An original synthetic population tool applied to Belgian case:VirtualBelgium
Dr Eric Cornelis Johan Barthelemy Laurie Hollaert Philippe Toint
naXys - GRT, University of Namur, Belgium
NTTS 2013, Brussels, March 5
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 1 / 21
Contents
1 Motivation
2 Classical approaches
3 Why classical approaches are not feasible for the Belgian case?
4 VirtualBelgium, an innovative method
5 Temporal evolution for the synthetic population
6 Application and perspective
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 2 / 21
Contents
1 Motivation
2 Classical approaches
3 Why classical approaches are not feasible for the Belgian case?
4 VirtualBelgium, an innovative method
5 Temporal evolution for the synthetic population
6 Application and perspective
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 3 / 21
Motivation
Micro-simulation
Agents based models
Disaggregated spatial meshing
BUT
Impossible or too expensive to get an exhaustive data set if the number of agents is large
Privacy rules
Need for SYNTHETIC POPULATION
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 4 / 21
Motivation
Micro-simulation
Agents based models
Disaggregated spatial meshing
BUT
Impossible or too expensive to get an exhaustive data set if the number of agents is large
Privacy rules
Need for SYNTHETIC POPULATION
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 4 / 21
Motivation
Micro-simulation
Agents based models
Disaggregated spatial meshing
BUTImpossible or too expensive to get an exhaustive data set if the number of agents is large
Privacy rules
Need for SYNTHETIC POPULATION
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 4 / 21
Motivation
Micro-simulationAgents based modelsDisaggregated spatial meshing
BUTImpossible or too expensive to get an exhaustive data set if the number of agents is largePrivacy rules
Need for SYNTHETIC POPULATION
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 4 / 21
Motivation
Micro-simulation
Agents based models
Disaggregated spatial meshing
BUT
Impossible or too expensive to get an exhaustive data set if the number of agents is large
Privacy rules
Need for SYNTHETIC POPULATION
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 4 / 21
Motivation
Micro-simulation
Agents based models
Disaggregated spatial meshing
BUTImpossible or too expensive to get an exhaustive data set if the number of agents is large
Privacy rules
Need for SYNTHETIC POPULATION
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 4 / 21
Motivation
Micro-simulation
Agents based models
Disaggregated spatial meshing
BUTImpossible or too expensive to get an exhaustive data set if the number of agents is large
Privacy rules
Need for SYNTHETIC POPULATION
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 4 / 21
Motivation
Micro-simulation
Agents based models
Disaggregated spatial meshing
BUTImpossible or too expensive to get an exhaustive data set if the number of agents is large
Privacy rules
Need for SYNTHETIC POPULATION
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 4 / 21
Contents
1 Motivation
2 Classical approaches
3 Why classical approaches are not feasible for the Belgian case?
4 VirtualBelgium, an innovative method
5 Temporal evolution for the synthetic population
6 Application and perspective
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 5 / 21
Classical approaches
IPFP: Iterative Proportion Fitting Process
Idea: Integrating an aggregate dataset (target) with a disaggregate dataset (seed).
+ ...Significant
sample
Set of consistent margins
Synthetic Population
”Cloning” entities from the sampleProblems for bi-level approach (e.g. individuals and households) but solved by (Guo and Bhat,2007)
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 6 / 21
Classical approaches
IPFP: Iterative Proportion Fitting Process
Idea: Integrating an aggregate dataset (target) with a disaggregate dataset (seed).
+ ...Significant
sample
Set of consistent margins
Synthetic Population
”Cloning” entities from the sampleProblems for bi-level approach (e.g. individuals and households) but solved by (Guo and Bhat,2007)
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 6 / 21
Classical approaches
IPFP: Iterative Proportion Fitting Process
Idea: Integrating an aggregate dataset (target) with a disaggregate dataset (seed).
+ ...Significant
sample
Set of consistent margins
Synthetic Population
”Cloning” entities from the sampleProblems for bi-level approach (e.g. individuals and households) but solved by (Guo and Bhat,2007)
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 6 / 21
Classical approaches
IPFP: Iterative Proportion Fitting Process
Idea: Integrating an aggregate dataset (target) with a disaggregate dataset (seed).
+ ...Significant
sample
Set of consistent margins
Synthetic Population
”Cloning” entities from the sample
Problems for bi-level approach (e.g. individuals and households) but solved by (Guo and Bhat,2007)
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 6 / 21
Classical approaches
IPFP: Iterative Proportion Fitting Process
Idea: Integrating an aggregate dataset (target) with a disaggregate dataset (seed).
+ ...Significant
sample
Set of consistent margins
Synthetic Population
”Cloning” entities from the sampleProblems for bi-level approach (e.g. individuals and households) but solved by (Guo and Bhat,2007)
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 6 / 21
Contents
1 Motivation
2 Classical approaches
3 Why classical approaches are not feasible for the Belgian case?
4 VirtualBelgium, an innovative method
5 Temporal evolution for the synthetic population
6 Application and perspective
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 7 / 21
Why classical approaches are not feasible for the Belgian case?
Generic problem of categories not present in the sample
No representative sample available for the Belgian case
Inconsistencies in the margins
Contingency table Source Margins Prop.municipality × gender × age GeDAP, 2001 405.491 1,00municipality × hh type GeDAP, 2001 380.653 0,94municipality × education level GeDAP, 2001 426.372 1,05municipality × status GeDAP, 2001 396.594 0,97district × hh type × age INS, 2001 357.884 0,88district × education level INS, 2001 398.582 0,98
Example of data inconsistencies for the district of Charleroi
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 8 / 21
Why classical approaches are not feasible for the Belgian case?
Generic problem of categories not present in the sample
No representative sample available for the Belgian case
Inconsistencies in the margins
Contingency table Source Margins Prop.municipality × gender × age GeDAP, 2001 405.491 1,00municipality × hh type GeDAP, 2001 380.653 0,94municipality × education level GeDAP, 2001 426.372 1,05municipality × status GeDAP, 2001 396.594 0,97district × hh type × age INS, 2001 357.884 0,88district × education level INS, 2001 398.582 0,98
Example of data inconsistencies for the district of Charleroi
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 8 / 21
Why classical approaches are not feasible for the Belgian case?
Generic problem of categories not present in the sample
No representative sample available for the Belgian case
Inconsistencies in the margins
Contingency table Source Margins Prop.municipality × gender × age GeDAP, 2001 405.491 1,00municipality × hh type GeDAP, 2001 380.653 0,94municipality × education level GeDAP, 2001 426.372 1,05municipality × status GeDAP, 2001 396.594 0,97district × hh type × age INS, 2001 357.884 0,88district × education level INS, 2001 398.582 0,98
Example of data inconsistencies for the district of Charleroi
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 8 / 21
Why classical approaches are not feasible for the Belgian case?
Generic problem of categories not present in the sample
No representative sample available for the Belgian case
Inconsistencies in the margins
Contingency table Source Margins Prop.municipality × gender × age GeDAP, 2001 405.491 1,00municipality × hh type GeDAP, 2001 380.653 0,94municipality × education level GeDAP, 2001 426.372 1,05municipality × status GeDAP, 2001 396.594 0,97district × hh type × age INS, 2001 357.884 0,88district × education level INS, 2001 398.582 0,98
Example of data inconsistencies for the district of Charleroi
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 8 / 21
Why classical approaches are not feasible for the Belgian case?
Generic problem of categories not present in the sample
No representative sample available for the Belgian case
Inconsistencies in the margins
Contingency table Source Margins Prop.municipality × gender × age GeDAP, 2001 405.491 1,00municipality × hh type GeDAP, 2001 380.653 0,94municipality × education level GeDAP, 2001 426.372 1,05municipality × status GeDAP, 2001 396.594 0,97district × hh type × age INS, 2001 357.884 0,88district × education level INS, 2001 398.582 0,98
Example of data inconsistencies for the district of Charleroi
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 8 / 21
Contents
1 Motivation
2 Classical approaches
3 Why classical approaches are not feasible for the Belgian case?
4 VirtualBelgium, an innovative method
5 Temporal evolution for the synthetic population
6 Application and perspective
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 9 / 21
Objectives
A Virtual population for Belgium
± 11.000.000 individuals
± 4.350.000 households
589 municipalities (LAU2)
Individuals and households interacting
Characterization by attributes influencing the travel behaviour
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 10 / 21
Objectives
A Virtual population for Belgium
± 11.000.000 individuals
± 4.350.000 households
589 municipalities (LAU2)
Individuals and households interacting
Characterization by attributes influencing the travel behaviour
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 10 / 21
Objectives
A Virtual population for Belgium
± 11.000.000 individuals
± 4.350.000 households
589 municipalities (LAU2)
Individuals and households interacting
Characterization by attributes influencing the travel behaviour
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 10 / 21
Objectives
A Virtual population for Belgium
± 11.000.000 individuals
± 4.350.000 households
589 municipalities (LAU2)
Individuals and households interacting
Characterization by attributes influencing the travel behaviour
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 10 / 21
Characteristics of our method
no need to be fed with a representative sample of the population
feasible for case where that could exist incoherencies amongst the margins
allowing building a synthetic population of individuals gathered in households according toinformation (margins) related to one level (individuals) or the other one (households).
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 11 / 21
Characteristics of our method
no need to be fed with a representative sample of the population
feasible for case where that could exist incoherencies amongst the margins
allowing building a synthetic population of individuals gathered in households according toinformation (margins) related to one level (individuals) or the other one (households).
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 11 / 21
Characteristics of our method
no need to be fed with a representative sample of the population
feasible for case where that could exist incoherencies amongst the margins
allowing building a synthetic population of individuals gathered in households according toinformation (margins) related to one level (individuals) or the other one (households).
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 11 / 21
Characteristics of our method
no need to be fed with a representative sample of the population
feasible for case where that could exist incoherencies amongst the margins
allowing building a synthetic population of individuals gathered in households according toinformation (margins) related to one level (individuals) or the other one (households).
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 11 / 21
Virtual Belgium: a new synthetic population generator
...
Source 1
Source 2
Inconsistencies
processing
Source n
Synth Pop Generator
Goal
A synthetic population generator han-dling the data inconsistencies due to theuse of various sources
Data sources:
INS, GeDAP (UCL), MOBEL, . . .
2 levels of aggregation:- Municipality- District
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 12 / 21
Virtual Belgium: a new synthetic population generator
...
Source 1
Source 2
Inconsistencies
processing
Source n
Synth Pop Generator
Goal
A synthetic population generator han-dling the data inconsistencies due to theuse of various sources
Data sources:
INS, GeDAP (UCL), MOBEL, . . .
2 levels of aggregation:- Municipality- District
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 12 / 21
Virtual Belgium: a new synthetic population generator (2)Main principles
General philosophy
Construct individuals and households by drawing their characteristics or members randomly
within the relevant distributions at the most disaggregate level available;
while maintaining known correlation structures.
A 3-steps procedure applied to each municipality
1 Generation of Ind : a pool of individuals.
2 Estimation of Hh: the households’ attributes joint-distribution.
3 Construction of the synthetic households by drawing individuals from Ind .
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 13 / 21
Virtual Belgium: a new synthetic population generator (2)Main principles
General philosophy
Construct individuals and households by drawing their characteristics or members randomly
within the relevant distributions at the most disaggregate level available;
while maintaining known correlation structures.
A 3-steps procedure applied to each municipality
1 Generation of Ind : a pool of individuals.
2 Estimation of Hh: the households’ attributes joint-distribution.
3 Construction of the synthetic households by drawing individuals from Ind .
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 13 / 21
Virtual Belgium: a new synthetic population generator (2)Main principles
General philosophy
Construct individuals and households by drawing their characteristics or members randomly
within the relevant distributions at the most disaggregate level available;
while maintaining known correlation structures.
A 3-steps procedure applied to each municipality
1 Generation of Ind : a pool of individuals.
2 Estimation of Hh: the households’ attributes joint-distribution.
3 Construction of the synthetic households by drawing individuals from Ind .
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 13 / 21
Example of outcomes
VirtualBelgiumNAXYS - 2011
0 20 40 6010
Km
12.3 to 19.2 %
19.3 to 20.8 %
20.9 to 22.1 %
22.2 to 23.5 %
23.6 à 31.8 %
60+ years old individuals
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 14 / 21
Contents
1 Motivation
2 Classical approaches
3 Why classical approaches are not feasible for the Belgian case?
4 VirtualBelgium, an innovative method
5 Temporal evolution for the synthetic population
6 Application and perspective
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 15 / 21
Temporal evolution for the synthetic population
Data source: Statbel
On a year per year basis
using
known fecundity and mortality rates, . . .
transition matrices
discrete choice models
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 16 / 21
Temporal evolution for the synthetic population
Data source: Statbel
On a year per year basis
using
known fecundity and mortality rates, . . .
transition matrices
discrete choice models
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 16 / 21
Temporal evolution for the synthetic population
Data source: Statbel
On a year per year basis
using
known fecundity and mortality rates, . . .
transition matrices
discrete choice models
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 16 / 21
Contents
1 Motivation
2 Classical approaches
3 Why classical approaches are not feasible for the Belgian case?
4 VirtualBelgium, an innovative method
5 Temporal evolution for the synthetic population
6 Application and perspective
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 17 / 21
Application to travel behaviours: Activity-based model
Activity chains assignment
1 Random draw of an activity chain A ∈ Ai = set of activity chains for the individual type i
2 Spatial and temporal localization ∀a ∈ A
NB: activity chains start and end at the individual’s house
Data source: MOBEL (2001)
10.000 different activity chain patterns and 12 activity purposes
192 individual types
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 18 / 21
Application to travel behaviours: Activity-based model
Activity chains assignment
1 Random draw of an activity chain A ∈ Ai = set of activity chains for the individual type i
2 Spatial and temporal localization ∀a ∈ A
NB: activity chains start and end at the individual’s house
Data source: MOBEL (2001)
10.000 different activity chain patterns and 12 activity purposes
192 individual types
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 18 / 21
Application to travel behaviours: Activity-based model
Activity chains assignment
1 Random draw of an activity chain A ∈ Ai = set of activity chains for the individual type i
2 Spatial and temporal localization ∀a ∈ A
NB: activity chains start and end at the individual’s house
Data source: MOBEL (2001)
10.000 different activity chain patterns and 12 activity purposes
192 individual types
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 18 / 21
Example of outcomes
Assignment of traffic due to activity chains
Namur case
using MATSIM framework for assignment
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 19 / 21
Example of outcomes
Assignment of traffic due to activity chains
Namur case
using MATSIM framework for assignment
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 19 / 21
Perspective
Virtual Belgium
Residential choices
Activity Chains :
- assignment
- localization (statistical sectors)
Social Networks
Dynamic evolution
Epidemiology
New variables
...
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 20 / 21
Thanks for your attention
Cornelis et al. (naXys -GRT, UNamur) VirtualBelgium NTTS2013 21 / 21