Delay at Signalized Intersection

Shivam Gupta
CE 683 Traffic Engineering

Indian Institute of Technology, Kanpur

Abstract
An intersection is a shared space that is used by more than one
approach at a time. A signalized intersection is one where the shared
space is used alternatively by a fixed number of approaches for a pre
defined time interval as per the phasing scheme used for the intersec-
tion. Delay and queuing process are the main characteristics of such
an intersection which help in analyzing and designing the intersection
for a particular level of service.

1 Introduction
Vehicle delay is the most important parameter used by transportation pro-
fessionals in evaluating the performance of a signalized intersection. This is
perhaps because it directly relates to the time loss that a vehicle experiences
while crossing an intersection (though we have not considered other prob-
lems like congestion due to queuing, extra fuel loss due to vehicle ignition
etc.). However delay is a parameter that is not easily determined due to
the non deterministic nature of the arrival and departure processes at the
intersection. But lot of research has been done in this field to define delay by
a number of analytical delay models, including deterministic queuing, steady
state stochastic and time dependent stochastic models. There are assump-
tions in these models that help in simplifying the complex flow conditions to
a quantifiable model which gives an approximate measure of average delay
faced by a vehicle crossing an intersection. Some research has been done that
tries to predict the variance of overall delay that individual vehicles may ex-
perience at signalized intersection due to large variation and randomness of
traffic arrivals and interruption caused by traffic signal controls.

1
2 Delay at Signalized Intersection
Delay at signalized intersection is computed as the difference in the departure
time and the arrival time of a vehicle. It can also be said equal to the total
extra time spent by a vehicle at the intersection than what is required if
the vehicle were allowed to pass the intersection without any hindrance. The
total delay time can be categorized into deceleration delay, stopped delay and
acceleration delay. The deceleration delay is the time loss that the vehicle
takes in slowing down to reach to a stoppage, in case the signal is red or to
a speed, in case there is a queue which is moving when the signal is green.
The stopped delay is the delay that the vehicle spends at the intersection
while it is standing in a queue waiting for the signal to turn green. While
most of the delay incurred at signalized intersections is directly caused by
the traffic signal operation, a fraction of the total delay is due to the time
required by individual drivers to react to changes in the signal display at the
beginning of the green interval, to mechanical constraints, and to individual
driver behavior. When headway of the departing vehicles are observed it is
seen that vehicles in front of queue take more time to cross a stationary point
than the following vehicles and reach saturation headway after some point.
To account for the additional delays due to driver reaction time and vehicle
acceleration constraints, the operation of a signalized intersection is usually
defined in terms of effective signal intervals instead of actual intervals in
delay estimation models. Instead of explicitly considering green, yellow and
amber intervals, delay calculations are typically performed by dividing the
signal cycle into effective periods of stopped and moving traffic within which
constant traffic characteristics can be assumed. The amount of difference
between the actual and effective timings will thus depend on the assumptions
regarding driver reaction time at the beginning of the green interval and
vehicle accelerations.
A final element that may affect the delays incurred at intersection ap-
proaches is the randomness in vehicle arrivals. If vehicles were to arrive at
uniform intervals, the delays incurred by vehicles within successive signal
cycles would be identical, as there would then be an exact replication of
the arrival and departure patterns. In this case the flow at one intersection
is independent of any other intersection upstream or downstream. In case
when the vehicles are arriving in a platoon, intersections may be designed in
such a way that the platoon keeps on moving from one intersection to an-
other without getting dissipated. Finally, under random arrival patterns, the
number of arrivals may fluctuate from one cycle to the other, thus resulting
in different queue lengths. This may in turn result in arrival demands that
occasionally exceed the approach capacity, and therefore, in higher delays.

2
3 Delay Models
3.1 Deterministic Queuing Model
Considering a single lane which is controlled by a traffic signal giving a frac-
tion of time available for movement and assuming that the arrival and de-
parture process are deterministic, this model tries to predict average delay
experienced by a vehicle within a signal cycle. First considering unsaturated
flow conditions where the vehicles arriving in a cycle are cleared before the
start of another cycle. The model is generated by first assuming that the
vehicles arrive at a uniform and constant flow rate. This ignores the ran-
domness in arrival and since we are considering only unsaturated conditions
it also overlooks on over-saturated flow conditions. A second assumption is
that that vehicles decelerate and accelerate instantaneously. This assump-
tion converts all deceleration and acceleration delays into equivalent stopped
delay, and thus allows a direct estimation of the total delay incurred by vehi-
cles attempting to cross an intersection. A final assumption is that vehicles
queue vertically at the intersection stop line. While this assumption does not
represent a normal queuing behavior and may not accurately represent the
exact number of queued vehicle at a given instant, it does not bias the delay
estimation process over an entire queue formation and dissipation process and
is therefore a valid simplification when only considering delay estimations.

C(1 − λ)2
du = (1)
2(1 − Xλ)
where du is average delay due to unsaturated flow, λ is fraction of effective
green to cycle length ( gCe ), X is approach volume (v) to capacity (c) ratio or
saturation ratio (hereX ≤ 1).
In over-saturated conditions, the number of vehicles reaching the inter-
section exceeds the number of vehicles that can be served by the traffic signal
(X ≥ 1). This causes a growing residual queue to occur. In this case the
total delay is categorized into two components, the delay due to unsaturated
component and the delay due to over-saturated component.
C − ge T
do = + (X − 1) (2)
2 2
where do is average delay due to over-saturated flow, T is the time for which
the flow is over-saturated.

3
3.2 Steady State Stochastic Model
While deterministic queuing assumes uniform arrival rate, in practice the ar-
rival process is random and more complex. In order to make it more close to
reality, attempts are made to account for randomness of arrivals. First con-
sidering that the number of arrivals in a given time interval follows a known
distribution, typically a Poisson distribution, and that this distribution does
not change over time. Second, they all assume that the headway between
departures from the stop line follow a known distribution with a constant
mean, or are identical. Third, while it is recognized that temporary over-
saturation may occur due to the randomness of arrivals, it is assumed that
the system remains unsaturated over the analysis period. Fourth, the system
is assumed to have been running long enough to allow it to have settled into
a steady state. Fifth, all these models still consider that vehicles decelerate
and accelerate instantaneously and thus, that all drivers behave similarly.
Webster’s model is one such model. The first term estimates the average
approach delay assuming uniform arrivals. The second term considers the
additional delays attributed to the randomness of vehicle arrivals. The third
term is an empirical correction factors that reduces the estimated delay by
5-15%, to be consistent with simulation results.

C(1 − λ)2 X2
1/3
c

d= + − 0.65 2 X 2+λ (3)
2(1 − Xλ) 2v(1 − X) v

3.3 Time Dependent stochastic Model

A main consequence of the stochastic delay modelling described in the pre-
vious section is that the estimated delays tend to infinity as traffic demand
approaches saturation (X ≥ 1). In order to explain more rigorously the delay
for all flow conditions, a general time dependent delay model is conceived pri-
marily based on empirical calculations. Considering the Highway Capacity
Manual (2000) model, the average delay can be given by

d = d1 · fpf + d2 (4)

(1 − λ)2
d1 = 0.5C
1 − λ · x1
 s 
8kX 
d2 = 900T (X − 1) + (X − 1)2 +
cT

4
 where fpf is delay adjustment factor for quality of progression and con-
trol type (= 1.0 for pre-timed non coordinated signals), c is the capacity of
lane (in vph) and k is an incremental delay factor dependent upon signal
controller setting (0.50 for pre timed signals), T is the time of evaluations
(in hr) and x1 is min(1.0, X). This model assumes steady-state traffic con-
ditions. It estimates delay under stochastic equilibrium conditions, when the
arrival and departure flow rates have been stationary for an indefinite period
of time. It also assumes that the number of arrivals in a given interval fol-
low a Poisson distribution and that the headway between departures have a
known distribution with a constant mean value. Similar to the earlier models,
it also assumes instantaneous acceleration and deceleration to simply data
estimation

4 Variability in Delay
The models that have been described give average delay. Since delay is an
important parameter that is used as a decision variable in designing the
intersection it is desired that a distribution is predicted that would approx-
imate the delay and match the average delay proposed by these analytical
models. Consider the cumulative arrival and departure of vehicles during
the time interval [0,T] The delay for a particular vehicle arriving at time t,
called overall delay and noted as D, is considered to include two components:
uniform delay (D1 ), when the approach is un-saturated and vehicles arrive
uniformly and overflow delay (D2 ), due to temporary overflow queues result-
ing from random nature of arrivals. Similarly the total variance of delay can
be divided into two components, var[D1 ] and var[D2 ].
V ar[D1 ], represents the variation of uniform delay that would be experi-
enced by vehicles arriving during time interval [0,T]. This variation results
from the uncertainty of the vehicle’s arrival time during each cycle of the
interval. The vehicle can arrive at any moment within a cycle and thus ex-
perience variable delays as a result of the signal control. An estimate of this
variance component can be obtained theoretically on the basis of a determin-
istic queuing model with vehicles arriving uniformly during the cycle

C 2 (1 − λ)3 (1 + 3λ − 4λx1 )
var[D1 ] = (5)
12(1 − λx1 )2
In order to establish a model for the variance of delay caused by an over-
flow queue, two extreme traffic conditions are first investigated: unsaturated
conditions (X ≤ 1.0) and over-saturated conditions (X > 1.0). For unsatu-
rated conditions, overflow delay experienced by a vehicle arriving during the

5
time interval [0,T] is mainly caused by occasional overflows of traffic from
each cycle. The relationship between the variance of this delay and the degree
of saturation can be approximated by

X(4 − X)
var[D2 ] = (6)
12c2 (1 − X)2
For X = 1, the variance becomes infinite which is incorrect. This ex-
pression gives good approximate for the variance for values of X < 1. If
the intersection approach is highly over-saturated during time period [0, T],
there is a high probability that an overflow queue always exists during the
period from time 0 to time T. Consider a vehicle arriving at time t during
time period [0,T]. The overflow queue for a vehicle arriving at time t, Qt ,
can be determined as the total arrivals minus the total departures and then
the delay D2 will then be given by the time taken for the queue to dissipate
with capacity flow rate.

Qt = Nt − c · t
Nt − c · t
D2 =
c
where Nt is a random variable with mean vt. The variance of delay D2 for
vehicles arriving during time interval [0, T] can be obtained by assuming that
the arrival time t is a random variable with known distribution. The term
D2 becomes random and which is dependent upon time. Therefore using the
total law of variance and law of conditional probability, variance of D2 can
be written as (for detailed description please refer Appendix (B)

E[var[Nt − ct|t]] + var[E[Nt − ct|t]]

var[D2 ] =
c2
E[var[Nt |t] − 0] + var[E[Nt |t] − ct]
=
c2
E[var[Nt |t]] + var[vt − ct]
=
c2
E[var[Nt |t]] + (v − c)2 var[t]
=
c2

Assuming that the arrival time is uniformly distributed during the time
interval [0,T] with mean equal to T /2 and variance equal to T 2 /12, and since
the arrival follows Poisson distribution (for which mean is equal to variance),
using var[Nt |t] = E[Nt |t]:

6
 T X T 2 (1 − X)2
var[D2 ] = + (7)
2c 12
It must be emphasized that Eq.(7) is valid only when an overflow queue
is present during the period from time 0 to time t. In reality, however, it is
possible that no overflow queue exists at time t, and consequently no overflow
delay is experienced. Therefore, it can be concluded that Eq.(7) represents an
upper-bound estimate of the variance of overflow delay. The actual variance
would be lower, but the prediction error should become smaller as the de-
gree of saturation increases and the associated likelihood of overflow queuing
increases. This can be done by multiplying Eq.(7) with a correction factor
x0 β
e−( X ) where values of parameter x and β are calibrated using simulated
0
data of variances.

References
[1] L. Fu, and B. Hellinga, Delay Variability at Signalized Intersections. In
Transportation Research Record: Journal of the Transportation Research
Board, No. 1710, TRB, National Research Council, Washington, D.C.,
2000, pp. 215-221.

[2] F. Dion, H. Rakha, and Y. S. Kang, Comparison of delay estimates

at under-saturated and over-saturated pre-timed signalized intersections,
Transportation Research Part B (38) (2004), pp. 99-122.

7
Appendices
A Law of Total Expectation
Assuming that the events X and Y are random events in the same probability
space, then the expected value of conditional expected value of X given Y is
the same as the expected value of X. This is proved as:
X
E [E[X|Y ]] = E[X|Y = y] · P (Y = y)
y
!
X X
= x · P (X = x|Y = y) · P (Y = y)
y x
XX
= x · P (X = x|Y = y) · P (Y = y)
y x
XX
= x · P (Y = y|X = x) · P (X = x)
x y
!
X X
= x · P (X = x) · P (Y = y|X = x)
x y
X
= x · P (X = x)
x
= E[x].

B Law of Total Variance

Assuming that X and Y are random variables in the same probability space
and using the law of total expectation given in Appendix A, the total variance
of X can be written as:

var[X] = E[X 2 ] − E[X]2

h i
= E E[X 2 |Y ] − E [E[X|Y ]]2
h i
= E var[X|Y ] + E[X|Y ]2 − E [E[X|Y ]]2
 h i 
= E [var[X|Y ]] + E E[X|Y ]2 − E [E[X|Y ]]2
= E [var[X|Y ]] + var [E[X|Y ]]