Introduction to Transportation Engineering Applications of Queueing Theory to Intersection Analysis Level of Service

Dusan Teodorovic and Antonio A. Trani Civil and Environmental Engineering Virginia Polytechnic Institute and State University

Blacksburg, Virginia Spring 2005

Virginia Tech

Material Covered
Application of deterministic queueing models to study intersection level of service Study various types of intersection controls schemes used in transportation engineering Most of the material applies to ground transportation modes (highways)

Virginia Tech

Basic Ideas
Trafc control represents a surveillance of the motion of vehicles and pedestrians in order to secure maximum efciency and safety of conicting trafc movements. Trafc lights or trafc signals are the basic devices used in trafc control of vehicles on roads. They are located at road intersections and/or pedestrian crossings. The rst trafc light was installed even before there was automobile trafc (London on December 10, 1868). The current trafc lights were invented in USA (Salt Lake City, (1912), Cleveland (1914), New York and Detroit (1920)).
Virginia Tech

Basic Denitions
Drivers move toward the intersection from different approaches Every intersection is composed of a number of approaches and the crossing area (see Figure) Each approach can have one, or more lanes. The trafc stream is composed of all drivers who cross the intersection from the same approach During green time, vehicles from the observed approach can leave the stop line and cross the intersection The corresponding average ow rate of vehicles that cross the stop line is known as a saturation ow.

Virginia Tech

Intersection Geometry
Approach

Crossing area

Figure 1. Typical Road Intersection.

Virginia Tech
5

Flow Conditions at an Intersection

In most cases, queues of vehicles are established exclusively during the red phases, and are terminated during the green phases. Such trafc conditions are known as a undersaturated trafc conditions. An intersection is considered an unsaturated intersection when all of the approaches are undersaturated. Trafc conditions in which queue of vehicles can arrive at the upstream intersection are known as a oversaturated trafc conditions.

Virginia Tech

Trafc Control Techniques

Trafc engineers apply various trafc control strategies (see the Figure) in order to minimize the total delay at the intersection and/or maximize the intersection capacity.
Phase 1 Phase 2 Phase 1 Phase 2

Cycle

Cycle

Time

Figure 2. Intersection with Two Phases.

Virginia Tech
7

Trafc Signal Control Strategies

Many isolated intersections operate under the fixed-time control strategies. These strategies assume existence of the signal cycle that represents one execution of the basic sequence of signal combinations at an intersection. A phase represents part of the signal cycle, during which one set of traffic streams has right of way. Figure 2 shows two-phase traffic operations for the intersection. The cycle contains only two phases. Phase 1 is related to the movement of the north-southbound vehicles through the intersection. Phase 2 represents the movement of the east-westbound vehicles.
Virginia Tech

Trafc Control Strategies

The cycle length c represents the duration of the cycle measured in seconds. The sum of the phase lengths represents the cycle length. For example, in the case shown in Figure 2., the cycle length could be 90 seconds, length of the Phase 1 could be 50 seconds, while the length of the Phase 2 could be equal to 40 seconds. The cycle length is a design parameter of the intersection as well as the green times allocated to each phase. Trafc engineers can modify the settings of intersection controllers based on demand needs at the intersection.

Virginia Tech

Control Strategies
Phase 1 Phase 2 Phase 3

Cycle

Figure 3. Intersection with Three Phases.

Virginia Tech

10

Control Strategies
Higher number of phases is usually caused by trafc engineers wish to protect some movements (usually leftturning vehicles) Protection assumes avoiding potential conicts with the opposing trafc movement, and/or pedestrians There is always a certain amount of lost time (few seconds) during phase change. For example, when the green light changes to red there is am amber light period to warn drivers of an impending change Obviously, the higher the number of phases, the better the protection, and the higher the value of the lost time associated with a phase change.
Virginia Tech

11

Control Strategies
Trafc signals are control devices. The typical sequence of lights at the intersection approach could be: Red, Red All, Green, Amber, Red, Red All,....
Flow [veh/h] Saturation ow

Time 0 Red Green Red All Effective green Amber

Figure 4. Denition of Green, Amber and Red Times.

Virginia Tech
12

Control Strategies
Green time, effective green,red time, and effective red are linguistic expressions frequently used by trafc engineers In theory, all drivers should cross the intersection during the green light. In reality, no one driver starts his/her car exactly in a moment of the green light appearance Similarly, at the end of a green light, some drivers speed up, and cross the intersection during the amber light Green Time represents the time interval within the cycle when observed approach has green indication. On the other hand, Effective Green represents the time interval during which observed vehicles are crossing the intersection.
Virginia Tech
13

Vehicle delays at signalized intersections: Uniform Vehicle Arrivals

For simplicity, let us assume for the moment that observed signalized intersection could be treated as a D/ D/1 (deterministic) queueing system with one server (hence the notation (D/D/1)) We assume uniform arrivals, and uniform departure rate (see Figure 5).

Virginia Tech

14

Queueing Theory Short-hand Nomenclature

Queues come in different avors as demonstrated so far Kendall developed a simple scheme to designate queues back in the early 50s. His nomenclature has been widely adopted Typically 6 parameters:

a/b/c/d/e/f a = inter-arrival time distribution (arrivals) b = service time distribution c = number of servers
Virginia Tech (A.A. Trani)

14a

Queueing Theory Short-hand Nomenclature

Typically 6 parameters:

a/b/c/d/e/f d = service order (i.e., FIFO, LIFO, etc) e = Max. number of customers f =Size of the arrival population

Virginia Tech (A.A. Trani)

14b

Queueing Theory Short-hand Nomenclature

Possible outcomes for (a) and (b) M = Times are neg. exponential (i.e., Poisson arrivals) D = Deterministic distribution Ek = Erlang distribution G = general distribution

Virginia Tech (A.A. Trani)

14c

 M/M/1/FIFO// Stochastic queue with neg. exponential time

between arrivals

Example Queueing Systems we Have Studied

Neg. exponential service times 1 server First in-rst out Innite no. of customers in system Innite arrival population
Virginia Tech (A.A. Trani)

14d

Example Queueing Systems we Have Studied

M/M/2/FIFO/15/15 Stochastic queue with neg. exponential time

between arrivals

Neg. exponential service times 2 servers (2 pavers) First in-rst out Up to 15 no. of trucks in system 15 trucks population
Virginia Tech (A.A. Trani)

14e

Denition of Queueing Terms for Intersection Analysis

Cumulative number of vehicles
Cumulative arrivals Cumulative departures

D(t)
C

A(t) g0
A

g
c Time

Red

Green

Figure 5. Arrivals and Departures at an Intersection.

Virginia Tech
15

Deterministic Queueing Analysis

Let us denote by vehicles arrival rate, and by vehicles departure rate during the green time period. In the deterministic case, the cumulative number of arrivals A ( t ) and the cumulative number of departures D ( t ) are:
A(t) = t D(t) = t
(1) (2)

where:
c r g

- the duration of the signal cycle - effective red - effective green

Virginia Tech
16

Deterministic Queueing Analysis

The duration of the signal cycle equals:
c = r+g
(3)

The formed queue is the longest at the beginning of effective green. The queue decreases at the beginning of effective green. We denote by g the time necessary for queue to dissipate (Figure 5). The queue must dissipate before the end of effective green. In the opposite case, the queue would escalate indenitely. In other words, queue dissipation will happen in every cycle if the following relation is satised:
0

Virginia Tech

17

Deterministic Queueing Analysis

g0 g
(4)

The relation (4) will be satised if the total number of vehicle arrivals during cycle length c is less than or equal to the total number of vehicle departures during effective green g , i.e.:
c g

dt dt
0 0

(5)

t ct 0 cg

g 0

(6) (7)

Virginia Tech

18

Deterministic Queueing Analysis

Finally, we get:
-- g - - c
(8)

Let us note the triangle ABC (Figure 5). Vehicles arrive during time period ( r + g ) . Vehicles depart during time period g . The total number of vehicle arrivals equals the total number of vehicle departures, i.e.:
0 0

( r + g0 ) = g0 ( ) g0 = r

(9) (10)

Virginia Tech

19

Deterministic Queueing Analysis

The time period g required for queue to dissipate equals:
0

r g 0 = ----------

(11)

We divide both numerator and denominator by . We get:

-- r g 0 = ---------- 1 -
(12)

Dene the utilization factor ( ) (or trafc intensity) of the intersection as = --- , we can write:
r g 0 = ----------1 Virginia Tech
(13)

20

Deterministic Queueing Analysis

The area A of the triangle ABC represents the total delay d of all vehicles arrived during the cycle. This area equals:
ABC

1 A ABC = -- r h 2

(14)

where h is the height of the triangle (ABC). The ratio

h = ----------------( r + g0 ) h ----------------( r + g0 )

represents the slope , i.e.:

(15)

Virginia Tech

21

Deterministic Queueing Analysis

The height h of the triangle ABC is:
h = ( r + g0 )

The area of the triangle ABC equals:

1 r 1 A ABC = -- r h = -- r ( r + g 0 ) = --------- ( r + g 0 ) 2 2 2
(16)

The total delay d of all vehicles arriving during the cycle equals:
r r r2 r d = --------- ( r + g 0 ) = --------- r + ----------- = ----------- 1 + ----------- 2 2 2 1 1
(17)

Virginia Tech

22

Deterministic Queueing Analysis

r2 d = ----------------------2 (1 )
(18)

The average delay per vehicle d represents the ratio between the total delay d and the total number of vehicles per cycle. The total number of vehicles per cycle equals c . Therefore the average delay per vehicle d is:
dd = --------c
(19)

or
r2 ----------------------2 (1 ) d = -----------------------c Virginia Tech
(20)

23

Simplifying the previous expression, average delay per vehicle is the average:
r2 d = ------------------------------2 c (1 )
(21)

Virginia Tech

24

Example Problem 1
The cycle length at the signalized intersection is 90 seconds. The considered approach has the saturation ow of 2200 [veh/hr], the green time duration of 27 seconds, and ow rate of 600 [veh/hr]. Analyze trafc conditions in the vicinity of the intersection. Calculate average delay per vehicle. Assume that the D/D/1 queueing system adequately describes considered intersection approach.

Virginia Tech

25

Problem 1 - Solution
The corresponding values of the cycle length and the green time are:
c = 90 [ s ]

;g

= 27 [ s ]

The ow rate ( ) and the service rate ( ) are:

veh 600 veh veh = 600 -------- = ----------- -------- = 0.167 -------hr 3600 s s veh 2200 veh veh = 2200 -------- = ----------- -------- = 0.611 -------hr 3600 s s

Trafc intensity equals:

Virginia Tech

26

Problem 1 - Solution
veh 0.167 -------s = -- = --------------------------- = 0.273 veh 0.611 -------s

The duration of the red light for the considered approach is:
r = c g = 90 27 = 63 [ s ]

The number of arriving vehicles per cycle is:

veh c = 0.167 -------- 90 [ s ] = 15.03 [ veh ] s

Virginia Tech

27

Problem 1 - Solution
The number of departing vehicles during green light is:
veh g = 0.611 -------- 27 [ s ] = 16.497 [ veh ] s

We conclude that the following relation is satised:

cg

This means that the trafc conditions in the vicinity of the intersection are undersaturated trafc conditions. The average delay per vehicle is estimated using:
r2 d = ------------------------------2 c (1 ) Virginia Tech
28

Problem 1 - Solution
63 2 d = -------------------------------------------- = 30.33 [ s ] 2 90 ( 1 0.273 )

Virginia Tech

29

Example Problem 2
The cycle length at the signalized intersection is 60 seconds. The considered approach has the saturation ow of 2200 [veh/hr], the green time duration of 15 seconds, and ow rate of 400 [veh/hr]. Analyze trafc conditions in the vicinity of the intersection. Assume that the D/D/1 queueing system adequately describes the intersection approach considered. Calculate: (a) the average delay per vehicle; (b) the longest queue length; (c) percentage of stopped vehicles.

Virginia Tech

30

Problem 2 - Solution:
(a) The corresponding values of the cycle length and the green time are:
c = 60 [ s ]

;g

= 20 [ s ]

The red time is:

r = c g = 60 20 = 40 [ s ]

The ow rate and the service rate are:

veh 400 veh veh = 400 -------- = ----------- -------- = 0.111 -------hr 3600 s s

Virginia Tech

31

Problem 2 - Solution
veh 2200 veh veh = 2200 -------- = ----------- -------- = 0.611 -------hr 3600 s s

The utilization factor for the queue is:

veh 0.111 -------s = -- = --------------------------- = 0.182 veh 0.611 -------s

The average delay per vehicle equals:

r2 d = ------------------------------2 c (1 )

Virginia Tech

32

Problem 2 - Solution
40 2 d = -------------------------------------------- = 16.3 [ s ] 2 60 ( 1 0.182 )

(b) The longest queue length L happens at the end of a red light (Figure 5). The quantity L is calculated as follows:
max max

veh L max = r = 0.111 -------- 40 [ s ] = 4.44 [ vehicles ] s

(c) Vehicles arrive all the time during the cycle. The total number vehicles arrived A during the cycle equals:
veh A = c = 0.111 -------- 60 [ s ] = 6.66 [ vehicles ] s

Virginia Tech

33

Problem 2 - Solution
All vehicles that arrive during time interval ( r + g ) are stopped. The total number of stopped vehicles S equal:
0

S = ( r + g0 )

The time period g required for queue to dissipate is estimated using equation:
0

r g 0 = ----------

We get:
r 0.111 40 S = ( r + g 0 ) = r + ----------- = 0.111 40 + -------------------------------- 0.611 0.111

Virginia Tech

34

S = 5.43 [ vehicles ]

The percentage of stopped vehicles equal: S 5.43 P = -- 100 = --------- 100 = 81.53 [%]
A 6.66

Virginia Tech

35

Example Problem 3
A simple T intersection is signalized. There are two approaches indicated in the gure. The cycle length at the signalized intersection (Figure) is 50 seconds.
Phase 1 Phase 2 Approach 1

Approach 2 Cycle

Virginia Tech

36

Example Problem 3
Approach 1 has the saturation ow of 2200 [veh/hr], the effective green time duration of 35 seconds, and the ow rate of 600 [veh/hr]. Approach 2 has the saturation ow of 2000 [veh/hr], the effective green time duration of 15 seconds, and the ow rate of 550 [veh/hr]. Assume that the D/D/1 queueing system adequately describes considered intersection approach. Calculate: (a) the average delay per vehicle for every approach; (b) Allocate effective red and green time among approaches in such a way to minimize the total delay of the T intersection.

Virginia Tech

37

Problem 3 -Solution
(a) Approach 1: The corresponding values of the cycle length and the green time are:
c = 50 [ s ]

;g

= 35 [ s ]

The red time equals:

r 1 = c g 1 = 50 35 = 15 [ s ]

The ow rate and the service rate are respectively equal:

veh 600 veh veh 1 = 600 -------- = ----------- -------- = 0.167 -------hr 3600 s s

Virginia Tech

38

Problem 3 -Solution
veh 2200 veh veh 1 = 2200 -------- = ----------- -------- = 0.611 -------hr 3600 s s

The utilization factor for the queue in approach 1 is:

1

veh 0.167 -------s 1 1 = ---- = --------------------------- = 0.273 1 veh 0.611 -------s

The average delay per vehicle equals:

r12 15 2 d 1 = -------------------------------- = -------------------------------------------- = 3.09 [ s ] 2 c ( 1 1 ) 2 50 ( 1 0.273 )

Virginia Tech

39

Problem 3 -Solution
Approach 2: The corresponding values of the cycle length and the green time are:
c = 50 [ s ]

;g

= 15 [ s ]

The red time is:

r 2 = c g 2 = 50 15 = 35 [ s ]

The ow rate and the service rate are:

veh 550 veh veh 2 = 550 -------- = ----------- -------- = 0.153 -------hr 3600 s s

Virginia Tech

40

Problem 3 -Solution
veh 2000 veh veh 2 = 2000 -------- = ----------- -------- = 0.555 -------hr 3600 s s

The utilization factor for the queue equals:

2

veh 0.153 -------s 2 2 = ---- = --------------------------- = 0.276 2 veh 0.555 -------s

The average delay per vehicle equals:

r22 35 2 d 2 = -------------------------------- = -------------------------------------------- = 16.92 [ s ] 2 c ( 1 2 ) 2 50 ( 1 0.276 )

Virginia Tech

41

Problem 3 -Solution
(b) The total delay per cycle of all vehicles on both approaches is the sum of the delays of every approach:
TD = 1 d 1 + 2 d 2

substituting the denitions of d and d (see equation 21)

1 2

r12 r22 TD = 1 -------------------------------- + 2 -------------------------------2 c ( 1 1 ) 2 c ( 1 2 )

Since:
r1 + r2 = c

after substitution, we get:

Virginia Tech

42

r12 ( c r1 )2 TD = 1 -------------------------------- + 2 -------------------------------2 c ( 1 1 ) 2 c ( 1 2 )

The total delay is minimal when:

d[ TD ] --------------- = 0 dr 1

After substitution, we get:

r12 ( c r1 )2 d 1 -------------------------------- + 2 -------------------------------2 c ( 1 1 ) 2 c ( 1 2 ) ---------------------------------------------------------------------------------------------------- = 0 dr 1 r1 ( c r1 ) 1 ------------------------- 2 ------------------------- = 0 c ( 1 1 ) c ( 1 2 )

Virginia Tech

43

Problem 3 -Solution
( 50 r 1 ) r1 0.167 ------------------------------------- 0.153 ------------------------------------- = 0 50 ( 1 0.273 ) 50 ( 1 0.276 )

After solving the equation, we get:

r 1 = 24 [ s ] g 1 = c r 1 = 50 24 = 26 [ s ] r 2 = 26 [ s ] g 2 = 24 [ s ]

These are the optimal values of green and red times to minimize the intersection delay.
Virginia Tech

44

Problem 3 -Solution
We can recalculate the average delays per vehicle: Approach 1
r12 24 2 d 1 = -------------------------------- = -------------------------------------------- = 7.92 [ s ] 2 c ( 1 1 ) 2 50 ( 1 0.273 )

Approach 2:
r22 26 2 d 2 = -------------------------------- = -------------------------------------------- = 9.34 [ s ] 2 c ( 1 2 ) 2 50 ( 1 0.276 )

These delays compare favorably with those obtained before (3.09 and 16.92 seconds, respectively for approaches 1 and 2).

Virginia Tech

45

Vehicle Delays at Signalized Intersections: Random Vehicle Arrivals

Trafc ows are characterized by random uctuations The delay that a specic vehicle experiences depends on the probability density function of the interarrival times, as well as on signal timings and the time of a day when the vehicle shows up Obviously, individual vehicles experience at a signalized approach various delay values.

Virginia Tech

46

Intersection with Random Arrivals

Cumulative arrivals

Cumulative number vehicles of

Overow Delay Uniform delay

Time

Red Green

Figure 6. Intersection with Random Arrivals.

Virginia Tech

47

Intersection with Random Arrivals

Let us calculate the delay D for the vehicle arriving at time (Figure 6). The overall delay D is composed of the uniform delay d and the overow delay d , i.e.:
R

D = d + dR

(22)

The uniform delay d represents delay that would be exp rienced by a vehicle when all vehicle arrive uniformly and when trafc conditions are unsaturated (see Equations in previous sections). Due to the random nature of vehicle arrivals, the arrival rate during some time periods can go over the capacity, causing overow queues.

Virginia Tech

48

Considering Random Arrivals

The overow delay d represents the delay that is caused by short-term overow queues. This delay can be easily calculated using queueing theory techniques.
R

Queueing System

Crossing area

Virginia Tech

49

Intersection with Random Arrivals

We assume that vehicle interarrival times are exponentially distributed. The service rate is deterministic (we denote by departure rate from the articial queue into the signal), and there is only one server. This means that the articial Queueing System is M/D/1 queueing system (single server with Poisson arrivals and deterministic service times). The average delay per customer in the M/D/1 queueing system equals:
2 d R = -------------------------------2 (1 )
(23)

Virginia Tech

50

Intersection with Random Arrivals

where:

- utilization ratio in the M/D/1 queueing system

The utilization ratio in the M/D/1 queueing system equals:

= -
(24)

The departure rate from the articial queue into the signal can be expressed in terms of departure rates from the trafc signal . The departure rate equals during green time. During red time, departure rate equals zero (see Figure 6).
Virginia Tech
51

Intersection with Random Arrivals

Service rate [veh/h] g

0 Red Cycle Green

Time

Figure 7. Service Rate Denition at a Trafc Intersection.

Virginia Tech

52

Intersection with Random Arrivals

Departure rate during
= 0r+g -------------------------c g = -c

whole cycle :
(25) (26)

The utilization ratio in the M/D/1 queueing system :

= = ---------- g -c
(27)

i.e.:

Virginia Tech

53

c = ---------g

(28)

The quantity is known as a volume to capacity ratio. The average vehicle delay is:
D = d + dR r2 2 D = ------------------------------- + -------------------------------2 c (1 ) 2 (1 )
(29) (30)

It has been shown by simulation that Equation (30) overestimate the average vehicle delay. The following two formulas for average vehicle delay calculation were proposed as a corrections of the equation (30):

Virginia Tech

54

Websters formula:
2 + --------cr ---- c D = ------------------------------- + -------------------------------- 0.65 2 2 c (1 ) 2 (1 ) 2 2 1 -3 5 g

(31)

Allsops formula:
9 2 r2 D = ----- ------------------------------- + -------------------------------10 2 c ( 1 ) 2 ( 1 )
(32)

Virginia Tech

55

Example Problem 4
Using data given in the Example Problem 1, calculate: (a) average delay per vehicle using Allsops formula. (b) Calculate duration of the green time necessary to achieve average delay per vehicle of 40 seconds. Solution: (a) The cycle length, green time, arrival rate, departure rate, trafc intensity, volume to capacity ratio, and red time duration are:
c = 90 [ s ] g = 27 [ s ]

Virginia Tech

56

veh 600 veh veh = 600 -------- = ----------- -------- = 0.167 -------hr 3600 s s veh 2200 veh veh = 2200 -------- = ----------- -------- = 0.611 -------hr 3600 s s veh 0.167 -------s = -- = --------------------------- = 0.273 veh 0.611 -------s veh 0.167 -------s --------------------------veh 0.611 --------s = -- = ---------------------------- = 0.273 = 0.91 -----------g 27 [ s ] 0.3 ------------c 90 [ s ]

Virginia Tech

57

Solution - Problem 4
r = c g = 90 27 = 63 [ s ]

The average delay per vehicle based on Allsops formula equals:

9 2 r2 D = ----- ------------------------------- + -------------------------------10 2 c ( 1 ) 2 ( 1 ) 90.91 2 63 2 D = ----- -------------------------------------------- + ------------------------------------------------10 2 90 ( 1 0.273 ) 2 0.167 ( 1 0.91 ) D = 52.083 [ s ]

(b) The average delay per vehicle is:

Virginia Tech

58

9 2 r2 D = ----- ------------------------------- + -------------------------------10 2 c ( 1 ) 2 ( 1 ) 10 r2 2 ------------------------------- = ----- D -------------------------------9 2 c (1 ) 2 (1 ) r = r = 10 2 [ 2 c ( 1 ) ] ----- D -------------------------------9 2 (1 ) 10 0.91 2 [ 2 90 ( 1 0.273 ) ] ----- 40 ------------------------------------------------9 2 0.167 ( 1 0.91 )

r = 47 [ s ] g = c r = 90 47 g = 43 [ s ]

Green time to achieve a delay of 40 seconds per vehicle.

Virginia Tech
59