Mobility & Transport Networks

Plan de Formación (40h)

Plan de Formación (40h)
(2021)

Demand Modelling
 Contents

TRANSPORTATION DEMAND.
Introduction to UTP model

1. Information needs
2. UTP modeling process
3. Trip generation
4. Spatial distribution
5. Modal split
6. Assignment
7. Software
8. References

2
 Transportation Demand Modelling

1. PROBLEM DEFINITION
2. OBJETIVE CONSTRAINTS AND STUDY METHODOLOGY
3. DEFINITION OF THE STUDY AREA, ZONING AND
TRAMIFICATION OF THE ROAD/INFRASTRUCTURE
4. AVAILABLE INFORMATION AND DATA GATHERING
5. FIELDWORK
6. METHODOLOGY FOR DEMAND MODELING
7. TRIP GENERATION AND ATTRACTION
UTP MODEL

8. TRIP DISTRIBUTION. TRIP MATRIX FOR YEAR BASE

9. MODAL SPLIT
10. TRAFFIC ASSIGNMENT TO ROAD NETWORK
11. MODEL CALIBRATION AND VALIDATION
12. FUTURE FORECAST OF TRIP MATRIX
13. FUTURE ASSIGNMENT
14. APPRAISAL

3
Information needs

4
 Information (1)
1. Current situation analysis:
• Description
• Management
• Operation
2. Monitoring and development and changing trends
• Evolution of socio-economic indicators  prevision
3. Knowledge of previous studies of future tendencies, performed
prognosis and analysis of the needs of change
4. Definition of parametric bases of the models

INFORMATION
QUALITATIVE QUANTITATIVE
BASIC SPECIFIC
5
 Information (2)

BASIC INFORMATION:  Public Administration

1. Socio-economic data Population

Motorization Index (# cars/1,000 inhabitants)
Per capita income
People who work / family

2. City planning data Population density

Number of homes in the area
Land uses

Traffic “spider webs”* (city) and Accidents - safety

3. Transportation “ladders”* (roads)
network data
Speed PT lines
Capacity Environmental impact

6
 Information (3)

Home Mobility Survey

• Phone survey (CATI) in 2006
to people >14 years old
• Sample of 100,000 people, in
Spring and Fall
• 7 million euros
• 27 questions in 3 sections:
– Trips of yesterday
– Socio-economic
characterization
– Opinion: use of transport
modes, satisfaction, reasons,
perceived costs, etc.

7
 Information (4)

SPECIFIC INFORMATION: surveys, traffic counts...

• Flows of people or goods in a section (stretcht) of the specific
network.
• Traveled distance by people, vehicles or goods and travel times.
• Traffic composition:
– Type of vehicle
– People (age, occupation)
– Purpose for travel
– O-D (origin-destination)
– Vehicle occupancy
– Distribution of travel demand by transportation mode
– Etc.

A lot of this information is obtained today automatically

thought mobile phones (bluetooth, GPS, etc.)
8
 Information (5)

SURVEYS:
“The trips done by a determined group of people are
repeated and are similar day by day”.
• EMO (Encuesta de Movilidad Obligada = Home-based mobility
REVEALED PREFFERENCES

survey). Survey of Commuting.

– Work and study. Catalunya 19812001
– Commuting today  50 % of the trips
– In general, TZ=Municipality (TZ = Transportation Zone);
large municipalities have 3-6 TZ (except for Barcelona)
• On public roads. 1 transportation mode. People reluctant
• Home: STATED PREFERENCES
– Surveys (1 day, weekly) (HYPOTHETICAL
– 2 % of the population SCENARIO )
– Memory? “Daily homework”
• Phone
• Mail (low response)
F. Robusté
9
 Information (6)
TRAFFIC COUNTS:
Evaluate traffic “intensity” and composition.
• Manually. Short periods, complete info. (Mistakes? Professional?)
• Automatic.
• Pneumatic tube
• Magnetic loop
• PeMS: Performance Measurement System (inductive loop
“signature”: it can classify vehicles)
• From vehicle
• Photographs (remember Eulerian vs Lagrangian observer)
• Video
• Bluetooth (BITCarrier): toll highways (AP-7, AP-2) in Catalonia
• GPS
F. Robusté
EVALUATION OF MESUREMENT ERROR??
10
 Information (13)

• Inherent traffic data errors:

– Sampling error
– Measurement error
– Error due to inadequate treatment process or data collection
• Specification errors
• Calibration errors
• Traffic forecast error (time)
• Transfer error (space)

ERROR  LACK OF PRECISION, RISK!!

11
Information (7)

12
 Information (8)

F. Robusté
13
 Information (9a)

F. Robusté
14
 Information (9b)

F. Robusté
15
 Information (10)

F. Robusté
16
 Information (11)

• Calculation of the sample size and error:

K= Multiplication factor of  in normal distribution corresponding to the

confidence level
Universe= Population (normally)
P= Distribution of the universe (Men/Women=50%, modal distribution for
example 30% in car)

K 2 P(1  P )Universe
Sample  2
e (Universe  1)  K 2 P(1  P )

K 2 P (1  P )(Universe  Sample)
Error 
Sample(Universe  1)

F. Robusté
17
 Information (12)
• Calculation Number of elements of a finite sample extracted from finite population:

Confidence 99.73 99 98 96 95.45 95 90 80 68.27

k 3 2,58 2,33 2,05 2 1,96 1,64 1,28 1

• Determining size of sample given error limit:

Level of confidence (%) 95,45 K 2
Population size 1000
P 50 Q 50
Sample size
Error limit (%) 5 286 (28,6%)

• Determination of error in sample given sample size:

Level of confidence (%) 95,45 K 2
Population size 10000
Error limit
P 50 Q 50
7%
Sample size 200

F. Robusté
18
 UTP process (Urban
Transportation Planning)

19
 UTP simulation process
m: purpose
UTP process: CATS, 1960 tmijkrh i: origin
j: destination
k: mode
Director = Bill Garrison r: route
(Geographer) h: period

Trip Generation Gi, Aj = f(P,R,E,S,…)

S
T Spatial distribution tij = Gi Pr(j І i )
E
P tijk = tij Pr(k І ij )
Modal choice*
S

Route assignment tijkr = tijk Pr(r І ijk )

F. Robusté
20
 Territory activities (1)

Territory

•Partition of the economic territory

•Transportation zones (TZ) and gates
– Size, information, accuracy
•Centroid (frees, or with restrictions)
•Land uses / socio-economic
characteristics / activities
F. Robusté
21
 Territory activities (2)

Area (Surface)
Population
Housing
Motorization
Employment
Shops / stores
Industry
Inter-zones trip modeling Services

22
 Zone definition (1)

• Zoning: Territory is
divided (partition) in
transportation zones
(TZ) so their socio-
economic and mobility
characteristics are
homogeneous.
• Size TZ: units of info.,
accuracy (variance
activities), computing
power (old times); less
distance= smaller TZ,
more distance = larger
TZ
23
Zone definition (2)

Zoning: 3 levels of
detail
-Region
-Metropolis
-City

24
 Zone definition (3)

F. Robusté
25
 Zone definition (4)

F. Robusté
26
 Zone definition (5)

• From the definition of

influence areas in
counties or large
“crowns”…

• … to fine-tuning
(smaller TZ) of the
area of interest
(taking into account
information units in
the municipality)

27
 Zone definition (6)

• From the municipality …

• … it is refined in the area

of interest zooming down
to the definition of
districts or
neighborhoods

28
 Zone definition (7)
• From the definition of districts/neighborhoods… it is refined in
the area of interest to de definition of transportation zones as
a grouping of census tracts

F. Robusté
29
Zone definition (8)
Interior zoning
in Barcelona

30
 Accessibility network (1)

•From transportation networks to graphs

–Nodes (decisions), archs
–Connectors (centroids-nodes)
•Archs with characteristics that condicionate
behavior (capacity, lenght, speed, flow, ramp/slope,
etc.)

31
 Accessibility network (2)

TRANSPORTATION
NETWORK

STUDY AREA

CONNECTOR

CENTROID

DOOR

32
Accessibility network (3)

33
 Accessibility network (4)

TRANSPORTATION
NETWORK
• Road corridors
through France –
Iberian Peninsula-
Maghreb for the
transport of goods

34
 Accessibility network (5)

TRANSPORTATION NETWORK
GRAPH

• Nodes: Trip Generators main

points (industrial centers) and
intersections between
corridors.
 Codification
• Links: Main corridors of
transportation and connectors
of industrial centers to the
general network .

F. Robusté
35
 Simulation (1)

•Traffic web
•Demand (flows)
•Level of service
•Time (speed)
•Consumption
•Emissions
•Accidents
•Etc.

F. Robusté
36
 Simulation (2)

ASSIGNMENT TO ROAD
NETWORK

• Determination of flow of
goods in each corridor.

37
Simulation (3)

38
 Simulation (4)

•Details of spins/turns
•O/D matrix in links

39
 Calibration (1)

y = f( , x) Causal Model y0 = f( , x0)  ^

0

Supply modeling Demand modeling

GRAPH: LAND USES/ACTIVITIES
-Archs
-Nodes GENERATION/ATTRACTION
-Centroids DISTRIBUTION
-Connectors MODAL CHOICE
-Generalized cost

ASIGNEMENT
Measurements Adjustments
CALIBRATION

40
 Calibration (2)

Spiess Method

0 10000 20000 30000 40000 50000 60000 70000 80000

80000 80000
Modeled flows

70000 70000
Volúmenes modelados después ajuste

60000 60000

50000 50000

40000 40000 10000

9000
30000 Linear Regression 30000
Y=A+B*X 8000
-----------------------------
20000
A 2326.41349
20000 7000
B 0.94914
6000

Volumen
10000 R-Square 0.96155 10000
5000
0 0
0 10000 20000 30000 40000 50000 60000 70000 80000 4000
Volúmenes observados 3000
Observed Flows 2000
1000
0
1 2 3 4 5 6 7 8 9 10 11 12 13
Observed counts (during Punto de Aforo
rush hour) vs model
Aforos de Hora Punta Asignación del modelo
assignment

41
 Calibration (3)

Spiess Method
F. Robusté
42
 Forecasting
^
y0 = f( , x0)  
y = f( , x) 0

^ ^ Future
y1 = f(0 , x1)
Future socio-
economic
t0ij characteristics
Generation/attraction
Spatial distribution
Modal choice
t1ij
Future
Future network
Forecasting

43
 Trip Generation Gi, Aj = f(P,R,E,S,…)

Spatial distribution tij = Gi Pr(j І i )

Modal choice tijk = tij Pr(k І ij )

Route assignment tijkr = tijk Pr(r І ijk )

Trip generation / attraction

Gi, Aj = f(P,R,E,S,…)

44
 Generation / attraction (1)

Growth factor Pif Iif Mif

Non-Linear effects Fi =
Multiplicative effects  Cobb-Douglas Pia Iia Mia

Analysis by categories Gi =  Ai(h) T(h)

Constant taxes of trip generation in time for hH
each category A: homes
T: trips/home
Crosstabulation (SPSS, SAS, S-PLUS, …)

Multiple linear regression k=n

k=n Gi = 0 +  k Xik
Cobb-Douglas Gi = 0  Xikk k=1
k=1

45
 Generation / attraction (2)

Examples of Generation
Trips / family

0 1 2 3 Family Members

46
 Generation / attraction (5)
• Examples of categories analysis (Kanafani, 1983):

47
 Generation / attraction (3)

Examples of Generation

Individual trips in TPC/year

Relation of trips per person
in transit (public
transportation) and
motorization

Cars/1000 inhabitants

48
 Generation / attraction (4)

• Examples of models calibrated using multiple linear regression:

49
 Generation / attraction (6)

Examples of Generation: ITE Parking Generation

log(Pi) = 0 + 1 log(Xi)

occupied in PH
Parking places
R2 = 0,939

ITE

Mall surface

50
 Generation / attraction (7)
• Airport parking. ITE “Parking generation”
Occupied places in PH vs. Passengers that land at the airport
MONDAY TO FRIDAY SATURDAY

Medium Range Standard Nº of Average Medium Range Standard Nº of Average Nº

Coef. of coef. des. studies Nº of Pax. Coef. of coef. des. studies of Pax.

0,44 0,16- 0,20 74 1624 0,89 0,16- 0,59 13 649

1,03 2,17

Ln ( P )  0 ,58 Ln ( X )  2 ,0
R 2  0 ,786

51
 Generation / attraction (8)
• Parking in a shopping mall. ITE (1987)
Places occupied in PH vs. 1,000 squared feet of leisure surface
MONDAY TO FRIDAY SATURDAY
Medium Range Standard Nº of Average
Coef. of des. studies surf. Medium Range of Standard Nº of Average
coef. Coef. coef. des. studies surf.
3,97 1,11- 1,13 178 836 3,23 1,02-6,17 1,20 141 635
6,06

52
 Generation / attraction (9)

Linear regression vs. Categories

Regression Categories
Sample size required Medium Very large
Model stability Good Good
Significance of the explanatory variables Very good Difficult
Correlation between explanatory variables Detectable Little
Availability of software Global Limited
Non-linearity and no additivity of impacts NO YES
Quality of the model Good Very good

53
 Trip Generation Gi, Aj = f(P,R,E,S,…)

Spatial distribution tij = Gi Pr(j І i )

Modal choice tijk = tij Pr(k І ij )

Route assignment tijkr = tijk Pr(r І ijk )

Spatial distribution

tij = Gi Pr(j І i )

54
 Trip distribution (1)

DEMAND ORIGIN-DESTINATION
Beckmann & Golob tij = Ai Bj Cij- tij = k Ai Bj Cij-
Pi Pj e-Dij
Wardrop tij =
kPk ei-Dik
CHOICE e-Vij
Logit tij = Oi
ke-Vik
OPPORTUNITY e-L V(j-1) – e-L V(j)
tij = Oi
Of intervention
Dj 1 - e-L V(j)
Of competence tij = Oi
 kD k

55
 Trip distribution (2)

 PHYSICAL INTERACTION SPACE

Gravitatory tij = k Oi Dj Cij-
tij = k Oi Dj e-Cij
Cobb-Douglas tij = k Oi Dj Cij-
tij = k Oi Dj e-Cij
Oi Dj Gij
tij = Calibration for method BPR
kDk Gik (Bureau of Public Roads)

Of entropy (Wilson, 1970)

tij = Ai Oi Bj Dj e-kCij
tij = Ai Oi Bj Dj Cij e-kCij Short trips (De Grande, ’06)
56
 Trip distribution (3)

Examples of gravity model

“Sterling route” in Spain log (dij)

tij = k Pi Pj dij

log (tij/PiPj)
Inter-municipal trips in Catalonia

tij = 58 (PiPj)0.35 (MjMj)1.05 (EiEj)0.16 (IiIj)1.52 Cij-2.03 (R2=0.81)

57
 Trip Generation Gi, Aj = f(P,R,E,S,…)

Spatial distribution tij = Gi Pr(j І i )

Modal choice tijk = tij Pr(k І ij )

Route assignment tijkr = tijk Pr(r І ijk )

Modal choice

tijk = tij Pr(k І ij )

F. Robusté
58
 Modal choice (1)
Factors that affect the modal choice
PASSENGER TRIP MODE
Vehicle availability Purpose Waiting time
Income Hour of the Travel time
day
Family structure Cost
Housing density Cost and parking facilities

Constraints of the rest of Comfort

the day
Regularity
Safety

59
 Modal choice (2)

DISCRETE CHOICE MODELS

Ui = Vi + i (i= 1,...,m modes)
Perceived Measurable Error
Utility Utility (random)
(random) (deterministic)

- Passenger chooses j if Uj = max [Ui]  i I

- Dependent categorical variable

- Randomness due to:

-  rules of rational choice
- Vi does not contain all the decision variables
-  perfect information about alternatives

60
 Modal split (3)

 
p
~ ~
Vi  , a   0i    ki a ki
k 1
Parameters Attributes
to be calibrated
(with max. likelihood)

p i  Pr U i  U j  j  i 
 Pr  j   i  V i  V j  j  i

Distribution of i ? Must be simple and it should make sense:

i Normal PROBIT (makes sense)

i Gumbel LOGIT (simple)

61
 Modal split (5)

LOGIT
F x   e e x

i ~ iid Gumbel () f x   e e  x e x

Average    0,577

Vi
e
pi  m

e
Vj

j 1

Daniel McFadden (1975)

Nobel Price for Economics, 2000

62
 Modal split (6)

LOGIT – IAI (Independence of the irrellevant alternatives)

Transformation logit : log(1-y/y)
IAI: Coupled routes:
Coupled
Independents T2 T3
Bus red-blue: 1 2 3 T1= T2= T3=T 1 2 3 p1=1/3
p1=1/3 But is observed
p1  1/2

Correct IAI
p1=2/3 p2=1/3
with hierarchic
or nested logit
p1=1/2 p2= p3=1/4
But is observed
p1=2/3

63
 Modal split (8)
HIERARCHIC OR NESTED LOGIT
Satisfaction:
S  maxU j   log  eVi
def m

j
i 1

S 1 Vj
V’
 e  pj (Daganzo)
nests
A’
V j e Vi

-Start for the lower nest (independent attributes)

-Calculate the nest utility: V’=  S +  
-Calibrate parameters of the nest 0<1 and 
-Climb to the upper nest MIXED LOGIT

64
 Value of time (1)

SUBJECTIVE VALUE OF TIME (VOT)

-Theoretical foundations: De Serpa, 1971
-Lineal specification of the measurable utility
-Same cost parameter for economic theory (independent
of j):
Vj = j -  cj - j Tj + ksk
VSTj = j / 

-Difference between predict (behavior: different VOT) and

evaluation (identical VOT for everybody)

65
 Value of time (2)

SUBJECTIVE VALUE OF TIME

Going to work  Return from work
Work  ? Other reasons
- Regular trip
- Depends on dates
Waiting  ?Access  In the car
K (level of income)
K (block of saved time)
K (time to a deadline)-1
Trip with stops  Non-stop trip
Infrequent trip  Frequent trip
Weekend  Working days
66
 Value of time (3)

SUBJECTIVE VALUE OF TIME

-2 modes or routes (fast but expensive, slow and cheap)

- Vmin, Vmax and Income  LOGNORMAL (empiric)

100%
Vmin VdT

Fast Standing
route Vmax
up
p Slow Seated
route
q Pax/m2

x VOT
Emilio Merino, 1997
Mateu Turró, 1979
67
 Value of time (4)
Value of Time by Country
60,000 20
18
50,000
16
14
40,000
GDP per Capita $

Value of Time $/hr

12
30,000 10
8
20,000
6
4
10,000
2
Netherlands

- 0
Belgium

Germany
Finland
Austria

Ireland
Denmark

Italy

Norway
France

Portugal

Canada
Argentina
USA

Chile
Sweden
Spain

UK
Country

"GDP per capita US$" VOT Business VOT General

8.7 euros/hour in Spain (Maria Gutiérrez-Domènech, 2008)

6.7 Euros/h (Barcelona, 2000)
8-11.5 Euros/h (1999)
10-11.5 US$/h for Spain

68
Demand Modelling Problem 1

69
 Initial assumptions
• Apply the demand modeling process (sequential model of the 4 steps) to the
urban area of Figure 1 composed of 7 different transport zones to determine the
traffic flow in each arc of the network in the future. The generation of travel is
proportional only to the population of each area, while the attraction of travel in
each area is a function of the number of jobs and the area of commercial malls.
• Travel generation and attraction data are detailed in Tables 1 and 2.

TABLE 1. BASE YEAR

Superficie comercial Llocs de treball
Zona Població
(1000m2)

TABLE 2. FUTURE YEAR

Superficie comercial Llocs de treball
Zona Població
(1000m2)

Fig.1

70
 STEP 1. GENERATION AND ATRACTION

• Step 1. Determine generation and atraction for each transport zone:

Gi  3,56 P  554
Ai  3,29 LT  64,68S C  519

GENERATION AND ATRACTION BASE YEAR

Superficie comercial Llocs de treball Generació Atracció
Zona Població
(1000m2)

G1  3,565000  554  16.500

A1  3,29420  64,6820  519  2800

GENERATION AND ATRACTION FUTURE YEAR

Superficie comercial Generació Atracció
G1  3,567000  554  24.366
Zona Població Llocs de treball
(1000m2)

A1  3,29790  64,6866  519  7.387

71
 STEP 2. DISTRIBUTION

• Step 2. The travel distribution matrix for the particular future year is calculated.
To determine it, we will need first to calculate the travel times between the
different areas. Figure 2 determines the travel times in different arcs of the
network for the future year. From these travel times, it is necessary to identify
the route of shortest travel time between the different areas of the transport
network, which are detailed in Table 3.

Fig 2. Travel time in the links of the Table 3. Travel time between transport zones
transport network

72
 STEP 2. DISTRIBUTION

• Subsequently, a gravitational model of travel distribution must be calibrated

where the impedance Fij is expressed as a function of travel time.

Gi A j Fij Gi A j C ij2
Tij 
A F j ij

A C j
2
ij
Fij  Cij2 where Cij is the travel time
j j

TRIP DISTRIBUTION FUTURE YEAR – ITERATION 0

The sum of the
travel generations
corresponds to the
values of the
generation model ...
BUT AND THE
SUM OF THE
ATTRACTIONS ??
Superficie comercial Llocs de treball Generació Atracció
Zona Població
(1000m2)

73
 STEP 2. DISTRIBUTION

• In order to adjust these values, a row-column factor must be applied, and

modified trips are recalculated from the following formulation:

Aj
Tij'  Tij where T’ij is the modified number of trips and Tij the initial value
T
i
ij

TRIP DISTRIBUTION FUTURE YEAR – ITERATION 1

In this iteration, the

atracted trips are
consistent with the
original values

Gi
We have to apply a row-column factor, Tij'  Tij
and we calculate again modified trips.  Tij
j

74
 STEP 2. DISTRIBUTION

FINAL MATRIX OF TRIPS FUTURE YEAR

75
 STEP 3. MODAL SPLIT

• Step 3. Determination of the modal split between the private vehicle and public
transport. It is based on the multinomial logit choice model (model calibration is
omitted) and the population is assumed to be divided into two market
segments. Utility functions are determined by:

Market segment 1: Market segment 2:

UVP 1  0,012t  0,40c  2,50 U VP  2  0,012t  0,35c  3,50
U TP 1  0,01t  0,50c  0,0 U TP  2  0,015t  0,35c  0,0

t= travel time UVP= utility of a trip by car

c= travel cost (out-of-pocket) UTP= utility of a trip by public transport

Travel time by car (min): Travel time by públic transport (min):

76
 STEP 3. MODAL SPLIT

Travel cost by car (€): Travel cost by PT (€):

With the application of the logit functions, the probabilities of making a trip with one
transport mode or another for a type of market to which each individual of the population
belongs are calculated. v
e i

pi  m


v
e j

j1

In that sense, for the calculation of the probability of travel of an individual of market
segment 1 between zones 1 and 2 by private vehicle would be carried out:
expU VP 1;1  2 
pVP 1;1  2  
expU TP 1;1  2  expU VP 1;1  2 
exp 0,0125  0,401,25  2,50  6,96
   0,9408
exp 0,0125  0,401,25  2,50   exp 0,017,5  0,501,5  0,0  6,96  0,44
77
 STEP 3. MODAL SPLIT

Probability of choosing car for an Probability of choosing car for an

individual from market segment 1 individual from market segment 2

• These probabilities are combined using the fractions of the population that belong to
each market segment in order to determine the total number of trips that are made by
car between areas i.j; according to the formulation:

S VP ;ij  pVP 1;ij mi1  pVP  2;ij mi 2

p VP-k;ij is the probability that a trip between i,j
Zona Segment de mercat 1
of an individual belonging to the market
Segment de mercat 2
segment k will take place by a private vehicle
mik is the fraction of the population of the area
and belongs to the market segment k
(attached table)

78
 STEP 3. MODAL SPLIT

• The values of the variable S in the previous equation will be multiplied by the
corresponding value of the travel matrix in the future year to give the expected
number of trips by private vehicle. In this case, it is necessary to increase the
number of passenger trips in car trips considering a vehicle occupancy of 1.4%.
Carrying out this calculation will determine the total number of car trips
between the different transport areas.

FLOW MATRIX OF CARS FUTURE YEAR

0,370,9408  0,630,9734
TVP ;1  2  12.951  8.893
1,4

79
Demand Modelling Problem 2

80
• A territory with 3 zones (i= 1,2,3) and populations Pi, matrix of total interzonal
trips tij, distance in km dij, travel times by car in minutes Tij,c, travel times by
bus in minutes Tij,b, and modal split by bus pij,b.

Total trips tij Distance (km) dij

Population Pi

Travel time by bus Tij,b Modal Split by bus pij,b

Travel time by car Tij,c

81
1. Formulate a logit binomial model for modal split assuming that travelers
determine the transport mode based only on travel time and costs of covering
the distance (measurable utility is proportional to the generalized costs with
the same proportionality constant θ for the two transport modes).

2. Calculate the generalized cost matrices CGb and CGc and calibrate the
parameter θ of the previous model. Assume a value of time 6 €/h, an
integrated fare system of 1 € for the bus independent from the traveled
distance and operating costs for car of 0,3 €/km.

3. Which is the aggregated market quote for bus mode for that territory and
travelers?

4. Draw the curve of probability of election of bus pb with regard to the difference
of generalized costs between bus and car: CGb-CGc.

5. If we implement a bus lane and the travel times by bus are reduced 30%, which
is the new modal split for the bus?

82
1. Logit model: 1 1 Vc   (CGc )  a
pb  
1 eVcVb 1 e (CGb CGc )a Vb   (CGb )

2. Calibration :
Matriz T bus actual Matriz T bus actual
Tbus= 20 30 40 Cgbus 3 4 5
30 25 50 4 3.5 6
40 50 30 5 6 4
Matriz T coche actual D
Tcoche= 10 20 30 Dcoche= 3 6 9 CG coche 1.9 3.8 5.7
20 5 15 6 2 5 3.8 1.1 3
30 15 2 9 5 1 5.7 3 0.5

CGb-Cgcoche 1.1 0.2 -0.7

0.2 2.4 3
-0.7 3 3.5

Caso O D Pb (1-Pb)/Pb ln((1-Pb)/Pb) CGb-CGc

1 1 2 0.2 4.00 1.386 0.2
2 1 3 0.4 1.50 0.405 -0.7
ln((1-Pb)/Pb) 3 2 3 0.05 19.00 2.944 3

3.5
3 y = 0.6553x + 1.0326
R² = 0.9753
2.5
Estimación lineal
2 tetha 0.655
ln((1-Pb)/Pb)
1.5 a 1.0326
Linear (ln((1-Pb)/Pb))
1
0.5
0 CGb-CGc
-1 0 1 2 3 4 83
3. Market share: BUS

Pb tij tij en bus

0.2 110 22
0.4 30 12
0.05 60 3
Suma 200 37
Cuota de mercado bus (%) 18.5

4. Curve of probability of election of bus (pb) with regard to generalized cost

difference:

1 1
pb  
1 eVcVb 1 e (CGb CGc )
pbus pbus
1
1
0.9
0.9
0.8 0.8
0.7 0.7
0.6 Pbus 0.6
Pbus

0.5 pbus 0.5

0.4
pbus
0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
-7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7
-7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7
CGbus-CGcoche CGbus-CGcoche
84
5. New modal split with bus lanes

Matriz T bus actual Matriz Tbus futura

Tbus= 20 30 40 T'bus= 14 21 28
30 25 50 21 17.5 35
40 50 30 28 35 21

Matriz CG bus
CG'bus=0,1*T'bus+1= 2.4 3.1 3.8
3.1 2.75 4.5
3.8 4.5 3.1

CGb-CGc theta*D CG pb pb*viajes

1//2 -0.7 0.5741 0.3603 39.6320
1//3 -1.9 -0.2119 0.5528 16.5833
2//3 1.5 2.0151 0.1176 7.0576
Suma 63.27
200
Cuota de mercado (%)= 31.64

85
Demand Modelling Problem 3

86
A hierarchical logit modal choice model compares car, taxi and carpool options in
the upper nest; In the lower nest of mass transport, the bus and metro modes are
contemplated. Calculate the numerical value of the probabilities of choosing each
mode of transport according to the following measurable utility functions and for the
average values of the explanatory variables that are presented in the table below.

 Vcoche= -0,03 Tcoche – 0,02 Ccoche + 1,25

 Vtaxi = -0,03 Ttaxi – 0,02 Ctaxi - 0,20
 Vtte. colectivo= -0,60 EMU
 Vbus = -0,04 Tbus – 0,03 Cbus + 0,5
 Vmetro = -0,04 Tmetro – 0,03 Cmetro

Modo k Tiempo (Tk) Coste (Ck)

Coche 4,5 23 Note: EMU= Expected Maximum
Taxi 5,5 15 Utility of the nest = ln( j e{Vj})
Bus 7,5 5,5
Metro 5,5 3,6

87
 • First, lower nest:

coche taxi tte. colectivo e Vbus

Pr coger el bus tte. colectivo 
bus metro e Vbus  e Vmetro

Vbus  0.04  Tbus  0.03  C bus  0.5  0.04  7.5  0.03  5.5  0.5  0.035
Vmetro  0.04  Tmetro  0.03  C metro  0.04  5.5  0.03  3.6  0.328

e 0.035
Pr coger el bus tte. colectivo   0.5898
0.035 0.328
e e

e Vmetro e 0.328
Pr coger el metro tte. colectivo     0.41
Vbus Vmetro 0.035 0.328
e e e e

88
 • Second, upper nest:

e Vcoche
coche taxi tte. colectivo Pr coger el coche 
Vtte. colectivo
e Vcoche  e Vtaxi  e
bus metro

Vcoche  0.03  Tcoche  0.02  C coche  1.25  0.03  4.5  0.02  23  1.25  0.655
Vtaxi  0.03  Ttaxi  0.02  C taxi  0.2  0.03  5.5  0.02  15  0.2  0.665

Vtte. colectivo  0.6  EMU Vtte. colectivo  0.6  0.563  0.3378

  
EMU  ln e Vbus  e Vmetro  ln e 0.035  e 0.328  0.563 
e 0.655
Pr coger el coche   0.6106  61.1%
e 0.655  e 0.665  e  0.3378

e  0.665
Pr coger el taxi    0.1631  16.3%
e 0.655  e 0.665  e  0.3378

e  0.1467
Pr coger el tte. colectivo   0.2263  22.6%
e 0.655  e  0.655  e 0.3378 89
 coche taxi tte. colectivo

bus metro

• Third, lower nest:

Pr coger el bus   Pr coger el bus tte. colectivo  Pr coger el tte. colectivo 

 0.2263  0.59  0.1335  13.35%

Pr coger el metro   Pr coger el metro tte. colectivo  Pr coger el tte. colectivo 

 0.2263  0.41  0.0928  9.28%

90
 Trip Generation Gi, Aj = f(P,R,E,S,…)

Spatial distribution tij = Gi Pr(j І i )

Modal choice tijk = tij Pr(k І ij )

Route assignment tijkr = tijk Pr(r І ijk )

Route assignment

tijkr = tijk Pr(r І ijk )

91
 Assignment (1)

time
FORMULATION
xa

min Z    t a ( w) dw
a 0

s.a.
C q
(1) f
k
k
rs
 qrs r , s
xa: flow in arch a

( 2) f k
rs
0 k , r , s fkrs: flow in route k
connecting r-s

(3) xa   f krs akrs a akrs: 1 if arch  route k

between r and s (0
r s k otherwise)

92
 Assignment (2)


  f  
t(f): Time (costs) in link with flow f
f: Flow in link
t ( f )  t0  1   t0: Travel time in link with free flow (f=0)
  cap  cap: Link capacity

Type of axes Alfa Beta 152

Links type autopista 0,13 9,5 150
tiempo GETRAM
148
Links type autovia 0,21 9,0 BPR estimada
146
Links type carretera 1+1 0,09 8,0 tiempo (seg)
144
Links type carretera 1+1 142
with complex orography 0,04 6,5 140
138
136
134
132
130
0,000 0,200 0,400 0,600 0,800 1,000 1,200
x = flujo / capacidad : [grado de saturación]

93
 Assignment (3)

EQUILIBRIUM
Of user (UE). Any traveler can improve his travel time
unilaterally
Of stochastic user (SUE). Any traveler believes that could
be improved ….
System optimum (SO). Total time is minimum
WARDROP PRINCIPLES(1952)
-1º (UE): Time used in all itineraries is the same and is lower than
the time required for not used itineraries.

94
 Assignment (4)

CLASIFICATION OF MODELS

ALL OR DIAL , BURRELL

NOTHING (AON)
ARCHS WITH
CAPACITY

NO
YES

INCREMENTAL SUE (Stochastic User

METHOD Equilibrium)

95
 Assignment (5)

METHODS
-ALL OR NOTHING (AoN)
Van = (1 - ) Van-1 + Fa
-INCREMENTAL ASSIGNMENT
with  = 1/n
-DIAL (1971): distribution logit among
logic paths (stochastic proporcional
 w C ijr
allocation) e
-BURRELL (uniform distribution) Tijr  Tij
e
 w C ijm
-SUE: Stochastic User Equilibrium
(Sheffi, 1985). Normal distribution for m
the variables of generalized cost.

96
 Assignment (6)

ADEQUACY OF METHODS

AON SUE
INCREMENTAL
Equilibrium under
congestion

q
Less accuracy

97
Bici-Vici: Bikes traffic in BCN

98
Traffic network result of assignment

99
Demand Modelling Problem 4

100
• Given a network with tis characteristics and an O/D matrix.
ARISTA LONGITUD VELOCIDAD
A 2.000 m. 40 km/h
B 400 m. 45 km/h
C 700 m. 30 km/h
D 400 m. 40 km/h
E 1.200 m. 30 km/h
F 400 m. 40 km/h
G 600 m. 40 km/h
H 1.900 m. 50 km/h
I 1.000 m. 50 km/h
J 1.400 m. 45 km/h
K 900 m. 40 km/h

O/D 1 2 3 4 5
1 - 357 428 475 112
2 348 - 406 303 167
3 430 421 - 210 195
4 461 288 211 - 187
5 115 159 195 183 -

• Note: matrix O-D includes travelers and speed per link of the table is the initial
speed without congestion. 101
• Average car occupancy with regard to the destination is:
Zona Medio de personas por vehículo
1 1,9
2 2,1
3 1,5
4 1,8
5 2,0

• Question: Traffic assignment with the all-or-nothing method and stochastic

method knowing that costs and times per link follow the equation:

 lijk 
t 1
ijk      0 

v
 k    (i k ) 
• Being: α = 0,0252 , β = 1,6298 , λ = 0,1264 , ρ = 0,2397

102
 Arista Longitud Velocidad Tiempo
• Solution: A 2.000 40 180
B 400 45 32
C 700 30 84
D 400 40 36
Time (sec) = 3600*Length (m)/1000/speed (km/h) E 1.200 30 144
F 400 40 36
G 600 40 54
H 1.900 50 137
I 1.000 50 72
J 1.400 45 112
K 900 40 81

• When we have the times, we identify minimum travel times (in seconds)
between pairs O-D: 1 2 3 4 5
1 211 252 112 81
2 211 103 293 292
3 252 103 191 263
4 112 293 191 72
5 81 292 263 72

• And the paths: 1 2 3 4 5

1 A-B A-D-F J K
2 B-A B-D-F B-D-F-G-H B-A-K
3 F-D-A F-D-B G-H G-H-I
4 J H-G-F-D-B H-G I
5 K K-A-B I-H-G I
103
• Assign the flows per link, All or nothing: Arista Todo-nada
A 1.889
B 2.449
C 0
D 2.276
Assumption: Flows independent of the direction; E 0
flows in one direction affect flows in the opposite F 2.276
direction G 1.402
H 1.402
I 760
J 936
K 553

• Finally, from travelers to cars based on car occupancy: Arista Todo-nada

A 1.024
B 1.289
C 0
D 1.288
E 0
F 1.288
G 790
H 790
I 423
J 507
K 276
104
• For the stochastic method, first we calculate the initial travel times without
congestion: Arista Longitud Velocidad Tiempo
A 2.000 40 180
B 400 45 32
C 700 30 84
D 400 40 36
E 1.200 30 144
F 400 40 36
G 600 40 54
H 1.900 50 137
I 1.000 50 72
J 1.400 45 112
K 900 40 81
e  tk
• Probabilities for each path using the formula: pk oshd  
• For centroid 2 as a destination, possible paths:
 e tk

k
Recorridos y tiempos de recorrido

We do not include paths 12 AB

212
JHEB
425
JHGFDB JHGFC KIHEB KIHGFDB KIHGFC
407 423 466 448 464
ADC
300
that are not a path of 32 FC
120
FDB
104
GEB
230
Minimum time or cost 42 JAB HEB HGFC HGFDB IKAB
324 313 311 295 365
clearly 52 KAB IHEB IHGFC IHGFDB
293 385 383 367
105
• We calculate the probabilities with the previous formula (logit):
12 AB JHEB JHGFDB JHGFC KIHEB KIHGFDB KIHGFC ADC
We obtain extreme 1,000 0,000 0,000 0,000 0,000 0,000 0,000 0,000
32 FC FDB GEB
Values (0 or 1) 0,000 1,000 0,000
42 JAB HEB HGFC HGFDB IKAB
0,000 0,000 0,000 1,000 0,000
52 KAB IHEB IHGFC IHGFDB
1,000 0,000 0,000 0,000

• Doing the same for all the destinations, we would get the total flow per link:
Arista Flujo
A 1.024

B 1.289
C 0
D 1.120
E 168
F 1.120
G 622
H 790
I 423
J 507
K 413

106
 • We calculate new travel times per link with the previous flows:
Arista ik α β λ ρ V0 l t1
A 1.024 0,0252 1,6298 0,1264 0,2397 40 2.000 312,09
B 1.289 0,0252 1,6298 0,1264 0,2397 45 400 55,29
 lijk  C 0 0,0252 1,6298 0,1264 0,2397 30 700 84,00
1
t ijk      0 
 D 1.120 0,0252 1,6298 0,1264 0,2397 40 400 62,52
 v k    (ik )  E 168 0,0252 1,6298 0,1264 0,2397 30 1.200 247,54
F 1.120 0,0252 1,6298 0,1264 0,2397 40 400 62,52
G 622 0,0252 1,6298 0,1264 0,2397 40 600 92,98
H 790 0,0252 1,6298 0,1264 0,2397 50 1.900 233,50
I 423 0,0252 1,6298 0,1264 0,2397 50 1.000 122,10
J 507 0,0252 1,6298 0,1264 0,2397 45 1.400 191,16
K 413 0,0252 1,6298 0,1264 0,2397 40 900 138,72

• With the new travel times, we use the logit formula again, obtaining a new
assignment of travelers and vehicles and new flows that give new travel times:
Arista ik α β λ ρ V0 l t2
A 1.024 0,0252 1,6298 0,1264 0,2397 40,0 2.000 312,09
B 681 0,0252 1,6298 0,1264 0,2397 45,0 400 54,83
C 608 0,0252 1,6298 0,1264 0,2397 30,0 700 147,31
D 512 0,0252 1,6298 0,1264 0,2397 40,0 400 61,83
E 168 0,0252 1,6298 0,1264 0,2397 30,0 1.200 247,54
F 1.120 0,0252 1,6298 0,1264 0,2397 40,0 400 62,52
G 622 0,0252 1,6298 0,1264 0,2397 40,0 600 92,98
H 790 0,0252 1,6298 0,1264 0,2397 50,0 1.900 233,50
I 423 0,0252 1,6298 0,1264 0,2397 50,0 1.000 122,10
J 507 0,0252 1,6298 0,1264 0,2397 45,0 1.400 191,16
K 276 0,0252 1,6298 0,1264 0,2397 40,0 900 138,08 107
• We iterate up to convergence:
Arista ik α β λ ρ V0 l t3
A 1.262 0,0252 1,6298 0,1264 0,2397 40,0 2.000 313,11
B 1.076 0,0252 1,6298 0,1264 0,2397 45,0 400 55,15
C 271 0,0252 1,6298 0,1264 0,2397 30,0 700 145,37
D 1.313 0,0252 1,6298 0,1264 0,2397 40,0 400 62,68
E 0 0,0252 1,6298 0,1264 0,2397 30,0 1.200 234,72
F 1.583 0,0252 1,6298 0,1264 0,2397 40,0 400 62,88
G 731 0,0252 1,6298 0,1264 0,2397 40,0 600 93,18
H 731 0,0252 1,6298 0,1264 0,2397 50,0 1.900 233,30
I 429 0,0252 1,6298 0,1264 0,2397 50,0 1.000 122,12
J 243 0,0252 1,6298 0,1264 0,2397 45,0 1.400 189,72
K 276 0,0252 1,6298 0,1264 0,2397 40,0 900 138,08

F t  F t 1

t 1
F
108
Simulation Software

109
 Software

Assignment models
G/A

Distribución

• Private transport Reparto Modal

• Public transport
• Bimodal assignment (EMME/2)
Asignación

• Integrated models macro-micro

(AIMSUN)
• Multimodal equilibrium models
(Estraus)
F. Robusté
110
 EMME/2
Bimodal assignment: paradigm EMME/2

Assignment
Asignación
Travel time auto (timau) used in impedance
Auto
auto traffic (segments tied to arch)

Assignment
Asignación
Vehicles equivalent auto (veq) are used in
impedance auto (volad, additional volume) transit
Traffic

111
 EMME/2
Modeling environment EMME/2

• EMME/2 is a multimodal urban transportation planning

system that uses interative figures/pictures of the state
of art

• It offers to the planner a complete and whose set of

tools to model the demand, modeling and analysis of the
multimodal network and for implementing procedures of
evaluation

• Panoramic

112
 TransCAD

Modeling environment CAD

• TransCAD is a geographic information system (SIG)
specially design for transport profesionals with the
object of store, show and analyse transportation data
• What is different of TransCAD among other
transportation software packages is that combines in a
single integrated platform all the SIG propperties and the
modeling capacities of transportation.
• Caliper Corporation
1172 Beacon Street • Newton MA 02461 • USA • Tel:
+1(617)527-4700 • informacion@[Link]

113
 TransCAD

Modeling environment TransCAD

• Powerful SIG platform with
specific extensions for
transportation models.

• Analysis tools designed for

transportation, maping and
displaying.

• Applications for generating

route moduls, trip demand
forecast, public transport,
logistics and territory
management.

114
 TransCAD

Modeling environment TransCAD

• Models of trip Generation-
Supply modeling and
Production.
transportation planning
• Models of Trip Attraction.
• The algorithms of trip adjustment are
• TransCAD is the only designed to adjust the total volume
planning program based in a of attractions to the productions.
SIG environment totally • Trip distribution models.
integer with tools
• Modal choice models.
and assignment proccess.
• P-A to O-D and Schedule Matrix
include tools to convert generation
• TransCAD includes all the and attraction in origin – destiny,
typical elements in a divide a schedule matrix in a group of
transportation planning timetables, transform travelers in
programm ... vehicles and apply peak-rush hours
factors.
115
 TransCAD

Modeling environment TransCAD

• The Advanced Models of Traffic Assignment included in TransCAD

are based in generalised cost functions, to evaluate problems of
multimodal transport or competence between different
transportation modes.

• TransCAD permits to apply a traffic assignment model that at the

same time asigns light vehicles, heavy vehicles and buses in the
road network. This procedure allows including tolls for type of
vehicle and use different speeds in the equation of generalised
costs. Restrictions for type of vehicle can be defined as well. Some
vehicles with fixed itineraries could be considered as precharges in
the network.

116
 TransCAD
Modeling environment TransCAD:
results

117
 TransCAD
Modeling environment TransCAD:
BCN Model

118
 TransCAD
TransCAD: results BCN model assignment

119
 TransCAD
Modeling environment TransCAD:
results
• The previous
network of
private
transport could
be assigned
with the
demand matrix
and we can get
results at arch
levels and OD
pattern.

120
 AIMSUN

Modeling Environment AIMSUN-NG

• AIMSUN-NG was developed at la UPC and

actually in TSS (Transport Simulation
System)

• An evoluted informatics environment that

integrates: microscopic simulation,
transportation planning and land uses.

121
References

122
 REFERENCES - Demand Modeling

Meyer, M. & E. Miller. Urban Transportation Planning,

Ed. Mc Graw Hill, 2001.
Oppenheim, N. Urban Travel Demand Modelling, John
Wiley & Sons, Inc.,1995.
Ortúzar, J.D. & L. Willumsen. Modelling transport.
John Wiley & Sons. 4th edition, 2011.
Sheffi, Y. Urban transportation networks: Equilibrium
analysis with mathematical programming methods.
Prentice-Hall, 1984.

123
Mobility & Transport Networks
Plan de Formación (40h)
Plan de Formación (40h)
(2021)

Demand Modelling