1. Demand-Responsive Transit (DRT) Service in the Stockholm Area
Group 1
Adeel Anwar_Alexander Jacob_Mahnaz Narooie_Ehsan Saqib _Annmari Skrifvare_Elisabetta Troglio
AG2421 – A GIS Project
Geoinformatics, KTH, Period 2, 2010
Gyozo Gidofalvi
T.A. Jan Haas
3. Introduction
Objective:
• Create a decision-making-support tool for finding the optimal area to implement
a pilot project for a taxi service in a DRT manner.
• DRT stands for demand responsive transit!
• Created service based on a demand model
• Model contains also distribution of demand in terms of trips between zones
• Core of our analysis is a database combining a variety of different data sources of
both spatial and non-spatial character.
4. Methodology
1. Literature review:
• Methodology overview
• Indicators
2. Data cleaning & selection (relevant data):
• Trip zones /OD matrices, road network cleaning
• Mosaic – finding the useful indicators according to literature
3. Data fusion – ArcGIS level
• Fusing mosaic data to trip zones
• Fusing population data to trip zones
5. 4. Database:
• Set up Postgres database
• Creation & import of tables (.shp)
• Population tool (JAVA) – (.csv) files into database
Data in the database:
• One common reference system
• Basemma.shp – trip zones as reference zone
• OD matrices – main information
• Own calculations
O/D Matrices Mosaic data
Public
Transports
Points of
interest
Road network Taxi data
External
Java
program
ArcGIS,
Microsoft
Access
Digitalization
P.T. system
Given
data
Given Data
Postgres import function
of shape files
Methodology
6. O/D Population segmentation ->
Potential customer
Potential customer flows
+
Extended O/D
Methodology
5. Demand generation and distribution (conceptual model):
7. 6. Visualization - Open Layers
• Dynamic map by changing parameters
Methodology
8. Data
Import
Database
Analysis preparation
Analysis
Results
O/D Matrices Mosaic data
Public
Transports
Points of
interest
Road network Taxi data
External
Java
program
Attraction
based on O/D,
aggregated on
flows
Gravity model
Trips with DRT service
ArcGIS,
Microsoft
Access
Digitalization
P.T. system
Trips produced by
Potential customers
AHP weighting
Literature
review
Clustering
10 TOP ZONES
Control with
Points of
interests
P.T.
Friction factor
Car travel time
peek (and off peek)
Given Data
Postgres import function
of shape files
Methodology
9. GRAVITY MODEL
• Attraction
• Friction factor (Travel time)
• Find Trips generated by potential customers
Gravity model gives the opportunity to analyze potential flows by clustering
analysis - find most interesting zones
Attraction
based on O/D,
aggregated on
flows
Gravity model
Trips with DRT service
Friction factor
Car travel time peek
(and off peek)
Trips produced by
Potential customers
AHP weighting
Demand generation and distribution
10. Areas with HIGH probability of car sharing members (similar group):
POPULATION BASED:
• Age distribution: 20-39 years
AND
• Level of education: University degree
AND
• Number of cars/household: 0-1
GEOGRAPHIC BASED:
• High density areas – Housing
SENSITIVITY ANALYSIS:
• Age distribution
AND
• Income
Trips produced by Potential customer
11. 20 - 39 40 - 59 150 - 399 400+
1 X X X X
2 X X
3 X X X
4 X X X
5 X X X
Defining potential customers
Age Income
12. AHP- weights
Age Income Education Housing
Age 1 1/0.144 0.208 0.488
Income 0.144 1 0.228 1/0.184
Education 1/0.208 1/0.228 1 0.357
Housing 1/0.488 0.184 1/0.357 1
0.199, 0.224, 0.309, 0.267
13. Attraction and friction factor
Sum(trips pointing to one zone)
All O/D demand included
Aggregated inflow per zone
1/ travel time
2
3
10 4
5
8
9
7
1
3
10 9 8 7 3 37 attraction
14. Gravity model
i j ij
ij
j ij
1
PA F
T
A F
n
j
Tij = Trips between i and j
Pi = Trips produced in zone i
Aj = Trips attracted to zone j
Fij = 1 / travel time
15. Clustering
• Heuristic based!
• For every zone a subset with the biggest
amount of trips to ,is selected and all inner
trips out of this “cluster” selection are
counted.
• Those are ordered by the inner-trip-count
and the top results are high-lighted on the
map
18. Selection of zones using extended flows
Top 3 clusters
1. Sollentuna (235 trips/day)
2. Hammarbyhöjden/Björkhagen (228 trips/day)
3. Södertälje (213 trips/day)
Parameters:
Type 4
3-8 km
0.5 minimum demand
Cluster size 10 zones 1
2
3
19. Selection of zones using exteflows
Top 3 clusters
1. Sollentuna
Cluster includes Greater Sollentuna, Kista, Akalla, Husby
20. Selection of zones using extended flows
Top 3 clusters
2. Hammarbyhöjden/Björkhagen
Cluster includes Älta, Kärrtorp , Bagarmossen
21. Selection of zones using extended flows
Top 3 clusters
3. Södertälje
Northern part of Södertälje
22. Resulting recomendation
Based on our analysis we suggest that the pilot project of the DRT service should
be located in Sollentuna and its neighboring areas.
It should however be noted that this is only one possible result, based on one
specific set of parameters. Different parameter sets might produce different
outcomes. We chose a set that we found reasonable based on some assumption
what range and cluster size is suitable for a taxi service pilot project as well as
the demographic group most promising from the literature review.
23. Discussion
• Travelling itself is usually no purpose
• Further analysis of characteristics of resulting zones can give clues of
more specific customer purposes (shopping, corporate, evening/night
etc.)
• POI (points of interest) can be used
• Price
• probably has a strong influence on acceptance of service
• should be oriented on competitors such as existing public transport
• maybe slighter higher due to better convenience
• Time
• Now using peak hour for worst case scenario, with possibility to extend the
analysis to off-peak hours
24. Discussion
• Data usage
• Not all data is used in the current analysis due to different problems:
1. Mosaic population profiles only in percentage for day and night but
amount of day and night population not given!
2. Taxi data neither includes all zones nor covers a 24h period, thus a
model first needs to be created to use them parallel to O/D matrices.
3. Public transport availability is high but not included in the analysis
• Clustering
• many possible solutions (e.g. Ripleys K, K – means, etc.)
• for most exact result every trip needs to be compared with every other.
• computational efficient – on the fly
25. Conclusion
• We created a web application that can be used for finding
suitable areas for a pilot project!
• It currently enables the customer to select from a set of pre-
calculated demand sets and perform simple clustering
based on mainly three parameters!
• This could be improved further by:
• for example creating demand sets freely based on all available
demographic indicators at run time.
• including a price based model
• more advanced clustering methods
• …
26. Thank you for your attention!
Feel free to open the discussion!