CE5205 – Trip Distribution

Associate Professor Lee Der‐Horng, PhD

Department of Civil Engineering
National University of Singapore

Trip Distribution Objectives

• Replicate spatial pattern of trip making
• Account for spatial separation among
origins and destinations (proximity in
terms of time, cost, & other factors)
• Account for attractiveness among TAZs
• Reflect human behavior

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 1

1
 Land use & socio economic
target-year allocation by zone

Calibrated trip generation

model

Total number of trips

produced/attracted

Inputs & outputs of trip generation

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 2

Trip-end estimates from

trip generation

Target-year
estimates of Calibrated trip
interzonal distribution
impedances dij model

Estimated target-year Tij

Inputs & outputs of trip distribution

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 3

2
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 4

Trip Distribution (TD)

• Purpose: to estimate the number of trips (Tij)
performed from origin zones to destination
zones, given aggregated trip numbers from
the previous step (TG).
• Techniques:
– Growth‐factor methods, Fratar method
– Gravity
G i models d l
– Intervening opportunities method

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 5

3
 Trip Distribution
• Convert Production and Attraction Tables into
Origin ‐ Destination (O ‐ D) Matrices

Destinations
TAZ P A
Origins 1 2 3 4 5 6 Sum
1 15 22
1 T11 T12 T13 T14 T15 T16 O1
2 13 6
2 T21 T22 T23 T24 T25 T26 O2
3 26 5
3 T31 T32 T33 T34 T35 T36 O3
4 18 52
4 T41 T42 T43 T44 T45 T46 O4
5 8 2
5 T51 T52 T53 T54 T55 T56 O5
6 13 6
6 T61 T62 T63 T64 T65 T66 O6
Sum 93 93
Sum D1 D2 D3 D4 D5 D6

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 6

Crude approximation for HBW

Crude OD
Production Attraction Calculation
Zone ID P A O=D=(P+A)/2
1 15 22 18.5
2 13 6 9.5
3 26 5 15.5
4 18 52 35
5 8 2 5
6 13 6 9.5
Total 93 93 93

4
 Trip Distribution, Methodology
• General Equation:
Tij = Ti P(Tj)
• Tij = calculated trips from zone i to zone j
• Ti = total trips originating at zone i
• P(Tj) = probability measure that trips will be attracted
to zone j
• Constraints:
• Singly Constrained
Sumi Tij = Dj OR Sumj Tij = Oi
• Doubly Constrained
Sumi Tij = Dj AND Sumj Tij = Oi

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 8

Trip Distribution Models

• Growth Factor / Fratar Model
Tij = Ti (Tj / T)
• Tij = present trips from zone i to zone j
• Ti = total trips originating at zone i
• Tj = total trips ending at zone j
• T = total trips in the entire study
Tij* = Tij (Fi Fj) / F
• Fi = Ti* / Ti
• Fj = Tj* / Tj
• F = T* / T
• * = estimated future trips
• Weak in areas experiencing land use changes
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 9

5
 Trip Distribution Models
• Gravity Model You can consider this as the
a probability spatial distribution
Aj / C
ij
P(Tj)
Tij = Ti a
Sum (Ax / Cix)

• Tij = trips from zone i to zone j

• Ti = total trips originating at zone i
• Aj = attraction factor at j
• Ax = attraction factor at any zone x
• Cij = travel friction from i to j expressed as a generalized
cost function
• Cix = travel friction from i to any zone x expressed as a
generalized cost function
• a = friction exponent or restraining influence
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 10

Trip Distribution Models

• Intervening Opportunities Model, what does
it mean??
– the probability of choice of a particular destination
(from a given origin for a particular trip purpose)
is proportional to the opportunities for trip
purpose satisfaction at the destination and
i
inversely
l proportional
ti l tto allll such
h opportunities
t iti
that are closer to the origin.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 11

6
 Trip Distribution Models
• Intervening Opportunities Model, what does
it mean??
– The basic idea behind the intervening‐
opportunities model is that trip making is not
explicitly related to distance (or cost) but to the
relative accessibility of opportunities for satisfying
th objective
the bj ti off the
th trip.
ti

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 12

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 13

7
 Trip Distribution Models
• Intervening Opportunities Model

Tij = Ti (e -LT ‐ e -LT(T + Tj) )

• Tij = trips from zone i to zone j
• T = trip destination opportunities closer in time to zone
i than those in zone j
• Ti = trip end opportunities in zone i
• Tj = trip end opportunities in zone j
• L = probability that any destination opportunity will be
chosen

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 14

Growth Factor Methods

• Similar than the growth method used in trip
generation
ti
• Assuming there is a trip matrix for base year,
with the growth rate to be expected in the
target year, the trip matrix of target year can
be estimated

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 15

8
 Growth Factor Methods
• Type of growth factor methods
– Uniform growth method
– Single constrained growth factor method
– Double constrained growth factor method

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 16

Uniform Growth Factor

• If the only information available is about a
generall growth t  for
th rate f the
th study
t d area, it
can be assumed that this factor will apply to
each cell in the trip matrix:
Tij =  tij for all i and j

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 17

9
 Single constrained growth factor
method
• Consider the situation where information
available
il bl on the
th expected
t d growthth iin ttrips
i
originating in each zone, for example,
shopping trips.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 18

Single constrained growth factor

method (cont’d)
• It will be possible to apply ‘origin specific’ (i )
t th
to the corresponding
di rows iin th
the ttrip
i matrix.
ti

Tij = i tij for origin (i) specific

factors

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 19

10
 Single constrained growth factor
method (cont’d)
j Sum Target
1 2 3 4
i over j Oi

1 5 50 100 200 355 400

2 50 5 100 300 455 460

3 50 100 5 100 255 400

4 100 200 250 20 570 702

Sum
205 355 455 620 1635 1962
over i
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 20

Single constrained growth factor

method (cont’d)
• Each row in the last trip matrix be multiplied
b a ratio
by ti off ttargett year Oi over the
th base
b year
total sum over j.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 21

11
 Single constrained growth factor
method (cont’d)
j Sum Target
1 2 3 4
i over j Oi

1 5.6 56.3 112.7 225.4 400 400

2 50.5 5.1 101.1 303.3 460 460

3 78.4 156.9 7.8 156.9 400 400

4 123.2 246.3 307.9 24.6 702 702

Sum of
257.7 464.6 529.5 701.2 1962 1962
i
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 22

Single constrained growth factor

method (cont’d)
• In a similar way, it will be possible to apply
‘d ti ti specific’
‘destination ifi ’ (j ) to
t the
th
corresponding rows in the trip matrix.

Tij = j tij for destination (j)

specific factors

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 23

12
 Single constrained growth factor
method
• When information is available on the future
number
b off ttrips
i originating
i i ti andd tterminating
i ti in i
each zone, it implies:
– Different growth rates for trips in and out of each
zone
– Therefore, having two sets of growth factors for
each zone, for example, i and j

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 24

Single constrained growth factor

method (cont’d)
• The application of an ‘average’ growth factor,
f example,
for 0 5 (i +j ) iis a poor
l Fij = 0.5
compromise as none of the two targets (at
origins and at destinations) will be satisfied.
• Some iterative methods have been proposed
to do the estimation.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 25

13
 Advantages of Growth Factor
Methods
• Simple
• Make direct use of observed trip matrices and
forecasts of trip‐end growth
• Preserve as much as is consistent with the
information available on growth rates

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 26

Limitations of Growth Factor

Methods
• Only reasonable for short‐term planning horizons
• They are heavily dependent on the accuracy of base
year matrix.
• They do not take into account changes in transport
costs due to improvements (or congestion) in the
network.
• Trip purposes are usually ignored.
• They are of limited use.
• What to do with zero cells in Furness Method?

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 27

14
 Furness Method
• Balancing factors, Ai and Bj
Tij = tij i j Ai Bj

• Or incorporating growth rates into new

variables ai and bj:
Tij = tij ai bj
where ai = i Ai
bj = j Bj

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 28

Furness Method (cont’d)

• The factors, Ai and Bj or ai and bj must be
calculated
l l t d so that
th t th
the following
f ll i constraints
t i t
can be satisfied:

T
j
ij  Oi

T
i
ij  Dj
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 29

15
 Furness Method (cont’d)

The iterative process:

Step 1 Set all bj =1.0 and solve for ai which

means to find the correction factors ai that
satisfy the trip generation constraints.
Step 2 With the latest ai , solve for bj , e.g., satisfy
the trip attraction constraints
constraints.
Step 3 Keep the bj’s fixed, solve for ai and repeat
Steps 2 and 3 until the changes are
sufficiently small.
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 30

Furness Method (cont’d)

In plain English…
1. The base year matrix cells for one row are multiplied
by the growth for that zone and all rows are done in
turn. The matrix so obtained will have its origin trip
ends matching the future year origin trip ends which
is what we wanted to achieve, however the column
totals will not in general match the future year
destination tripp end,, so:

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 31

16
 Furness Method (cont’d)

In plain English…
2. The matrix cells for each column are multiplied by the
ratio of the future year destination trip end to the
column total achieved in 1 above so that the resulting
matrix will have its column total matching its future
year destination trip end.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 32

Furness Method (cont’d)

In plain English…
3. However its row total will not generally match its
future year origin trip end so Steps 1 and 2 are
repeated successively until the row and column total
are both close to the future year origin and
destination trip ends. The process stops when they
are close enough (e.g., to within a few trips).

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 33

17
 j Sum Target
1 2 3 4
i over j Oi

1 5 50 100 200 355 400

2 50 5 100 300 455 460
3 50 100 5 100 255 400
4 100 200 250 20 570 702
Sum
205 355 455 620 1635
over i
Target
260 400 500 802 1962
Dj
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 34

Furness Method (cont’d)

• The solution to this example, after 3 iterations
on rows andd columns
l (three
(th sets
t off
corrections for all rows and three for all
columns), will be:

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 35

18
 j Sum Target
1 2 3 4
i over j Oi

1 400
2 460
3 400
4 702
Sum
1962
over i
Target
260 400 500 802 1962
Dj
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 36

Weaknesses of Furness
• If a cell in the matrix is zero, no matter how much it
is factored it always remains zero. Quite often a zone
would have few trips in the base year because it
does not have many people living there, nor jobs nor
shops etc so not many trips come from or go to
there. Whereas in the future the zone may be fully
developed with houses, shops, factories etc. No
matter how many trips are forecast to originate or
be destined for it, if most of the zone's cells are zero
in the base year, with Furness, they will remain zero
in the future.
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 37

19
 Weaknesses of Furness
• One method for getting round this is to 'seed' all the
zero cells with a value (e
(e.g.,
g one trip
trip, or to assume a
distribution of trips from the zone in question to
every other zone and from every other zone to the
zone in question ).
• The resulting matrix for this zone is therefore
dependent on the input assumptions and it puts the
responsibility on the modeler to get the seeding
right.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 38

Weaknesses of Furness
• The second weakness of the Furness is that it is not
sensitive to changes in the transport system
system. It is
known that if the transport system is improved
people will adjust their choice of destination to make
the most of additional destination opportunities
which have suddenly become much more accessible.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 39

20
 Weaknesses of Furness
• For example if a new road is built which now
connects you to a superstore
superstore, you are more likely to
use the new superstore.
• This second weakness is more intractable and one
generally adopts a more sophisticated model such as
the gravity model to overcome this weakness.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 40

Gravity Models
• Gravity models are based on Newton’s law
of gravitation
gravitation.
• Gravity models estimate trips for each cell
in the trip matrix without directly using the
observed trip pattern. That’s why they are
sometimes called synthetic models.
• Gravity
G it models d l ill
illustrate
t t ththe macroscopic
i
relationships between places (e.g., homes
and workplaces).
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 41

21
 Gravity Models (cont’d)
• The number of trips performed between O
andd D are directly
di tl proportional
ti l to
t the
th relative
l ti
attraction of each zone and inversely
proportional to some function of the spatial
separation between the two zones.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 44

The Function Form

 Pi Pj
Tij  c
(1)
d ij
where
Pi trip production of origin i
Pj trip
p attraction of destination j
dij impedance between i and j
c parameters (constant)
 proportionality factor
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 45

22
 Gravity Models (cont’d)
• The trip‐production balance constraint:

Pi   Tix (2)
x

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 46

Gravity Models (cont’d)

• Eq. (2) states that the summation over all trip‐
attracting
tt ti zones j off the th interchange
i t h volumes
l
that share i as the trip‐producing zone must
equal the total productions of zone i.
• Eq. (2) ensures the model will distribute to
competing
p g zones j ((x)) exactlyy as manyy as are
produced by zone i.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 47

23
 Gravity Models (cont’d)
• Substituting Eq. (1) into (2) and taking the
t
terms nott involving
i l i theth index
i d x outside
t id the
th
summation:

Px
Pi   Pi  (3)
x d ixc

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 48

Gravity Models (cont’d)

• Solving for  :
-1
 Px 
   c  (4)
 x d ix 
Eq. (4) is the expression for  that ensures
• Eq
that the trip balance Eq. (2) is satisfied.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 49

24
 Gravity Models (cont’d)
• Substituting Eq. (4) into (1) leads to the
classical
l i l form
f off gravity
it model:
d l

 c 

Tij  Pi 
Pj d ij 
(5)
 P dc 
 x ix 
 x

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 50

Gravity Models (cont’d)

• The bracket in Eq. (5) is the proportion of the
trips produced by zone i that will be attracted by
zone j in competition with all trip‐attracting
zones x.

 c 
 Pj dij 
 P dc 
 
x
x ix

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 51

25
 Gravity Model (cont’d)
• The estimated target‐year trip attractions of a
t d as Pj* ) may b
zone j (denoted
(d be computed
t db
by
applying the following trip‐attraction balance
equation to the results of the model:

Pj*   Txj (6)

x
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 52

Gravity Model (cont’d)

• The gravity formula is often written
alternatively
lt ti l as:
 
Tij  Pi  
P f
j ij
(7)
 Pf 
 
x
x ix

1
where f ij  c
d ij
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 53

26
 Gravity Model (cont’d)
• Finally, a set of socioeconomic factors kij can
b iintroduced:
be t d d

 
 Pj f ij kij 
Tij  Pi    Pi pij (8)

Pj f ij kij

 j 

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 54

Gravity Models (cont’d)

• In Eq. (8), the pij denotes the probability that a
trip generated by zone i will be attracted by zone
j.

 
 Pj f ij kij 
Tij  Pi    Pi pij ((8))

Pj f ij kij

 j 

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 55

27
 Generalization

Tij   Oi D j f (cij ) (9)

• f(cij) is a generalized function of travel cost with
one or more parameters for calibration.
• f(cij) is usually referred as deterrence function
because it represents the disincentive to travel as
distance (or time, or cost) increases.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 56

Generalization (cont’d)

f (cij )  exp (- cij ) exponential function

f (cij )  c ij n power function
f (cij )  c ijn exp (- cij ) combined function

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 57

28
 Classical version of double
constrained gravity model
• To ensure trip conservation constraints, the
proportionality factor  is replaced by two
sets of balancing factors Ai and Bj as in the
Furness model:

Tij  Ai Oi B j D j f (cij )
Tij  ai b j f (cij )
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 58

Single constrained gravity model

• Single constrained gravity model, either
origin or destination constrained, can be
produced by making one set of balancing
factors Ai or Bj equal to one.
• For an origin‐constrained model, Bj equal to
one for all j,j and

Ai  1/  D j f (cij )
j
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 59

29
 Double constrained gravity model
• In the case of double constrained model
th values
the l off th
the balancing
b l i factors
f t are:

Ai  1/  B j D j f (cij )
j

B j  1/  Ai Oi f (cij )
i

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 60

Balancing factors
• Ai and Bj are independent.
• An iterative process similar to Furness’s can
be adopted: given set of values of deterrence
function f(cij), start with all Bj = 1, solve for Ai
and then use these values to re‐estimate the
Bj’s,, repeat
p until convergence
g is reached.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 61

30
 Gravity Model Process
1. Create Shortest Path Matrix
‐ Minimize link cost between centroids
2. Estimate Friction Factor Parameters
‐ Function of trip length characteristics by trip purpose
3. Calculate Friction Factor Matrix
4. Convert Productions and Attractions to Origins and
Destinations
5. Calculate O‐D Matrix
6. Enforce Constraints on O‐D Matrix
‐ Iterate between enforcing total origins and destinations

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 62

Shortest Path Matrix

• Matrix of Minimum Generalized Cost from any
Zone i to any Zone j
– Distance, time, monetary cost, waiting time, transfer
time, etc.. may be used in generalized cost
– Time or distance often used
– Matrix not necessarily TAZ TAZ ID
symmetric (effect of ID 1 2 3 4 5 6
1 C11 C12 C13 C14 C15 C16
one ‐ way streets)) 2 C21 C22 C23 C24 C25 C26
3 C31 C32 C33 C34 C35 C36
4 C41 C42 C43 C44 C45 C46
5 C51 C52 C53 C54 C55 C56
6 C61 C62 C63 C64 C65 C66
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 63

31
 Example: travel time matrix for 6 TAZs

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 64

Friction Factor Models

• Exponential:
f(cij) = e- c (cij) c>0

• Inverse Power:
f(cij) = cij- b b>0

• Gamma:
f(cij) = a cij- b e- c (cij) a > 0, b > 0, c > 0

Trip Purpose a b c
HBW 28507 0.020 0.123
HBP 139173 1.285 0.094
NHB 219113 1.332 0.010
ref. NCHRP 365 / TransCAD UTPS Manual pg. 80

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 65

32
 Example friction factors using
travel times alone

Zone ID 1 2 3 4 5 6
1 1.349859 1.822119 2.459603 3.320117 4.481689 6.049647
2 1.822119 1.349859 1.568312 1.822119 2.459603 3.320117
3 2.459603 2.459603 1.822119 1.822119 1.822119 2.459603
4 3.320117 3.320117 2.459603 1.349859 1.822119 2.459603
5 4.481689 4.481689 3.320117 2.459603 1.349859 1.822119
6 6.049647 6.049647 4.481689 3.320117 3.320117 1.349859

Friction ij = 1/exp(‐0.03* Timeij)

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 66

Friction Factor Matrices

• Matrix of friction from any zone i to any zone
j, by trip purpose
– Each cell of a friction factor matrix is a function of the
corresponding cell of the shortest path matrix
– Each trip purpose has a separate friction factor matrix
because trip making TAZ TAZ ID
behavior changes for ID 1 2 3 4 5 6
1 F11 F12 F13 F14 F15 F16
each trip purpose 2 F21 F22 F23 F24 F25 F26
3 F31 F32 F33 F34 F35 F36
4 F41 F42 F43 F44 F45 F46
5 F51 F52 F53 F54 F55 F56
6 F61 F62 F63 F64 F65 F66
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 67

33
 Trip Conversion (Approximate)
• Home Based Trips: Non ‐ Home Based Trips:
Oi = (Pi + Ai) / 2 Oi = Pi
Di = (Pi + Ai) / 2 D i = Ai
• Oi = origins in zone i (by trip purpose)
• Di = destinations in zone i (by trip purpose)
• Pi = productions in zone i (by trip purpose)
• Ai = attractions in zone i (by trip purpose)

• Note: This only works for a 24 hour time period

• If our models are for one period in a day we
prefer to work directly with Origins‐Destinations
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 68

O‐D Matrix Calculation

• Calculate Initial Matrix By Gravity Equation, by
Trip Purpose
– Each cell has a different friction, found in the
corresponding cell of the friction factor matrix
• Enforce Constraints in Iterative Process
– Sum of trips in row i must equal origins of TAZ i
• If not equal, trips are adjusted proportionally
– Sum of trips in column j must equal destinations of
TAZ j
• If not equal, trips are adjusted proportionally
– Iterate until no adjustments required

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 69

34
 O‐D Matrix Calculation
Destinations
Origins 1 2 3 4 5 6 Sum
1 T11 T12 T13 T14 T15 T16 O1
2 T21 T22 T23 T24 T25 T26 O2
3 T31 T32 T33 T34 T35 T36 O3
4 T41 T42 T43 T44 T45 T46 O4
5 T51 T52 T53 T54 T55 T56 O5
6 T61 T62 T63 T64 T65 T66 O6
Sum D1 D2 D3 D4 D5 D6

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 70

Attraction/friction matrix
Aj/friction factor Table

Zone ID 1 2 3 4 5 6 Total
1 16.298 3.293 2.033 15.662 0.446 0.992 38.724
2 12.074 4.445 3.188 28.538 0.813 1.807 50.865
3 8.945 2.439 2.744 28.538 1.098 2.439 46.203
4 6.626 1.807 2.033 38.523 1.098 2.439 52.526
5 4.909 1.339 1.506 21.142 1.482 3.293 33.670
6 3.637 0.992 1.116 15.662 0.602 4.445 26.453

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 71

35
 Gravity model probability
Gravity Model Probability
Zone ID 1 2 3 4 5 6 Total
1 0 421
0.421 0.085
0 085 0.052
0 052 0.404
0 404 0.012
0 012 0.026
0 026 1
2 0.237 0.087 0.063 0.561 0.016 0.036 1
3 0.194 0.053 0.059 0.618 0.024 0.053 1
4 0.126 0.034 0.039 0.733 0.021 0.046 1
5 0.146 0.040 0.045 0.628 0.044 0.098 1
6 0.137 0.037 0.042 0.592 0.023 0.168 1

Aj / frictionij
Sum (Ax / frictionix)

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 72

Trip Interchange ‐ iteration 1

TIJ iteration 1 Target
Zone ID 1 2 3 4 5 6 Total Oi
1 7.8 1.6 1.0 7.5 0.2 0.5 18.5 18.5
2 2.3 0.8 0.6 5.3 0.2 0.3 9.5 9.5
3 3.0 0.8 0.9 9.6 0.4 0.8 15.5 15.5
4 4.4 1.2 1.4 25.7 0.7 1.6 35 35
5 0.7 0.2 0.2 3.1 0.2 0.5 5 5
6 1.3 0.4 0.4 5.6 0.2 1.6 9.5 9.5
Total 19.5 5.0 4.5 56.8 1.9 5.3 93
Dj 18.5 9.5 15.5 35 5 9.5
Ratio 0.949099 1.907311 3.47064 0.615986 2.630148 1.77888

For each cell value we apply the gravity equation once ‐ in this
iteration ‐ after this we use the ratio to adjust the values in the
cells ‐ until row and column targets are satisfied

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 73

36
 Trip interchange ‐ iteration 2
TIJ iteration 2
Zone ID 1 2 3 4 5 6 Total Oi Ratio
1 7.4
7 4 3.0
3 0 3.4
3 4 4.6
4 6 0.6
0 6 0.8
0 8 19 8
19.8 18.5
18 5 0 9355882
0.9355882
2 2.1 1.6 2.1 3.3 0.4 0.6 10.1 9.5 0.9430955
3 2.8 1.6 3.2 5.9 1.0 1.5 15.9 15.5 0.9732926
4 4.2 2.3 4.7 15.8 1.9 2.9 31.8 35 1.1000923
5 0.7 0.4 0.8 1.9 0.6 0.9 5.2 5 0.956075
6 1.2 0.7 1.4 3.5 0.6 2.8 10.2 9.5 0.9329658
Total 18.5 9.5 15.5 35 5 9.5 93

Using the ratios from before we succeed in getting the targets for the
sums off cells
ll ffor each
h column
l - look
l k att the
th other
th ratios
ti

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 74

Trip interchange ‐ iteration 3

TIJ iteration 3
Z
Zone ID 1 2 3 4 5 6 T
Total
t l
1 6.9 2.8 3.2 4.3 0.5 0.8 18.5
2 2.0 1.5 1.9 3.1 0.4 0.6 9.5
3 2.8 1.5 3.1 5.7 0.9 1.4 15.5
4 4.6 2.5 5.2 17.4 2.1 3.2 35
5 0.7 0.4 0.7 1.8 0.6 0.8 5
6 1.2 0.6 1.3 3.2 0.5 2.6 9.5
Total 18 1
18.1 93
9.3 15 4
15.4 35 6
35.6 50
5.0 94
9.4 93
Ratio 1.020289 1.016843 1.004978 0.982477 0.991224 1.007046

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 75

37
 Trip interchange ‐ iteration 4
TIJ iteration 4
Zone ID 1 2 3 4 5 6 Total Oi Ratio
1 7.1 2.9 3.2 4.2 0.5 0.8 18.6 18.5 0.993094
2 2.1 1.5 2.0 3.0 0.4 0.6 9.5 9.5 0.997665
3 2.8 1.5 3.1 5.6 0.9 1.4 15.5 15.5 1.000101
4 4.7 2.6 5.2 17.1 2.1 3.2 34.9 35 1.003991
5 0.7 0.4 0.7 1.8 0.5 0.8 5.0 5 1.001637
6 1.2 0.6 1.3 3.2 0.5 2.7 9.5 9.5 1.000214
Total 18.5 9.5 15.5 35 5 9.5 93

We get both rows and columns to produce the sums we want!

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 76

Example 1
• Perform an application of the basic gravity model for
trips between a proposed site (attractor) and three
surrounding zones. This site receives 1,524 incoming
trips during morning peak. Assume that all incoming
trips are from the surrounding zones. Average travel
times from each zone to the site as well as the
population of each zone are given in the figure next
page Estimate the number of trips from each zone
page.
to the site.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 77

38
 Zone 1 Zone 2
Population 3000 Population 4500

ts1 =12 ts2 =16

Site

ts3 =15

Zone 3
Population 7500

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 78

Solution:

(1524)(3000)
1 
G1s  381,000 
12

(1524)(4500)
G2 s    428,625 
16

(1524)(7500)
G3 s    762,000 
15
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 79

39
 The sum of all “gravities” should be equal to the
number of trips destined to the site. Thus,

G1s  G2 s  G3 s  1524
which yields   0.0009697, thus
G1s  369
G2 s  416
G3 s  739
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 80

Zone 1 Zone 2

369 416
Site

739
Zone 3

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 81

40
 Example 2
• The target year productions and relative
attractiveness
tt ti off a 4‐zone
4 city
it hhas b
been
estimated to be as follows:

Zone Productions Attractiveness

1 1500
500 0
2 0 3
3 2600 2
4 0 5
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 82

Example 2 (cont’d)
• The calibration of the gravity model for this city
estimated the parameter c to be 2 2.0
0 and all
socioeconomic adjustment factors to be 1.0. Apply
the gravity model to estimate all target interchanges
Tij and to estimate the total target‐year attractions
of each zone given that the target‐year interzonal
impedances
p dij will be shown in the followingg table.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 83

41
 i j 1 2 3 4

1 5 10 15 20

2 10 5 10 15

3 15 10 5 10

4 20 15 10 5

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 84

Solution: The gravity model calculations of the

Interchange volumes are shown as follows for
the two trip-producing zones (i = 1 and i =3).

j Pj f1j k1j Pj f1j k1j p1j T1j

1 0 0.0400 1 0 0 0

2 3 0.0100 1 0.0300 0.584 875

3 2 0.0044 1 0.0089 0.173 260

4 5 0.0025 1 0.0125 0.243 365

i =1, P1 =1500 0.0514 1.0 1500

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 85

42
 Solution: i =3, P3 =2600

j Pj f3j k3j Pj f3j k3j p3j T3j

1 0 0.0044 1 0 0 0

2 3 0.0100 1 0.03 0.188 488

3 2 0.0400 1 0.08 0.500 1300

4 5 0.0100 1 0.05 0.312 812

i =3, P3 =2600 0.16 1.0 2600

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 86

To find the total target-year trip attractions of

the non-residential zones (j =2, 3, and 4), apply
the trip-attraction balance (Eq. 6) to get:

P2*  875  488  1363

P3*  260  1300  1560
P4*  365  812  1177
Pj*   Txj (6)
x
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 87

43
 The solution can be summarized by the
following trip table:

i j 1 2 3 4 Sum

1 0 875 260 375 1500

2 0 0 0 0 0

3 0 488 1300 812 2600

4 0 0 0 0 0

Sum 0 1363 1560 1177 4100

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 88

Entropy Maximization Method

• Originally developed in statistical mechanics,
it iis a ‘‘general’l’ method
th d
• Entropy is the degree of likelihood of the final
state of a system.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 89

44
 Derivation

3 1 1 5 4 0 1 5
5 2 1 8 3 2 3 8
2 2 4 8 3 3 2 8
10 5 6 21 10 5 6 21

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 90

Derivation (cont’d)
• Assume that Oi and Dj are known, there exist
many possible trip matrices that add up to the
number of Os and Ds.
• The last two tables are only two of a large
number of possible Tij .
• Each Tij represents a macro‐state that can be
produced
d d with ith a large
l number
b off micro‐states.
i t t

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 91

45
 Derivation (cont’d)
• The problem is to find the Tij that is associated
with
ith th
the maximum
i number
b off micro‐states,
i t t
assuming a priori that they are all equally
likely.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 92

Derivation (cont’d)
• Consider the following constraints:

T j
ij  Oi (10)

T i
ij  Dj ((11))

T i j ij cij  C (12)

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 93

46
Defining T as the total number of trips :
T   Tij , the total number of combinations
ij

that can give rise to Tij can be calculated as

follows : first select T11 from T , then T12 from
T  T11 ; and so on. So that Eq. (13) is as follows :
T! (T  T11 )! T!
W ... 
T11 (T  T11 )! T12 (T  T11  T12 )!  ij Tij !

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 94

The Tij which maximizes W subject to Eqs (10) - (13)

can be obtained with the standard method for
constrained maximization. The Lagrangean :
L  ln W    i' (Oi   Tij )    "j ( D j   Tij ) 
i j j i

  (C   T c
j ij
ij ij ) (14)

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 95

47
  i' and  "j are " Lagrangian multiplers. The
Tijj s which maxmize L and which are therefore
the most probable distribution of trips are the
solution of :
L
  ln Tij   i'   "j   cij  0 (15)
Tij
Therefore, Tij  exp( i'   "j   cij ) (16)

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 96

Substituting Eq. (16) into Eqs. (10) and (11) :

exp( i' )  Oi  exp(
j
"
j  cij ) 
1
(17)

exp(  "j )  D j  exp(   c )

i i
'
ij
1
(18)

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 97

48
 For convenienc e, we define :
Ai  exp(  i' ) / Oi (19)
B j  exp(  "j ) / D j (20)
So, Tij  Ai Oi B j D j exp(   cij )  0 (21)
Eq. (21) is actually t he classical gravity model.
The value of balancing factors can be derived :
Ai   B D exp(   c )
j j j ij
1
(22)

Bj   A O exp(   c ) 
i i i ij
1
(23)

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 98

Derivation (cont’d)
• Eqs. (21)‐(23) represent the entropy
maximizing
i i i spatial
ti l interaction
i t ti model
d l in
i its
it
most general form (Wilson, 1970, 1974).

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 99

49
 Properties of Gravity Model
• A more rigorous representation
• Tools available in mathematical programming
• Physical interpretation
• match the assumption

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 100

Calibration
• Parameters: Ai , Bj ,  , etc.
• Calibration for linear regression model can be
solved by a relatively easy minimization of the
sum of squared derivatives.
• Calibration of gravity model is an iterative
process
process.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 101

50
 Calibration (cont’d)
• Step 1: Set initial parameter values
• Step 2: Assessed by best known base year
data
• Step 3: Compare computed and observed
values (trip‐length frequency distribution)
• Step 4:
4 Convergence

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 102

Model Comparison
Model Advantages Disadvantages
Growth Factor Simple Does not reflect changes in the
Easy to balance origin and frictions between zones
destination trips at any Does not reflect changes in the
zone network
Gravity Specific account of friction Requires extensive calibration
and interaction between Long iterative process
zones
Intervening - Does not require origin - Accounts for only relative
Opportunities destination data changes in time - distance
Claimed to bear a better relationship between zones
“fit” to actual traffic Arbitrary choice of
probability factor
New: Destination choice models build on intervening opportunities

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 103

51
 Practical Considerations
• Sparse matrices
– An attribute usually found in realistic matrices
– Likely to cause problem in matrix operation and
computer implementation
– Fail to capture the truth when it is being
expanded

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 104

Practical Considerations (cont’d)

• Treatment of external zones

– Synthetic model is accepted as a suitable
approach in a study area.
– Synthetic model is sensitive to distance‐based
measures, thus it is debatable if external trips
are considered.
– In practice, external trips are separated from
the synthetic modeling process and are
forecasted by method such as growth factor.
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 105

52
 Practical Considerations (cont’d)

• Intra‐zonal
Intra zonal trips
– In practice, intra‐zonal trips are taken out from
considerations.
– Intra‐zonal trips are not loaded to the network
in later steps (TA).
– Require special treatment only for rather
coarse zoning systems.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 106

Practical Considerations (cont’d)

• Trip purposes
– In practice, trips to work/school are modeled
by double constrained gravity model;
shopping/social trips are modeled by single
constrained model.
– The
Th reason, accurate t ttrip
i attractions
tt ti are
difficult captured straightforward.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 107

53
 Practical Considerations (cont’d)

• Generations/Attractions;
Origins/Destination
– The synthetic model is developed under the
assumption that each trip has a generation and
an attraction end.
– Remember that the generation end for HB
trips is always the home.
– Inconsistent with the realization of O & D.
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 108

Production Attraction

Home Production Attraction Work

Home based trip

Production Attraction
Work
W Shop
S p
Attraction Production

Non-home based trip

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 109

54
 Practical Considerations (cont’d)

• Generations/Attractions;
Origins/Destination
– In practice, it is necessary to convert into
proper OD matrices for shorter period of time
(e.g., peak periods) based on single purpose
(e g to
(e.g., to‐work).
work)

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 110

Practical Considerations (cont’d)

• K Factors
– The purpose is to consider some special
associations of OD pairs. The concerned
associations may have overwritten the fact
modeled by gravity model.
– Be careful in applying K factors.

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 111

55
 K Factors
• Even after calibration, there will typically still be
discrepancies between the observed & calculated data
• To “fine‐tune”
fine‐tune the model
model, some employ socioeconomic
adjustment factors, also known as K‐Factors
– The intent is to capture special local conditions between some
zonal pairs such as the need to cross a river
1-Xi
Kij = Rij
1 - XiRij

Rij = ratio of observed to calculated Qij (or Tij)

Xi = ratio of the base-year Qij to Pi (total productions of
zone i)
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 112

K Factor Example
Observed Qij 300
Rij = = R13 = = 1.20
C l l t d Qij
Calculated 249

Base-Year Qij 300

Xi = = X1 = = 0.60
Pi 500
1-Xi 1-0.6
Kijj = Rijj = K13 = 1.20 = 1.71
1 - XiRij 1 (0 6)(1 2)
1–(0.6)(1.2)

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 113

56
 The Problem with K‐Factors
• Although K‐Factors may improve the model
in the base year
year, they assume that these
special conditions will carry over to future
years and scenarios
– This limits model sensitivity and undermines the
model’s ability to predict future travel behavior
• The need for K
K‐factors
factors often is a symptom of
other model problems.
– Additionally, the use of K‐factors makes it more
difficult to figure out the real problems
CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 114

Practical Considerations (cont’d)

• Errors in modeling
– Sampling errors
– Try to identify the source of errors (input,
processing, mis‐representation of reality)
– Statistical methods
– Simulation and activity‐based
y
– Estimation from traffic counts

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 115

57
 Practical Considerations (cont’d)
• Multiple Matrices
– For each trip purpose obtain different Origin‐
Destination Tij matrices
– Usually these are 24 hour Matrices (number of
trips from one zone to another in a 24 hour
period)
– In assignment we will need a matrix of vehicles
moving from a zone to another during a specific
period (peak usually) in a typical day

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 116

Limitations of the Gravity Model

• Too much of a reliance on K‐Factors in
calibration
• External trips and intrazonal trips cause
difficulties

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 117

58
 Limitations of the Gravity Model
• The skim table impedance factors are often
too simplistic to be realistic
– Typically based solely upon vehicle travel times
• At most, this might include tolls and parking costs
– Almost always fails to take into account how things
such as good transit and walkable neighborhoods
affect trip distribution
– No obvious connection to behavioral decision‐
making

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 118

Limitations of the Gravity Model

• The model fails to reflect the characteristics of the
individuals or households who decide which destinations
to choose in order to satisfy their activity needs
White Collar Jobs
Zone 3
Income=20000
Zone 2
Zone 1
Zone 4
Income=75000
Blue Collar Jobs
 The gravity model does not take this type of situation
into account without using K-Factors… which leads
back into another whole set of problems

CE5205 - Trip Distribution DH Lee © 2013, All rights reserved 119

59
End of This File

60