wide and of acceptable, it is accepted, and the departure
sufficient length to accommodate the longest time of the N-S vehicle entering the intersec-
queue expected. In addition, a queue counter tion is computed as its effective arrival time
indicates the number of vehicles currently in plus the appropriate starting delay. Its delay
the queue. As each new vehicle joins the queue, in queue is computed by subtracting its arrival
the queue counter is first increased, then the time from its departure time. This delay is
row of the array corresponding to the count is added to the cumulative record of delay. After
selected. Into the three columns of this row are correcting the queue for the departure of one
placed the actual arrival time, the effective vehicle,- the N-S count is increased and the
arrival time (if it can currently be computed), simulation loop starts again by the generation
and the departure time (if it can currently be of a new N-S vehicle.
computed). As the first vehicle in the queue Two important points are illustrated by
moves out of the queue and across the inter- this example. First, in addition to random
section, the first row in the array is removed, generation of arrivals on two approaches there
each of the remaining entries is moved up one is random generation of gap acceptance. Sec-
row, the place vacated by the last vehicle is set ond, the "clock" is not advanced by a uniform
to zero, and the queue counter is decreased. At periodic interval; instead the examination
the same time any further computations of moves from one important time to another
departure and actual arrival times are per- important time. These differences in the meth-
formed. ods of "review" or "scanning" are known,
The gap in the E-W stream acceptable to respectively, as "periodic scan" and "event
the first-in-queue N-S vehicle is determined. scan."
A random fraction is generated; this fraction is
then compared with the percentage values in
9.5.3 Simulation Example: Freeway
Table 9.2 (expressed as fractions). The gap
Merging Area
corresponding to the random fraction is then
designated as the minimum gap that will be This example, based on a paper by Woh1,1s
accepted by the N-S vehicle. consists of a slightly more complex simulation
Now begins an extensive series of tests -that of a freeway merging area. Figure 9.7
concerning gaps in the E-W stream. First, test shows the schematic layout of the merging area.
MODEL FORMULATION 185
Next Freeway
Arrival Gap vailable Entrance Lag
-
Next Ramp Arrival Gap
his expected arrival time at the nose of sec) plus 5-sec acceleration delays on
the ramp and the availability of a proper entering the freeway flow. Thus the
freeway gap. If at the time the ramp total delay is 16 sec.
vehicle arrives at the nose there is an The second ramp vehicle cannot
acceptable gap in the freeway flow, he enter the freeway until some time fol-
will enter. If not, he will be delayed and lowing the departure of the first ramp
must wait for a later gap in the freeway vehicle. However, the second ramp ve-
flow. Figure 9.8(b) shows the positions of hicle may have arrived on the ramp
the first ramp and freeway vehicles at the earlier than the departure time of the
time the ramp vehicle arrives at the nose. first and been waiting in queue. Since
Because the first freeway vehicle had we are primarily concerned with delays
passed the nose prior to the arrival of the to ramp vehicles that enter the freeway
first ramp vehicle, the next freeway ar- flow, the original times of arrival of the
rival gap (7 sec) and its arrival time at ramp vehicles at the nose must be com-
the nose (0012 sec) must be determined; puted as well as the times at which the
also, a check must be made to ensure that ramp vehicles actually entered the free-
this freeway arrival time (0012 sec) is way flow. Thus, the first two steps of
later than the ramp arrival time (0008 Figure 9.8(e) are to "generate" the next
sec). Because it is, the available entrance ramp arrival gap ( 6 sec) and compute
Jag or gap (that is, the gap available for arrival time of the second ramp vehicle
the ramp vehicle to enter the freeway (0014 sec). Following this, it must be
flow) must be computed (it is 4 sec). (If determined whether or not the ramp ve-
the freeway arrival time is not later than hicle was delayed by the previous ramp
the ramp arrival time, successive freeway vehicle; that is, Is the arrival time (of
arrival times must be determined until the the second ramp vehicle) earlier than the
first one with an arrival time later than departure time of the previous ramp
that of the ramp vehicle is located.) Fol- vehicle? If the ramp vehicle was delayed
lowing this, the acceptable LRS entrance by the previous one, the earliest time that
gap (5 sec) for the ramp vehicle must be the ramp vehicle can actually begin
"generated." looking for an acceptable freeway gap
Since the available entrance lag is must be calculated on the basis of the
not acceptable, the ramp vehicle will be departure time of the previous ramp ve-
delayed at least until the second freeway hicle. Consequently, the so-called "look-
vehicle passes. Figure 9.8(c) shows the ing-for-gap" time of the second ramp
vehicle positions at the arrival time of the vehicle ( 0022 sec) is equal to the de-
second freeway vehicle at the nose (0012 parture time of the previous ramp vehicle
sec); at this time the ramp vehicle ex- (0019 sec) plus operating delays (3 sec)
amines the next freeway arrival gap ( 4 to allow for perception and reaction
sec), which in this case is also the avail- times, etc.
able entrance gap. Since the ramp The next step is to determine the
vehicle has been delayed, we must deter- earliest freeway vehicle arrival time that
mine its acceptable HRS gap ( 6 sec). is later than the look-for-gap time of the
Again, the available entrance gap is not second ramp vehicle. In this case the last
acceptable to the ramp vehicle, and computed freeway vehicle arrival time
another freeway gap must be examined. (0026 sec for the fourth freeway vehicle)
Figure 9.8(d) shows the vehicles at is later than the looking-for-gap time of
the time when the third freeway vehicle the second ramp vehicle (0022 sec); there-
arrives at the nose. The next freeway fore, the available entrance lag is the dif-
arrival gap (10 sec) is determined and the ference between these two times, or 4 sec.
freeway arrival time of the fourth free- Based on earlier assumptions, an accept-
way vehicle (0026 sec) is computed. The able LRS entrance gap for the waiting
available entrance gap (10 sec) is larger ramp vehicle (3 sec) must be computed.
than the acceptable HRS gap ( 6 sec); The available entrance gap is acceptable,
therefore, the ramp vehicle can accept and the second ramp vehicle enters the
the gap and enter the freeway. The de- freeway; see Figure 9.8(f). Its departure
parture time (0019 sec) for the ramp
time (0024 sec) is the looking-for-gap
vehicle is the third freeway vehicle ar-
rival (0016 sec) plus the operating delays time (0022 sec) plus operating delays of
(3 sec) to account for the starting reac- 2 sec; its delay ( 15 sec) is the departure
tion delay, etc. The final step is to com- time (0024 sec) minus arrival time (0014
pute the ramp vehicle delay, which is its sec) plus 5-sec acceleration delays.
departure time (0019 sec) minus its These steps can be repeated for as
original arrival time at the nose (0008 many simulation trials as are desired; the
Initialization
SET (1) Volume Levels,
(2) Gap Acceptance Data, t
etc. ,_.
00
00
0
z
0
'Tl
Generate LRS acceptable
entrance gap
,..,
I':'
>
'Tl
'Tl
Is freeway veh. arrival time Increase no.
later than looking-for11ap of freeway (3
time of ramp vehicle 7
Yes I I No
Generate next freeway
veh. by one
-
'Tl
0
$2 t
arrival gap and compute
its arrival time
•
Increase no. of ramp Generate HRS acceptable
vehs. by one entrance gap
Yes B
Compute total delay for
all ramp vehs. that have Compute no. of ramp vehs.
entered freeway waiting in line and their
delays; add to total delay
~
0
Yes No
Yes No
00
Re-set ramp volume to IC!
initial value
•
Chapter 10
EPILOGUE
Although there is at present no unified preparation of this document, for instance, the
theory of traffic flow, there are many theoretical new literature has been extensive, and there
approaches to a variety of traffic phenomena. have been several new theoretical developments
A substantial portion of these theoretical ap- that could not be adequately covered. Never-
proaches have made important contributions to theless, the need for further research is great,
practical solutions of traffic problems. and all who find the theory of traffic flow in-
It should be evident from the discussions teresting are seriously urged to consider under-
that traffic flow theory is an evolving science. taking research.
Each year sees research that results in impor- It is hoped that a unified theory of traffic
tant theoretical advances. During the course of flow will not be long in coming.
197
Appendix A
DATA SUPPLEMENTARY TO CHAPTER 2*
A-1 RELATIONSHIPS BETWEEN TIME But
AND SPACE SPEED STATISTICS
Time Mean Speed from Space Mean Speed
by definition of mean and
Using the method of Wardrop, 1 segregate
total flow into m subflows by speed. Define Lf;'(ui-u.)2 =a-.2
(2.7)
and
But
iit= i"' uft(u)du
where
Thus,
ft(u) =speed density in time and
"k I
L ;U; =k """~
2 I 2 f.(u) =speed density in space.
k kq L q An important relationship given in Haight and
Mosher 11 but proved by Breiman 13 is:
where
uJt(u) =uf.(u)
After multiplying both sides by u and integrat-
ing over the entire range of u:
But
q= k u.; Thus
u. i"' uft(u)du= i"' u 2 f 8 (u)du
- k"
ut= L -f!u/ 1 "t '[-
k - =-=-Li
u.
u.+ (ui-us
u.
- )]?- u.ut= i"' u 2 f.(u)du (A)
Define
199
.. ..
V m =-
1
-L,
N
(Xi- M) 2 =variance about arith-
-(M~N )(O)
N i=i
metic mean +~NL (Xi-M)4 . .
N
-N-- = h armomc. mean. It can be assumed that
1
L.x.
f=l i
M 5N» L (Xi-M) 4
since Eq. C is converging.
Expand in Taylor series: Thus the last term can be neglected, as can all
later terms in the expansion.
;. = A
i
0 +A 1 (Xi-M) +A 2 (Xi-M) 2 Then
+A3(Xi-M)3+. 1~ 1 - 1 vm - M 2 V,,.
Evaluate constants by differentiating,
N ?X.-
1=1 i
M + M 3 --w-
1 M3 M
H= (M + V)IM 3 2 M + V,,.
2
1+ v,,,
M2
for
M( 1-~)
1
+
---(X I -M)
M4
3
t (C)
v
=M(l- MZ 2
V111)=M- V,,,
M.
*
Then 1--"-'·
M'
lNl lNl 1
Converting to traffic notation gives:
N ix.=N
1=1 t
I M - NM 2 :L<Xi-M)
l=l
(2.8)
+ N~3 L (Xi-MF as an approximate method for use in traffic
engineering practice. Note that this relationship
__1 _"""'(X.-M)3 when combined with Eq. 2.7 implies that
NM' L.., '
1
+ -NM-"""'(X
5 L..,
-M) '+ I
Thus, in using Eq. 2.8 one must be willing to
accept this assumption.
'''Note the use of approximate methods in
this development. * Yule and Kendall 1° suggest this relationship
t Eq. C converges for 0 < x < 2M, which is for cases where deviations are small compared to
usually the case for traffic. the mean.
DATA SUPPLEMENTARY TO CHAPTER 2 201
For i=2,
(3.30)
dP 2 (t) =><[P 1 (t)-P 2 (t)]
dt
Now, P_ 1 (t) =0 (i.e., impossible to have
<0),
(i.e., no arrivals up to * Any standard method may be used for solu-
time t=O), tion of this differential equation. The method of
operators is particularly simple. See any standard
Pi(O) =0 for i"?. l (zero probability 1
of i arrivals at time t=O). text, such as Ford. 58 The form Y= D+Au(x)
results in a solution
Setting i=O in Eq. 3.30,
202
DATA SUPPLEMENTARY TO CHAPTER 3 203
Thus
(3.32)
1
P 9 (t) =--A(At)e-At
- D+A
2m 2 e-1n 3m 3 e-m
= O+me-11i+ 2!
+---
3!
But
=me-{l+m+ ;,
2
+-};~ J
== me-me·m,
=m (3.33)
= L (x-µ,) 2
P(x) (3.35)
x=O
(3.2)
Because the population mean is m, this variance
may be stated:
If At= m, the result is the most familia1 form
a- 2 = 2,(x-m)2P(x)
of the Poisson distribution:
= 2,(x2 -2xm+m 2 )P(x)
P(x)=--1-
mxe-m
(3.1)
= 2,x P(x)-2m2,xP(x) +m 2,P(x)
2 2
..
J
L,t(x)
x=O
-m
-
2
e-"'[l+m+m?
2! . .J
where f(x) is the frequency of occurrence of x.
For the Poisson distribution, substitute
2,x 2 P(x) =m 2
+m
mxe-111· a-2 = [mz + m] - [2m2] + [m2]
P(x)=--,-
m. er= m (3.36)
204 DATA SUPPLEMENTARY TO CHAPTER 3
Thus, for the Poisson distribution the popula- by application of a similar argument the dis-
tion variance equals the population mean. tribution for the whole population is found to be
µ.= f 00 tp(t)dt
p(t) = -=--e-11-T•/•T-n.
dt -T
1
00
1
µ,= ( ---e-rf-T>/(T-T•dt
and integrating by parts and inserting limits, 00
T-r
gives
1
00
1
=0+ t ·--e-U-r>/CT-r,•dt
µ.=T ( 3.41) T-T
erf<T-r) "'
=·- · - - [ - 7 ( T-T)e-rl<T-T1(T-T)2e-r/<T-T)]
Population Variance of Negative (T-r)
T
=
J -e-t!Tdt-2
oo("
T
1"' o
te-t!Tdt+ J"'
o
Te-t/Tdt
P?pu_lati~n Variance Shifted Exponential
D1stnbut1on (about Origin)
(]""= T2 (3.42)
1
00
1
<T2= (t-T)Z--e-U-rl/<T-r•dt
T-r
Shifted Exponential Distribution
In considering Figure 3.5 in text, assume = e ~T
T
_T
T
{1°" T
t 2 e- T-T
I dt-2T 1°"
T
te- T-T
t dt
an origin through the point of contact of the
curve and the t (horizontal) axis. About this 00 t
axis
P(h < t) = 1-e-t'/T' (3.43)
+ y2
1 e T-T df
}.
Integrating by parts and simplifying,
where t' =abscissa of any point from the origin
at the point of contact of the curve and the (3.44)
vertical axis; and T' =abscissa of the mean or
center of gravity with respect to the same
origin. B-4 LOGNORMAL DISTRIBUTION
Now construct a vertical axis normal to
the t axis and a distance r to the left of the point The normal distribution is used to describe
of contact of the curve and the t axis. Now all systems where the measured variable is nor-
abscissas t are measured from this new axis. mally distributed; lognormal distribution is used
In Eq. 3.43, each abscissa t' is now replaced to describe systems where the logarithm of the
by ( t - r) . The value T now represents the measured variable is normally distributed. 6 1-63
abscissa of the center of gravity of the curve Because log is not defined for arguments equal
from the new axis, and the value of T' in Eq. to or less than zero, the lognormal distribution
3.43 is now replaced by (T-r). Thus, for the is defined only for positive measured variables.
shifted exponential, (In the present discussion, use of natural logs
is assumed; use of common logs (to base 10)
P(h < t) = 1-e-<t-T>!<T-r> for t?.r (3.22) would require only a change of scale.) To
206 DATA SUPPLEMENTARY TO CHAPTER 3
obtain a straight line for the cumulative log- Estimation of the population parameters by the
normal distribution, use graph paper having a maximum liklihood method may be simpler 61 :
log scale on one axis .and a normal probability
scale on the other. (Figure B.2 shows several 0-2 == 52 (3.49)
curves plotted on normal probability paper; •
Figure B.3 shows the same curves plotted on
lognormal paper.) Fitting then proceeds in the same manner as
The parameters of the lognormal distribu- when fitting a normal distribution, making use
tion may be computed as follows 61 : of normal distribution tables. That is, the prob-
ability cp(x;) is obtained from normal probabil-
Measured variable=yi; xi=log Yi, for xi ity tables using an argument Z;, where
assumed to be normally distributed.
x,-µ,
- 1 N 1 N Zi=-- - (3.50)
er
x=N"2.,xi=N:L:,logyi (3.45)
t=l i=l
% 0 2 3 4 5 6 7 8 9
0 2.67 2.95 3.12 3.25 3.36 3.45 3.52 3.59 3.66
10 3.72 3.77 3.82 3.87 3.92 3.96 4.01 4.05 4.08 4.12
20 4.16 4.19 4.23 4.26 4.29 4.33 4.36 4.39 4.42 4.45
30 4.48 4.50 4.53 4.56 4.59 4.61 4.64 4.67 4.69 4.72
40 4.75 4.77 4.80 4.82 4.85 4.87 4.90 4.92 4.95 4.97
50 5.00 5.03 5.05 5.08 5.10 5.13 5.15 5.18 5.20 5.23
60 5.25 5.28 5.31 5.33 5.36 5.39 5.41 5.44 5.47 5.50
70 5.52 5.55 5.58 5.61 5.64 5.67 5.71 5.74 5.77 5.81
80 5.84 5.88 5.92 5.95 5.99 6.04 6.08 6.13 6.18 6.23
90 6.28 6.34 6.41 6.48 6.55 6.64 6.75 6.88 7.05 7.33
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
99 7.33 7.37 7.41 7.46 7.51 7.58 7.65 7.75 7.88 8.09
"From Finney."
DATA SUPPLEMENTARY TO CHAPTER 3 207
99.95
99.8
99.5
99
95
90
a= 0.5
70
x
a.. 50
cf?-
30
10
5
1
0.5
0.2
0.05
0 10 20 30 40 50
x
Figure B.2 Three cumulative lognormal distributions, all with log x = 1.0 and standard deviations of
log x= 0.1, 0.3, and 0.5 plotted on probability paper."
99.95
99.8
99.5
99
95
90
70
x
0:::- 50
'#. 30
10
5
0.5
0.2
0.05LL.~-L.~...J_-L.---L---L-LJ.J.~L__J-..J.._L_L-~-L~..l--L.-L-L.....L---l.__J---l._.J..._L.J
1 2 3 4 6 8 10 20 30 40 60 80 100
x
62
Figure B.3 Same curves as those in Figure B.2 but plotted on logarithmic probability paper.
208 DATA SUPPLEMENTARY TO CHAPTER 3
B-6 FITTING GENERALIZED POISSON of statistical significance of the fit are available.
DISTRIBUTIONS USING INCOMPLETE These tests permit selection of one of the two
GAMMA FUNCTION TABLES decisions: (1) It is not very likely that the true
distribution (of which the observed data consti-
The procedure for fitting generalized tute a sample) is in fact identical with the
Poisson distributions using tables of the incom- postulated distribution; (2) the true distribu-
plete gamma function is as follows: tion (of which the observed data constitute a
sample) could be identical with the postulated
1. Enter graph (text Figure 3.13) with distribution.
mean and variance. Select value for k, which
It can be seen that either decision can be
need not be an integer but may be any value
erroneously made. Decision 1 can be wrong
greater than 1. (If k = 1, the usual "simple"
if in fact the postulated distribution is the true
form of the Poisson should be used.)
distribution. On the other hand, decision 2
2. Compute A from Eq. 3.29. can be wrong if the true distribution is in fact
3. Form the table: different from the postulated distribution. Sta-
tistical tests of significance allow for specifying
x (p+l) u p l(u,p) 1-I(u,p) P(x) the probability (or risk) of making either of
these types of error. Usually, the probability
0 of making the first type of error (incorrectly
1 rejecting the postulated distribution when in
2 fact it is identical with the true distribution) is
3 specified, and no statement is made with regard
etc. to the second type of error. The specification
of the first type of error is expressed as a
"significance level." Common significance levels
The following equations are used: are 0.01, 0.05, and 0.10. Thus, when a test is
made at the 0.05 (5%) level, the engineer takes
(p+l)= (x+l)k the chance (risk) that 5% of the rejected postu-
1
U= 'J.../ (p+ 1 ) /z
lated distributions are in fact identical with the
p=(p+l)-1 corresponding true distributions. For a com-
plete discussion of the theory underlying this
4. Values of I(u,p) are obtained from and other statistical tests of significance, the
tables.17 reader is referred to any standard text on
5. The entries in the column [1-/(u,p)) statistics.
are the cumulative probabilities. The prob-
abilities of the individual values of x-i.e.,
P(x)-are obtained by subtraction: Chi-Square (x 2 ) Test
P(O) = 1-I(u,p) 0 The best known test of goodness-of-fit is
P(x) = [l-/(u,p),.)-[l-I(u,p),._ 1 ) the chi-square (x 2 ) test, which is described as
= I(u,p),._ -I(u,p).,.
1 follows:
Table 3.11 in text contains observations Let f =observed frequency for any group
during 64 15-sec counting intervals on a free- or interval and F =computed or theoretical fre-
way during the morning peak period. A gener- quency for the same group. Then, by definition:
alized Poisson distribution has been fitted to
the data. Table B.2 illustrates the method of
(3 .52)
fitting by means of the incomplete gamma
function: here the value of k is 2.
where g is the number of groups.
Expanding,
8-7 GOODNESS-OF-FIT TESTS 0
~ [ f·"
When comparing the fit between a theo- x~~
-L - - - 2/
I
-- F c -F-"]
! I 1 I
i=l F; F; F;
retical distribution and a set of experimental
data, it is desirable to have some method of
quantitative evaluation of the fit. Several tests
DATA SUPPLEMENTARY TO CHAPTER 3 209
In the fitting process, the total number of
theoretical observations is set equal to the total
number of experimental observations. Thus,
., (!I f;")
x·= LT
i=l l
-n ( 3 .53) ---
-r-t-Ot-MOO\N"<l-10\0_.:
OON\Ci~NNocii.r)~ 'I
Distribution A v
Poisson 1 g-2
Negative binomial 2 g-3
Generalized Poisson 2 g-3
Binomial 2 g-3
-
O\r-...('i")V)O\Vi.......cQ\\O\r')('i")
further stipulates that the total number of ob- or.'.'6vi-.t-.t-.t.,.;.,.;.,.;.,.;
servations be at least 50. When the number of
theoretical occurrences in any group is less than
5, the group interval should be increased. For
the lowest and highest groups this may be N
a:i
* In using a table of x 2, care should be exer- w
N'<l"\OOOON'<!"\OOOON
.-t....-11"""'1...-1.......iNM
10.20
2.5
13
4.7
7.1
8.8
l 15.5 7.81
11.37
>
Cl.l
c
"t)
'"d
t""'
trl
t
7 11 77 539 12.3 9.84 9.4 12.87 ~
8 10 80 640 12.1 8.26 8.8 11.37 trl
z
'l
9 11 99 891 9.4 12.87 7.3 16.58
10 9 90 900 5.4 15.00 ~
11
12
1
1
11
12
121
144
2.8
51.0
10.0 12.10
3.7}
2.3 8.7 0.46
~
>12
TOTAL
-
64
0
-
478
0
--
0
3,822
0.4
---
64.0
--
64.90
--
2.7
64.0
--
75.46
a
(")
::c:
m= 478/64=7.469 ii= 0.46 m=7.469 >
'"d
( 478) 2 n= 16.08 e-"'= 0.00057 ....,
3822-~ 'l:f/IF,--:Z.f,= 64.90-64.00 'l:f.2/F,-'l:f,= 75.46-64.00 trl
s- 63 =0.90 = 11.46 :;>;;
..,,
v= 6-3=3 v=7-2=5 t
= 3.999 2 2
S-lm= 0.535 X o.oo= 7.81 b X o.os= 11.07 c
•
DATA SUPPLEMENTARY TO CHAPTER 3 211
accomplished by making these groups "all less TABLE B.4 Values of the Kolmogorov-
than" and "all greater than," respectively. Smirnov Test Statistic "d" a for Various
Table B.3 demonstrates the computations Sample Sizes and Levels
for the x2 test. of Significance b
g(T)dr=f(u)du=f( ~) (~)ctr
integration being over all values of t for which
(The minus sign of the expression for du is ql(t) >0.
dropped because probabilities are always posi- For real cases histograms result rather
tive.) Now, because speeds are assumed to be than continuous curves. Thus the following
normally distributed, expression is employed,
(u;~) ]
2
1 q2(j) = L,qi(i)g(j-i)
f(u) = exp [ -
<T\}21T (]" i
where (]" is the population standard deviation of where i and j are discrete intervals of the histo-
speed. Therefore, gram.
It should be noted that the foregoing anal-
* Reference citations are listed in Chapter 7. ysis assumes that passing is possible at will.
213
..
Appendix D
DATA SUPPLEMENTARY TO CHAPTER 9*
D-1 DISCUSSION OF SIMSCRIPT Exogenous event An event affecting the
AND GPSS program from outside the simulation.
Many simulation languages exist, espe-
cially when one includes those written in SIMSCRIPT
Europe; the two best known in the United
States are SIMSCRIPT and GPSS. Before pro- SIMSCRIPT is a compiler-type language
ceeding with brief descriptions and a compari- based on FORTRAN 19 - 23 • In SIMSCRIPT the sys-
son of these two languages, it will be helpful to tem is described in terms of entities, attributes,
consider the following glossary. sets, and events. For instance, cars, intersec-
Assembler A computer programming lan- tions, streets, etc., would constitute entities.
guage that provides a one-to-one correspon- Attributes of each car would include destina-
dence between mnemonic symbols (source lan- tion, gap acceptance criteria, normal accelera-
guage) and the natural machine language tion, desired cruise speed, etc. Sets would in-
commands (object language), but usually with clude all cars in queue on the northbound leg
the automation of such tasks as the assignment of intersection 23, all cars moving as a platoon
of memory addresses. The program written in from intersection 18 to intersection 19, etc. An
assembly language is converted to a program event represents one or more actions that take
in machine language by an assembly program. place instantaneously at a given time (e.g.,
The assembled program is then used for com- main street signal turns green). An activity is
puter runs. an occurrence that takes place between two
Interpreter A computer programming events (e.g., car has left intersection 14 but
language that provides the capability for pro- has not yet reached intersection 15). The
gramming extensive computations by a simple state of the system at any time is given by at-
(source) language. The conversion to the tribute values and set memberships of all
machine (object) language usually involves the individual entities. The state is changed by the
provision of many machine-language com- occurrence of an event. Changes in state may
mands for each source command. This con- be responses to commands such as
version is performed during the computation
CREA TE or DESTROY an individual
run and is repeated each time the source lan- entity.
guage calls for a particular function or com-
ALTER set membership of an indi-
putation. Thus, the interpretation program vidual entity.
must occupy part of the computer memory dur-
CHANGE numerical value of an attri-
ing running, and the interpretation during each
bute.
iteration can greatly increase the computation
time. Events are handled by FORTRAN-like subrou-
Compiler A computer programming lan- tines. Endogenous events are triggered dy-
guage in which programs are written in a namically from within the model. Exogenous
source language closely adapted to the type of events may be scheduled prior to the start of
computation to be performed and in which the the simulation run. Statistics may be accumu-
conversion to machine (object) language in- lated. Output is obtained through the use of
volves the compiling of a number of library a report generator whose format is defined on
subroutines. The compilation is performed a format definition form. Input consists of
prior to the computation run. definition cards, initialization cards, and sub-
Endogenous event An event affecting the programs.
program from inside the simulation. The various inputs are handled as follows:
Defined items are converted directly into
*Reference citations are listed in Chapter 9. machine-coded subroutines that will control the
214
DATA SUPPLEMENTARY TO CHAPTER 9 215
storage and retrieval functions; machine-coded 5. Computational elements
dynamic control (timing) subroutines are also Arithmetic variables
produced directly. s1MSCRIPT subprograms are Functions
converted into FORTRAN source language sub-
routines. These FORTRAN subroutines are then Transactions, temporary elements having
compiled by a standard FORTRAN compiler. eight parameters and eight priority levels, move
Initialization cards are treated as data. from one block to another in the model. As a
transaction enters a block a subroutine asso-
ciated with the bloc;k type causes a change in
GPSS (General Purpose System
the state of the system. Transactions are
Simulator)
created by ORIGINATE and GENERATE blocks and
GPSS is an interpreter-type simulation lan- are destroyed by TERMINATE and ASSEMBLE
guage that describes the system in terms of blocks. Time is advanced in discrete steps, the
blocks, transactions, and equipment using block value of the step being designated by the user.
diagrams 22 - 2 ". Outputs consist of a standard set of sta-
GPSS provides the following elements: tistics written on the output tape at the end of
the run.
1. Basic elements Inputs consist of definition cards, each of
Blocks which defines a block, function, table, arith-
Transactions metic variable, or storage capacity.
2. Equipment elements
Facilities
Comparison of SIMSCRIPT and GPSS
Storages
Logic switches Table D.l compares SIMSCRIPT with GPSS,
3. Statistical elements based on the work of Murphy 26 and Teichrow
Queues and Lubin. 23 To summarize, sIMSCRIPT is
Distribution tables more efficient than GPSS but requires a much
4. Reference elements higher level of skill on the part of the user; thus,
"Sa vexes" GPSS is easier for use by the beginner.
00385 El=l.O
00390 .Al=0.00115*V
00395 .A2=1-.Al
00400 BC1)=3.8
00405 BC2)=6.9
00410 BC3)=9·6
00415 8(4)=12.0
00420 8C5)=14.2
00425 FOk I=6 TO 30
00430 BCI>=l4.2+2.l*CI-5>
00435 NEXT I
00440 C2= Cl+ Hl
00445 D4=C2
00450 HEM
00455 HEM
00460 REM GEN EH.ATE .AhhI VAL
00465 h2=HNDCX>
00470 h3=HNDCX)
00475 IF R2<=Al THEN 00490
00480 H=T2*C-LOGCh3))
00485 GO TO 00495
00490 H=Tl*C-LOGCH3))+El
00495 C3=C3+H
00500 GO TO 00685
00505 REM ENTER .ARhIV.AL INTO QUEUE LIST
00510 ACLl+l>=C3
00515 Ll=Ll+l
00520 IF Ll=<Ql THEN 00530
00525 Ql=Ll
00530 GO TO 00535
00535 HEM PROCESS DEP.AHTUhES
00540 IF Gl=O THEN 00670
00545 IF D3>Cl THEN 00670
00550 IF D3>C2 THEN 00670
00555 D4=D3
00560 hEM CHECK IF MOVING
00565 IF Xl=l THEN 00655
00570 Sl=Sl+D3-.AC1>
00575 N=N+ 1
00580 FOR I=l TO Ql
00585 .ACI>=ACI+l)
00590 NEXT I
00595 L2=L2+1
00600 IF Ll=O THEN 00615
00605 IF Ll<O THEN 00920
00610 Ll=Ll-1
00615 GO TO 00620
00620 IF Ll>O THEN 00640
00625 Xl=l
00630 D3=D9
00635 GO TO 00665
00640 REM
00645 D3=G3+BCL2)
00650 GO TO 00665
00655 M=M+l
00660 D3=D9
00665 REM
00670 REM CHECK IF TIME 10 GENERATE NEW AHRIVAL
00675 IF Cl>=C3 THEN 00460
218 DATA SUPPLEMENTARY TO CHAPTER 9
00680 REM
00685 REM CHECK COMPLETION OF. RUN AND OUTPUT RESULTS
00690 IF Cl<D2*3600 THEN 00770
00695 Q2=CN+M>ID2
00700 PRINT " SIGNALIZED INTERSECTION SIMULATION"
00705 PRINT " INPUT VOLUME="V;"VEHICLES PER HOUR"
00710 PRINT "SIG.CYCLE";Rl;"SEC• RED",G2J"SECo GREEN+AMBER"
00715 PRINT "MIN. HDVJY FOR MOVING PASSAGE=";H2J"SECONDS"
00720 PRINT " RESULTS"
00725 PRINT
00730 PRINT "FLOVJ LEAVING INTERSECTION="JQ2;"VEHICLES PER HOUR"
00735 PRINT "MAXIMUM QUEUE LE.NGTH=";Ql
00740 PRINT "TOTAL DELAY"JSU"VEHICLE SECONDS FOR RUN OF";D2.i"HOURS"
00745 PRINT "TOTAL GREEN INTE.RVALS="JU3
00750 PRINT "NUMBER OF. GREEN INTERVALS NOT FULLY UTILIZE.D=";U2
00755 GO TO 00925
00760 HEM
00765 REM INCREMENT CLOCK. CHECK SIGNAL STATUS; CHANGE IF READY.
00770 Cl=Cl+Dl
00775 IF C2>Cl THEN 00855
00780 IF Gl=O THEN 00815
00785 REM CHANGE TO RED
00790 Gl=O
00795 Xl=O
00800 C2=C2+Rl
00805 GO TO 00860
00810 REM CHANGE TO GREEN
00815 Gl=l
00820 Ul=l
00825 U3=U3+1
00830 L2= l
00835 G3=C2
00840 C2=C2+G2
00845 D4=G3
00850 D3=G3+BC1>
00855 hE.M CHECK IF TIME TO PROCESS NEW ARRIVAL
00860 IF Cl<C3 THEN 00535
00865 IF Xl=O THEN 00505
00870 IF Ll>O THEN 00905
00875 IF C3<=CD4+H2) THEN 00905
00880 IF Ul=O THEN 00895
00885 Ul=O
00890 U2=U2+1
00895 D3=C3+T3
00900 GO TO 00535
00905 Xl=O
00910 GO TO 00505
00915 REM
00920 PRINT "QUEUE INDEX E.hhOR"
00925 STOP
00930 PHINT "VOLUME 100 GREAT"
00935 STOP
00940 END
Appendix E
ILLUSTRATION CREDITS AND
COPYRIGHT PERMISSIONS
Grateful acknowledgment is made to the re- generally for copyrighted material, are for this
spective publishers for permission to use the use only, and in no way are to be construed as
following items as the basis of certain of the il- extending to further use or reproduction of the
lustrations and other material prepared for material covered without express written per-
use in this publication. The permissions are mission of the copyright holder.
219
...