Diapositive 1

1
Design approach of behavioural models
of traffic network users
Choice of one transport means
Boussier 1,2
Estraillier 1
Jean-Marie
Pascal
Augeraud 1
Sarramia 1
Michel
David
1 Laboratory L3I - University La Rochelle
- France
2 EIGSI ( Engineering School ) , La Rochelle
- France
Laboratoire Informatique-Image-Interaction
2
Framework 3D urban traffic simulator
activity-based model and multi agent
technology
Requirements
Modelling
Simulation
MAS Architecture
Interactions modelling
UTS information systems
semi automatic generation
Ground
Services information systems
Urban Transport System
General Structure
( UTS )
Composing
Signals
Observation/ management
Buildings
Behavioural models
and
Model checking
Underrun Verifications
Adequacy
3
Making of a virtual city our ground model
for activity location
building, park, car park
for pedestrians
3D ground model
for vehicles
2D Map of the virtual city
2D car park
with square blocks
a pattern of a square block
an assembling of 4 square blocks
Map of the studied city
Each block contains predefined areas, graphs,
signals, parking places
Blocks can be reused, modified, specialized
Areas used by agents - Areas only for
displaying scenery
4
Making of a virtual city our 3D
graphical tool
EDITING
a BLOCK
a made square block
a basic square block
EDITING
a SYSTEM
3D representation
Assembling of square blocks
graphical editor produces 3D graphical
representation from templates
set up a system of procedural construction of
buildings
in order to adapt their forms and their
functions to available space
5
Design approach of behavioural models
average behaviour for socio-demographic
categories
Identification of Factors and Response
Design Of Experiments
Selection of the Array and its Linear Graph
MODELLING of
RESPONSE
Design of the Model with Analysis Of Variance
Vigiers Model
Evaluation of Travel
6
Design approach of behavioural models
average behaviour for socio-demographic
categories
Identification of Factors and Responses
Coupling of 4 concepts
Design Of Experiments
Selection of the Array and its Linear Graph
MODELLING of
CLASSIFICATION of RESPONSES
RESPONSE
Dempster-Shafers Theory
Fusion of Answer Data
Design of Models with Analysis Of Variance
Vigiers Model
Choice of the most robust Response
Robustness Indicator
Evaluation of Travel
In our case and here, the responses are the
transport means
7
Design Of Experiments making of a
questionnaire
Example choice of one transport means (on foot,
car, bus, bicycle)
Level 1
Level 2
(input) Factors
favourable
not favourable
Weather
low
high
Risk of incidents
short
long
Distance
Travel conditions
with parcel
without parcel
Timing
to be in hurry
not to be in hurry
8
Design Of Experiments making of a
questionnaire
Example choice of one transport means (on foot,
car, bus, bicycle)
Level 1
Level 2
(input) Factors
List of possible answers
favourable
not favourable
Weather
low
high
Risk of incidents
Linguistic evaluation
Crisp value
short
long
Distance
Travel conditions
with parcel
without parcel
very rarely
1
. . . . .
sometimes
2
. . . . .
Timing
to be in hurry
not to be in hurry
frequently
3
. . . . .
very frequently
4
. .
32 4 trnsp means x 8 runs
9
Design Of Experiments making of a
questionnaire
Example choice of one transport means (on foot,
car, bus, bicycle)
Level 1
Level 2
(input) Factors
List of possible answers
favourable
not favourable
Weather
low
high
Risk of incidents
Linguistic evaluation
Crisp value
short
long
Distance
Travel conditions
with parcel
without parcel
very rarely
1
. . . . .
Timing
sometimes
2
. . . . .
to be in hurry
not to be in hurry
frequently
3
. . . . .
very frequently
4
. .
Taguchis orthogonal array L8(2)7
Factors (and interactions)
A
B
C
D
E
F
G
Linear graph
Run
1
1
1
1
1
1
1
1
A
1
1
1
2
2
2
2
2
1
2
2
1
1
2
2
3
1
2
2
2
2
1
1
4
E
C
2
1
2
1
2
1
2
5
F
2
1
2
2
1
2
1
6
B
D
2
2
1
1
2
2
1
7
2
2
1
2
1
1
2
8
G
interactions
10
Two Uses of the questionnaire
Modelling of a response answers are the average
of crisp evaluations
Factors
Answers of an asked person
Weather
Risk of
Distance
Travel
Timing
Bicycle
Question N
incidents
(D)
(B)
(A)
(C)
conditions
(G)
(V)
(Run)
with
to be in
1
favourable
low
short
2
parcel
hurry
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .

not
without
to be in
6
low
long
1
favourable
parcel
hurry
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .

Choice of one transport means answers obtained
by the information fusion
Factors
Answers of an asked person
Weather
Risk of
Distance
Travel
Timing
Bicycle
Car
Question N
incidents
(D)
(B)
(A)
(C)
conditions
(G)
(V)
(C)
(Run)
with
to be in
1
favourable
low
short
sometimes
sometimes
parcel
hurry
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .

not
without
to be in
6
low
long
very rarely
frequently
favourable
parcel
hurry
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .

11
Fusion of Information our framework
Use of Dempster-Shafers theory
V
C
B
F
T Bicycle, Car, Bus, on Foot
Frame of discernment
T
2 F , Bicycle, , Bicycle ? Car, , T
Set power
Mass assignement
Example each asked person is one source of
information
Bicycle
Car
Bus
On foot
for the scenario y
sometimes
frequentely
very frequentely
very rarely
0.2 , 0.3 , 0.1 , 0.4
for an asked person giving the same answer
for several transport means
a normalization and a redistribution of
masses are done
a discounting operation is done
, n being the response number with the same
evaluation
where
N being the number of responses
Here, response number number of transport means
12
Fusion of Information Calculation of Masses
Orthogonal Sum
S1
S1
S2
S3
SM
SM
m m ? . . . ? m
m ? m ? m ? . . . ? m
T
T
T
T
T
T
T
a source
Fusion of 2 sources
For a conflict
0 if no conflict
13
Fusion of Information Pignistic
probabilities
Use of the Transferable Belief Model ( T B M )
cardinality of A
By using the pignistic probabilities ,
masses are redistributed on singleton
hypothesises .
ignorance
sub-set coupling hypothesis
singleton hypothesis
F V B C
F?V
F?B
F?C
V?B
V?C
B?C
F?V?B
F?V?C
F?B?C
V?B?C
F?V?B?C
survey results
17 11.1 12.4 3.9
14.8
7.5
0
14.8
0
0
14.8
0
0
0
3.7
3.9
33
28.4
34.3
Pignistic transform
F on Foot - V Bicycle - B Bus - C Car
F on Foot - V Bicycle - B Bus - C Car
14
Vigiers model
for each scenario j and each transport means k, a
score is defined
j
k
S ? . PT (Hk )
l is an arbitrary constant
j
a1b1 a1b2
S Smean (a1 a2)A (b1 b2)B . . .
AB . . . e
a2b1 a2b2
ai Smean (Ai ) - Smean
mean effect of a factor
with
ai bj Smean (Ai , Bj ) - Smean - ai - bj
mean effect of an interaction
Example of a behavioural model bicycle
students - age 22.3 -
1,472 answers
IT IS A SYMBOLIC WRITING
after ANOVA, only A (weather), C (distance), D
(travel conditions)
are significant as factors for the category of
student agents
For each category of agents, the model can be
different
Possibility to reduce the number of information
for an agent
15
Concept of robustness
(1)
Feedback used for the memory effect
Noise Factors
example bicycle, student
Control Factors
Scenario
Travel Conditions (D)
(Run n)
the first score used by an agent is in the
first column wherein are
the best external conditions (e.g. score of
scenario 1 S 2.8 )
if no information is given during the travel
and the road safety
is bad, the score of the next travel (same
scenario) will be S 2.1
for the next travel (same scenario), the agent
"individual" will select another
transport means if the score of the bicycle is
lower than those of others
16
Concept of robustness
(2)
Selection of the most robust transport means
Given a scenario,
the scores of two transport means are
nearly equal
k
S
k
S
mean
j
20 log

( signal-to-noise ratios)
k
N
s
j
j
The agent "individual" selects the transport
means when
S
is maximal
N
because
the choice of one transport means is not affected

by external conditions
17
Conclusions
Design of models describing average behaviours
of individual agents
classified in various socio-demographic
categories (Vigiers model)
Determination of the sufficient number of
information used by an
agent individual (thanks to ANOVA, the
initial number can be reduced)
Possibility to upgrade models (Dempster-Schafers
theory)

• new sources of information (people answers) can
  enhance the database

• the number of transport means can be changed

• conflicts between people answers are taken into
  account

Feedback for the memory effect (robustness
indicator)
Selection of the most robust transport means
(robustness indicator)
Perspectives
Use of Dezert-Smarandaches theory for the fusion
of information
( Better taking conflicts between people answers
into account )
Taking answers of asked people into account as
fuzzy values
18
Design approach of behavioural models
of traffic network users
Choice of one transport means
Estraillier 1
Boussier 1,2
Pascal
Jean-Marie
Augeraud 1
Sarramia 1
Michel
David
1 Laboratory L3I - University La Rochelle
- France
2 EIGSI ( Engineering School ) , La Rochelle
- France
Laboratoire Informatique-Image-Interaction
19
Design Of Experiments (D.O.E.)
back
Field of J. Louviere Marketing research
Stated Preference Methods
Rating
Ranking
Stated Choice
Other Choice Methods
Referendum Contingent Valuation
Attribute Based Stated Choice
About 7 days before my paper proposaI,
I discoveried one paper of Louviere
Warren F. Kuhfeld January 1, 2005 -
Marketing Research Methods in SAS
Approach of Taguchi optimization of industrial
process
Some STATISTICIANS hate Taguchis approach
This declaration was on a slide of a conference
in USA
Other methods and approaches
20
(No Transcript)
21
Design Of Experiments (D.O.E.)
Taguchis method
In some countries (e.g. USA), some people wont
hear of Taguchi
Some STATISTICIANS hate Taguchis approach
Research field of Louviere
Fractional factorial designs
About 7 days before I sent the paper, I
discoveried one paper of Louviere In this last
one, About fractional factorial designs (arrays
orthogonal )
Warren F. Kuhfeld January 1, 2005 SAS 9.1 Edition
Marketing Research Methods in SAS