KTH-Texxi Project 2010

AG2421 – A GIS Project
Geoinformatics, KTH, Period 2, 2010
Gyözö Gidofalvi
T.A. Jan Haas
Demand-Responsive Transit (DRT) Service in the Stockholm Area
Group 1
Adeel Anwar, Alexander Jacob, Mahnaz Narooie, Ehsan Saqib, Annmari Skrifvare, li e oglio

Demand responsive Transit Service in the Stockholm Area
2 | P a g e
Table of Contents
Chapter 1 Introduction
Chapter 2 Study Area and Data Description
Chapter 3 Methodology
Chapter 4 Results
Chapter 5 Discussion
Chapter 6 Conclusion
References
Appendix

3 | P a g e
Chapter 1 Introduction
Demand Responsive Transport, known also as Demand-Responsive Transit (DRT), is an
advance form of public transport, which is characterized by being user-oriented and flexible
in routing and schedule. The service is normally run by public transport society / local transit
authority (or municipalities) or co-founded by the public sector.
DRT is used to achieve the need of transport in scattered-low density areas whit low
passengers demand, where scheduled bus lines are hence not feasible (EU report, 2002). The
system uses small or medium vehicles, operating in share-ride mode, in which pick-up and
drop-off locations are optimized on passenger needs.
For the GIS project the feasibility for starting a DRT service in the Stockholm area has been
analyzed, based on the TEXXI concept (The Transit Exchange for the XXIst
Century / The
Shared Taxi you Text). The main goal is to support the implementation of such a service with
a tool for finding the best areas for a pilot project. The presentation layer of this tool is an
interactive web frontend, which is connected to a spatial database enriched by a lot of self
written functions to serve the special needs of this project.

4 | P a g e
Chapter 2 Study Area and Data Description
2.1 Study Area
Stockholm County consists of 26 municipalities and covers an area of 6519 km2. In 2008 the
county had 1 949 516 inhabitants. This makes it the most densely populated area in Sweden.
(Nationalencyklopedin) Public transport in Stockholm includes subway, commuters train,
light rail and buses covering large areas of the county from Bålsta and Märsta in the north to
Södertälje and Nynäshamn in the south. (Storstockholms Lokaltraffik)
Figure 1 Study area (map from Eniro Dec 2010)

5 | P a g e
2.2 Data description
2.2.1 Election statistics
National electoral data from 2010 was provided. Administrative borders of different scale
were given in vector shape files, in SWEFER99. Excel files with the corresponding attribute
data were also given. We decided that it was hard to make any well founded assumptions
about travel behavior and political party and did not use this data.
2.2.2 Extended mosaic
Mosaic data of geodemographics were provides in vector shape files on two different levels,
postal code and mosaic area. The attributes of the geodemographics were age, education, type
of economic or industrial sector, healthcare, income, housing, cars and day and night
population etc. Year of acquisition is unknown.
2.2.3 Points of interests
Points of interest were provided by Dong Fang in point data format. The points included
amusement, restaurants and bus stops. Finally we did not use the data.
2.2.4 Road network
Road network data and land cover data were provided in vector shape files in SWEFER99
TM, RH 2000. The data includes road, railroad, built up area, water areas etc covering
Stockholm County. Year of acquisition is unknown.
2.2.5 Simulated travel demand
Simulated demand was given in CSV matrix files either car trips or non car trips. The files
contain information about origin and destination of trips for 24 hour period and cover purpose
and mode. Peak or off-peak hour information was also given. Corresponding geographical
information was made available in shape files.
2.2.6 Taxi positions & journeys
The taxi data set was provided directly into a database. It covered 1500 taxis and their GPS
position during certain hours of the day. Since taxi data neither covered our area spatially (all
trip zones) nor in time (24 h period), we chose not to use it.

6 | P a g e
Chapter 3 Methodology
3.1. Literature review
The DRT run by TEXXI is oriented towards market results and the aim is not the one that
characterizes the service provided by local transport authority. In order to investigate the
opportunity/feasibility for this service in the Stockholm area, a literature review has been
carried out in order to identify the potential costumers.
Since the service is run y p iv e comp ny, he po en i l co ume ’ indic o h ve een
based on DRT studies, combined with car sharing one. DRT literature enlighten the concept
of the service and the indicators used in different case studies, which are useful to indentify
the system profile, but are mainly oriented towards the public offer of transport means.
Therefore some indicators (age, car ownership, income) have been checked and combined
with the one coming from the car-sharing/pooling field, based on similar concept of sharing
ride, but better oriented towards profitability.
Acknowledgements
Valmyndigheten – Electoral data 2010
Experian - Mosaic data
openstreetmap.org - Points of interests
Lantmäteriet Metria – Road network
3.2 Data preparation
3.2.1 Cleanup & Selection
All data needed to be checked for its usability and validity. To also increase the clarity in the
data, a process of selection of relevant data needed to be performed. The shape files
containing the road network data for example contain a lot of irrelevant data for us who just
want to use it as a background for our images. The O/D matrices forming the simulated travel
demand covered a larger area than the corresponding shape file. For us only data that is
spatially referencable was of interest. Due to that fact we removed those pairs of origin and
destination which were not referencable.
We l o couldn’ find in e e ing co el ion e ween vo ing eh vio nd he u ge of
DRT-services in our literature review and discarding this data set completely.
Finally we matched the factors found from the literature review against our demographic data
source namely the mosaic data and selected only those which were of relevance for the
project.

7 | P a g e
3.2.2 Fusion
All of our analysis is centered at trip zones level. The demographic data such as population,
age level, etc was not provided at that level so a matching/fusion was required to have the
data at the trip zone level. The data which was available to us was in two forms: value data
and percentage data. By value data we mean that we were provided a whole value such as
total population in a certain area. As for percentage data we had the percentage of people in
different categories. An example of percentage data is the percentage of people in different
age categories such as Age-0-9, Age-10-19 etc. Due to different nature of data, it required
two different methods of fusion as shown in figure xx. Both the methods assume uniform
distribution of data.
For the value data first the spatial intersection was calculated between trip zones and
mosaic/postal code zones. Secondly a portion of value based on area of intersection from the
postal/mosaic was assigned to trip zone. For example if a mosaic area A has population 10
and if it lays 30% in trip zone X and 70% in trip zone Y then the population will be divided
in such a way that X will have 3 people while Y will take the rest 7.
As for percentage data, first all the areas which are intersecting a particular trip zones are
added. If we have two areas of intersection a1 and a2, we get sum S = a1+a2. Secondly a
weight is assigned to each of area with respect to its contribution to the sum. The weight w1
for area a1 will be calculated by dividing its area by sum: a1/S. In the third step a percentage
value of each category in an area is multiplied with the weight calculated in previous step. In
the final step all the values calculated for a particular category are summed to get the final
value for the trip zone.
How to match mosaic data into trip zones:
Figure 2 Conceptual matching – contribution to each trip zone: value (left) and percentage (right)

8 | P a g e
3.3. Database:
3.3.1 Set up database
The database chosen for this project is PostgreSQL (8.4.1) in combination with its spatial
extension PostGIS (1.5.3).
3.3.2 Creation & import of tables
Almost all provided data was migrated into one big database to make inter relations easily
possible. All the spatial data was additionally transformed to trip zone level beforehand, as
explained above.
Figure 3 Database and import of data
3.3.2.1 From shape files
All data, existing in ESRI shape format, was imported into postgresql using a plug-in for that
purpose which is provided with PostGIS for the pgAdmin III tool. With this tool you create a
new spatially extended table for saving the geometries of the shapes as well as all there
related non-spatial attributes. For the geometry also the reference system needs to be
specified to create corresponding constraints on the data set.
3.3.2.2 From csv files
To import the simulated travel demand, which was provided in matrix form, a small java tool
was written. This tool can operate in two different modes. It can connect directly to the
database and then reads all the information from the different file matrices populating the
database with insert and update queries or it writes back the data in a new file that is in
relational format and thus can easily be imported in the database using the copy command.
The latter is to be preferred for large data sets due to the significant faster import time.

9 | P a g e
3.4. Demand generation and distribution of DRT service
Figure 4 Conceptual matching – contribution to each trip zone: percentage (left) and values (right)
The figure above shows the conceptual tasks to fulfill to create a dynamic model of the
demand expressed in terms of trips per day between the trip zones. The O/D matrices give an
a-priori information about the travel behavior. That means we can get an idea from where to
where people like to go and what time it takes. From the demographic data the potential
customer can tried to be found and give some clue how many people would use the new
service which can be interpreted as demand for the service. Combining those two sources in a
meaningful way opens the possibility to then study the travel behavior of the potential
customer. For this we chose the approach of the gravity model!

10 | P a g e
3.4.1 Gravity model
The general idea of a gravity model is that a trip from one zone i will be attracted to another
zone j depending on the extent of activity in zone j related to a trip purpose and a friction
factor between the zones. (Kutz, M, 2004)
Tij = Trips between i and j
Pi = Trips produced in zone i
Aj = Trips attracted to zone j
Fij = 1 / travel time
A gravity model gives the opportunity to analyze potential flows by clustering analysis to
find most interesting zones.
Figure 5 Gravity model parameters
i j ij
ij
j ij
1
PA F
T
A F
n
j



11 | P a g e
3.4.1.1 Attraction
The attraction is calculated as the sum of trips that are pointing to one zone, that is the total
demand of every individual zone.
Figure 6 Attraction
3.4.1.2 Friction factor (Travel time)
The friction between two zones is approximated with the inverse of the travel time.
Longer time gives the higher friction.
3.4.1.3 Find Trips generated by potential customers
The main variables set identified as fundamental to point out a potential costumers group for
DRT service are:
- Individuals aged 21-39: they represent the working class, with high mobility needs.
- High proportion of renters, non-family households and single-person households: this
parameter is related with urban density, enlarging the numbers of people living in an
area, and therefore the potential costumers.
- Average vehicles per household. Researchers have shown the correlation between
high density and limited numbers of car owners due to parking costs/congestion
charges and availability of public services/shops. Dense cities, pedestrian and bicycle
friendly are a key aspect for the success of sharing transport service (Andrew et al.,
2006).
- Level of education: people having high level of education shown a general inclination
towards innovation, which characterize the DRT service.

12 | P a g e
- Level of income: DRT service is less expensive compared to taxi, but still more than
public transport. People having medium-high level of income are likely to use more
this service than others.
Find the function: pattern by indicators (mosaic data attributes)
Population Based
Age distribution: 20-39 years
Level of education: Bachelor or higher
Number of cars/household: 0-1
Housing: Apartment block (indicator for high density districts)
Income level: medium-high
Geographic Based
High density areas
Land use mix/point of interest (bu ine cen e , comme ci l e , lei u e, chool, …)
Although a potential customer can be found, five different potential customer profiles were
created with varying age and income levels. This enables the choice of seeing how different
profiles behave with different sets of parameters in a sensitivity analysis.
Age Income
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
Table 1 Potential customer profiles

13 | P a g e
3.4.1.4 AHP - weighting
The attributes of our potential customer are correlated. To be able to determine an absolute
number of people in one zone, without accounting for someone twice, AHP weighting was
introduced. The pair wise comparison is set up with the criterion of correlation in an n*n
matrix. Traditionally the scale for pair wise comparison is 1-9 implying an ordered scale but
correlation takes values [1,-1] and we consider each attribute equal in value and it is only the
weight that is our unknown parameter. By normalized columns averaging and normalize the
rows of the matrix we can arrive at the priority matrix which gives the weight for each
attribute. The sum of he weigh i lw y 1 nd he pe cen ge ep e en e ch i u e ’
relative value. (Karlsson J, Ryan K, 1997)
It can be interpreted as that each attribute is one layer and each trip zone is one unit in the
layer. The weights of the priority matrix are multiplied which respective attribute value and
summed up to get number of absolute trips generated by potential customers per zone.
Table 2 AHP example (priority matrix and resulting weights)

14 | P a g e
3.5 Clustering
Once the demand of the new service is distributed using the gravity model, the question rises
how to interpret this data best and how to utilize this information to find the best areas for
testing the service? This is where clustering is used. The problem we want to solve with our
clustering is to find a small network of trip zones that have a strong possible usage of our
service. The key features to be clustered are strong flow and a spatial component, that puts
weight on the fact, that zones should neither be too close nor too far away from each other.
Several methods for this purpose are available within literature for example the Ripleys K
index or the K-Means clustering method. None of those was however directly available in the
database. The first one was made available through an interface to use R a statistical
programming tool within the database, but it turned out that this interface is far too slow to
operate on our large data set. Due to this fact, we started developing our own clustering
methods.
3.5.1 Clusteredness as a quality measure
The two ideas we had for clustering are both based on the same idea of maximizing the flow
within the cluster. That means the sum of all trips between the trip zones that are part of a
cluster. From the figure below you can see an example of two clusters and how they would
rank in relation to each other.
Figure 7 Cluster ranking
3.5.2 Two own methods of clustering
The two methods we developed are mainly differing in how to select those zones that are part
of the cluster. The first proposed method starts from a zone and looks for the zone with the

15 | P a g e
highest flow to. This zone and the original zone are then considered as part of the cluster and
the next zone is selected in a way that from both those zones the flow to the new zone is
maximized again. This process continues until the final cluster size is reached. The second
method is stronger based on heuristics and takes from the original zone the top destinations
and calculates the clusteredness from this set of zones.
The latter is computational more efficient but of course more of a guess than the first one.
Comparison between both methods showed however that the resul don’ diffe oo much
and thus the second one was chosen for the online application.
Both methods can in theory be used in the application because the signature of their functions
are identical in terms of input and output parameters. The figure below shows those
parameters.
Figure 8Clustering parameters and result
The distance filter’ pu po e i o limi p i l ex ent of a cluster to reasonable extend and
thus omit clusters scattered over the whole Stockholm area. The demand filter mainly omits
negative demand created from a car density of more than one car per person in a zone. The
cluster size finally also gives some limitations to the spatial extent.

16 | P a g e
3.6. Visualization - Open Layers
Apache Web server, PHP and open layers were used to create the web mapping application for
visualization. Both spatial and non spatial data was fetched from PostgreSQL database in the form
of xml. At the start of the application, upon client request all the basic data from trip zones is
brought from the database by Apache Web Server and PHP. Afterwards the client uses open
layers to renders the map. The spatial data in brought only once at the start. In the later stages
of analysis only values of some attributes are dynamically changed to create different
thematic maps.
The visualization supports two main analyses. First one is concerned with viewing best ranked zones
while the second one shows the cluster for a selected zone. The user is requested to give input for
carrying out the analysis. The input parameter includes:
Input variable Comments
Demographic type Select one of the five types described earlier
Top Best The number of best zones to display
Distance Min, Max Distance filter to limit too close or too far zones
Demand Min A threshold on minimum demand
Cluster size The size for the cluster
When a user provides these inputs, a sql based request is send through PHP to the database and the
results are brought back to client in the form of xml. The client reads the xml and updates the values
of attributes of trip-zone layer thus changing the thematic map to indicate the requested result.
The visualized map also supported tools such as navigate, zoom, pan, and query.
Figure 9 Interface in web application

17 | P a g e
Chapter 4 Results
Since the main result of our project is not the analysis itself but a tool to perform it we show
here in this section an example of one possible outcome of an analysis performed with our
tool. Additionally also some of the preprocessed data that is major input to this analysis is
presented.
4.1Analysis scenario
As presented in the methodology section several different O/D matrices as an outcome of the
gravity model were computed to represent the travel behavior of potential customers of the
service. 5 different types are currently available. For the scenario type 4 was chosen because
it has the strongest similarity to the type of customer found from the literature review. This is
a young well educated person living in dense populated areas with middle to high income. To
then perform the clustering analysis to find most suitable areas for the implementation of a
pilot project several more parameters needs to be defined by the user. Those include the
spatial extend and density of the cluster as well as the minimum demand for specific pairs of
origin and destination. To make potential trips for the service not too short, a minimum
average distance of 3 km was defined and to keep the extend still limit to a local scale of only
a few neighborhoods, the maximum distance was limited to 8 km. To exclude all negative
demand and very sparse used trips the demand minimum was set to 0.5 trips per day. The
cluster was also set to consist of no more than 10 trip zones.

18 | P a g e
4.2Scenario results
Using this setup the following top-clustered areas were found:
1.) Sollentuna
2.) Hammarbyhöjden/Björkhagen
3.) Södertälje
Figure Top 3 clusters (map from Eniro Dec 2010)

19 | P a g e
4.2.1 Sollentuna result in detail
The greater Sollentuna area had the highest demand with 235 trips/day. The cluster also
includes interesting areas as Kista, Akalla and Husby and also Norrortsleden (road 265). To
name a few features in the area of Kista hosts KTH and Stockholm university divisions.
(kth.se, 2010) Perhaps foremost it is the home the Kista science city which is a world known
ITC cluster. There is also a mall for shopping and restaurants. (kista.com, 2010). This makes
it a center for many different groups of people as students, shoppers and businessmen apart
from that the cluster also includes large residential areas.
This variety of people would likely benefit a DRT service and covers a range of purposes.
Kista area is connected by the close by passing of motorways E4 and E18, which is an
advantage for the DRT service. On the downside, the southern zones of this cluster are
already quite well connected by public transport as it lies along the blue subway line.
Figure 10 Trip zone map with Sollentuna cluster highlighted (yellow)

20 | P a g e
Figure 11 Top five flows in Sollentuna cluster

21 | P a g e
4.2.2 Hammarbyhöjden/Björkhagen result in detail
Hammarbyhöjden/Björkhagen cluster is has the second highest demand of 228 trips/day. The
origin zone and some of the other cluster zones Bagarmossen are along the green subway line.
What is interesting with Hammarbyhöjden/Björkhagen cluster is that we can tell that some of
the highest demand comes from Älta area and north of Älta (Lovisedal, Kolarängen) which
are surrounded by a large nature area. These particular zones are connected by road 260
(north/south) and 229 (east/west passing Skärpnäck) but not by subway or commuter train
which makes it slightly isolated. This is in our favor since the only public transport in these
areas is bus.
Figure 12 Trip zone map with Hammarbyhagen cluster highlighted (yellow)

22 | P a g e
Figure 13 Top five flows in Hammarbyhagen cluster

23 | P a g e
4.2.3 Södertälje result in detail
The origin zone of the third cluster with 213 trips/day is the area north of Södertälje. Most
probably this is the demand of work related commuters of the western and northern outskirts
of Södertälje. The cluster itself include zone with road E20 (East/West) and zones west of the
central bridge. This indicates that there is demand both within Södertälje but also a possible
false demand if E20 is included, people who are just passing Södertälje or start and will
continue past Södertälje is included.
Figure 14 Trip zone map with Södertälje cluster highlighted (yellow)

24 | P a g e
Figure 15 Top five flows in Södertälje cluster

25 | P a g e
4.3General statistics of generated demand
The two following tables show the absolute demand generated by our model as well as the
percentage of population using our service according to it! A per trip zone aggregation can be
found in the figure.
Table 3 Absolut values for all types of generated demand
Table 4 Relative values for all types of generated demand
Figure 16 Aggregated in and outflow per trip zone using type 4 demand (other types can be found in
appendix).

26 | P a g e
Chapter 5 Discussion
5.1 Resulting recommendation
The scenario analysis suggested the Sollentuna solution to be the most suitable one for the
implemen ion of pilo p ojec . hi i even hough i ’ w ju cen io ce inly one of
the most interesting areas. And the parameters chosen for this scenario were not random but
those that to our knowledge based from the literature review and discussions with Gyözö
suits the wishes of the client best.
One should still be aware of the fact that this is only one of many possible outcomes using
our decision support tool. Another set of parameters might yield a completely different
solution. At least regarding the Sollentuna area in contrast to for example Södertälje we
found that it is strong in all demographic demand types which can be seen from the maps in
the Appendix.
5.2 Methodology
5.2.1 Data Usage
In the current model for the demand not all data available was used for different reasons, but
if a suitable way to utilize this data as well can be found, then probably even better results
can be achieved.
5.2.1.1 Mosaic
The Mosaic data contains the very interesting profiles of night and day population as
percentages of the mosaic population groups. Those groups give very interesting information
of what kind of people you can find in which areas. Due to the fact that this information is
given in pe cen ge of num e we don’ know, n mely he c u l popul ion d y ime o
at night time, it is impossible to translate this number into absolute demand as we did in the
current model. We were thinking about using at least for night time population the given
population of a postal code which we transformed on trip zone level as well as the
percentages but we came to the conclusion that this can only be valid for mainly residential
areas where it can be assumed that the biggest part of the population is staying at home at
night time.
5.2.1.2 Taxi
The Taxi data is very valuable in the sense that this is real data not coming from a model and
thus is not based on any assumptions. The problem with this data is, that it is not available for
all zones it is thus spatially incomplete as well as that it only covers daytime and thus also is
temporal incomplete. We tried however to get some estimates for taxi travel time for all
zones by finding a relation between car travel time and taxi travel time and then use this
relation to calculate the travel times for zones not covered in the taxi data set. It turned out
however, that the correlation between those to variables was very low (around 0.2) so that it
is impossible to calculate statistical reliable estimates from that information.

27 | P a g e
5.2.1.3 Public transport
We have the locations of a great number of bus stops as well as for the subway and trains and
this information shows clearly that they are well distributed over the whole city. Areas which
are already well connected through public transport are probably more unlikely to use the
new DRT service and some analysis maybe based on the distance to public transport stops
should be included to give this credit and make the results thus more reliable.
5.2.2 Clustering
The clustering method currently used is heavily based on heuristics and therefore c n’ e
treated as a exact measure. There are many other clustering methods available and it might be
worthwhile to put effort into evaluating those further. We see clustering as one of the most
important methods to find well connected zones and thus a perfect tool to select the pilot area.
Maybe it is even worth to rewrite some of the existing algorithms in a database function to
increase the computational efficiency. As stated before the approach to utilize external
programs through interfaces turned out to be far too slow to be considered useful.
5.2.3 Travelling is no purpose
The analysis now is based on the travel pattern provided from the O/D matrices. It is however
not known to us under which exact assumptions those are created and thus not possible to
derive more distinct purposes of travelling from those patterns. The knowledge about purpose
connected to the information of which kind of people are living in specific zones can give
very deep insight about the travel behavior. A possible solution might be to use the point of
interest data or even an extended version of it since the current version only includes data
about night time activities. Using this data for the attraction in a gravity model which
distinguishes between the different population groups can give a very distinct purpose based
travel pattern.
5.2.4 Price politics
A very important component for the acceptance of this service is also the financial
component for the customer. It should take into account the prices for public transport and
other similar services in the study area. The price can of cause be higher than of public
transport given credit to the higher convenience factor.
5.2.5 Time
In the current model the peak hour time is used as the friction factor between zones in the
gravity model. A more detailed analysis taking into account the two peaks during a day and
the off peak hours can give a more realistic model about when people are where. However,
fo he ove ll dyn mic of he zone i doe n’ pl y n impo n ole nd does not change the
outcome of our current analysis.

28 | P a g e
Chapter 6 Conclusion
We can finally conclude that we succeeded in creating a web application for finding suitable
areas for a pilot project of a DRT service in the Stockholm area! The current version gives
the customer the possibility to run the sensitivity analysis on 5 different data sets. Based on
the chosen clustering parameters he can find both a top ranking of best zones as well as some
more detailed information about the clusters those top zones belong to.
Of cause there is still a lot of room for improvement both in the interaction with data over the
interface as well as with the underlying data itself.
One nice future feature would be to generate the demand on the fly based on a selection of
demographic parameters as well as some other influencing factors like the price and
accessibility of public transport etc. For both the latter factors realistic models needs first to
be created.
Another point is the clustering which is now a very fast but not very reliable clustering. The
potential of other methods should be investigated, both in terms of the clustering result itself
as well as the performance when applied to the complete data set.

29 | P a g e
References
Andrew J., Douma F., Developing a Model for Car Sharing Potential in Twin Cities
Neighborhoods, Transportation Research Board 85th Annual Meeting, 2006 (www.trb.org);
available at www.mdt.mt.gov/research/docs/trb_cd/Files/06-2449.pdf.
Celsor C., Millard-Ball A., Where Does Car-Sharing Work? Using Gis To Assess Market
Potential, Annual Meeting of the Transportation Research Board, 2007
EU Project Penelope (Promoting ENergy Efficiency to Local Organisations through
dissemination Partnerships in Europe), Demand Responsive Transit service (DRTs):
PersonalBus - Tuscany - Florence – Italy, Report, 3 September 2002
Horn M.E.T., Multi-modal and demand-responsive passenger transport system: a modeling
framework with embedded control system, Transport Research Part A, 36, 2002, Pages 167-
188
Hua T.-L., Sheu J.-B., A fuzzy-based customer classification method for demand-responsive
logistical distribution operations, Fuzzy Sets and Systems, Volume 139, Issue 2, 16 October
2003, Pages 431-450
Karlsson J, Ryan K, A Cost-Value approach for prioritizing requirements, IEEE Software,
1997, pp 67-74
Kista Science City, http://www.kista.com, 2010-12-20
KTH, http://www.kth.se/ict/campusitu/campus-kista-1.7534, 2010-12-20
Kutz, M, Handbook of transportation Engineering, New York: MacGraw-Hill, 2004
Mageean J., Nelson J. D., The evaluation of demand responsive transport services in Europe,
Journal of Transport Geography 11, 2003, Pages 255–270
Palmer K., Dessouky M., Abdelmaguid T., Impacts of management practices and advanced
technologies on demand responsive transit systems, Transportation Research Part A, 38, 2004,
Pages 495–509
PostGIS, http://www.postgis.org , 2010-12-19
PostgreSQL, http://www.postgresql.org , 2010-12-19
Quadrifoglio L., Xiugang L., A methodology to derive the critical demand density for
designing and operating feeder transit services, Transportation Research Part B, 43, 2009,
Pages 922–935
Quadrifoglio L., Maged M. Dessouky, Fernando Ordonez, Transportation Research Part A,
42, 2008, Pages 718–737

30 | P a g e
Stockholms län. http://www.ne.se/kort/stockholms-län, Nationalencyklopedin, 2010-12-17.
Storstockholms Lokaltrafik, www.sl.se, 2010-12-17.
A simulation study of demand responsive transit system design.
Yim Y.B., Khattak Asad J., Personalized Demand Responsive Transit Systems, California
Path Program Institute Of Transportation Studies University Of California, Berkeley, October
2000

31 | P a g e
APPENDIX:

32 | P a g e

KTH-Texxi Project 2010

More Related Content

What's hot (20)

Viewers also liked (17)

Similar to KTH-Texxi Project 2010 (20)

More from Texxi Global (20)

Recently uploaded (20)

KTH-Texxi Project 2010