CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

MODULE 3

3.1 Intervening Opportunities Model:

The gravity model is by far the most commonly used aggregate trip distribution
model. But the gravity model does not exhaust all the theoretical possibilities. Intervening
opportunities model which although much less used; offer real alternatives to the gravity
model.
The basic idea behind the intervening-opportunities model is that trip making is
not explicitly related to distance but to the relative accessibility of opportunities for
satisfying the objective of the trip. The original proponent of this approach was Stouffer
(1940), who also applied his ideas to migration and the location of services and
residences. But it was Schneider (1959) who developed the theory in the way it is
presented today.
Consider first a zone of origin i and rank all possible destinations in order of
increasing distance from i. Then look at one origin-destination pair (i, j), where j is the
mth destination in order of distance from i. There are m-1 alternative destinations actually
closer (more accessible) from i. A trip maker would certainly consider those destinations
as possible locations to satisfy the need giving rise to the journey: those are the
intervening opportunities influencing a destination choice.
The basic hypothesis of this model given by Stouffer (1940) is that the number of
trips from an origin zone to a destination zone is directly proportional to the number of
opportunities at the destination zone and inversely proportional to the number of
intervening opportunities. This hypothesis may be expressed as:

tij = (k.aj/vj)

Where;
aj = the total number of destination opportunities in zone j
vj = the number of intervening destination opportunities between zones i and j
k =a proportionality constant to ensure that all trips with origins at zone i are
distributed to destination opportunities.

56
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Schneider (1959) has proposed the following modification of the Stouffer

hypothesis. The Schneider hypothesis states that the probability that a trip will terminate
in some volume of destination points is equal to the probability that this volume contains
an acceptable destination times the probability that an acceptable destination closer to the
origin of the trips has not been found. This hypothesis many be expressed as:

pr (dv) = [1 pr (v)]. ldv

Where;
pr (dv) = the opportunity that a trip will terminate when dv destination
opportunities are considered.
pr (v) = the cumulative probability that a trip will terminate by time v possible
destinations are considered = (1 ki exp (-lv))
ki = a constant for zone I to ensure that all the trips produced at zone i are
distributed to zone i
v = the cumulative total of the destinations already considered.
l = a constant probability of a destination being accepted if it is considered.
The trip interchange between an i-j pair is given by:
tij = pi (probability of trip terminating in zone j)

= pi (pr(vj+1) pr(vj))

Where;
pr(vj) = the probability that a trip will have found a suitable destination in the
opportunities already considered up to zone j
pr(vj+1) = the probability that a trip will have found a suitable destination in the
cumulative opportunities considered up to and including zone j.

From figure it may be restated as: the probability of locating within the dv
opportunities being considered is equal to the product of the probability of not having
located within the v opportunities already considered, and the probability of finding an

57
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

acceptable location within dv opportunities, given that a location has not already been
found.

Boundary of Total
Destination Opportunities

v Destination
Opportunities
dv Destination Opportunities in Given
Travel-Time Interval from i

The intervening-opportunities model is interesting because it starts from different

first principles in its derivation: it uses distance as an ordinal variable instead of a
continuous cardinal one as in the gravity model. It explicitly considers the opportunities
available to satisfy a trip purpose at increased distance from the origin. However, the
model is not often used in practice, probably for the following reasons:

The theoretical basis is less well known and possibly more difficult to understand
by practitioners;
The idea of matrices with destinations ranked by distance from the origin (the nth
cell for origin i is not destination n but the nth destinations away from i) is more
difficult to handle in practice;
The theoretical and practical advantages of this function over the gravity model
are not overwhelming;
The lack of suitable software.

58
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

This is based on the concept that a trip cannot always go to the nearest destination and
stop, it must consider the nearest destination and if that is not acceptable consider the
next nearest and so on.

Lmax

P ( I L)
U j V j
Tij = i
U
( I L)
Lmin

Tij = Total number of trios generated from zone I with a particular L value .
(I-L)U = Probability that a given trip will get to a given destination.
U j V j
( I L) = Probability that a given trip will not find a satisfactory destination in j and
go on to next zone.
L= Probability of a particular trip origin stopping at any randomly chosen destination.
V= Number of destination found closer to the point of origin I than zone j.
Vj = Number of destinations in zone j.

3.1.1 Example problem:

Find the trip interchange for the given figure?

P= Production
A=attraction

59
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

P=1250 P=1850
1 A=800 A=1200 2

P=500 P=450
A=1950 A=3900
3 4

Iteration 1:

Zone Productions Attractions

1 1250 800
2 1850 1200
3 450 3900
4 500 1950

Let l=0.00055

T11 = 1250 (e0-e (-0.00055*800) ) = 445

T12 = 1250 (e (-0.00055*800) e (-0.00055*2000) ) = 389
T13 = 1250 (e (-0.00055*2000) e (-0.00055*5900) ) = 367
T14 = 1250 (e (-0.00055*5900) e (-0.00055*7850) ) = 32
Therefore

T1j= 445+389+367+32=1233

T21 = 1850 (e (-1200*0.00055) e (-0.00055*2000) ) = 894

T22 = 1850 (e0-e (-0.00055*1200) ) = 340
T23 = 1850 (e (-0.00055*2000) e (-0.00055*5900) ) = 544

60
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

T24 = 1850(e (-0.00055*5900) e (-0.00055*7850) ) = 47

Therefore

T2j=340+894+544+47=1825

T31 = 450 (e (-3900*0.00055) e (-0.00055*5100) ) = 10

T32 = 450 (e (-0.00055*5100) e (-0.00055*5900) ) = 25
T33= 450 ( e0- e (-0.00055*3900) ) = 380
T34= 450 (e (-0.00055*5900) e (-0.00055*7850) ) = 12

Therefore

T3j=380+25+10+12=427

T41 = 500 (e (-1925*0.00055) e (-0.00055*3150) ) = 84

T42 = 500 (e (-0.00055*3150) e (-0.00055*3950) ) = 31
T43 = 500 (e (-0.00055*3950) e (-0.00055*7850) ) = 50
T44 = 500 (e 0- e (-0.00055*1950) ) = 329

Therefore

T4j=84+31+50+329=494

Values of l1, l2, l3, l4 :

1 pm
li = . log m i (I=j)
v ij log e p i t ij
Substituting the respective values
l1=0.00059
l2=0.00056
l3=0.000565
l4=0.00056
Using the modified values of lij, the next iteration is done.

Iteration 2:

61
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

L1=0.00059

T11 = 1250 (e0-e (-0.00059*800) ) = 470

T12 = 1250 (e (-0.00059*800) e (-0.00059*2000) ) = 395
T13 = 1250 (e (-0.00059*2000) e (-0.00059*5900) ) = 345
T14 = 1250 (e (-0.00059*5900) e (-0.00059*7850) ) = 26

Therefore

T1j= 470+395+345+26=1236.

T21 = 1850 (e (-1200*0.00056) e (-0.00056*2000) ) = 376

T22 = 1850 (e0-e (-0.00056*1200) ) = 905
T23 = 1850 (e (-0.00056*2000) e (-0.00056*5900) ) = 537
T24 = 1850(e (-0.00056*5900) e (-0.00056*7850) ) = 45

Therefore

T2j=905+376+537+45=1863

T31 = 450 (e (-3900*0.000565) e (-0.000565*5100) ) = 24

T32 = 450 (e (-0.000565*5100) e (-0.000565*5900) ) = 9
T33= 450 ( e0- e (-0.000565*3900) ) = 400
T34= 450 (e (-0.000565*5900) e (-0.000565*7850) ) = 10

Therefore

T3j=24+9+400+10=443

T41 = 500 (e (-1925*0.00056) e (-0.00056*3150) ) = 84

T42 = 500 (e (-0.00056*3150) e (-0.00056*3950) ) = 30
T43 = 500 (e (-0.00056*3950) e (-0.00056*7850) ) = 48
T44 = 500 (e 0- e (-0.00056*1950) ) = 332

Therefore

T4j=84+30+48+332=496

62
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Values of l1, l2, l3, l4 :

1 pm
li = . log m i (I=j)
v ij log e p i t ij

Substituting the respective values

l1=0.000597
l2=0.000565
l3=0.000591
l4=0.000567

Using the modified values of lij, the next iteration is done.

Iteration 3:

L1=0.000597

T11 = 1250 (e0-e (-0.000597*800) ) = 474

T12 = 1250 (e (-0.000597*800) e (-0.000597*2000) ) = 396
T13 = 1250 (e (-0.000597*2000) e (-0.000597*5900) ) = 341
T14 = 1250 (e (-0.00059*5900) e (-0.00059*7850) ) = 25

Therefore

T1j= 474+396+341+25=1236

L2=0.000565

T21 = 1850 (e (-1200*0.000565) e (-0.000565*2000) ) = 341

T22 = 1850 (e0-e (-0.000565*1200) ) = 905
T23 = 1850 (e (-0.000565*2000) e (-0.000565*5900) ) = 531
T24 = 1850(e (-0.000565*5900) e (-0.000565*7850) ) = 44

Therefore

63
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

T2j=341+905+531+44=1821

L3=0.000591
T31 = 450 (e (-3900*0.000591) e (-0.000591*5100) ) = 23
T32 = 450 (e (-0.000591*5100) e (-0.000591*5900) ) = 8
T33= 450 ( e0- e (-0.000591*3900) ) = 405
T34= 450 (e (-0.000591*5900) e (-0.000591*7850) ) = 9

Therefore

T3j=23+8+405+9=445

L4=0.000567

T41 = 500 (e (-1925*0.000567) e (-0.000567*3150) ) = 84

T42 = 500 (e (-0.000567*3150) e (-0.000567*3950) ) = 31
T43 = 500 (e (-0.000567*3950) e (-0.000567*7850) ) = 48
T44 = 500 (e 0- e (-0.000567*1950) ) = 334

Therefore

T4j=84+31+48+334=497

Values of l1, l2, l3, l4 :

1 pm
li = . log m i (I=j)
v ij log e p i t ij

Substituting the respective values

l1=0.000604
l2=0.00057
l3=0.000597
l4=0.000567

64
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Using the modified values of lij again, further iterations are carried out until
consecutive li values are same.

3.2 Competing opportunities Model:

Tomazinis proposed the competing opportunities model as follows:

tij=pibij=pi(praj)(prsj)
where

praj= the probability of attraction to zone j

= The destination opportunities in zone j divided by the sum of the destination

opportunities available in time bands up to and including m

aj
= n

a
x =1
x

prsj=Probability of trip end allocation satisfaction in zone j

=1- the sum of the destination opportunities available in time bands up to n including
band m divided by the sum of total destinations in the study area j

65
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

x =1
ax
=1- n

a
x =1
x

x=any time band.

m=Time band into which zone j falls
ax=The destination opportunities available in time band x
n= The last time band as measured from an origin zone i

3.3 Linear Programming Approach:

the objective is to minimize the total amount of travel time of trip makers in moving
between origin and destination pairs . Blunden , Colston Blunden have formulated the
trip distribution
The objective function is to minimize the total vehicle travel time:

minl(i).a(i, j, k).V( j, k)
i j k

Where
l (i ) =Travel time of link i
V ( j , k ) =Volume of link i
a(i, j , k ) =1 if link i lies on the path j to k
=0, if not
Step 1: -Determining the basic feasible solution by Least Square Method.

66
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Step 2: -There will be 3+2-1=4 basic variables. Checking for optimality,

1. Only simplex method is used to determine the entry variable. If the optimality
condition is satisfied, stop if not go to step 2.
2. Determine leaving variable using the simplex feasibility condition.
By the method of multipliers, for each basic variable:

u i + v j = cij Letting u 3 = 0 , I can solve for the remaining as shown in table below

67
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

3.3.1 Example problem:

Assign the traffic to the various links for the network shown below.
E1=4000 E2=5000
Table. Travel Time Matrix
1 2
D 1 2
O
R5=3000
3 11 22
5
4 16 18
4
5 21 10 R2=3000
3
R3=3000

The objective function is to minimize the total vehicle travel time:

min l (i).a(i, j, k ).V ( j, k )
i j k

Where
l (i ) =Travel time of link i
V ( j , k ) =Volume of link i
a(i, j , k ) =1 if link i lies on the path j to k
=0, if not
subject to constraints
T31 + T32 = 3,000
T41 + T42 = 3,000
T51 + T52 = 3,000
T31 + T41 + T51 = 4,000
T32 + T42 + T52 = 5,000
As this is a balanced transportation problem, we can solve in using transportation label
model. The travel times are in the N-E corner of the each table.

Step 1: -Determining the basic feasible solution by Least Square Method.

68
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

1 2
11 22
3 3000 3000

16 18
4 1000 2000 3000

21 10
5 3000 3000

4000 5000

1. Cell (3,1) has the least travel time. Assign 3000 to this cell. Row I1 is satisfied,
satisfy column 1.
2. Next the cell with the least travel time is (5,2). Assign 3000 to this cell and
adj8ust the total of column 2 to 5000-3000=2000
3. Next cell with least travel time is (4,1) and allocate to it 1000
4. Allocate 2,000 to cell (4,2)

Step 2: -There will be 3+2-1=4 basic variables. Checking for optimality,

3. Only simplex method is used to determine the entry variable. If the optimality
condition is satisfied, stop if not go to step 2.
4. Determine leaving variable using the simplex feasibility condition.
By the method of multipliers, for each basic variable:

u i + v j = cij Letting u 3 = 0 , I can solve for the remaining as shown in table below

69
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Basic Variable ( u, v) Solution

(3,1) u 3 + v1 = 11 v1 = 11

(4,1) u 4 + v1 = 16 u4 = 5
(4,2) u 4 + v 2 = 18 v 2 = 13
5,2) u 5 + v 2 = 10 u 5 = 3

I can use the tabulated solutions to get the values of the non-basic variables.

Non-basic Variable u i + v j cij Comment

(3,2) u 3 + v 2 c32 = 0 + 13 22 = 9 The starting solution is

(5,1) u 5 + v1 c51 = 3 + 11 21 = 13 optimal

As the problem is minimization, the starting points are the solution as the non-basic
variables has non-positive values.

The solutions are:

T31 T41 T42 T52

Solution 3000 1000 2000 3000

The assigned trips can shown as: E1=4000 E2=5000

1 2
D 1 2 3000
O 1000
3000 2000
3 11 22 R1=3000
5
4 16 18
4
5 21 10
3
R2=3000
R3=3000

70
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

3.4 Wilson Modified Entropy Model:

The urban and regional scientist faces a number of theoretical problems. His activity is
often a multi disciplinary one in the sense that he uses to concepts from several
disciplines- economics, geography, and sociology and so on. The concept of entropy has,
until recently, been used primarily in the non social science. It has hoped that entropy
enables the social scientist to take some of his basic problems in a fruitful way, and thus
to make progress which might not be possible so easily which more orthodox tools.

Applications of Entropy:
1. It is used for theory building hypothesis (model) development.
2. Used in expressing the laws about system dynamics.
3. For interpretation procedures for the theories.

The main views of entropy is

1. The relationship of entropy to probability and uncertainty.
2. The entropy of probability distribution.
3. Entropy and Bayesian statistics.

Gravity Model
- Based on land use and transportation network
- Calibrated for many urban areas
- Simple
- Accurate
- Supported by the USDOT

"The number of trips between 2 zones is directly proportional to the number of trip
attractions at the destination zone and inversely proportional a function of the travel time"

71
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Tij = Trips produced in zone i and attracted to zone j

Pi = Trips produced in zone i
Aj = Trips attracted in zone j
Fij = Friction factor for impedance (usually travel time) between zones i and j
Kij = Socioeconomic adjustment factor for trips produced in i and attracted to j

How do we determine values for the variables?

- Recall Ps and As come from trip generation

- The sum of productions has to equal the sum of attractions

- Ks are used to force estimates to agree with observed trip interchanges

(careful! do not use too many of these! Have a good reason for using them!)
- Fs are determined by a calibration process (by purpose), and depend upon
the willingness of folks to make trips of certain lengths for certain purposes
recall... trip purposes

HBW - home based work

HBO - home based other

NHB - non-home based

HBS - home based school

Derivation of Gravity model:

m1 m2
Fij=
d122

O1 D2
T12= k 2
d 12

Double the values of , Oi, Dj

72
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

O1 D2
T12= k 2
d 12

Trip end balance is required

Balancing factors are Ai & Bj
Take the impedance in the form of generalized function
=ln W(Tij)+i(1)(Oi Tij)+j(2)(Di Tij)+(C TijCij)
j i i j

Tij=exp(i(1)j(2)(Cij))

j
Tij=Oiorexp(i(1)) exp (j(2)(Cij))
j

exp(i(1))=Oi/Ai exp(j(2)(Cij))1
j

exp(i(2))=Dj/Bj exp(j(1)(Cij))1
j

Tij=AiOjBjDJexp(Cij)
1
exp ( i )
Ai=
Oi
( 2ji )
exp
Bj=
Dj

Ai= [ D exp(j(2)(Cij))1]
j
j

Bj=[ O Aiexp((Cij))1]
i
i

1
Ai=
B
j
j D exp( C ij )

73
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

3.5 Trip Assignment:

3.5.1 All Or Nothing Assignment Model:

All or nothing assignment is basically an extension of finding the minimum paths through
a network. It is called all or nothing because every path from origin zone traffic to a
destination zone has either all the traffic (if it is assumed as minimum paths) or none of
the traffic.

The steps followed are:

1. Find the minimum path tree from each of the zone centroid nodes to all other
nodes.
2. Assign the flow from each origin to each destination node obtained from the trip
table to the arcs comprising the minimum path for that movement.
3. Sum the volume on each arc to obtain the total arc volume. If (undirected) link
volume is desired, sum the flows on the two arcs that represent bi-directional link.

The All or Nothing Traffic Assignment is illustrated using the following example.

10

1 11 2

7 8 7 5

4 6 3

10

Home Node Minimum Path tree

0 11
1
7

74
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

17

2 10 0

13 7

14 5
3

0
6

4 8 15

10
0

Inter node volume, veh/hr is given belo

1 2 3 4
1 500 750 350
2 275
1050 475
3 650 1870 950

4 1250 350 2050

275

1 500 2

1100 1900 1525 2220

2075
4 3
315

LINK 4-3 IS CONGESTED LINK

75
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

3.5.2 Capacity Restrained Method:

This model involves two-travel time versus volume relationships used iteratively to arrive
at predictions of volumes on up to four separate routes between any two zones. The first
equation utilizes route volume as the dependent variable.
The symbolical representation of the graphical portrayal is:

d (Vr Vrc ) Lr
t r = t rc + ------------------------------------------------------------------(1)
Vrc

where t r =travel time on route r (minutes)

Vr = Volume of traffic on route r (veh/hr/lane)
Vrc = Critical volume for the route r (veh/hr/lane)

t rc = unit travel time at the critical volume (min/mile)

Lr = length of the route r (miles)

d = delay parameter (min/mile)
=0.50 for Vr < Vrc

=10.0 for Vr Vrc

The second relationship used for predicting the volume on the route r given travel time:

1/ tr
Vr = m
V ---------------------------------------------------------------------------(2)
1 / t
r =1
r

where V =the total volume of trips from zone I to j on all routes

t r =travel time on route r (minutes)
Vr = Volume of traffic on route r (veh/hr/lane)

Equation (2) divides up the volume of trips from zone I to j among the various routes in
accordance with the reciprocal of the travel times.

76
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

The steps which are followed will be:

i. Find t r from equation (1)
ii. Find Vr by inserting t r in equation (2)
iii. Enter the value of Vr in equation (1)
iv. Repeat the same procedure until the changes in volumes or travel times
become negligible

Example of the TRC Trip Assignment model

The interplay between two TRC Trip Assignment equations can be demonstrated through
a n example in which a pair of zones is connected by the two routes whose characteristics
are shown in table 1 below.

Table 1 Example Route Characteristics

Route No. of Speed limit Signals/ Length Critical Critical Travel time
No. lanes (mph) mile (mile) volume travel time with no volume

1 1 30 1 3 600 3 2.5
2 1 50 1 4 1100 2 1.5

Iteration 1: - starts with the ideal travel times for the entire length of each route. Thus, if
no traffic were:
On route 1, the travel time would equal: 2.5min/mileX3miles = 7.5min
On route 2, the travel time would equal: 1.5min/mileX4miles = 6.0min
This leads to:

1 / 7.5
V1 = (1200) = 532vph / lane and
1 / 7.5 + 1 / 6.0

1 / 6.0
V2 = (1200) = 668vph / lane
1 / 7.5 + 1 / 6.0
From equation (1) taking d = 0.5 since both the routes have volumes less than their
respective critical volumes, we get:

77
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

0.50(532 600)
t1 = 3.0 + 3 = 8.82 min and
600

0.50(668 1,100)
t 2 = 2.0 + 4 = 7.20 min
1,100
Iteration 2: -
1 / 8.82
V1 = (1200) = 536vph / lane and
1 / 8.82 + 1 / 7.72

1 / 7.20
V2 = (1200) = 664vph / lane
1 / 8.82 + 1 / 7.72
Then the travel times will be:
0.50(536 600)
t1 = 3.0 + 3 = 8.85 min and
600

0.50(664 1,100)
t 2 = 2.0 + 4 = 7.18 min
1,100

where d = 0.5 again since both the routes have volumes less than their respective critical
volumes.
Since the last set of travel times do not differ significantly from the previous, the
procedure can be terminated and hence the final results are:
V1 = 536vph / lane and V2 = 664vph / lane
t1 = 8.85 min and t 2 =7.18 vph / lane
The number of iterations is less in this particular case. If the inter zonal volume falls
within a range of 300vph / lane of the sum of the critical volumes which in example is
600 + 1,100 = 1,700 vph / lane . The reason for this is that d parameter will jump from
0.5 to 10.0 and back, causing corresponding fluctuations in the related travel times.

78
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

3.5.3 Multipath Assignment Model :

McLaughlin developed one of the first multi-path traffic assignment techniques.

A driver route selection criteria is used by McLaughlin which is a function of:
Travel time
Travel cost
Accident potential
The minimum resistance paths between each origin and destination pair are calculated
with all the link resistances set to values that correspond to a zero traffic volume. The
minimum resistance value between an origin and destination pair is increased by 30%.
All the paths between the origin and the destination pair with the resistance values less
than this maximum value are identified.

McLaughlin used certain principles of linear graph theory to accomplish the multi-path
assignment. Using an electrical analogy it is possible to identify a through variable y that
corresponds to current, or traffic flow. A cross variable x may be identified that
corresponds to the potential difference, or traffic pressure.

Two postulates from linear graph theory may be introduced that are known as the vertex
and circuit postulates.
At any vertex
e

aiyi = 0
i =1

where e = the number of oriented terminal graphs, or elements

yi = the through variable of the ith element
ai = 0 if the ith element is not connected to v
=1 if the ith is oriented away from V
=-1 if the ith element is oriented toward
For any circuit,
e

bixi = 0
i =1

79
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

where e = the number of oriented terminal graphs, or elements

xi = the cross variable of the ith element
bi = 0 if the ith element is not in the jth circuit
=1 if the ith element orientation is the same as the jth circuit
=-1 if the ith element orientation is opposite to the jth circuit

A sub-graph is then established for each origin and destination pair with these
representing two vertices. The connecting elements are the acceptable paths between the
vertices plus one flow driver element that corresponds to the car travel demand between
the origin and destination pair. The travel demand is assigned among the potential paths
building phase. The traffic assigned to each alternative path must be such that the
alternative paths have an equal across variable value.

The cross variable X , the resistance value R ( y ) and the through variable y for each
path are assumed to be related as follows:

X = R( y ) y
The above equation is analogous to Ohms Law in that the potential is equal to the
resistance times the flow. In this case the resistance along a path is assumed to be a
function of the flow along that path

The figure 1 below represents the way McLaughlin illustrated the assignment method. A
schematic two-way street system is shown along with the link descriptions and then trip
table. The minimum path trees were determined for all origin-destination pairs using the
resistance function for the zero flow, and these are given in table 1 below. Multiplied the
minimum path resistance values and the paths whose resistance values were less than this
higher value were determined and these paths are also given in table 2 below.

The trip table inputs were assigned directly for the origin-destination pairs with only one
path. Sub-graphs were formed for the remaining trip table inputs and solved by the chord

80
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

formulation of linear graph theory. Figure 2 shows the sub-graph for the origin1 and
destination 3.

The circuit equations may be represented in the general form as:

B11 B12 u 0 X b 1
X
b2
B21 B22 0 u X c i

X c2
where B11 =coefficient matrices corresponding to the branches
u =a unit matrix corresponding to the chords
X b 1 =the column matrices of the branches

X c 1 ..= the column matrices of the chords

1 2 3

81
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Link Characteristics
No. Link Length (miles) Maximum Capacity Free speed
(veh/hr) (mph)
1 1-2 1.0 1200 40
2 1-4 2.0 4000 50
3 2-1 1.0 1200 40
4 2-3 0.5 1000 30
5 3-2 0.5 1000 30
6 3-4 0.5 1000 30
7 4-1 2.0 4000 50
8 4-3 0.5 1000 30

Trip Table
Origin Destination
1 2 3 4
1 100 100 2500
2 300* 100 800

Fig 1. Multi-path assignment example problem

Table 1 Paths for multi-path example

Origin Destination Minimum Diversion Minimum Diversion
path* path* Path Path
2 1,2 - 1,2 -
1 3 1,2,3 1,4,3 1,2,3 -
4 1,4 - 1,4 1,2,3,4
1 2,1 - 2,1 -
2 3 2,3 - 2,3 -
4 2,3,4 - 2,3,4 -

82
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

* Based on the resistance function = s ( p)t ( p )

based on the resistance function = S ( p )

Where s ( p) = travel cost (operating, accident, and comfort) as a function of the

volume -to-capacity ratio p
t ( p) = Travel time as a function of the volume-to-capacity ratio p
S ( p) = Travel cost (operating, accident, comfort, and time) as a function of the
volume -to-capacity ratio p

1 1 3

Branch Element 1- path 1,2,3

Chords 2- path 1,4,3
3-trip table input
Fig 2.Subgraph of the origin 1 and destination 2

The first term is non-existent in this system and the circuit equations are:

B12 u 0 X b2
X =0
c 1
B22 0 u X c 2

The terminal equations of the street components may be represented by:

83
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

X b 2 Rb 2 0 Yb 2
=

X c 1 0 Rc 1 Yc 1

Where Rb 2 = the sum of the link resistances corresponding to branch paths

Rc 1 =The sum of the link resistances corresponding to chord paths

Yb 2 =The flow on the branch path

Yc 1 =Flow on the chord paths

For the demand assignment:

X 1 9.51 0 Y1
=

X 2 0 12.17 Y2

The sub-graph fundamental circuit equations are then substituted into the chord set of
equations

0 B12 u Rb 2 0 B12T T
B22 Yc 1
X + =0
c2
u B22 0 0 Rc 1 u 0 Yc 2

Where B12 =a column matrix with coefficients equal to -1; the number of rows in this
matrix correspond to the number of non driver chords in the sub-graph; or it corresponds
to the number of paths less 1 between an origin-destination pair
B22 =+1, corresponding to the driver or trip table input
Yc 1 =The unknown flows for the non driver chord elements

Yc 2 =The through driver, or trip table input

84
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

The specific formulation for this example is:

0 1 1 9.51 0 1 1 Y2
X + =0
3
1 1 1 0 12.17 1 0 100

Taking the first set of the above equations, the solution is Y2 =44 vehicles per hour, and
the flow on element 1 is solved by subtraction; Y1 =100-44=56veh/hr. The results of the
demand assignment are presented in table 2

Table 2 Link volumes for multi-path example

No. Link R0 Y1 R1 y2 y3 y4

1 12 4.90 156 4.98 138 1,290 625

2 14 7.56 2,544 8.42 2,562 1,410 2,075
3 21 4.90 300 5.00 300 300 300
4 23 4.61 956 16.40 938 2,090 1,425
5 32 4.61 0 4.61 0 0 0
6 34 4.61 800 5.80 1,890 1,890 1,225
7 41 7.56 0 7.56 0 0 0
8 43 4.61 44 4.61 62 0 0

In McLaughlins assignment procedure, new link and path resistance values are
calculated for the capacity restraint assignment that corresponds to the flows obtained
from the demand assignment. The procedure described above is employed again to
calculate the restrained volumes. If these volumes are within tolerable limits of the
demand volume then the restraint assignment is complete, otherwise an iterative is
required.

An iterative solution is achieved by averaging the link volumes according to the

following expression:
n yi
Y= ]
i =1 n

85
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Where Y = the average assigned volumes

Yi =The trips assigned to the links during the ith iteration of the linear
graph procedure including iterations
n= the number of linear graph iterations

Burrell has proposed a technique for generating multiple paths through a traffic network.
This method assumes that the user does not know the actual travel times on the links but
associates a supposed travel time on each link that is drawn at random from a distribution
of times. It assumes that the user finds and uses a route that minimizes the sum of the
supposed link times.

Burrell assumes that a group of trips originating from a particular zone have the same set
of supposed link times and consequently there is only one tree for each zone of
production. A rectangular distribution that could assume eight separate magnitudes was
assumed and the ranges of distributions for each of the links were selected so that the
ratio of the mean absolute deviation to actual link time was the same for all links. The
demand or capacity restrained assignments are then made to the paths selected in the
above manner.

Another multi-path assignment technique has been proposed by Dial. With this technique
each potential path between a particular origin and destination pair is assigned a
probability of use that then allows the path flows to be estimated.

86
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

3.6 GRAPH THEORY APPROACH

Graph theory is basically a branch of topology. Geometric structure of a transport
network, which is the topological pattern formed by nodes and routes is studied by graph
theory. The use of graph theoretic approach in road network analysis is to compare and
evaluate one network analysis with the other which may be in the same region or in
different regions. It can be also used to check connectivity and accessibility level of
different nodes.
Connectivity:
Its concept involves following terms:
1. Degree of vertex: Number of edges meeting at the vertex.
2. Path: Collection of vertices and a subset of their incident edges so that degree
of each internal vertex is two or more and the degree of each terminal vertex
is one.
3. Circuits: Closed path where all vertices are of degree two or more.
4. Connected and Unconnected Graph: Connected if there exist at least one
path between any pair of vertices in graph. In unconnected, there are pairs of
points or vertices which cannot be joined by a path.

Structural and geometrical properties of alternative transport networks can be measured

in terms of following graph theoretic measures:
1. Beta Index: Ratio of total number of links to the total number of nodes in
network.
Mathematically: = (e/v)
Where: e and v are, respectively, number of edges and vertices in network.
2. Cyclomatic Number: A count of the number of fundamental circuits existing
in the graph. It is an measure of redundancy in the system.
Mathematically: = e (v p)
Where: p is number of maximal connected sub graph.
3. Gamma Index: Ratio of the observed number of edges in network to
maximum number of edges which may exist between specified number of
vertices.

87
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Mathematically: = e x 100/ (3(v 2)

4. Alpha Index: Ratio of the observed number of fundamental circuits to the
maximum possible number of complete circuits.
Mathematically: = x 100/(2v 5)
5. Associate Number: The number of links needed to connect a node to the
most distant node from it. The node which has low associate number is most
accessible.
6. Shimbel Index: Measure of accessibility which indicates the number of links
needed to connect any node with all other nodes in the network by the shortest
path. The node having lower shimbel index is the most accessible.
7. Dispersion Index: It is the measure of connectivity of transport network and
obtained by sum of the shimbel index.
v v
Mathematically, DI = d ij
j =1 i =1

8. Degree of Connectivity: Ratio of maximum possible number of routes to

have
Complete connectivity to observed number of routes in network.
Degree of Connectivity = ((v(v 1)/2)/e
Example 1: D C
E
B
F
A
G

The solution for the problem can be done in tabular form as follows:

88
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Table giving Solution for above figure

A B C D E F G A.N. S.I

A - 1 2 2 2 1 2 2 10
B 1 - 1 2 2 1 2 2 9
C 2 1 - 1 2 1 2 2 9
D 2 2 1 - 1 1 2 2 9
E 2 2 2 1 - 1 1 2 9
F 1 1 1 1 1 - 1 1 6
G 2 2 2 2 1 1 - 2 10
Total 13 62
Mean Associate Number/Vertex = 13/7 = 1.8571

Mean Dispersion Index/Vertex = 62/7 = 8.85

As mean associate number and mean dispersion index for this network is
less. So from accessibility point of view, this network is better.

Node link incidence matrix:

It is an n x l matrix E whose element in the row corresponding to node i and the column
corresponding to the link (j, k) is defined to be:
+1 if i = j,
-1 if i = k, and
0 otherwise.
By multiplying node link incidence matrix (E) with chain flow vector (f), we get O D
flow vector. So mathematically,
Exf=g
Where:
E is the n x l node link incidence matrix,
f is the l x 1 link flow vector, and
g is the n x 1 O D flow vector.

89
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

3.7 Flows and Conservation Law (Kirchhoffs Law)

Fundamental to the theory of flow of electric currents in electrical networks, water in

pipe networks, or traffic in transportation networks , is kirchhoffs law , which is a
conservation law stating that, for steady state conditions imply for traffic applications that
we are not concerned with the microscopic and stochastic characteristics of a traffic
stream of individual vehicles traveling at random or in platoon on a city street network,
but rather with the gross macroscopic behavior of traffic as, for example, on a main road
network. We ignore fluctuations over time.

Kirchhoffs law states that the sum of all flows leaving an intermediate node equals the
sum of all flows entering the node.
Kirchhoffs law then states that the sum of all flows leaving the centroid equals the flow
produced at the centroid, and the sum of all flows entering the centroid equals the flow
attracted to the centroid.

We shall adopt for general transportation network the terminology of centroids and
intermediate nodes to distinguish between nodes where traffic may be, and may not be,
produced or attracted. In many other applications, the centroids are called source and
sinks. We shall adopt the following notation .The link flow on the directed link (i, j) will
be denoted by fij, the flow produced at a centroid i by ai, and the flow attracted i by bi.
The quantities fij , ai, bj are assumed to be nonnegative. It is convenient to define A (i)
and B (i), the set of nodes after and before node i by

A(i) = {j/j N,(i, j)L}, (1)

B(i)= {j/j N, (j, i)L}. (2)

Kirchhoffs law for a directed transportation network [ N,L] can then be written in the
form of conservation equations as follows:
If i is a centroid then the following formulae should be satisfied

f
A( i )
ij = ai (3)

90
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

f
B (i )
ji = bi (4)

if I is an intermediate node then the following formulae should be satisfied

f ij f
B (i )
ji =0, (5)
A (i )

For these equations to have solutions, the total production, a i= r say, must be equal to
the total attraction bi. Since the number of links is generally at least twice the number of
node in a network, the number of unknowns in Eqs. (3)- (5) greatly exceeds the number
of equations and the equations are rich in solutions.

Figure 1 illustrates a transportation network with two centroids and two intermediate
nodes. For the intermediate node 2, Kirchhoffs law can be easily verified:

i=2, A(2) = {3,4}, B(2) = {1,3},

f
A( 2 )
2j f j 2 = f 23 + f 24 f 12 f 32 = 5 + 1 3 3 = 0
B ( 2)

91
 CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

3.8 Min Cut Max Flow Theorm:

In max flow min cut theory, a cut may be defined as any set of directed arcs containing at
least one arc from every directed path from source to sink. There normally are many
ways to slice through a network to form a cut to help analyze the network. For any
particular cut, the cut value is the sum of the arc capacities of the arcs (in the specified
direction) of the cut. The max-flow min-cut theorem states that, for any network with
single source and sink, the maximum feasible flow from the source to the sink equals the
minimum cut value for all cuts of the network.

3
A
0 1
0
5 0 0
0 4 9 T
O B `D
7
5 0 0
4 2 0

0
0
1
0 C 4 0 E
6

Considering above network, one interesting cut through this network is shown by dotted
line. Notice that the value of the cut is 3 + 4 + 1 + 6 = 14, which was found to be the
maximum flow value. So this cut is a minimum cut.

92
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

3.9 Dynamic Stochastic Assignment Model with Many Routes

between a Single O-D Pair
Let D be the demand for travel from the origin zone h to the destination zone d and
assume that the period between the earliest and latest possible times departure from h is
considered ax a series of equal spaced epochs, representing intervals of length w .

Suppose the available routes are numbered 1,2, r , .I and that Dr(k,j) is the number
of vehicles for unit time departing from h , on route r at epoch k on day z.

Let Xa(k,z) and la denote the number of vehicles on any link, a, at epoch k on day z and
the link length respectively.

The number of vehicles leaving link a per unit time at epoch k on day z is

Va ( k , z ) = X a ( k , z ) * S a ( k , z )
la
The flow conservation equations for link a may be written as
z
X a (k + 1, z ) = X a ( k , z ) + w{ Dr (k , z ) ar Va (k , z )} For a Ih
1

Where ar= 1 if link a is on route r

0 otherwise
X a (k + 1, z ) = X a ( k , z ) + w{( Pam (k , z )Vm (k , z ) Va (k , z ) if a isnt on the route h
m

where Ma is the set of links having as final node the initial node of link a, and Pam(k,z) is
the proportion of the vehicles leaving link m at epoch k on day z which enter link a.
In a test with a small network Vythoulkas was not able to contain a stable situation ; the
departure pattern tended to oscillate about a possible equilibrium distribution .

93
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

3.10 Dials Model:

Dials model is based on the hypothesis that there should be a non-zero probability of use
of all efficient paths. He proposed two definitions of efficient path, namely:
1. a path in which every link has its initial node closer than its final node to the
origin and has its final node closer than its initial node to the destination; and
2. a path in which every link has its initial node closer than its final node to the
origin.

Conceptually the first of these definitions is more attractive but it has the drawback that
in any single run trees can be built from origin zone to one destination zone only because,
for each node in the network, the minimum cost to each destination zone must be known
as well as that from the origin zone under consideration. The second definition lends
itself to the usual once through approach.

Basically, trips are allocated to any efficient route R so that,

V R = V * exp( (c R c * )) = V * exp( R ) --------------------------------------------(1)

Where R = c R c * is the excess cost in using route R rather than the minimum cost

route, the cost on which is c * , and V R and V * are the flows on route R and the
minimum cost route respectively.

The value of determines the proportions of the trips allocated to the efficient paths; if
= 0 , then the trips are shared equally between them but a high value of produces a
heavy bias towards the cheapest routes.

Dials route (vine) building and link loading algorithm

The following algorithm may be used to simultaneously assign trips from the origin node
to all destination nodes in accordance with Dials second definition of the efficient paths:

94
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Step 1:
Determine the minimum path costs from the origin node h to all nodes in the network.
Let c *j denote the minimum path cost from h to node j and let dm be the destination node

with greatest minimum path cost from h.

Step 2:
i. Initialize all nodal weights:
Set all node weights w j =0; set origin node weight wh = 1.

ii. Consider each node j in the network, in order of increasing minimum path cost, c *j ,

until dm is reached, as indicated below.

a. For each link (i,j) with final node j:
Calculate the effectiveness

exp( ij ) if

eij = ci* < c *j -----------------------------------------------------------(2)
0 otherwise

where cij is the cost on link (i, j), and ij = (ci* + cij* ) c *j is the excess cost involved in

going from origin node h to node j via link (I, j) rather than by the minimum cost path;
and
Calculate its link weight wij = wi eij ----------------------------------------------------(3)

b. Determine the node weight w j = wij -------------------------------------------(4)

i

wij wij
(the ratio = is effectively the probability of a trip from origin h
w i
ij wj

arriving at node j via the predecessor link (i, j), rather than via the other links with
final node j)

95
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Step 3: (Backward pass)

a. Initialize all nodal volumes
Set v j = 0 for all nodes j; and

Set v d = Th d , the trips from h to d, for all destination nodes d.

b. Starting with destination node d m , consider each node j in the network in the order

of decreasing c *j as indicated below.

For each link (i, j) with final node j:

wij
i. Derive the link volume vij = vi ; and
wj

ii. Set vi = vi + vij

Terminate when j = the origin node, h.

Note: - if P is any path from the origin node h to destination node d, via nodes i, j, k, l, ---
, y, z, say, and p is the excess cost on the path, then:

p = (c hi + cij + c jk + c kl + ... + c yz + c zd ) c d*

*
c h* + c hi ) ci* ) + ((ci* + cij )) c *j ) + ((c *j + c jk ) c k* )
= - c + ((
h
+ ... + ((c *z + c zd ) c d* )

= hi + ij + jk + ... + zd , Since c h* = 0

Following step 2 above, the probability of choosing path P from h to d is:

W zd W yz W xy Wij Whi
Pr ( p) = . . .... .
Wd W z W y Wj W i

W z .e zd W y .e yz W x .e xy Wi .eij W h .ehi
= . . .... .
Wd Wz Wy Wj Wi

1
= exp( ( zd + yz + xy + ... + ij + hi )) Since Wh = 1.
Wd

96
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

1
= exp( ( p ))
Wd

Ep
= where E p = exp( ( p )) is the effectiveness of path p------------(5)
Wd

If S is the set of all efficient paths from h to d, then P ( p) = 1

PS
r

Ep
Therefore, W
PS
= 1 and Wd = E p
PS
d

Since, for the minimum cost path, p * , say, E *p = 1, wd may be regarded as the effective

number of efficient paths from h to d.

Example 1:
In the network shown in figure 1 below, in which all the links are on one way, using
Dials method assign 4,000, 2,000 and 1, 000 trips from node 1 to nodes 6, 8 and 9
respectively. Assume = 1.

(4) (4)
1 2 3

(3) (5) (3) (2)

4 (2) 5 (2) 6

(2) (2) (5) (4)

(4) (3)
7 8 9
Figure.1.1 Simple network for the Example

97
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Table: Network Table

Link No 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Initial Node 1 1 1 2 2 3 4 4 5 5 5 6 7 8
Final Node 2 4 5 3 5 6 5 7 6 8 9 9 8 9
Cost 4 3 5 4 3 2 2 2 2 2 5 4 4 3

Step 1: The minimum path costs from origin 1 to all other nodes can be determined by
inspection. In a normal-sized network they could be determined by any of the tree
building algorithms.

Table: Nodal Weights

Node j Wj

Initial Final*
1 1.0
2 0 1.0000
3 0 1.0000
4 0 1.0000
5 0 2.1353
6 0 2.1353
7 0 1.0000
8 0 2.2706
9 0 5.1915
* Final W j = Wij
i

98
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Table: Link Weights

Node j c *j Links in Initial Node eij wij

F j (a ) of link in F j ,i

1 0
4 3 2 1 1.0000 1.0000
2 4 1 1 1.0000 1.0000
5 5 3 1 1.0000 1.0000
5 2 0.1353 0.1353
7 4 1.0000 1.0000
7 5 8 4 1.0000 1.0000
6 7 6 3 0 0( b )
9 5 1.0000 2.1353
8 7 10 5 1.0000 2.1353
13 7 0.1353 0.1353
3 8 4 2 1.0000 1.0000
9 10 11 5 1.0000 2.1353
12 6 0.3679 0.7856
14 8 1.0000 2.2706

Note: a. Fj is the set of all links with final node j

b. Link no.6 is not efficient since c3*  c6* . It may be ignored hereafter

99
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Table: Nodal Volumes

Node j Vj

1 7000.2
2 434.3
3
4 145.2
5 6854.8
6 151.3
7 145.2
8 2437.4
9 1000

Differs from 7000 due to rounding

Table: Link Volumes

Node j cj Links in Initial Node wj wij vij

Fj link in F j ,i

9 10 11 5 5 2.1353 411.3
12 6 1915 0.7856 151.3
14 8 2.2706 437.4
3 8 4 2 1.0000 1.0000 0
8 7 10 5 2.2706 2.1353 2292.2
13 7 0.1353 145.2

100
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

6 7 9 5 2.1353 2.1353 4151.3

7 5 8 4 1.0000 1.0000 145.2
5 5 3 1 2.1353 1.0000 3210.2
5 2 0.1353 434.3
7 4 1.0000 3210.2
2 4 1 1 1.0000 1.0000 434.3
4 3 2 1 1.0000 1.0000 3355.7
1 0 1.0000

The link loads in the final column relate to the links indicated in the third column

Selected link analysis for Dials assignment procedure (second definition of

efficient path)

Although the efficient paths between individual origin-destination node pairs, h-d, are
not explicitly defined when Dials assignment procedure is used, it is possible to
determine the volume on any link (i, j) arising from the trips h to d using an algorithm
proposed by Van Vliet (1981)

If ci* > c *j , where ci* and c *j are the minimum path costs from the origin node h to nodes I

and j respectively, then the link (i, j) is excluded from the efficient paths from h and
hence there are no trips from the origin node h on link (i, j).

Assuming ci* < c *j , we can proceed as follows:

Let X i denote the set of all efficient paths, p, from origin node h to node i and let Y j

denote the set of all efficient paths, Q, from node j to the destination node d.

101
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Let Wh i = E
p X i
p and W j d = E
QY j
Q , where E p and EQ are already defined, but in

relation to paths from h to i and from j to d respectively.

Van Vliet has shown that if trips have been assigned from origin node h to destination
node d in accordance with Dials algorithm, then the probability of a trip from h to d
passing along link (i, j) is given by:

Pr ((i, j ) /(h d )) = eijWh iW j d / Wd ----------------------------------------------------(6)

If there are Th d trips to be assigned from origin node h to destination node d, then the

number of these trips passing along link (i, j) is Th d Pr ((i, j ) /(h d ).

On the basis of this relationship, the loading on link (i, j) arising from the assignment of
the trips Th d may be determined as follows: -

Step1: (A standard forward pass)

a. Carry out steps 1 and 2 as for the standard Dial algorithm.
In course of step 2: eij is determined; Wh i is determined as Wi ; and the Wd are

determined.
b. Set Wd' = Wd for all destination nodes d.

Step 2: (Modified forward pass to determine W j d )

a. Repeat step 2(a) of the standard Dial algorithm but with W j = 1, not Wh = 1.

b. Calculate link weights Wij and node weights W j as steps 2(b)(i) and (ii) of the

standard Dial algorithm but when considering each node in the network in order
of increasing cost from the origin node h, consider only those nodes coming after
node j.

102
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

(The values of Wd obtained from step 2 are the required values of W j d , i.e, set

W j d = Wd from the table of nodal weights at the end of step 2).

Step 3: (Determination of link volumes)

a. Substitute the values of eij ,Wh i and Wd = Wd' from step 1 and W j d from step in

equation (?) to obtain, for each h-d pair, the value of Pr ((i, j ) /(h d )).
b. Calculate Vij (h d ) = Th d Pr ((i, j ) /(h d )) and output results.

Note: Van Vliet (1981) actually considered the select link analysis for a set of links, i.e,
link (i,j) in the above is replaced by a sub-route from i to j.

Example 2: Determine the loading on link (4,5) in Example 1.

We have: origin node, h = 1; d j = 6,8,9; T16 = 4000; T18 = 2000; T19 = 1000

Step 1: From the previous example of table 2, Wh i = W1 4 = W4 = 1.0000; for the

destination nodes- W6' = 2.1353;W8, = 2.2706; and W9' = 5.1915 . From table 3,

e45 = 1.0000

Step 2: (Modified forward pass)

Table: Nodal Weights

Node s Ws (= W5 s )
Initial Final
1 0
2 0
3 0
4 0

103
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

5 1.0000
6 0 1.0000
7 0
8 0 1.0000
9 0 2.3679
Final Ws (= W5 s ) = Wrs from table 1.6 below.
r

Table: Link Weights

Node s c s* Links in Fs Initial Node of ers Wrs

link in Fs ,r

7 5 8 4 1.0000 0
6 7 6 3 0 0
9 5 1.0000 1.0000
8 7 10 5 1.0000 1.0000
13 7 0.1353 0
3 8 4 2 1.0000 0
9 10 11 5 1.0000 1.0000
12 6 0.3679 0.3679
14 8 1.0000 1.0000

Table: Select Link Loading for Link (4,5)

i j eij W1i d W j d Wd' Pr ((i, j ) /(1 d )) T1 d Vij (1 d )

4 5 1.0000 1.0000 6 1 2.1353 0.4683 4000 1873.3

8 1 2.2706 0.4404 2000 880.8
9 2.3679 5.1915 0.4561 1000 456.1

W j d = Wd from the table 1.6

104
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Table: Output

Load on Link (4,5) Origin Node Destination Node

1873.3 1 6
880.8 1 8
456.1 1 9

Total 3210.2

105
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

3.11 Gunnarssons Model:

Gunnarsson (1972) said of his model that it is seemingly close in intent to Dials model
but vastly different in both strategy and tactics. In concept, the method differs from that
of Dial in that whereas Dial considered the entire routes from origin to destination when
making a choice, Gunnarsson assumed that a driver makes a choice of route at each node
arrived at, independently of any previous decision. He considered a reasonable route from
origin node to destination node d to be one such that every link in it has its final node
nearer to d than its initial node and is such that the cost on the route is within chosen
limits of the cost on the minimum cost path.

He assumed that at any node, i, the basis of the choice of the next link, (i, j) say, in the
path to the destination node d, is the cost from i to d via link (i, j), which he referred to as
the resistance of link (i, j), denoted by rij . Letting d *j and cij denote the minimum cost to

d from any node j and the cost on link (i, j) respectively.

rij = cij + d *j --------------------------------------------------------------------------------(7)

The probability of use of link (i, j) is then given by:

Pr (i, j ) = f (i, j ) / f (r
( i , k )I i
ik ) --------------------------------------------------------------(8)

where f (rij ) is some function of rij and I i is the set of all links with initial node i.

Following traffic studies, he used f (rij ) = rij8 . To prevent excessively costly routes being

used, he introduced an acceptable prolongation factor, w, for a route. To be a part of a

reasonable route, a link (i, j) should satisfy the following conditions.

1. d *j < d i* ; and

2. rij wd i*

106
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

A vine building and link loading algorithm for Gunnarssons model

The following is an algorithm, for simultaneously assigning trips from all origin nodes to
destination node d, in accordance with his definition of a reasonable path: -
(It is assumed that the value of the acceptable prolongation factor, w, and the form of the
probability function f (rij ) have been specified).

Step1: (Determination of d i* )

Set cij' = c ji for all links (j, i).

Using the link cost matrix {cij' } determine the minimum path costs, d i* , from d to all

nodes, i. Let hm be the origin node with greatest minimum path cost to d.

Step 2: (Determination of link and nodal attractivities Aij and ai .

Starting with hm , consider each node I in the network in order of decreasing d i* until d
is reached.
For each node i:
a. For each link (i,j) with initial node i, i.e, for all (i,j) I i ,

i. determine its resistance rij = cij + d *j (= c *ji + d *j ) ,and

ii. determine its attractivity (weight) Aij where

Aij = f (rij ) if d *j < d i* and rij < wd i* , otherwise Aij = 0;

b. Determine the total attractivity of node i, using ai = A

( i , k )I i
ik

Step 3: (Determination of nodal volumes V j and link volumes Vij )

a. Initialize all nodal volumes:

Set Vi = 0 for all nodes i; and

Set Vh = Th d , the trips from h to d, for all origin nodes h.

b. Starting with origin node hm , for each node i in order of decreasing d i* :

107
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

i. for each link (i,j) in I i , derive the link volume Vij = Vi Aij / ai ; and

ii. set V j = V j + Vij

Stop when node d is the node considered.

Gunnarssons model may be brought closer to Dials by specifying:

f (rij ) = exp( (rij d i* )) = exp( ij ),

where ij = (cij + d *j ) d i* is the excess cost involved in proceeding to destination node

from node i via link (i,j) rather than by the minimum cost path from i. With this form for
f (rij ), the probability that, of all the links available at node i, link (i,j) is chosen for the

continuation of the journey to d is given by:

f (rij ) exp( ij ) Aij

Pr (i, j ) = = = ------------------------------------------(9)
f (r )
( i , k )I i
ij exp(
( i , k )I i
ik ) ai

Example1:
In the network in Example 1, reverse all a link directions and assign 4000, 2000 and 1000
trips to node 1 from nodes 6, 8 and 9 respectively using Gunnarssons model with W = 1
and f (rij ) = exp( ij ).

Step 1: The minimum path costs to node 1 from all other nodes are the same as those
from node 1 to the other nodes using the un-reversed link cost.

108
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Step 2:

Table: Link and Nodal Attractivities Aij , ai

Node i d i* Final Node of cij dj rij ij Aij ai

link in I i , j

9 10 5 5 5 10 0 1.0000 2.3679
6 4 7 11 1 0.3679
8 3 7 10 0 1.0000
3 8 2 4 4 8 0 1.0000 1.0000
8 7 5 2 5 7 0 1.0000 1.1353
7 4 5 9 2 0.1353
6 7 3 2 8 0 1.0000
5 2 5 7 0 1.0000
5 5 1 5 0 5 0 1.0000 2.1353
2 3 4 7 2 0.1353
4 2 3 5 0 1.0000
7 5 4 2 3 5 0 1.0000 1.0000
2 4 1 4 0 4 0 1.0000 1.0000
4 3 1 3 0 3 0 1.0000 1.0000
1 0

109
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Table: Nodal Volumes

Node j Vj

1 7000.0
2 425.3
3
4 3431.7
5 6711.3
6 4155.4
7 288.7
8 2422.3
9 1000

Table : Link Volumes

Node i d i* Final Node of Aij ai Vij
link in I i , j

9 10 5 1.0000 2.3679 422.3

6 0.3679 155.4
8 1.0000 422.3
3 8 2 1.0000 1.0000 0
8 7 5 1.0000 1.1353 2133.6
7 0.1353 288.7
6 7 5 1.0000 1.0000 4155.4
5 5 1 1.0000 2.1353 3143.0
2 0.1353 425.9
4 1.0000 3143.0
7 5 4 1.0000 1.0000 288.7
2 4 1 1.0000 1.0000 425.3
4 3 1 1.0000 1.0000 3431.7

110
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Selected Link Analysis for Gunnarssons Assignment Procedure

The procedure proposed below for determining the trips from the various origin zones h
which include link (i,j) in their paths to destination node d, is an adaptation of the method
for carrying out a selected link analysis with GMTU assignment procedure.

The procedure for determining the movements to destination node d which include link
(i, j) in their paths is as follows: -

Step 1: (Determination of the minimum path costs, d i* , from all nodes, i, to the
destination node d).

Step 2: (Determination of the link attractives)

a. Repeat step 2 of the algorithm but terminate when node i has been
processed. Results in the link attractives matrix (LA) are, for each node m
for which d m* d i* in decreasing order of magnitude of d m* : the node
number m; a list of pairs of values of successor node n (the final node of a
link with initial node m) and Amn ; and a m = A
( m , n )I m
mn

b. Check if j is a successor node of I; if it is not then no minimum paths to d

pass along (i,j); if it is store Pr (i, j ) = Aij / ai

Step 3: (Setting up the pointer array)

Starting at the top of array LA, i.e., with node hm , and proceeding down the rows for
each row:
a. read node number, m, the list of pairs of successor nodes n and corresponding
Amn , and a m .

b. if for any successor node, n, d n* < d i* , delete n and Amn ;

c. replace all remaining values of Amn by Pr (m, n) = Amn / a m ;

d. for each successor node n remaining in the list, enter m and Pr (m, n) in row n of
the pointer array.

111
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Step 4: (Setting up an array N, consisting of nodes n which are such that at least one path
from node n to d passes through node i)
a. Set up array X with elements x r = 0 for r=1 to N nodes .

b. Enter I in array N and set xi = 1

c. Read the next entry, n, in N (n=i initially); sto if there are no more nodes in N.
d. Read the entries in row n of the pointer array. If there are no entries, go to c,
otherwise, for each node m entered:
If x m = o, set x m = 1 and enter m in array N; if x m 0 do nothing.
e. Go to (c)

Step 5: (Setting up an ordered pointer array)

Arrange the nodes, n in N in order of increasing d n* and enter nodes n together with the
entries in row n of the pointer array in an ordered pointer array.

Step 6: (Setting up a node weighing array, W, containing for each node n in N, the
probability that a path from n to d passes along link (i,j)).
a. Initialize array W
Set Wi = Pr (i, j ); set Wn = 0 for all other nodes n in N.
b. From each row of the ordered pointer array in turn:
i. read the node pointed to, n, (equal to i to start with), and the list of pairs of
predecessor nodes m and probabilities Pr (m, n);
ii. for each predecessor node m, calculate Wm = Wn .Pr (m, n) and set

W m = wm + W m .

Step 7: (For all origin nodes h, calculation of the trips to destination node d from h
passing along (i,j) and output of the link loading information).
a. From the trp matrix, read off the origin nodes, h, and Th d , the trips from h to d.

112
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

b. Consider each origin node, h, in turn. If h is not in array N select the next origin
node, if any, else if h is in N: -
i. Calculate Vij (h d ) = Wh .Th d and,

ii. Output a message indicating that there are Vij (h d ) trips on link (i,j) from

origin node h to destination node d.

Example 2:
Determine the loading on link (4,1) in Example 3 above.
Step 1: The minimum path costs to destination node d=1 from all other nodes are the
same as those from node 1 to the other nodes in Example 1.
Step2: The table link attractivities, LA, is the same as table 2.1 for the nodes down to and
including node i=4. j=1 is a successor node of i=4; Pr (4,1) = A41 / a 4 = 1.0000.
Step3:
Table 1.2.4 Pointer Array

Node (row), n Predecessor nodes, m, and corresponding Pr (m, n)

1
2 3(1.0000) 5(0.0634)
3
4 5(0.4683) 7(1.0000)
5 9(0.4223) 8(0.8808) 6(1.0000)
6 9(0.1554)
7 8(0.1192)
8 9(0.4223)
9

113
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Step 4:
Table X , Arrays

r 1 2 3 4 5 6 7 8 9
Xr 0 0 0 1 1 1 1 1 1

Table: N Array

n 4 5 9 8 6

Step 5:

Table: Rearranged Array N

d n* 3 5 5 7 7 10

n 4 5 7 8 6 9

114
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Table: Ordered Pointer Array

Node, n Predecessor nodes, m, and corresponding Pr (m, n)

4 5(0.4683) 7(1.0000)
5 9(0.4223) 8(0.8808) 6(1.0000)
7 8(0.1192)
8 9(0.4223)
6 9(0.1554)
9

Steps 6 & 7:

Table: Node Weighting Array and Calculation of V41 (h 1)

n/h 4 5 7 8 6 9
Wn (initial) 1.0000 0 0 0 0 0

Updated Wn 0.4683 1.0000 0.4125 0.4683 0.1978

(= Wn ) 0.5317 0.4223

0.4951
Th d 2000 4000 1000

Vij (h 1) = WhTh d 1063.4 1873.2 495.1

115
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Table: Output

Load on Link (4,1) Origin Node Destination Node

1063.4 8 1
1873.2 6 1
495.1 9 1
Total: 3431.7

116
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

3. 12 The Greater Manchester Transportation Unit (GMTU)

The GMTU model, Randle (1979), incorporates a deterministic method of producing
multiple routing through a network. Calculations of exact probabilities of link usage are
facilitated by restricting to two, the number of links via which paths from a particular
origin may arrive at a network node. Consider the movements from origin node h to
nodes d 1 and d 2 in the network Figure 1.3.1 below.

(5) d1
q (4)
(5) (5)n
(6) (9) q
(25) X (6)

0 (45)
(50) d2

Figure 1.3.1: A simplified Network for illustration of GMTU Model

Note: Figures in brackets are costs on sections.

For both movements, there is a choice between two routes with costs of 45 and 50 units
respectively. However, in travelling from h to d 1 the choice of route is actually made at
node x and is based on costs of 15 and 20 units, which is likely to lead to a split between
the two routes quite different from that arising from a choice between routes with costs of
45 and 50 units. It is clearly desirable to that in any route choice model, the choice
between routes should be based on mutually exclusive components, an aim which Dials
nor Gunnarssons model attempt to achieve. To accommodate this and other requirements
the GMTU model is based on two considerations:

117
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

1. travelers do not all perceive identical routes and hence there should be a
wide range of sensible routes between every O-D pair wherever possible;
and
2. it should be possible to calculate the probability of usage of every link in
the sensible paths and such calculations should consider only mutually
exclusive sections of paths used.

To satisfy the first consideration, it is assumed that on any link with actual cost c , the

distribution of perceived costs is uniform with mean c and range 2k (c) , where the
spread factor, k, is chosen by the model user.

If, in accordance with the second consideration, choices are to be based on mutually
exclusive sections of route, then they should be between pairs of routes; choices between
pairs of routes also facilitates direct calculation of route usage probabilities.

In the GMTU model, this is accomplished by storing, for each node n having two or more
predecessor nodes, both the minimum cost path predecessor node, p, herein referred to as
the best predecessor node, and the second-best predecessor node, q. The minimum
paths from the origin node, h, to p and q are then retraced from p and q towards h until
they meet, at node x say, as in figure 1.3.1.

Apart from the fact that at most two predecessor nodes are stored for any node, the only
other restriction on route usage is one which is necessary for the vine loading algorithm;
if the minimum path cost to q is greater than the minimum path cost to n, then the
probability of usage of link (q,n) is set to zero. This measure, which is effectively the
same as Dials requirement that on all links used, the initial node must be nearer to the
origin than the final node, also prevents U-turns loops.

118
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

GMTU Method of Assignment Using the Data from a Tree Table with Two
Predecessor Nodes

In the following, it is assumed that a Tree Table from origin h is available, containing, for
each node n:
i. The best predecessor node, p, and the minimum path cost from h to n via node
p, denoted by c n*( P ) ;

ii. The second-best predecessor node q, c q* (the minimum path cost from h to q)

and c n*(q ) (defined as for node p above), if there is no second-best predecessor

node then q, c q* and c n*(q ) are all zero; and

iii. If required , R, the order of removal of node n from the loose ends table.

Step 1: (Determination of the probability of usage of link (p,n))

For each node n in the Tree Table: -
a. From the read values of p, c n*( p ) , q, c n*( q ) and c q* ;

b. If q=0 or c q* > c n*( p ) (= c n* ) set Pr ( p, n) = 1.0; otherwise:

i. Using the minimum path predecessor nodes in the Tree table, retrace the
minimum cost paths to p and q backwards, towards h, until they meet, at node
x say, and read c *x from the table;
ii. Calculate the actual cost from x to n via p and q by

c ( p) = c n*( p ) c *x and c(q) = c *n ( q ) c *x respectively; and,

iii. if c( p) + k /(c( p )) < c(q) k /(c(q )), set Pr ( p.n) = 1.0, else set

Pr ( p, n) = 1.0 [c( p) c(q) + k{/(c( p)) + /(c(q))}]2 /{8k 2 /(c( p).c(q ))}
if q 0 and 1.0- Pr ( p, n) < suitable small value, set q=0 in the tree table and set
Pr ( p, n) = 1.0.

Step 2: (Initialization of the nodal volumes Vi and link volumes Vij )

119
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

a. i. set Vi = o for all nodes i; and

ii. set Vd = Th d , the number of trips from h to destination node d, for all
destination nodes d.
b. For all links (i,j), set the assigned link load V ij= 0.

Step 3: (Assignment of the trips from origin h to the network).

Starting with the node, n, having the greatest value for the order of removal, R, i.e. with
the greatest value of c n* , and proceeding in order of decreasing R until node h is reached,
for each node n: -
a. Read p and q from the Tree table;
b. Set V pn = Vn Pr ( p, n) and set V p = V p + V pn

c. If q 0, set Vqn = Vn V pn and set Vq = Vq + Vqn .

Note: - If the secondary analysis may be required at some future date, the values of
Pr ( p, n) should be added to the Tree Table.
Example 3:
Using the GMTU assignment model with k=1.2, assign 100, 300, 500 and 600 trips from
origin node to destinations 3,4,5,and 7 respectively in the network shown in the figure
1.3.2 below. Assume that the actual costs, shown in () on the links are the same in both
directions.
2 3
(20) (10)

(10) (8) (11) 4

1
5
(12) (20)

(9)
(17)
6 7
Figure 1.3.2 A simple Network for the GMTU Assignment Example 5

120
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Table 1.3.1 Network Table

Link No Initial Final Link Link No Initial Final Link

a Node Node cost, c a a Node Node cost, c a

1 1 2 10 11 5 2 8
2 1 6 9 12 5 3 11
3 2 1 10 13 5 6 12
4 2 3 20 14 5 7 7
5 2 5 8 15 6 1 9
6 3 2 20 16 6 5 12
7 3 4 10 17 6 7 17
8 3 5 11 18 7 4 20
9 4 3 10 19 7 5 7
10 4 7 20 20 7 6 17

Table :Tree Table

Node Best Predecessor Second- Predecessor F Array Order of

n Node, p c n*( p ) (= c n* ) best c n*(q ) Removal,
Node, q R
1 0 0 0 1 1
2 1 10 0 999 3 3
3 5 29 2 30 6 6
4 3 39 7 45 9 7
5 2 18 6 21 11 4
6 1 9 0 999 15 2
7 5 25 6 26 18 5

121
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Step 1: Calculation of Pr ( p, n)

Table :Calculation of Pr ( p, n)

Node Predecessor Intersection c *x c( p) c(q) Pr ( p, n)

n Nodes
p q Node, x
1 0 0 0
2 1 0 1.0
3 5 2 2 10 19 20 0.5898
4 3 7 5 18 21 27 0.8801
5 2 6 1 0 18 21 0.7428
6 1 0 1.0
7 5 6 1 0 25 26 0.5791

Step 3: Link Loading

Table :Node Loading (a)

Node No. n Node Load, V

1 1500.1(b)
2 953.8
3 364
4 300
5 1083.0
6 267.8 546.3
7 636.0

122
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Note:
(a) Nodes in the order of decreasing R: 4, 3,7,2,6,1
(b) Difference from 1500 due to rounding

Table: Link Loading

Initial Final Node, Link Load, Initial Final Node, Link Load,
Node, i j Vij Node, i j Vij

1 2 953.8 5 2 0
1 6 546.3 5 3 214.7
2 1 0 5 6 0
2 3 149.3 5 7 368.3
2 5 804.5 6 1 0
3 2 0 6 5 278.5
3 4 264.0 6 7 264.0
3 5 0 7 4 36.0
4 3 0 7 5 0
4 7 0 7 6 0

Selected Link Analysis for the GMTU model

To check whether or not trip from origin node h contribute to the flow on any link (i,j), it
is necessary to check whether i is a predecessor node to j in the Tree Table constructed
when determining minimum paths from h. For this, Randle and Turner (1979) can be
used. Basically, the Tree Table for the origin node h is transformed into a forward
pointing list of sub-lists, the sub-list for each node, n, being either empty (when n has no
successor nodes i.e. n is not a predecessor node) or consisting of its successor nodes (i.e.
those nodes which do have n as one of their predecessor nodes).

The procedure for determining the movements from the origin node h loading link (i,j) is:

123
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Step 1:
From the Tree table, check that i is a predecessor node to j; if it is, store Pr (i, j ) ,
otherwise, consider the next origin node or stop.

Step 2: (Setting up the pointer array)

c. Initialize the pointer array. In each row enter 0 for the first node and 0.0
for the corresponding probability.
d. For each node (row), n, in the Tree Table: -
i. Read p, q and Pr ( p, n) and enter n and S (n, p ) = Pr ( p, n) in row of p
of the pointer array, (if p = zero, i.e. n is the origin node, consider the
next node in the tree table);
ii. If q 0 , enter in row q of the pointer array, node n and
S (n, q ) = 1 Pr ( p, n)

Step 3: (Setting up array N consisting of nodes, n, which may be arrived at on paths from
h passing through node j)

Step 4: (Setting up an ordered pointer array)

Arrange the nodes n in N order of increasing c n* and enter n together with the entries in
row n of the pointer array in the ordered pointer array.

Step 5: (Setting up a node weighting array, w, containing for each node n in N, the
probability that a path to n from h passes along link (i,j)).
a. Initialize array W .
Set W j = Pr (i, j ); set Wn = 0 for all other nodes in N .

b. From each row in turn of the ordered pointer array: -

i. read the pointing node n (n=j initially), and the list of pairs of successor
nodes s and probabilities S ( s, n) ; and

124
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

ii. for each of the nodes pointed to, calculate Ws = Wn .S ( s, n) and set

W s = ws + Ws
Step 6: (For all destination nodes, d, calculation of the trips to d from h passing along
link (i,j) and out put of the link loading information).
a. From the trip matrix, read off the destination nodes, d, and Th d , the trips from h
to d.
b. Consider each destination node, d, in turn: -
If d is not in array N , consider the next destination node, if any, else, if d is in N :
i. calculate Vij (h d ) = Wd .Th d ; and

ii. out put a message indicating that there are Vij (h d ) trips on link (i,j) from

the origin node h to destination node d.

Example 4:
In Example 5, determine the movements from node 1 loading link (1-6)

Step 1: From the Tree table 1.3.2, in the row for the node j=6, p=1=i. From table 1.3.3,
row j=6, Pr (1,6) = 1.0. (Note that Pr (i, j ) values would normally be stored in the Tree
table if any secondary analysis is anticipated).
Step 2:
Table: Pointer Array

Node (row), n Successor nodes, s, and corresponding (s(s,n))

1 2(1.0000) 6(1.0000)
2 3(0.4102) 5(0.7428)
3 4(0.8801)
4 0.0
5 3(0.5898) 7(0.5791)
6 5(0.2572) 7(0.4209)
7 4(0.1199)
Values of S ( s, n) = Pr (n, s ) are taken from table 3.3.

125
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Step 3:

Table 1.3.7: x Array

r 1 2 3 4 5 6 7
xr 0 0 1 1 1 1 1

Table 1. 3.8: N Array

n 6 5 7 3 4

Step 4:

Table: Ordered N Array

c n* 9 18 25 29 39

n 6 5 7 3 4

Table: Ordered Pointer Array

Node, n Successor nodes, s & (S(s,n))

6 5 (0.2572) 7 (0.4209)
5 3 (0.5898) 7 (0.5791)
7 4 (0.1199)
3 4 (0.8801)
4 0 (0.0000)

126
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Step 5:

Table: Node weighting Array & Calculation of V16 (1 d )

n/d 6 5 7 3 4

Initial Wn 1 0 0 0 0

Updated Wn 0.2572 0.4209 0.1517 0.0683

(= Ws ) 0.5698 0.2018

Th d 500 600 100 300

Wd Th d 128.6 341.9 15.2 60.5

Table: Output

Load on Link (1,6) Origin Node Destination Node

128.6 1 5
341.9 1 7
15.2 1 3
60.5 1 4

127
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Comments on the Dial, Gunnarsson and GMTU Models

GMTU model is similar to Dials in that all links included in paths have their
initial node closer than their final node to the origin node h but, in that the
probability of usage of any link is determined at the node at which the choice of
including the link arises.
In GMTU model, the division of the trips from origin node h arriving at any node
i between its predecessor links (i, j) etc is based on explicitly on the costs on
mutually exclusive sections of the minimum cost paths from h to i via these
predecessors; the probability of usage being zero for links other than those with
initial node k as the best or second-best predecessor node to i. Gunnarsson splits
the trips leaving any node i for destination node between the successor links (i,j)
on the basis of the costs on the minimum paths from i to the destination node d
via these links.
In using the multinomial logit model as the basis for the division of trips between
routes, Dials model implicitly assumes constant variance in perceived costs on all
route sections; the same is true of Gunnarssons model if the exponential form is
used for f (rij ).

Route choice probabilities in Dials method are based on numbers of routes

available, ignoring completely the fact that many routes may be identical apart
from minor diversions. Gunnarssons and GMTU make choices of links to be
used at the nodes at which the choice arises and are also sensitive to the network
structure.

128
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

3.13 SCATA, an Arithmetical Approximation Approach to

Multiple Route Choice
If there are relatively few route choices, all of which can be explicitly enumerated,
arithmetical approximation methods of estimating route choice probabilities, and hence
link flows, are more accurate than simulation methods. Such circumstances are extremely
scarce in traffic assignment studies and hence the techniques are rarely used in practice;
however, they can be used to provide an insight into the characteristics of solutions
obtained by the simulation techniques that are not apparent from their application.

SCATA assumes a uniform distribution. Robertson and Kennedy (1979) suggested that
the assignment of traffic in a network by simulation methods is relatively insensitive to
the shape of the probability distribution assumed. SCATA, first proposed by Robertson
(1977) and then refined by Robertson and Kennedy (1979), has a composite cost structure

Simple Choice Algorithm for Trip Assignment (SCATA)

In the two route choice situation with uniform distributions of perceived link costs, the
proportions of trips choosing the dearer route, R2 , given in terms of the actual route

costs c1 , c 2 and spread factor, k , may be written in terms of c1 and c 2 and the ranges
of the perceived cost distributions, r1 and r2 , as:

Pr ( R 2 ) = (c1 c 2 + (r1 + r2 ) / 2) 2 /(2r1 r2 )

---------------------------------------------(11)

Assuming that the perceived costs on competing routes are independent of one another,
the variance in the perceived cost on R1 relative to R2 , called the relative variance, is:

Var1, 2 = Var1 + Var2 ----------------------------------------------------------------------(12)

where Var1 and Var2 are variances of perceived costs on R1 and R2 respectively.

129
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

As approximation, it may be assumed that this relative variance is divided equally

between the two routes, so that the variance of the perceived cost on each link is:

Var = 0.5(Var1 + Var2 ) ------------------------------------------------------------------(13)

Assuming that the common range of the distribution is r then, since Var = r 2 / 12,

r 2 = 6(Var1 + Var2 ) = 6Var1, 2 -----------------------------------------------------------(14)

Substituting r from the above equation for ?

Pr ( R2 ) 0.5(1 (c 2 c1 ) / (6Var1, 2 )) 2 -----------------------------------------------(15)

Note: Pr ( R2 ) = 0 if the term in parenthesis is negative. This facilitates the elimination of

unused routes.

Choice between three or more parallel routes

The calculation can be simplified by converting the single problem in n dimensions to (n-
1) problems in two dimensions. This is achieved by a procedure which enables a pair of
parallel links to be replaced by an equivalent single link. Values are calculated for the
the reduced cost and reduced variance, the mean and variance respectively of the
distribution of perceived costs on the equivalent link, so that the traffic assignment in the
remainder of the network is unaffected by the substitution.

Reduced Cost
Assuming c1 0.5r1 c 2 0.5r2 c1 + 0.5r1 c 2 + 0.5r2 , the probability density function
for the distribution of perceived costs, c , on links (routes) actually used is:

f (c) = 1 / r1 for c1 0.5r1 c c 2 0.5r2 --------------------------------------------(16)

and f (c) = (c1 + c 2 + 0.5(r1 + r2 ) 2c) /(r1 r2 ) for c 2 0.5r2 c c1 + 0.5r1

130
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

The mean perceived cost on the route actually chosen, i.e, the reduced cost, is given
approximately by:

c1+ 2 c1 {4( Pr ( R2 )) 3 Var1, 2 / 3} -----------------------------------------------------(17)

This indicates the important fact that the mean perceived cost of travel on routes actually
used through a network is less than the mean perceived cost of travel on the minimum
cost path.

Reduced Variance
Robertson and Kennedy (1979) gave the following approximation on the reduced
variance:

Var1+ 2 (1 1.3Pr ( R2 ))Var1 + Pr ( R2 ) Pr ( R2 ) + 0.2)Var2 ----------------------------(18)

From the above equation, if Pr ( R2 ) = 0.5,Var1+ 2 = 0.35(Var1 + Var2 )

In general, the variance of perceived costs on routes actually used is also less than the
variance of perceived costs on the minimum path.

Example 1 :
1000 vehicles per hour travel from h to d between which the three possible routes R1 , R2
and R3 , with no overlapping sections. The distributions of perceived costs on the routes

are assumed to be uniform with means, c, of 1.0, 1.2, and 1.4 equiv. Mins on R1 , R2 and

R3 respectively, and ranges 2 k (c ) where k = 1.2 when c is measured in tenths of

equiv. Mins. Estimate the mean and variance of the reduced costs on the routes actually
used and the flow of vehicles on each route.
In the following, costs are in units of tenths of equiv.mins.

Since Var (c) = (range) 2 / 12 = (2k (c )) 2 / 12 = 0.48c, we have:

c1 = 10;Var1 = 4.8; c 2 = 12;Var2 = 5.76; c3 = 14;Var3 = 6.72.

131
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

Consider the replacement of routes R1 and R2 by the equivalent route e ,

Var1, 2 = 4.8 + 5.76 = 10.56

Pr ( R2 ) = 0.5{1 (12 10) / (6(10.56))}2 = 0.2803

ce = c1+ 2 = 10 {4(0.2803) 3 (10.56) / 3} = 9.443

Vare = Var1+ 2 = (1 1.3(0.2803))(4.8) + 0.2803(0.2803 + 0.2)(5.76) = 3.826

Consider the replacement of e and 3.

Vare ,3 = 3.826 + 6.72 = 10.546

Pr ( R3 ) = 0.5{1 (14 9.443) / (6(10.546))}2 = 0.0912

c e+3 =(1-1.3(0.0912))(3.8260)+0.0912(0.0912+0.2) 6.72=3.551

Thus the mean perceived cost on the routes actually used is c e +3 =0.934 equiv.mins.
The variance of the reduced costs on the routes actually used is,
c e +3 = 3.551X 10 2 (equiv. min s ) 2
The flow on R3 is 1000x0.0912=91.2,

The flow on R2 is 10000(1-0.0912)(0.2803)=254.7,

The flow on R1 is 1000(1-0,0912)(1-0.2803)=654.1veh/h.

Table: The Effect of the Order of Combing Routes when Traffic is Assigned to
three routes in parallel (Example 7)

Proportion of Order of combing Routes Max.

Traffic Routes (1+2)+3 (1+3)+2 (2+3)+1 Deviation from
R1 , R2 and R3 Average (%)

Pr ( R1 ) 0.6541 0.6429 0.6635 1.6

Pr ( R2 ) 0.2547 0.2571 0.2370 5.0

Pr ( R3 ) 0.0912 0.1000 0.0995 9.4

Reduced Cost 9.340 9.330 9.312 0.2

132
CE -751, SLD, Class Notes, Fall 2006, IIT Bombay

In Langdon (1982), it is suggested that if it cannot be assumed that the introduction of an

additional route (mode in fact) does not change the relative proportions of trips on the
other routes, then it is necessary to carry out the calculations three times with each route
in its turn becoming the last one introduced.

133