Professional Documents
Culture Documents
Fulltext01 PDF
Fulltext01 PDF
2014-10-10
Cisilia Hildebrand
Stina Hörtin
Norrköping 2014-10-10
Upphovsrätt
Copyright
The publishers will keep this document online on the Internet - or its possible
replacement - for a considerable time from the date of publication barring
exceptional circumstances.
The online availability of the document implies a permanent permission for
anyone to read, to download, to print out single copies for your own use and to
use it unchanged for any non-commercial research and educational purpose.
Subsequent transfers of copyright cannot revoke this permission. All other uses
of the document are conditional on the consent of the copyright owner. The
publisher has taken technical and administrative measures to assure authenticity,
security and accessibility.
According to intellectual property law the author has the right to be
mentioned when his/her work is accessed as described above and to be protected
against infringement.
For additional information about the Linköping University Electronic Press
and its procedures for publication and for assurance of document integrity,
please refer to its WWW home page: http://www.ep.liu.se/
Cisilia Hildebrand
Stina Hörtin
The work presented in this thesis has been carried out in the Division of Communications-
and Transport Systems at Linköpings University and at WSP Analysis & Strategy.
First of all we want to thank our supervisor Ellen Grumert and examiner Anders Pe-
terson at Linköpings University for their feedback during this project. We would also
like to thank our colleagues at WSP for their support. A special thanks to our super-
visor at WSP Analysis & Strategy, Christian Nilsson, that has guided us through this
thesis. Finally, we want to thank our families and friends for their support during the
years.
i
Abstract
Macroscopic traffic simulations are widely used in the world in order to provide as-
sistance in the traffic infrastructure development as well as for the strategic traffic
planning. When studying a large traffic network macroscopic traffic simulation can be
used to model current and future traffic situations. The two most common software
used for traffic simulation in Sweden today are Emme and Visum, developed by INRO
respective PTV.
The aim of the thesis is to perform a comparison between the software Emme and Visum
with respect to the assignment of public transport, in other words how passengers
choose their routes on the existing public transport lines. However, in order to make
a complete software comparison the run-time, analysis capabilities, multi-modality,
capacity to model various behavioural phenomena like crowding, fares etc. this will
not be done in this comparison. It is of interest to study the differences between the
two software algorithms and why they might occur because the Swedish Transport
Administration uses Emme and the Traffic Administration in Stockholm uses Visum
when planning public transport. The comparison will include the resulting volumes on
transit lines, travel times, flow through specific nodes, number of boarding, auxiliary
volumes and number of transits. The goal of this work is to answer the following
objective: What are the differences with modelling a public transport network in Emme
and in Visum, based on that the passengers only have information about the travel times
and the line frequency, and why does the differences occur?
In order to evaluate how the algorithms work in a larger network, Nacka municipality
(in Stockholm) and the new metro route between Nacka Forum and Kungsträdgården
have been used. The motivation for choosing this area and case is due to that it is
interesting to see what differences could occur between the programs when there are a
major change in the traffic network.
The network of Nacka, and parts of Stockholm City, has been developed from an
existing road network of Sweden and then restricted by "cutting out" the area of interest
and then removing all public transportation lines outside the selected area. The OD-
matrix was also limited and in order no to loose the correct flow of travellers portal
zones was used to collect and retain volumes.
To find out why the differences occur the headway-based algorithms in each software
were studied carefully. An example of a small and simple network (consisting of only a
iii
start and end node) has been used to demonstrate and show how the algorithms work
and why volumes split differently on the existing transit lines in Emme and Visum.
The limited network of Nacka shows how the different software may produce different
results in a larger public transport network.
The results show that there are differences between the program algorithms but the
significance varies depending on which output is being studied and the size of the
network. The Visum algorithm results in more total boardings, i.e. more passengers
have an optimal strategy including a transit. The algorithms are very similar in both
software programs, since they include more or less parts of the optimal strategy. The
parameters used are taken more or less into consideration in Emme and Visum. For
example Visum will first of all focus on the shortest total travel time and then consider
the other lines with respect to the maximum waiting time. Emme however, first focuses
on the shortest travel time and then considers the total travel time for other lines with
half the waiting time instead of the maximum wait time. This results in that less
transit lines will be attractive in Emme compared to Visum. The thesis concludes that
varying the parameters for public transport in each software algorithm one can obtain
similar results, which implies that it is most important to choose the best parameter
values and not to choose the "best" software when simulating a traffic network.
Keywords: assignment, Emme, macroscopic traffic simulation, public transport, Vi-
sum
Contents
Preface i
Abstract iii
List of tables ix
1 Introduction 1
1.1 Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Paper outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Research contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Travel forecasting 7
2.1 The four step travel forecasting model . . . . . . . . . . . . . . . . . . 8
2.1.1 Trip generation . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.2 Trip distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.3 Mode choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.4 Route assignment . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Public transport assignment . . . . . . . . . . . . . . . . . . . . . . . . 11
3 Software products 13
3.1 Overview of macroscopic software products . . . . . . . . . . . . . . . . 13
3.2 Emme 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2.1 Public transport assignment . . . . . . . . . . . . . . . . . . . . 15
3.2.2 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 Emme 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4 Visum 13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.4.1 Public transport assignment . . . . . . . . . . . . . . . . . . . . 22
3.4.2 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.5 Assignment parameters and examples . . . . . . . . . . . . . . . . . . . 28
3.5.1 Assignment parameter settings in Emme and Visum . . . . . . . 28
3.5.2 Example with a small network, Emme . . . . . . . . . . . . . . 30
v
3.5.3 Example with a small network, Visum . . . . . . . . . . . . . . 33
3.5.4 Main differences between the algorithms . . . . . . . . . . . . . 35
3.5.5 Comparison between literature examples . . . . . . . . . . . . . 37
4 A case study 41
4.1 Model building in Emme . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.1.1 Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2 Model building in Visum . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.2.1 Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3 Model analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.3.1 Parameter analysis . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.3.2 Line run time analysis . . . . . . . . . . . . . . . . . . . . . . . 53
4.3.3 Algorithm analysis with 100 demand . . . . . . . . . . . . . . . 54
4.3.4 Node analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4 Emme 4, extended transit assignment . . . . . . . . . . . . . . . . . . . 56
5 Results 57
5.1 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.1.1 Emme 3 and Visum: base scenario . . . . . . . . . . . . . . . . 57
5.1.2 Emme 3 and Visum: future scenario . . . . . . . . . . . . . . . 60
5.2 Model analysis results . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.2.1 Results from parameter analysis . . . . . . . . . . . . . . . . . . 63
5.2.2 Results from line run time analysis . . . . . . . . . . . . . . . . 69
5.2.3 Results from algorithm analysis with 100 demand . . . . . . . . 72
5.2.4 Results from node analysis . . . . . . . . . . . . . . . . . . . . . 73
5.3 Emme 4, extended transit assignment, results . . . . . . . . . . . . . . 76
6 Analysis 79
6.1 Comparison of the simulation results . . . . . . . . . . . . . . . . . . . 79
6.2 Software sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
6.2.1 Comparison of the parameter analysis . . . . . . . . . . . . . . . 82
6.2.2 Comparison of the line run time analysis . . . . . . . . . . . . . 89
6.2.3 Comparison of the algorithm analysis with 100 demand . . . . . 89
6.3 Comparison of the node analysis . . . . . . . . . . . . . . . . . . . . . . 90
6.4 Comparison of the algorithms for public transport assignment . . . . . 97
6.4.1 Extended transit assignment . . . . . . . . . . . . . . . . . . . . 97
References 103
Appendix 105
List of Figures
1 Illustration of the four step travel model and what is included regarding
the decision in each step . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 The small network in Emme (not to scale) . . . . . . . . . . . . . . . . 30
3 Graphic result from transit assignment of the small network in Emme . 32
4 The small network in Visum (not to scale) . . . . . . . . . . . . . . . . 33
5 Graphic result from transit assignment of the small network in Visum . 35
6 Comparison between the reproduced example results in both Emme and
Visum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
7 The network limitation area of Nacka municipality, from Google maps . 41
8 The chosen alternative for Nacka metro . . . . . . . . . . . . . . . . . . 42
9 The network built in Emme . . . . . . . . . . . . . . . . . . . . . . . . 46
10 Transit lines in the base scenario (each line is a separate color) . . . . . 46
11 Transit lines in the future scenario (each line is a separate color) . . . . 47
12 The network built in Visum . . . . . . . . . . . . . . . . . . . . . . . . 49
13 Areas that the 100 demand will be assigned between . . . . . . . . . . 54
14 The circled nodes that will be analysed . . . . . . . . . . . . . . . . . . 55
15 Simulation results from the base scenario in Emme . . . . . . . . . . . 59
16 Simulation results from the base scenario in Visum . . . . . . . . . . . 59
17 Simulation results from the future scenario in Emme . . . . . . . . . . 62
18 Simulation results from the future scenario in Visum . . . . . . . . . . 62
19 The number of boardings . . . . . . . . . . . . . . . . . . . . . . . . . . 73
20 Diagram over boarding difference between base and future scenario in
respective software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
21 Diagram over boarding difference between Emme and Visum in respec-
tive scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
22 Graphs comparing the result from the parameter analysis of boarding
time weight in each software . . . . . . . . . . . . . . . . . . . . . . . . 83
23 Graphs comparing the result from the parameter analysis of wait time
factor in each software . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
24 Graphs comparing the result from the parameter analysis of wait time
weight in each software . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
25 Graphs comparing the result from the parameter analysis of auxiliary
time weight in each software . . . . . . . . . . . . . . . . . . . . . . . . 87
26 Graphs comparing the result from the parameter analysis of n.o. transfer
penalty in each software . . . . . . . . . . . . . . . . . . . . . . . . . . 88
vii
27 Comparisons between the base and future scenario in Emme and Vi-
sum at T-centralen with respect to the number of passengers boarding,
alighting or passing through the station. . . . . . . . . . . . . . . . . . 90
28 Comparisons between the base and future scenario in Emme and Visum
at Slussen with respect to the number of passengers boarding, alighting
or passing through the station. . . . . . . . . . . . . . . . . . . . . . . . 91
29 Comparisons between the base and future scenario in Emme and Visum
at Sofia with respect to the number of passengers boarding, alighting or
passing through the station. . . . . . . . . . . . . . . . . . . . . . . . . 92
30 Comparisons between the base and future scenario in Emme and Visum
at Kungsträdgården with respect to the number of passengers boarding,
alighting or passing through the station. . . . . . . . . . . . . . . . . . 93
31 Comparisons between the base and future scenario in Emme and Visum
at Nacka Forum (bus stop) with respect to the number of passengers
boarding, alighting or passing through the station. . . . . . . . . . . . . 94
32 Node analysis made in the base scenario . . . . . . . . . . . . . . . . . 95
33 Node analysis made in the future scenario . . . . . . . . . . . . . . . . 96
List of Tables
ix
29 Results from the parameter analysis in Emme . . . . . . . . . . . . . . 63
30 Results from the parameter analysis in Visum . . . . . . . . . . . . . . 63
31 The difference between Visum and Emme with respect to the difference
between the original results . . . . . . . . . . . . . . . . . . . . . . . . 64
32 Results in Emme with the original parameter settings . . . . . . . . . . 64
33 Parameter analysis results for the boarding time weight in Emme . . . 64
34 Parameter analysis results with wait time factor in Emme . . . . . . . 65
35 Parameter analysis results with wait time weight in Emme . . . . . . . 65
36 Parameter analysis results with auxiliary time weight in Emme . . . . . 66
37 Results in Visum with the original parameter settings . . . . . . . . . . 66
38 Parameter analysis results with boarding time penalty in Visum . . . . 66
39 Parameter analysis results with the formula for origin and transfer wait
time in Visum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
40 Parameter analysis results with factor for origin and transfer wait time
in Visum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
41 Parameter analysis results with factor for access, egress and walk time
in Visum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
42 Parameter analysis results with factor for number of transfers (NTR) in
Visum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
43 The new run times in Visum compared to the original run times in Emme 69
44 Line boardings with the new run times in Visum and the original run
times in Emme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
45 Line boardings with the new and the original run times in Visum . . . 71
46 Mean in-vehicle time (minutes) for the five tests with 100 demand . . . 72
47 Number of boardings per line for the five tests with 100 demand . . . . 72
48 Node results from the base scenario in Emme . . . . . . . . . . . . . . 74
49 Node results from the future scenario in Emme . . . . . . . . . . . . . 74
50 Node results from the base scenario in Visum . . . . . . . . . . . . . . 75
51 Node results from the future scenario in Visum . . . . . . . . . . . . . 75
52 The result from using the standard transit assignment in Emme 4 and
extended transit assignment (without any additional settings) . . . . . 76
53 Result from using the option flow distribution at origins . . . . . . . . 76
54 Result from using the option flow distribution at regular nodes with
auxiliary transit choices . . . . . . . . . . . . . . . . . . . . . . . . . . 76
55 The result when using the additional setting to use flow distribution
between transit lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
56 Comparison between Emme and Visum results from the base scenario . 79
57 Comparison between Emme and Visum results from the future scenario 79
58 Comparison of the results obtained from Emme with the Trafikverket
parameter values and from Visum with the Trafikförvaltning parameter
values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
59 Absolute boarding difference between base and future scenario in respec-
tive software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
60 Boarding difference between Emme and Visum in respective scenario . 81
61 Absolute difference between the mean in-vehicle time from the analysis
with 100 demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
62 A concluding comparison between the algorithms . . . . . . . . . . . . 100
63 Difference in the number of boardings between base and future scenario
in Visum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
64 Difference in the number of boardings between base and future scenario
in Emme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
xi
1
Introduction
When studying a large traffic network macroscopic traffic simulation (with aggregated
traffic flow relations) can be used to model the current and future traffic situation.
Macro simulation is often used as a part of travel forecasting at regional and national
traffic planning authorities and companies. An advantage with this type of simula-
tion is that one can analyse and investigate a larger traffic network without investing
in expensive infrastructure first. To obtain realistic results, the model must reflect
the reality accurate enough. This is done by calibrating the models, i.e. adjusting
model parameters until the results resemble the observed or estimated data. Several
macroscopic traffic simulation tools have been developed, with various advantages and
disadvantages. For example Emme (see manual [1]), Visum (see manual [2]), Aimsun
(see website [3]), TransModeler (see website [4]) and VIPS (see report [5]).
In Sweden the most commonly used commercial macroscopic traffic simulation software
products are Emme and Visum. The Swedish Transport Administration (Trafikverket)
and several municipalities use Emme. Visum is the main macro simulation software at
the Traffic Administration in Stockholm (Trafikförvaltningen) along with other traffic
planning companies and municipalities. Emme and Visum are the two main com-
petitors of the traffic software market in Sweden and they are therefore of interest to
compare.
This project will provide knowledge of how the macroscopic traffic simulation software
Emme and Visum differ regarding traffic assignment in terms of public transport. In
order to compare these software algorithms, the existing traffic network of Nacka (a
Stockholm municipality) is studied. There are plans for an expansion of the existing
metro in Stockholm. This is an infrastructure investment by the Swedish government,
Trafikverket, Trafikförvaltningen, Stockholm County Council, Stockholm and Nacka
municipalities with on-going preliminary studies and is therefore of interest to study
further how it will affect the public transport system. This thesis will use the extension
of the metro as an example of a project often used within traffic planning. With the
help of macroscopic traffic simulation one can investigate how the metro will affect
other parts of the traffic network. The models in Emme and Visum need to be verified
and then the infrastructure project, to build a metro to Nacka, added to the modelled
networks. The results for both the scenario with and without metro will be compared
between the software products in terms of assignment of public transport. Input data
needed for both scenarios will be collected, in collaboration with WSP Analysis &
Strategy and Trafikverket, from their earlier traffic prediction studies in Sweden.
1
1 INTRODUCTION
1.1 Aim
The aim of this master thesis is to perform a comparison between the software products
Emme and Visum with respect to the assignment of public transport. Headway based
assignment in these programs will be used, which means that the travellers will only
have information about the travel times and the line frequency.
The traffic network that will be the study area is Nacka municipality and the new
stretch of the metro between Nacka and Kungsträdgården. The comparison shall con-
sist of both the traffic network before the metro extension (base scenario) and the traffic
network after opening the new metro (future scenario). Both the base and future sce-
nario will have the same traffic volume, which are based on a future travel forecast
from Trafikverket. The goal is to provide a greater understanding of how these two
macroscopic traffic simulation software products performs and what their differences
are regarding the assignment of public transport.
The aim can be described by the following objective:
What are the differences with modelling a public transport network in Emme and in
Visum, when using headway based assignment, and why do the differences occur?
To provide an answer to this question the two simulation models in both base and future
scenario are required to be as similar as possible, which means that the networks in
Emme and Visum could be in need of adjustment with respect to pathways, metro and
bus etc. In order to evaluate the software products sensitivity regarding the public
transport assignment parameters this will also be analysed in the base scenarios.
1.2 Limitations
Data collection will not be done in this thesis, the relevant data is already available
from previous projects in the Stockholm area. There is a ready-made model in Emme
from traffic prognoses and this will be imported to Visum in order for the models to be
as similar as possible. In the two future scenarios the metro will be added to the base
scenarios and the same input data will be used to be able to compare all scenarios.
If the entire traffic network of Stockholm were to be studied, the project would become
too extensive. Therefore the area will be limited to Nacka municipality and the area
along the route of the proposed metro. This means that the OD-matrices that are
available needs to be adapted to this area. In Chapter 4 there is a more detailed de-
scription of the traffic network and available data. The results of this example network
may or may not extend to other networks, therefore other examples of smaller networks
helps to explain the actual differences along with a description of the algorithms.
The report will only include studies of public transport, i.e. bus, light rail and metro,
instead of using demand matrices for public transport, car and truck which would make
the project too extensive. The specific assignment procedure that will be analysed in
this thesis is called headway-based, which is suited for public transport areas with
high frequency transit lines. The headway can be explained as the time between two
vehicles of the same transit line serving a node. This type of assignment procedure
requires only a few types of input data, i.e. line frequencies and travel times. Since
2
1.3 Method
the analysis regards the future traffic situation there are only frequencies and travel
times available and this assignment method is therefore suitable for this procedure.
The headway assignments for Emme and Visum are based on the optimal strategies,
where the passengers choose the first transit line that arrives from an optimal set of
lines.
Since calibration is not the main focus in this thesis, only a comparison of the origi-
nal models and the base/future scenario model will be performed so that the models
produce realistic results. Due to this, the future scenario results will not show how the
metro actually affects travelling in Stockholm and Nacka. They will at best show an
elasticity measurement of the movements from bus to metro etc. However, an impor-
tant analysis will be the differences in public transport assignment between the software
models with and without the new metro, which can be done without calibration of the
models.
There is no known scientific basis for these parameters collected from Trafikverket
(Emme) or Trafikförvaltningen (Visum). Due to this one cannot draw any conclusions
regarding what parameters that are best at representing the reality, since the models
used in this thesis are based on future prognoses and cannot be validated. Further
studies are then needed concerning calibrations or evaluations of the assignment pa-
rameters, mentioned in section 2.2.
In order to make a complete software comparison the run-time, analysis capabilities,
multi-modality, capacity to model various behavioural phenomena like crowding, fares
etc. this will however not be done in this comparison.
1.3 Method
The main objective is to compare the macroscopic traffic simulation software products,
Emme (version 3) and Visum (version 13), with respect to the public transport as-
signment and examining the reasons for the result differences. By adjusting the model
parameters such as weights for waiting and boarding time, and by adding the new
metro line a more thorough comparison can be made. The future scenario models will
be used for further comparison between the software products and only to some extent
used for evaluation of the distribution of public transport passengers.
An evaluation of the extended transit assignment in Emme 4 will also be performed
in order to determine if there are any significant differences between using standard or
extended transit assignment. It is also interesting to see if the difference between the
extended assignment and the assignment in Visum.
In order to perform a comparison between the two software algorithms, a traffic network
with the same conditions is created. To ensure that both the road network and public
transport routes are consistent in both Emme and Visum, a network is developed in
Emme and then imported to Visum. See Chapter 4 for a more detailed description
of the adjustment of the road network, travel matrix and transit lines. In this thesis
the most interesting outputs are in-vehicle time (how long time the passenger spends
in transit vehicles), origin wait time (how long time the passenger waits at the origin
transit stop), transfer wait time (how long time the passenger waits at transfer stops),
3
1 INTRODUCTION
total transit volume, total number of transfers (transfer volume), transit volume on
different lines, the number of passengers that walks the whole way from origin to
destination, and average number of transfers per passenger. These outputs will be
used in the comparison between the software algorithms.
To compare the software algorithms, theory regarding macroscopic traffic simulation
and the theoretical models in each program will be studied. The manuals that describe
and explain the underlying mathematical methods in both software algorithms have
been a key part of the comparison. As a complement to the manuals, scientific articles
with relevant content have also been used to gain a broader and deeper understanding
of the assignments.
In order to ensure that this thesis is of good quality regarding the technical content,
the report will be revised by the developer of Emme, Michael Florian and Hans-Jürgen
Don from Visum’s Traffic Customer service. This is done to make sure that nothing
important will be overlooked or misinterpreted.
4
1.5 Research contributions
results of the simulations from the previously mentioned chapter. A comparison of the
transit assignment algorithms can also be found in this section.
In Chapter 7 the final conclusions are stated and directions for future research, are
made within the subject area.
5
2
Travel forecasting
Travel forecasting can be used for predicting changes in a traffic network. A common
method for performing travel forecasting is the four step model where a macroscopic
simulation software tool is a part of the procedure. A travel or traffic forecast is a
prediction of how the traffic will change in the future and are always based on external
conditions and factors. To create a model with good accuracy the model primarily need
to include socio-economic data and a transportation network. This chapter aims to
provide a short description regarding travel forecasting including the four step model
and an overview of how the public transport assignment works. For more information
about travel forecasting and the four step model see Ortuzar and Willumsen [6], Hydén
et.al. [7], WSP [8], National Cooperative Highway Research Program [9], California
department of transportation [10], and Lind [11].
The traffic forecasting can partly be carried out using a computer based software with a
model of the traffic network. Regarding the transportation network and specific public
transport network (transit network) the coding can be complex. There can be a big
range of available modes such as local bus, express bus, light rail, commuter rail and
bus rapid line. The lines service is often different in peak and off-peak hours during
the day. The total flow is studied in a macroscopic simulation and the individual
behaviour of the vehicles is not taken in consideration. The road network is of a larger
scale and may include traffic network for an entire city or country and the simulation is
made section by section. Emme and Visum are macroscopic simulation software used
when carrying out a travel forecast. The model that is created in the traffic simulation
software is a simplified version and representation of the real world network that is of
interest.
The forecasting method can be used for more than prediction of the future, it is also
used to investigate how the travel pattern will change when modifying or changing
anything in the traffic infrastructure. The model may be useful when analysing the
effect of editing the road design, altering the public transport supply or implementing
tolls. The result of prediction models can be a part of the decision making process
when deciding about changes regarding the traffic system. By forecasting it is possible
to compare the effects of alternative actions so the decision maker easier can determine
which action or actions will affect the transport system in the desired direction. The
most common reason and aim for a traffic prediction is to investigate changes in the flow
(traffic volume), after modifying the traffic network. This shows how the modification
has affected the traffic system in terms of more, less or shifted traffic volume. It can
7
2 TRAVEL FORECASTING
also be the number of trips and vehicle kilometres for different transport modes on
different roads and transit lines. It may as well provide an indication of the expected
traffic congestion in the area.
In the next section the four step travel model is described shortly, where the software
programs Emme and Visum are used in step 3 and 4 of the travel model.
The four step model is a commonly used approach for traffic modelling and consists of
four distinct steps which are well described in the literature, for example in Ortuzar and
Willumsen [6], see Figure 1 for an illustration of how the four step model works.
1. Traffic Generation
2. Trip distribution
3. Mode choice
4. Route assignment
8
2.1 The four step travel forecasting model
Figure 1: Illustration of the four step travel model and what is included regarding the
decision in each step
This step is described in [9] and the objective is to estimate the number of trips of each
purpose type that begin or ends in each zone. The estimation is based on the amount
of activity in the zone. The number of trips generated in this step are the flow used
in the model. Usually the trips are vehicle and person trips (auto or transit) which
often includes both walking and bicycle modes. The results and the outputs from the
trip generation model are trip productions and trip attractions for each zone and trip
purpose.
This step calculates the percentage of the total traffic that will travel between each
pair of zones. The result can be presented in a matrix where each row and column
represents a zone. Each value in a cell in the matrix represents the number of trips
between the zones and this is called a travel matrix or OD matrix, where Tij is the
demand from zone i to zone j. Often the travel demand varies over time and different
matrices may be used to study different time periods.
The purpose of this step is to split the trips between the zones by different travel modes.
The definition of modes depend on the areas supply of transportation and what type
of transportation analysis that is required. The modes can commonly be divided into
three groups: auto-mobile, transit and non-motorized. The choice of mode can depend
9
2 TRAVEL FORECASTING
on the range of transport modes, travel time with the particular mode and the cost.
There are different approaches for the mode choice but according to Hägerwall Stein,
[12], the most common is the logit model. There are a number of logit models but one
of the more basic versions is the so called linear model, see the formula below:
∑
e ai xi (j)
Pj = ∑ ∑ (1)
ai xi (k)
ke
The route assignment calculates how the forecasted travellers will be distributed among
different links in the transport network (included non-motorized links) or the transit
lines. How the route assignment works depend on the software used and what is being
analysed. There are different types of assignments and the two main ones are: the auto-
mobile assignment that handles routing of vehicles and the public transport (transit)
assignment that deals with routing of linked passenger trips (which include walks,
boarding and alighting). Depending on the software there can be more alternatives,
variants or combinations of these two assignments.
The flow unit in an auto-mobile assignment is the number of vehicles and in a transit
assignment the number of passengers. Another difference between the two assignments
regards that the transit assignment have line routes that consists of a set of links,
called line segments. When determining the perceived travel time of the passengers
an impedance function is computed. In the route assignment this function is used
in order to divide the demand on each route. The impedance function reflects the
unwillingness to travel and increases with longer total travel time. The impedance
function in transit assignment, compared to the auto-assignment, also contain level
of service variables that are not included in the auto-mobile assignment such as wait
time, boarding time, and auxiliary time (walk time). The trip between two nodes can
be served by more than one transit line and the lines can have different modes (e.g.
city bus, express bus, metro).
When the travellers have decided on using public transport, the demand is assigned
to different transit lines. There are different types of transit assignments depending
on the environment and available time table of the public transport. The assignment
procedures available vary depending on the software and an example of the most com-
mon are: transport system-, headway- and timetable-based. When the purpose is to
10
2.2 Public transport assignment
evaluate the entire system instead of analysing individual transit lines the transport
system-based procedure is used. This requires no transit line network and is used to
create a public transport network where the passengers chose the shortest routes. A
timetable-based procedure should be used for transport systems that have lines with
long headways, e.g. long-distance trains or transit lines in a rural area. To be able to
perform a timetable-based assignment it requires complete timetable information, trip
arrivals and departure times. Headway-based assignment is based on optimal strategy
theory which requires frequencies and travel times for the relevant public transport
lines. Since this type of procedure does not demand exact timetables it is only appro-
priate for long-term transport planning when the schedules are undetermined.
11
2 TRAVEL FORECASTING
area. Therefore the correct way to handle these parameters is to adjust them according
to different situations. No previous parameter translations between Emme and Visum
exists, at least according to the authors knowledge, in the headway-based assignment.
Therefore an assumption will be made from studying the manuals and examining how
the parameters work in each software. The parameters that will be translated are
weights, factors and/or penalties for how the passengers perceive boarding, walking,
travelling with a public transport vehicle, and waiting time compared to the time they
could spend in an auto-mobile instead. This translation is presented in Section 3.1,
Table 4.
Logit distribution is a probability distribution (see equation 1) where the flow are
split up according to the assigned percentage. The flow will be adjusted according to
equation 2, which considers the distribution on different modes, connectors between
orgin nodes and the network etc. By using a logit distribution one will force the demand
to chose different connector links between origin nodes and the network, boarding
a transit line versus walking to another station in order to obtain a shorter travel
time, or transferring to another transit line instead of staying on-board. According to
Florian and Constantin, [20], the logit distribution will even out the transit travellers
on different paths in several situations where they would all choose the same travel
strategy. This leads to a more realistic result where passengers choose different routes
from origin to destination.
12
3
Software products
This chapter contains information about different macroscopic simulation tools and es-
pecially focuses on the two most commonly used, i.e. Emme and Visum. More detailed
descriptions of these software products are therefore presented with respect to public
transport assignment and the algorithms available. The Section 3.2 and 3.4 describes
how the public transport assignment works in the respective software. These sections
also contain descriptions of the transit assignment algorithms. Visum have several
algorithms, however, only the algorithm that corresponds to the Emme 3 standard
algorithm will be thoroughly defined. Other algorithms are also available in Emme 4
and they will be presented in Section 3.3. Then, in Section 3.5.4, the mathematical
dissimilarities of the software algorithms are reviewed. The software manuals, Emme 3
[1], Emme 4 [21] and Visum 13 [2], have been used in this chapter and are the sources
if nothing else is stated.
In the article by Florian and Spiess, [22], there are an explanation of how the optimal
strategies work and an example with a small public transport network. Some previous
comparisons between Emme, Visum and similar software have been studied and used as
a foundation to this thesis. The conclusions of these comparisons have been of interest
for further investigations. Also, identification of weaknesses in the comparisons has
been important. Some of the comparisons studied are Johansson’s report [5], Larsen’s
report [23] and Hägerwall Stein’s master thesis [12], where the first two papers examines
both Emme and Visum (based on VIPS algorithms) with respect to public transport.
However, no thorough comparison of the algorithms has been made. [12] focuses on the
auto-mobile assignment and has therefore been used for software facts and comparison
method.
13
3 SOFTWARE PRODUCTS
website [4] states that the software contains simulation models such as toll facilities, on-
street parking, signal control, dynamic traffic assignment and a combination of micro
and macro (the areas of most interest are modelled on micro level and the rest at
macro level). There are also public transport assignments that are based on headways
or timetables. This software is not commonly used for public transport assignment
in Sweden. There are however some Swedish traffic projects that uses TransModeler
for dynamic auto-mobile assignment. VIPS (Volvo Interactive Planning System for
public transport), described in Johansson [5], was developed by Volvo Transportation
System in Gothenburg and have previously been frequently used at Trafikförvaltningen.
VIPS is a macroscopic transport simulation software that was incorporated with Visum.
Version 13 of Visum, contains some parts of the VIPS algorithm and there are relatively
few traffic analysts that still uses VIPS. In Sweden most municipalities and counties
uses traffic planning tools such as software with macroscopic traffic simulation for future
traffic predictions etc. Trafikverket, for example, have incorporated Emme in their
traffic prognosis software that they use for all parts of Sweden and Trafikförvaltningen
uses Visum instead. Since these two essential infrastructure operators in Sweden uses
Emme and Visum they are of interest for further investigation. They have previously
been compared in Johansson’s report [5], Larsen’s [23], and Hägerwall Stein’s master
thesis [12] that will be described below.
A comparison was made in 1984 between Emme and VIPS, which was incorporated into
a previous version of Visum, described in [5] at Stockholm county council (Trafikontoret
Stockholms läns landsting) by Johansson. The software VIPS was used to analyse
changes in the transit network, however the new launched Emme software had also
recently been installed at the office so it was of interest to evaluate both programs.
The aim was to compare the result of the chosen itineraries by the software programs
against a made survey. In the survey 100 persons per node-pair were asked about
their itinerary. The same transit line network over central parts of Stockholm was used
together with an OD-matrix for time period of an average hour between 07 : 00 − 09 : 00
in VIPS and Emme. According to the results VIPS generated more volumes (+12%)
on the buses compared to the survey and Emme lower volumes (−7%). Passengers split
up more on different itineraries between an origin and destination in VIPS compared to
Emme. The average number of transit per passenger is 0.63 in VIPS and 0.59 in Emme.
In Emme almost every one chose the same itinerary. The average absolute difference
between the survey and the models regarding the number of boardings on each bus line
is 30% for VIPS and 15% for Emme. Johansson, [5], also made a comment on that the
penalty of transfers of five minutes is calibrated for VIPS against the survey results.
The author mentions a desire to continue the comparison with calibrating the transfer
penalty against Emme.
In 2011 a comparison between Emme and Visum was made regarding the transit as-
signment by Larsen, [23]. He mentions some differences but does not really give an
explanation of how he comes to certain conclusions. There are some examples but
the calculations are missing and only the final answer is stated. However, the idea of
having a simple example to point out how the algorithms works is a good idea and will
be used in this master thesis as well.
The traffic (auto-mobile) assignment with network equilibrium was compared, between
Emme and Visum, and evaluated by Hägerwall Stein in 2007, [12]. Even though a
different assignment was compared the same method as in this report was used. A
14
3.2 Emme 3
limited part of a network was developed in Emme and later imported into Visum.
The same OD-matrix was used in both software programs. His conclusion was that
the result is similar regarding the flows on the links and routes which implies that
both Emme and Visum are based on the same algorithm (network equilibrium). The
differences were mainly caused by the rounding of the numbers in the OD-matrix.
3.2 Emme 3
Emme (Equilibre Multimodal, Multimodal Equilibrium), described in the Emme man-
ual [1], developed at the Centre for Research on Transportation (CRT) at the University
of Montréal in the seventies. In the eighties the first commercial Emme version was
developed at the CRT, called Emme 2. Professor Michael Florian is one of the founders
of INRO and the software Emme, and he had a key role in developing the modules used
regarding the transit assignment. The Emme 4 manual [21] states that improvements
have been made and new versions of the software Emme have been released. Emme
3 includes a graphic interface for network editing, more tools for simulation, analysis
etc. In Emme 4 there is a congestion assignment tool that models crowding, discom-
fort on vehicles, capacity limits and increasing waiting time. There are continuously
ongoing development regarding the interface, analysis, implementation of virtually and
zone-level travel demand model etc. Emme is used in over 85 countries, including Aus-
tralia, Canada, USA, South Africa, Central and South America, across Asia and most
European countries.
The macroscopic traffic simulation software Emme is used for modelling urban, regional
and national traffic systems. Emme is a traffic analysis tool that is used by trans-
portation planners and traffic analysts around the world. This section will describe the
software and some of its features with respect to the public transport assignment.
The transit assignment is based on the theory of optimal strategies approach by Florian
and Spiess, [22]. The public transport assignments in Emme consists of headway-based,
headway/time-based and timetable-based transit assignments, however the transport
system-based assignment is not included in the currently marketed versions.
The transit network consists of centroids (zones), regular nodes, links and a set of
transit lines. A transit line contains of a set of nodes and a set of links. The se-
quence of nodes represents the itinerary and where the travellers may board or alight.
Each link can have more than one transit line, which consists of several transit line
segments.
The transit assignment algorithm aims to minimize the total travel time containing;
wait time, auxiliary time, boarding time and in-vehicle time. The time components
are weighted to compare these times with the in- vehicle time. The total travel time
is converted into a general cost (T T T ), in other words the traveller wants to minimize
the total cost. The time is defined for each line segment and ads up to the total travel
time for the entire trip. The wait time factor scales the time a traveller has to wait
15
3 SOFTWARE PRODUCTS
at a node for an attractive line. The factor is also used to define the waiting times
together with the waiting time at a specific node. The value can differ between 0.01
to 1 and can either be node specific or the same for the entire network. The wait time
weight, that describes how much the wait time is valued compared to the in-vehicle
time, can be set between 0 − 999.999, as well as the parameter for boarding time. The
value can be the same throughout the whole network or be node/line specific. The
boarding time weight is also set between 0 − 999.999 and should represent, in relation
to the in vehicle time, how much boarding or a transfer is worth. The auxiliary time
weight and spread factor must also have the value 0 to 999.99.
After each transit assignment one can obtain an assignment report and matrices with;
transit times, in-vehicle times, auxiliary transit times, total waiting times, first waiting
times, boarding times, and average number of boardings. Graphical results are also
available in Emme 3, with options for comparisons between two scenarios.
A line i is attractive if the travel time of that line is lower than another attractive lines
total travel time, including wait time. This means that it is more profitable to board
line i if it arrives directly than it is to wait for a faster attractive line. The waiting
time at a node, twt , depends of the combined frequency (λi ) for the attractive lines (in
the optimal set of lines i ∈ I ∗ ), see equation (3) obtained from Nilsson’s educational
material, [24]:
1
twt = ∑ (3)
i∈I ∗ λi
16
3.2 Emme 3
3.2.2 Algorithm
Some of the notations mentioned in this section are described in Table 1 below.
Table 1: Line specific notation description for the algorithm section in Emme
The passengers want to minimize the travel time for the entire trip. The formula for
the total expected travel time (TTT) is shown below in equation (4).
The educational report by Matti Pursula et. al., [25], states that the assignment is
performed in two parts, first computing the optimal strategy to reach the destination
from each origin and second is to assign the demand according to the strategy.
The different options for how the passengers can reach the destination are saved in a
set of strategies. A strategy can be explained as rules that allow the traveller to make
feasible decisions and reach the destination node. An example of a strategy, according
to the Emme manual [1]: At node 1, take the line that arrives first of the attractive
lines 1 and 2. If line 1 was taken, alight at node 2. If line 2 was taken alight at node
4. At node 2, take the line that arrives first of the attractive lines 3 and 4. If line 3
was taken, alight at node 4. If line 4 was taken, alight at node 3. At node 3, take the
attractive line 5 and alight at node 4.
17
3 SOFTWARE PRODUCTS
The more information the traveller has the more complex the strategies become. The
traveller knows the distribution of inter-arrival times for the transit lines for a specific
node and the travel time between nodes. The traveller receives the information when
reaching the node and the distribution of passengers is also known. Together it is
possible to calculate the combined accepted waiting time for arrival of the first vehicle
in the set of transit lines passing the node and the probability of each line to arrive
first. The chosen routes will depend on what transit lines arrives first at the nodes.
According to Nilsson, [24], the travellers wait time at a node depends on the combined
frequency of the attractive lines that serves the node, see equation (3). This part is
reversely calculated, starting in a destination node and continuing backwards to all
affected origin nodes. A transportation network G consists of a set of nodes and a set
of links, G = (N, A). A trip is defined by a sequence of nodes, n ∈ N , via links a ∈ A.
A link a is assigned a link travel time ca and a distribution of the waiting time. The
result is the optimal strategy Anr (a sequence of links) with expected total travel time
∗ from each node n ∈ N to destination r.
T T Tnr
The first part of the algorithm initializes the expected travel time to reach r, T T Tnr ,
to infinity for all nodes except for the destination node passengers T T T which is set
to zero. The frequency variable, λi , for all i ∈ I ∗ , contains the combined frequencies of
the attractive lines and is initialized to zero. The set S is used to identify links that
have not yet been examined, and it is initialized with all the links in A. The set A is
initialized to an empty set and is used to identify the optimal strategy.
The second part of the algorithm starts with checking if the set S contains any non-
examined links, if it is empty then the algorithms first part will be stopped. If S is
′
not empty the link a closest to the destination r is selected. The time T T Tnr + ca is
considered to be the time from node n to the destination r without including waiting
time at node n. If this time is smaller compared to the previous time at n, T T Tnr the
link a is included in the optimal strategy and both λi and T T Tnr are updated to the
new combined total travel time of the attractive transit lines. It is important to know
how T T Tnr changes. The first time it will be λi T T Tnr = ”0·∞” (which is not defined),
in order to make the algorithm more compact the convention ”0 · ∞” = τ is assumed,
where τ is the waiting time factor.
To obtain the probability that a line i will be boarded is, according to [24] by Nilsson,
defined as the ratio between the line frequency and the combined frequency of the
attractive lines:
λi
πi = ∑ (5)
j∈I ∗ λj
The line headway is used in Emme when the optimal strategy is computed for passen-
gers. According to Matti Pursula et. al [25] the headway can be the actual headway
or the perceived headway of a specific transit line or segment (user-determined). The
headway in Emme is used to define the waiting times but also used for dividing the
passengers on attractive transit lines.
In the second part of the algorithm, the demand from node i to the destination r, gir ,
is assigned according to the optimal strategy A. The proportion of the demand of the
18
3.3 Emme 4
node i at the links a ∈ A corresponds to its frequency, see equation (5). The volumes
can be updated simultaneously because the links are evaluated in reverse topological
′
order (decreasing T T Ti + ca ) and therefore it is possible to examine every link only
ones.
The attractiveness test can be described by inequality (6) below and states that line i
(second choice, the line with next shortest travel time compared to the first choice) is
attractive if:
′
T T Tfirst choices > T T Ti,second choice (6)
This means that the travel time for line i is lower than the total expected journey
time for the first choice. Therefore it will be better to board line i if it would arrive
at the stop now than to wait for the first choice line. In order to calculate the final
expected total journey time for all the combined attractive lines the following equation
is formulated.
∑ ′
T T Ti∗ = πj T T Tj + twt wwt (7)
j∈I ∗
3.3 Emme 4
If Trafikverket will choose to upgrade their Emme version to Emme 4 there will be some
new functions available and more settings that can be used when calibrating models.
The most relevant assignment procedure for larger cities is, according to the Emme 4
manual [26], the extended transit assignment. Therefore this will be more thoroughly
described than the other assignments in Emme 4 and all facts are based on the prompt
manual, [26], and the scientific report by Cepeda, Cominetti and Florian, [27], which
describes some of the new features in Emme 4.
Extended transit assignment
This assignment is based on the standard transit assignment and the theory of optimal
strategy but in this extended version it is possible to model a connector choice. In other
words the travellers can be divided among more connectors instead of only choosing
the shortest connector. Also the choice of route is more sensitive to travel times (in
addition to the headway), so lines with lower frequencies and shorter travel times still
can be an attractive option.
In the extended transit assignment there are still the parameters used in the standard
transit assignment and some extra optional parameters such as boarding, in-vehicle and
auxiliary transit cost. The boarding cost is a penalty associated with every boarding
that is done (both initial and transfer). The in-vehicle cost will be multiplied with
the in-vehicle time weight and can be constant, segment, link, node or transit line
specified. The cost will be added to the total travel time. The auxiliary transit cost
can be constant, node or link specified and is multiplied with the weight and will be
added to the total auxiliary time in the TTT-equation.
19
3 SOFTWARE PRODUCTS
There is an optimal choice at the origin where a logit distribution can be used to split
the flows on different connectors. If the logit distribution is not used all travellers will
leave their origin node through the connector that have least impact on the travel time.
The logit distribution is specified at choice points and can be defined for all origins or
the ones with a special attribute. At the choice points the distribution may be applied
on all connectors or just the efficient ones, which means the connectors that brings the
traveller closest to their destination node. When using the logit distribution for nodes
with specified attributes it makes it possible to use different choice sets at different
origins (1 indicates to use logit on all connectors and −1 to apply only to the efficient
ones).
The logit function in the program contains two parameters, scale and truncation. The
scale is used in the computation of the likelihoods which the proportion are based on.
The scale parameter has to be 0 or greater. If it is 0 the proportions for all the con-
nectors in the choice set is the same. Larger values will give the best connector higher
proportions. The truncation parameter is used to drop connectors that have propor-
tions that are considered being too small. The connectors with smaller proportions
than the given truncation parameter are not included until the remaining connectors
have proportions larger than the parameter value.
The proportions computed for the connectors can be changed to fixed proportions using
a user-defined link attribute. The proportions must be between 0 and 1 and the sum
of for all the connectors from an origin must be 1. This can only be done on a subset
of origins and for the rest of the origin the link attribute must be −1.
In the extended transit assignment there is also possible to have a logit distribution at
the regular nodes with auxiliary transit choices. When using the logit assignment the
travellers at a node considering:
• Leave the node by the best auxiliary transit link or any efficient auxiliary transit
link
All travellers in an assignment without logit wait for a vehicle or leave the node by an
auxiliary transit link. When using the logit assignment it is also possible to split the
travellers between stay on board and alighting. The travellers that alight must leave
by an auxiliary transit. So there will only be a split if:
• The line on which the travellers are travelling on is also attractive at the node
The proportions of the alighting or the once who stays on board are computed based
on the impedance to the destination. The same logit function parameters are used as
in the function for original nodes.
In the standard transit assignment the flow distribution is based on the frequency but
in the extended assignment there is a choice to use a distribution based on frequency
and transit time to the destination. This means that fast lines with lower frequency
are more attractive, which results in smoother flow changes. This option can be chosen
in the whole network or for certain nodes.
20
3.4 Visum 13
It is also possible to prohibit connector to connector paths, which means that the trav-
ellers can not travel via only connectors to reach their destination. This suits networks
with a large amount of zones or a dense population.
3.4 Visum 13
Visum is a software product that has the same field of application as Emme and is
often used by traffic analysts in Sweden and around the world. The software devel-
opers of Visum are PTV Group in Karlsruhe, Germany. The company currently has
600 employees that works on improving the different software packages. The Visum
manual [2] states that the latest edition of Visum is version 13 which includes features
21
3 SOFTWARE PRODUCTS
such as trip distribution of the four step model, line cost calculations, fare calculations
and timetable-based assignment for public transport, and a traffic safety module that
contains historical data of accidents etc. PTV also develops microscopic and meso-
scopic traffic simulation software products. Today the company is located in America,
Latin America, Asia Pacific, Austria, Benelux (Belgium, Netherlands and Luxemburg),
China, France, Italy, the Middle East, and UK. PTV Group was founded by Dr.-Ing.
Hans Hubschneider at the Karlsruher University in Germany. However, all information
concerning the assignment procedures are presented in the manual, [2], and therefore
this will be the main source regarding Visum.
When performing a public transport assignment there are three different approaches;
transport system-, headway- and timetable-based. As previously mentioned the headway-
based assignment, that will be studied in this thesis, depends on the frequency of each
transit line, i.e. how often the line departures.
There are many output matrices available in the headway-based assignment procedure
in Visum. They are calculated in result matrices and the following outputs can be
obtained; Journey time, in-vehicle time, origin wait time, transfer wait time, walk time,
access time, egress time, perceived journey time, and number of transfers etc. Apart
from result matrices there is also a list of public transport assignment assignment
statistics containing information about the mean and total values of the previously
mentioned time components. Another output analysis tool that can be used is to view
different lists or graphic link and connector bars. In the lists and bars there are a
number of output choices including transit volumes on specific links, modes or transit
lines.
Perceived journey time All passengers wants to minimize this time, which consists
of all variables mentioned above together with the weights
and factors stated previously in this chapter
Impedance The objective for PuT-assignment, minimize weighted PJT
The perceived journey time (s) is calculated in the Visum headway-based assignment
in order to minimize the expected travel time for all demand. The function s consists of
several time variables, weights and penalties that affects the generalized journey time
in minutes. The different time components and factors are displayed in Table 2. In the
assignment settings the factors are multiplied with the corresponding time components
and all penalties are set to a user specified value. Some of the time components have
22
3.4 Visum 13
both factors and attributes, which are both multiplied with the time variable. Regard-
ing the Origin wait time (OWT) there is a formula that includes several attributes,
which can be multiplied or added to the OWT.
Impedance is a measure of how much the perceived journey time (s) is weighted, which
results in a value (in minutes) of how unwilling the travellers are to travel with a
specific line or mode. In Visum the impedance is calculated by multiplying a user
specified factor with s (the total expected travel time), see equation (8). The total
expected travel time is calculated with the combined parameters and time components
that corresponds to the Emme parameters.
These equations are used in the assignment procedure and all factors, penalties and
attributes can be determined by the user. All time components, stated in Table 2,
are calculated automatically and cannot be changed manually. However, it is possible
to control the passenger arrival rate so they do not arrive randomly at the stop area.
By adjusting the origin wait time factor one can obtain a transport system where the
passengers have more information about the timetable, which can result in a more
realistic assignment in a system with longer headways.
23
3 SOFTWARE PRODUCTS
3.4.2 Algorithm
Table 3: Line specific notation description for the algorithm section in Visum
For the headway-based assignment procedure there are three steps performed. In order
to calculate which route or routes will be the most attractive for the travellers there
are five types of passenger information systems that the user can apply to the model,
stated under choice models in the list below.
– From the mean wait time stated in the timetable, this is the default setting
24
3.4 Visum 13
∑k
1+ l=1 λl sl
uk = ∑k (9)
l=1 λl
For the route to be in the optimal set the remaining journey time of route with index k
needs to be less or equal to the constraining factor of the previous route in the sorted
set, see equation (10).
sk ≤ uk−1 (10)
If the route to be calculated is listed first there are no other routes to compare with and
therefore this route is directly set to be in the optimal set. However, the other optimal
routes can still turn out to be non-optimal in a later stage of the assignment procedure.
The routes are sorted according to the remaining journey time, in descending order,
and are used in the following assignment computation stage.
The next step is to filter out the unattractive routes based on a comparison between
the examined ranking index k and the previous routes with index 1 to k − 1. This
makes sure that all attractive lines have an expected remaining journey time that is
25
3 SOFTWARE PRODUCTS
lower than the combined remaining journey time for the other attractive lines. The
′
expected remaining journey time for index k is referred to as sk and is stated in
equation (11).
′
sk = waux taux + wboarding tboarding + ttravel + wtransf er N T R (11)
The share of each attractive line is based on the headways of all optimal routes and
the equation describes the probability that the passengers will board this specific line
route first. The line probability is therefore dependent of how frequently the transit
line departures and can be seen in equation (12).
λi
πi = ∑ (12)
j∈I ∗ λj
The combined routes for ranking index 1 to k − 1 will have an expected remaining
journey time that consists of the combined wait time (equation (3)) with the wait
factor wwt and the combined travel time. The equation for combined wait time and
′
remaining journey time, ck , are calculated according to equation (13).
′ 1 ∑
k−1 ′
ck = ∑k−1 wwt + πl sl (13)
l=1 λl l=1
Since the set of optimal routes are listed in descending order the first route will be the
best with respect to the total remaining journey time and is therefore always a part
of the final optimal set. To decide whether or not the other routes actually belongs to
the optimal set (I ∗ ) the constraint in equation (14) needs to be fulfilled.
′ ′
sk ≤ ck (14)
All the lines in the optimal set have a share of the assigned demand πi , see equation (12),
where i is one of the optimal lines.
This type of choice model suits situations where the headways are irregular and there
is no passenger information that can lower the uncertainty.
26
3.4 Visum 13
remaining journey time sk . The demand share is also calculated differently compared
to the previous information type, although it is still based on the probability that line
i departs first, for all i in the optimal set of lines.
The choice model with no information and constant headways are suitable in those
cases where the headways are fairly regular and the passengers have no information
about the transit lines actual arrival times.
In order to calculate whether a route is worth taking, even though the travel time
is longer than another line of the optimal set, tj needs to be computed. When the
passenger has waited tj time units or more, the remaining travel time for line j is
longer than the expected remaining wait time including the remaining travel time for
line j − 1. Therefore line j is excluded from the optimal set of lines at the exact time tj .
Complete information
The passengers have information regarding the departure times at all stops before
deciding which route to choose. This choice model suits transport systems where
all passengers have full access to departure times and always choose the route that
minimizes the total travel time.
27
3 SOFTWARE PRODUCTS
When performing any type of assignment there are user-determined values for adjusting
the model results. These are referred to as assignment parameters and can differ
between software products. The parameters mentioned below are used in the public
transport assignment in Emme:
• Boarding time (min)
• Wait time factor
• Weight factors:
– wait (origin and transfer wait time)
– auxiliary (walk time at the connectors)
– boarding
• Spread factor
With Visum there are several settings that can be varied and these are stated in the
following list:
• Boarding penalty (min), PuT (public transport)
• Boarding penalty (min), PuT-Aux (public transport auxiliary modes)
• Mean delay (min)
• Formula/attribute for origin/transit wait time
• Weight factors:
– origin and transit wait (OWT and TWT)
– access, egress and walk (ACT, EGT and WT)
– transfer (NTR, in minutes)
– in-vehicle (IVT)
– PuT-Aux ride time
– Fares
With the help of both software manuals the assignment parameters could easily be
determined according to the translation in Table 4. The following can be found in the
Emme [1] and Visum [2] manual and have been interpreted by the authors:
28
3.5 Assignment parameters and examples
• boarding time and boarding penalty PuT both implies an addition to the per-
ceived travel time
• wait time factor in Emme is described to be represented in Visum by an attribute
for the origin and transfer wait time
• wait time weight in Emme corresponds to the factors for origin and transfer wait
time in Visum
• aux time weight in Emme corresponds to the factors for access, egress and walk
time in Visum
• boarding time weight in Emme do not exist explicitly in Visum, it can however
be added to the boarding penalty PuT if needed
• spread factor in Emme do not exist explicitly in Visum, it can however be mul-
tiplied to the origin and transfer wait time if needed
• factor for NTR, number of transfers, do not exist in Emme
• fares for public transport lines in Visum (and cost penalty in Emme 4), do not
exist in Emme 3
Boarding time (Emme) and boarding penalty PuT (Visum) implies the same effect
on the network, for each boarding (transfers included) the passengers are punished
with a time penalty. This is done so that the impedance (resistance) increases for
boarding transit lines and the passengers try to minimize their number of transfers.
This parameter will be referred to as tboarding and is complemented with a weight of
how the boarding time is perceived, wboarding .
When a passenger waits for a transit line there are both a wait time factor and a
wait factor weight in the Emme assignment. The wait time factor is used in order
to obtain arrival distributions and the wait factor weight is used to compare how
much impedance that should be applied to the wait time compared to the in-vehicle
time (ttravel ). In Visum the corresponding time components are origin and transfer
wait time, each with a separate wait time factor and wait factor weight. These time
components together form the notation twt with the weight factor wwt . According to
the model obtained from Trafikverket and Emme manual the wait time factor is set
to 0.5, which corresponds to a uniform distribution of the passenger arrival times and
a regularly spaced service. Lower values can be used when for example the headways
are long and passengers know the timetables.
To obtain a spread of travellers on different transit lines Emme uses a parameter called
spread factor. Spread factor can in some situations be used in addition to the weights
29
3 SOFTWARE PRODUCTS
by increasing or decreasing the attractive trip options for the travellers. The factor
is multiplied with the product of the waiting time and waiting time weight, and is
included in wwt . There is a corresponding formula in Visum that achieves a variance in
the waiting times if desired, which also leads to a various amount of attractive transit
lines. A lower spread factor means less attractive transit lines. According to the model
from Trafikverket the spread factor is set to the default value 1, which is the default
value.
The time that passengers spend on connector links (from origin or destination to the
network) and between transfers is called auxiliary time in Emme. In Visum it is divided
into access (from origin), egress (from destination) and transfer walk time and together
they correspond to the auxiliary time in Emme. This variable will be mentioned as
taux and its associated weight factor is waux . The weight factor is a measurement of
how much the passengers wants to avoid walking compared to travelling with a transit
vehicle (in-vehicle time).
The headway (h) and frequency (λ) variables are both used in Emme and Visum, where
the frequency of a line j is inversely proportional to the headway, λj = h1j .
To exemplify how the algorithm for optimal strategies work a small network will be
used and the results calculated both by hand and in Emme. In Table 5 the used values
are stated and the network can be seen in Figure 2.
1 16.66 50 20 10
2 5 50 6 5
3 4.17 50 5 5
Zone 4 to node 1 - 5 1.2 -
Zone 4 to node 2 - 5 12 -
Node 3 to zone 5 - 5 3 -
The red nodes are zones from which the trips are generated. In this example zone
number 4 is the origin and zone number 5 is the destination for the 100 trips assigned.
There are three different public transport lines, line 1 goes from node 1 to node 3
directly, line 2 goes from node 1 to node 2 and line 3 goes from node 2 to node 3. The
30
3.5 Assignment parameters and examples
headways are set to 10, 5 and 5 which means that line 1 departs with a 10 minute
interval and line 2 + 3 departs every 5th minute.
The parameters used in Emme originates from Trafikverket’s assignment settings for
public transport. These values are well established and the most commonly used values
are presented in the list below. No scientific basis or any type of documentation for
choosing these specific values have been found. Since the thesis does not focus on
calibration of assignment parameters the following values will be used without further
consideration.
• Boarding time: 5.00
• Wait time factor: 0.50
• Weight factors:
– wait = 1.50
– auxiliary = 2.00
– boarding = 1.00
• Spread factor: 1.00
This leads to T T T = 1 · ttravel + 2 · taux + 1.5 · 0.5 · twt + 1 · tboarding . In order to exem-
plify the algorithm described in the previous section the optimal strategies have been
calculated by hand. The computations are reversed, i.e. it starts at the destination
node and moves backwards to the origin node. The network characteristics are stated
in Table 6.
3 - - 6 - -
2 3 5 6 3.75 5
1 1 Line 1: 20 Line 1: 6 Line 1: 7.5 Line 1: 5
2 Line 2: 11 Line 2: 6 Line 2: 3.75 Line 2: 5
Zone 4 - - To node 1: 2.4 At node 1: 2.5 -
To node 2: 24 At node 2: 3.75
There are no lines departing at node 3 and therefore it does not require any computa-
tions, apart from the travel time (walk time) with the walk time weight.
′
Node 2 have only one line, i.e. line 3, with the travel time T T Ti = 16 (see step 1 in
Table 7) and expected travel time 19.75 minutes (step 2 in Table 7). Since only this line
departs from node 2 that line is considered to be attractive (step 3 in Table 7).
At node 1 there are two lines available, line 1 and line 2, where the second line requires
a transfer. The expected travel time for the transfer line, 19.75 minutes, will be added
to the travel time for line 2. This is done in order to attain the total travel time from
node 1 if line 2 arrives now, i.e. without wait time at node 1, 30.75 minutes (step 1).
The first line is a direct line and therefore its total travel time only consists of the
in-vehicle and boarding time, 31 minutes (step 1). The line with the shortest travel
time will be the first choice if both lines arrive at the same time. This leads to that the
first choice is line 2 and after calculations according to the total travel time function
(with combined wait time) the expected travel time is determined as 34.5 minutes, see
31
3 SOFTWARE PRODUCTS
equation (7) and step 2 in Table 7. If line 1 arrives directly it will take 31 minutes
to arrive at destination, which is still attractive compared to 34.5 minutes in expected
travel time if the passenger waits for line 2 instead of boarding line 1. This can be
observed from the attractiveness test in equation (6) and step 3 in Table 7 where the
′
line i in this case is the first choice (shortest T T Ti among the lines departing from
that node).
Table 7 and 8 shows the complete results when following the reversed assignment
algorithm approach of the optimal strategies in Emme (i.e. begins with destination
and moves backwards to the origins).
Table 7: Algorithm first three steps for computing the optimal strategies in Emme
′
Node Step 1: Travel time (T T Ti ) Step 2: Expected travel time Step 3: Attractiveness test
without wait time (equation (4), first choice) (equation (6))
3 6 - -
2 16 19.75 Only line 3 is attractive
1 Line 1: 31 (Line 1: 38.5) 34.5>31,
Line 2: 30.75 (first choice) Line 2: 34.5 both lines are attractive
Zone 4 Node 1: 2.4 (first choice) Node 1: 4.9 4.9<24, only the link from
Node 2: 24 (Node 2: 26.75) zone 4 to node 1 is attractive
Table 8: Algorithm last two steps for computing the optimal strategies in Emme
Node Step 4: Probabilities for attractive lines Step 5: Expected travel time
(equation (5)) T T Ti∗ (equation (7))
3 - -
2 π3 =1 19.75
1 π1 =0.33 Line 1 and 2: 33.33
π2 =0.67
Zone 4 π4 =1 Line 1 and 2 via node 1: 35.73
This leads to the conclusion that both line 1 and 2 are attractive lines (step 3), which
also is the result from the assignment made in Emme (Figure 3). When calculating the
line shares for the attractive lines the line frequencies are used as previously mentioned
in the algorithm section (equation (5)) and step 4 in Table 8. The probability that
line 1 is chosen is 33% and 67% that the passengers choose line 2 instead. Step 5 in
Table 8 is done in order to determine the total expected travel time from zone 4 to
zone 5 with a combination of the attractive lines, which leads to 35.73 minutes. This
was also generated in the Emme assignment, in Figure 3.
Figure 3: Graphic result from transit assignment of the small network in Emme
32
3.5 Assignment parameters and examples
The assignment procedure in Visum with no passenger information and where the
transit lines have exponentially distributed headways, similar to the optimal strategies,
will be exemplified in this section. Since the headways are predetermined the headway
calculation is not described in this example. The used values are stated in Table 9 and
the network can be seen in Figure 4.
1 16.66 50 20 10
2 5 50 6 5
3 4.17 50 5 5
Zone 1 to node 1 - 5 1.2 -
Zone 1 to node 2 - 5 1.2 -
The blue nodes are zones from which the trips are generated. In this example the first
zone is origin and the second zone is destination for the 100 trips assigned. There are
three different transit lines, line 1 goes from node 1 to node 3 directly, line 2 goes from
node 1 to node 2 and line 3 goes from node 2 to node 3. The headways are set to 10, 5
and 5 which means that line 1 departs with a 10 minute interval and line 2 + 3 departs
every 5th minute. The line times are assigned by adding run times in the time profile
edit dialogue for each line.
The parameters used in Emme have been somewhat translated through experiments
and theory from the software manuals, see Table 10. Equation (15) shows the formula
for s with these translated parameter values.
33
3 SOFTWARE PRODUCTS
To exemplify the algorithm described in the previous section the first information level
has been calculated by hand. The calculation results were equal to the assignment
distribution of the demand in Visum. The used network data with weights can be seen
in Table 11 and 12.
Table 11: In-vehicle, walk and origin wait times for the different routes
Route Line In-vehicle time Weighted Un-weighted
walk time origin wait time
1 1 20 2.4 10
2 2+3 11 2.4 5
3 3 5 24 5
Table 12: Boarding, transfer and total wait times for the different routes
Route Line Un-weighted Total weighted tboarding wboarding
transfer wait time wait time +N T Rwtransf er
1 1 0 7.5 5
2 2+3 5 7.5 10
3 3 0 3.75 5
Step 1 is performed in order to calculate the remaining journey time from equation (15),
where the routes are ranked from the lowest remaining journey time to the highest.
From this ranking uk , equation (9), is calculated and if the route with ranking index k
fulfils the constraint sk ≤ uk−1 one can consider the route to be a part of the optimal
set, see Table 13 for the calculation results of the first step.
Table 13: First step when computing the optimal strategies in Visum
s uk In the optimal set of lines?
Route Line Ranking Index k
(equation (15)) (equation (9)) (equation (10))
This results in that route 1 and 2 are considered to be a part of the optimal set of lines,
although more tests are required to determine whether the routes truly are attractive
compared to each other. This will be done in step 2 below.
There will always be at least one attractive route, and that is the route with ranking
index 1 (with the lowest sk ). The other routes (in this case only route 2) needs to be
compared to those routes that have less sk time, i.e. all routes in the optimal set that
have ranking index from 1 to k − 1. When comparing ranking index k to the other
routes one will need:
• Line probability (combined frequency if there are transfers involved, equation (5))
πl , i ∈ I ∗ : i < k − 1
1
• Combined remaining wait time (first part of equation (13)) ∑k−1 wwt
l=1 λl
∑k−1 ′
• Combined remaining travel time (second part of equation (13)) l=1 πl sl
When the above mentioned equations are calculated for each route the combined travel
and wait times for all routes in the optimal set (ranking index from 1 to k − 1) is
34
3.5 Assignment parameters and examples
obtained, equation (13). To decide if the route is attractive the attractiveness test
′ ′
mentioned in equation (14) needs to be fulfilled, i.e. sk ≤ ck . See Table 14 for the
calculation results.
Table 14: Second step when computing the optimal strategies in Visum
Ranking Route Line The expected Comb. Line Comb. Comb. wait Attractiveness Share
index k remaining journey remaining probability remaining time and test
time for index k, wait time journey remaining
without wait time time journey time
In this specific example the optimal routes are the first and second, with ranking index
2 and 1. The graphic result can be seen in Figure 5. Compared to the results from
corresponding Emme example this leads to the exact same optimal route distribution
and line shares.
Figure 5: Graphic result from transit assignment of the small network in Visum
The difference between the software assignments is how the first and second attractive-
ness test is performed. The criteria in Emme (step 1, equation (6)) is based on that
the line with the shortest expected arrive time at destination will be the first choice
if it would arrive at the examined node directly (i.e. without wait time). In Visum
the total expected time including wait time is considered when limiting the number of
lines that can be chosen from the optimal set (from equation (10)).
The constraining factor in Visum, uk−1 , determines if a line is attractive in the first
attractiveness test. This factor is equal to twt plus the total expected journey time, s,
for all lines that are in the attractive set. This factor is the combined wait and travel
time for all the optimal lines, i.e. the lines with shorter expected journey time than
line k. When using equation (10), the total expected journey time of route with index
k must be less or equal to the combined expected journey time mentioned above as
uk−1 , which considers all lines with ranking number 1 to k − 1. The factor states the
longest time one should have to wait for one of the attractive lines and in order for line
k to be attractive its travel time must be lower than the longest time of all combined
optimal lines. The same type of criteria is stated in Emme, however the total expected
′
journey time does not include the wait time, T T T , and therefore states the travel time
if the line arrives now. Also, this travel time is calculated for the "second choice" (the
35
3 SOFTWARE PRODUCTS
line with next shortest travel time at that stop). The travel time for the second choice
must be strictly less than the total expected remaining journey time of the first choice,
i.e. the line with shortest travel time at that node. Both first attractiveness tests in
Emme and Visum aims to separate the lines with too long remaining travel time.
In Visum there are another attractiveness test performed in order to double check if a
line have too long travel time. Since the first test was focused on detecting lines with
too long journey time (low line frequency) this test needs to examine the travel time if
a line arrives directly at a node. In contrast to Emme, which first of all tests whether
or not the lines have low remaining travel time and secondly tests if the other lines
have shorter travel times compared to the total journey time of the first choice line.
This means that it would pay off to board another line rather than to wait for the line
with shortest travel time (takes the line frequency into account).
To summarize the differences between the two algorithms first and second attractiveness
tests these examples are stated for each software:
• Emme, example of the first and second attractiveness test:
– There are two lines, 1 and 2, departing from node A
– The first line has an expected remaining travel time of 20 minutes if it would
arrive at node A now and an expected remaining journey time of 27 minutes
(i.e. including wait time)
– The second line has an expected remaining travel time of 25 minutes if it
would arrive at node A now and an expected remaining journey time of 26
minutes (i.e. including wait time)
– When determining which line will be the first choice the algorithm in Emme
compares the travel time of line 1 and 2
– Line 1 will be the first choice since 20 < 25 (first attractiveness test in
Emme)
– To examine if line 2 also is attractive (if it would arrive first) the total
expected remaining journey time of line 1 will be compared to the expected
remaining travel time of line 2
– Line 2 will be the second choice since 25 < 27 (second attractiveness test in
Emme)
• Visum, example of the first and second attractiveness test:
– There are two lines, 1 and 2, departing from node A
– The first line has an expected remaining travel time of 20 minutes if it would
arrive at node A now and an expected remaining journey time of 27 minutes
(i.e. including wait time)
– The second line has an expected remaining travel time of 25 minutes if it
would arrive at node A now and an expected remaining journey time of 26
minutes (i.e. including wait time)
– When determining which lines that will be attractive all lines are ranked in
descending order with respect to the expected remaining journey time
36
3.5 Assignment parameters and examples
– Line 2 will be the first choice since 26 < 27 (first attractiveness test in
Visum)
– To examine if line 1 also is attractive (if it would arrive first) the total
expected remaining travel time of line 1 will be compared to the combined
expected remaining journey time of the already attractive lines (in this case
only line 2)
– Line 1 will be the second choice since 20 < 26 (second attractiveness test in
Visum)
Regarding that Emme includes the wait time in the flow distribution and in the second
attractiveness test but not in the first attractiveness test, the headway influence on
attractive lines are smaller than in Visum. In Emme 4, one can use a combined headway
and travel-time distribution so that more people will choose the line departing less often
and have a short travel time (the travellers adapt more to the timetable). Where in
Visum the wait time is included in the flow distribution and both attractiveness tests.
This leads to that the headway affects the choice of attractive lines more than in
Emme and therefore lines with longer headways are excluded. That problem can be
avoided if one uses other types of information settings available in Visum. The issue
with taking the wait time into account several times can lead to that more people
are choosing to change transit line in Visum as more routes are attractive which can
contain more transfers. An approach that can be used to prevent this is to apply a
transfer penalty.
This section will describe two examples stated in the reports by Nilsson [24] and Larsen
[23]. The example in [24] consists of a network in Emme and therefore a similar network
was built in Visum in order to evaluate the algorithm differences further. In Larsen’s
report [23] there are two networks built in Emme and Visum, however the results did
not coincide with the calculations made by the authors of this thesis. Therefore a
reproduction of the networks was made.
According to the example in Nilsson’s report [24] 50 % will choose line red and 50
% line green from the first node. From the last node before destination 8 % travels
with line blue and 42 % with line black. When calculating the results with the Visum
algorithm (and running the simulation) 50 % chose red and green. However, the blue
line was in this case determined as unattractive. Instead the demand was spilt equally
on the black and grey line, i.e. 25 % of the total demand on respective line. The
reasons why the blue, black and grey line differs between the software assignments are
the attractiveness tests and that Emme calculates the expected remaining journey time
compared to Visum that calculates the remaining journey time for all routes according
to the information available. Regarding the attractiveness tests they differ due to
that Emme will choose the line with the shortest travel time if the line arrives now
("first choice") and Visum will choose the line with the shortest total time (including
wait time etc.). That is why the blue line, with shortest travel time and longest
headway, will be chosen in Emme and not in Visum. Since the results from following
the stated assignment algorithm in Visum led to the same demand distribution when
37
3 SOFTWARE PRODUCTS
simulating the network one can assume that the algorithm is correctly interpreted from
the manual. The results from this reproduced example can be seen in Figure 6 (the link
colors represents each transit line), Table 15 and Table 16. Note that the percentage
in Figure 6 is the share of travellers from the previous node, not the share of the total
demand.
Figure 6: Comparison between the reproduced example results in both Emme and
Visum
When reproducing the test network in Chapter 5 from Larsen’s report [23] both the
simulation and calculation results were equal between Emme and Visum. However, the
results did not coincide with the Visum results in [23]. The reason why the results are
not the same depends on that the settings for headway-based assignment in Visum are
different. The simulation run by Larsen[23] used the RDT-based assignment ("random
departure time" from VIPS, or "no information and constant headways" in Visum),
which uses constant headways instead of exponentially distributed headways, while
the reproduced test network used exponentially distributed headways. The differences
stated by Larsen ([23]) are that the algorithms differ when deciding the attractive
38
3.5 Assignment parameters and examples
routes, as this thesis also states in the previous section 3.5.4, and how the travellers
are distributed on the attractive lines. The distribution however, does not coincide
with the reproduced results since the algorithm settings were different.
39
4
A case study
A traffic network representing the Stockholm area is available from earlier studies
along with transit lines and demand matrices. In order to obtain as similar models as
possible, the network will be built in Emme and then imported to Visum.
The large networks were obtained from Trafikverket and contain the transport system
of Sweden in a future scenario with and without the metro to Nacka, and represents the
traffic situation at 07 : 00 − 09 : 00 in the morning. In this thesis the traffic prediction
without metro is called the base scenario and the traffic prediction that contains the
metro route from Kungsträdgården to Nacka Forum is called the future scenario.
The existing network of Sweden that was obtained from Trafikverket will have to be
limited in order for the software differences to be more traceable. When only studying
a smaller part of Sweden it will be easier to identify the differences and similarities
between the transit assignment in Emme and Visum. To receive a more accurate
breakdown of public transport lines, the area includes the entire municipality of Nacka,
large parts of Södermalm and the area towards Stockholm Central Station. The lines
going into and out of the chosen area must be cut in order to function in the new area.
A series of assembly nodes will need to be positioned in order to capture the flows with
destination outside the area. The chosen network area can be seen in Figure 7.
Figure 7: The network limitation area of Nacka municipality, from Google maps
41
4 A CASE STUDY
The assignment parameters for public transport are determined for the Emme model
and have been interpreted to the Visum model by studying the manuals and performing
experiments, see Chapter 3. In the Sections 4.1 and 4.2 there are more detailed descrip-
tions of the used parameters and how the models where built and verified. According
to Florian and Spiess [22] urban environment and short headways in a public transport
network would be best suited for the headway-based assignment procedure when per-
forming long term traffic predictions. Stockholm is an urban environment with a public
transport system which have short headways and therefore this assignment procedure
is the most appropriate. Headway-based assignment is also used by Trafikverket and
Trafikförvaltningen when modelling Stockholm’s public transport system.
In Figure 8 the chosen metro alternative, according to the pre-study [28], can be seen
and consists of a connection to the existing blue line at Kungsträdgården. The reason
for choosing this alternative is due to that Stockholm City and the municipalities
involved decided that this was the most socio-economically efficient alternative. The
following stations will be a part of the metro extension to Nacka:
• Kungsträdgården (connection with the existing blue line)
• Sofia
• Hammarby Kanal
• Sickla
• Saltsjö-Järla
• Nacka Forum (end station)
42
All bus lines that will be of importance in the base scenario are listed in Table 17 with
line numbers, start and destination station. The important lines are those that pass
through a large part of the study area or those considered to have an essential roll in
the public transport supply (lines with a lot of boardings). When limiting the network
the existing public transport lines needs to be cut off and therefore several new origin
and destination nodes will be created.
The same type of table can be seen, with all relevant trail traffic lines (metro and light
rail) for the base scenario, in Table 18.
Table 18: Relevant metro/light rail lines for the base scenario
Line number Line origin Line destination
43
4 A CASE STUDY
Table 19 contains the transit lines for the future scenario, since some lines will be
changed or removed due to the new metro line.
Table 19: Relevant bus lines for the future scenario (the new or changed transit lines
are stated in italics)
Line number Line origin Line destination
43 Ruddammen Tanto
53 Roslagstull Henriksdalsberget
55 Hjorthagen Sofia
59 Karolinska sjukhuset Norra Hammarbyhamnen
66 Reimersholme Sofia
74 Mariatorget Sickla köpkvarter
76 Ropsten Norra Hammarbyhamnen
401 Slussen Älta
402 Slussen Kvarnholmen
403 Slussen Hästhagen
409 Nacka forum Ektorps C
410 Nacka forum Saltängen
411 Nacka forum Skuru
413 Nacka forum Björknäs C
420 Sickla bro Gustavsberg C
422 Sickla bro Gustavsberg, Lugnet
425X, 428X Slussen Gustavsberg, Björkvik
430, 430X Slussen Eknäs brygga
431, 433, 434, 435, 436, 437,
Värmdö Slussen
438, 440, 446, 462, 474
434 Slussen Sollenkroka
435, 437 Slussen Hemmesta vägskäl
442 Nacka forum Orminge C
443 Nacka forum Jarlaberg
444 Nacka forum Västra Orminge
445 Nacka forum Orminge Ö
448 Nacka forum Gustavsvik
449 Nacka forum Ektorp
471 Nacka forum Västra Orminge
474 Nacka forum Hemmesta vägskäl
821 Tyresö C Nacka sjukhus
840 Handenterminalen Nacka strand
Table 20 shows the trail traffic lines for the future scenario, where the new metro line
are stated in italics.
Table 20: Relevant metro/light rail lines for the future scenario (the new or changed
transit lines are stated in italics)
Line number Line origin Line destination
44
4.1 Model building in Emme
The parameter settings in Emme are, as previously mentioned, obtained from Trafikver-
ket and can be seen in the list below.
• Weight factors:
– wait = 1.50
– auxiliary = 2.00
– boarding = 1.00
In order to limit the network a graphic modification tool from Emme 2 was used,
and the limitation method follows the manual description of additional assignment
with traversal matrix from the manual [1]. The graphic modification tool can define
an area in which the chosen links or nodes are set to contain a user specified value.
The links in the desired area were marked and then all links, nodes and zones were
exported to a new database. This database was dimensioned to contain 161 zones,
600 regular nodes, 3 000 links, 100 transit lines and vehicles, 2 000 transit segments
and 10 full matrices. The network now consists of the links, nodes and zones from the
marked area. By combining transit lines that have the same route path the frequency
of that combined line can be increased, according to the headway-based assignment
algorithm. All lines that did not affect the study area were eliminated and the number
of transit lines reduced to about 30 lines (both directions). These final transit lines
were adjusted with respect to the new area, by adding a line segment attribute that
stated if a segment was in the area or outside the area.
A problem that arises when cutting a network is to adjust the OD matrix so there still is
the same amount of travellers going to the corresponding origin and destination zone.
This was solved by using an additional assignment for the auto-mobile mode called
traversal matrix. The traversal matrix ensures that no transit volume that travels in
to, out from or inside the desired area will change from the previous matrix. All links
going in or out to a zone within the area were marked with that zone number (portal
zones), in a link attribute. Links going in to the zone was assigned the positive zone id
and links going out from the zone was assigned the negative zone id. This procedure
leads to that all transit volume passing these marked links will be assigned a new
origin or destination of the portal zone number from that link attribute. The result
after an assignment of this type is a new matrix containing all demand travelling inside
or in to the limited network area. This traversal matrix was then imported to the new
database and a complete smaller network was obtained by adding the zones used in
the traversal matrix. The resulting network in Emme can be seen in Figure 9 and the
transit lines for the base and future scenario in Figure 10 and 11, where each transit
line is represented by a separate color.
45
4 A CASE STUDY
The base scenario includes more transit lines that passes Nacka Forum and further on
to Slussen. In the future scenario the transit lines have been altered and more buses
have Nacka Forum as the end station instead of Slussen. Also the new blue metro line
is added in the future scenario and can be spotted passing the water from Sofia to
Kungsträdgården.
Figure 10: Transit lines in the base scenario (each line is a separate color)
46
4.1 Model building in Emme
Figure 11: Transit lines in the future scenario (each line is a separate color)
4.1.1 Verification
With the limited network finalized it is of importance to check the model for errors
and to make sure that the results are realistic in comparison to the original models.
A number of verifications were performed such as comparing the large scale network
with the small using Emme 3 graphic result for transit and auxiliary transit volumes.
If the difference in transit and auxiliary transit flow is low and the connectors inside
the network have no difference then the network is considered to function according to
the original network. In order to verify the traversal matrix a comparison between the
origin and destination sum matrices were done. The traversal matrix was correct since
all zone demands inside the network were the same and all portal zone demands were
larger than or equal to the original demand. Another network check was to compare
the origin and destination sum matrices with the assigned auxiliary transit volumes.
The model was corrected by adding connectors from some portal zones to one of the
accessible nodes inside the network.
47
4.2 Model building in Visum
The network in Emme was completed and then the network, transit line, vehicle, and
mode files were exported and then imported to Visum through the import selection
"Emme/2 - import parameters". In order for the network to function properly a few
adjustments needed to be made and a PuT-assignment to be created in the procedure
sequence. First of all the walk links between origin/destination nodes and the rest of
the network were not stated as connectors when importing to Visum. The solution
to this problem was to add a separate node for each connector and thus create links
between the connectors and the rest of the network (an option when importing the
network from Emme). Another problem that occurred was that the walk times, travel
times with PuT and vehicle speed were incorrectly set. To adjust those issues one can
use the multi-edit tool for links and lines or manually change each line run time. It is
also important that the walk times are correctly set by editing the links and connectors.
To be able to run the assignment a demand matrix needs to be added, in this case the
traversal matrix created in Emme was used. In the option for OD demand data the
demand segments was set to be collected from the traversal matrix. Another setting
that is required for the assignment is demand models, which consists of the standard
4-step procedure mentioned in Chapter 2, Section 2.1.
From translating the assignment parameters in Emme to Visum the following values
were obtained. Since there are no previous translations of these parameters an as-
sumption was made from studying the manuals and how the parameters work in each
software, see Chapter 3.
• Weight factors:
– transfer = 0.00
– in-vehicle = 1.00
When the network was completed a procedure sequence was created and set to the
headway-based assignment. The traversal matrix containing the trips made in the
limited network was used as input together with additional settings such as, simula-
tion time interval, headway calculation and logit model options for distribution on the
connectors. Before running the simulation all assignment parameters were inserted
according to the list above (not mentioned parameters were kept as the default value).
There are also choice model information available here, which was described in Chap-
ter 3, Section 3.4.1. The settings for all impedance parameters that are used to adjust
the assignment procedure are also changed in the assignment procedure dialogue.
4.2.1 Verification
When analysing the software algorithms the conditions in both Visum and Emme have
to be the same regarding network coding, frequency, in-vehicle and walk times etc. In
this thesis the line run times and all other model coding are based on the future traffic
prognoses in Emme obtained from Trafikverket. Therefore a verification of the Visum
model is done and several tests were made such as analysing the assignment statistic
results and matrices for in-vehicle time, origin wait time and walk time. The transit
lines were studied as well, especially the total volume and travel time on each line.
When comparing travel times with Emme (which in this case is considered to be the
reality) some of the lines did not match. In certain situations there were stop points
missing in the time calculation. This was corrected by adding the missing stop points
in the time profile for each line and updating the run time so that the correct travel
time was copied from the Emme data. Graphic parameters for link and connector
bars were also used in order to evaluate the model. The transit and walk volume was
assigned to the link bars and compared to the Emme model. An important setting for
49
4 A CASE STUDY
the boarding penalty PuT is the data format, which have the default format of seconds.
In this case the data format "precise duration" was used instead of "integer" in order to
obtain the penalty in minutes. One must also change this value for each time profile
(i.e. each line) so that all lines obtains the correct penalty. In all cases where there are
time profile or stop area attributes, which the users can add themselves, it is implied
that those values should be added in the respective menu list.
A small difference between the total walk, transit and transfer volume in the models was
observed and further analyses of the assigned demand were required. When comparing
the input matrix and the assigned demand matrix the difference only consisted of the
intrazonal demand, i.e. all demand which was going in and out through the same node.
This was the case for Emme as well and therefore the differences ought to depend on
the route assignment algorithms or parameter settings, i.e. not a problem with the
input matrix.
The line run times for both Emme and Visum are verified and can be seen in Table 21
(base scenario) and Table 22 (future scenario). The line direction is stated with "r" or
"t" depending on whether it goes to or from the city center, and if there are several
line routes the letters "a" and "b" are used. From these tables it can be concluded that
Visum have a slightly lower line run time in both scenarios. This can either depend
on rounding of the line run times or that Visum have less dwell time. In the model
analysis the line times will be adjusted so that they are equal in order to identify if the
differences depends on the line run times.
50
4.2 Model building in Visum
Table 21: Line run times from Emme and Visum (minutes) in the base scenario
Line EMME Visum Difference
51
4 A CASE STUDY
Table 22: Line run times from Emme and Visum (minutes) in the future scenario
Line EMME Visum Difference
52
4.3 Model analysis
It is of interest to find out how sensitive the results are with respect to the assignment
parameters. When analysing public transport systems there are standard values used
to reflect the impedance for the respective time component (e.g. wait time). When a
model is used to predict the future with estimated values it is hard to calibrate these
parameter values because there are no data to compare to. Also, when using limited
networks and OD-matrices, as in this thesis, calibration and validation can only be
performed in very general terms. Due to these facts a sensitivity analysis has been
performed. The analysis has been done with the base scenario in both Emme and
Visum. First the results from the respective software will be presented and then a
comparison is stated between the two.
According to a model from Trafikförvaltningen, they use different standard values for
the transit assignment parameters in Visum compared to the translated parameters
listed in Table 4. Therefore it is of interest to evaluate the differences, both between
the Visum scenarios and between the Emme scenarios. The values used at Trafikför-
valtningen can be seen in the list below.
• Mean delay: 0.00
• Formula for origin and transit wait time: 1
• Weight factors:
– origin and transit wait = 2.00
– access, egress and walk = 2.00
– transfer = 5.00
– in-vehicle = 1.00
– PuT-Aux ride time = 1.00
As previously mentioned the line run times are slightly different in Emme and Visum.
Therefore an evaluation will be made of the run times, the results are presented in
Section 5.2.2.
53
4 A CASE STUDY
In order to identify if the algorithms function in the same way this type of demand
analysis could be of importance to the thesis. When assigning only 100 trips between
two zones the passenger choice of optimal strategy will be easier to follow and evaluate.
There have been five different tests performed between the orange circled areas in Fig-
ure 13; Nacka Forum to Slussen (13a), Värmdö to Slussen (13b), North of T-centralen
to Gullmarsplan (13c), Saltsjöbaden to Gullmarsplan (13d), and Kvarnholmen to the
north of T-centralen (13e).
Figure 13: Areas that the 100 demand will be assigned between
54
4.3 Model analysis
It is of interest for the thesis to see the volume at a specific node and how it may change
between the base- and future scenario. The chosen nodes are the ones that represents
the public transport stops at T-centralen, Kungsträdgården, Slussen, Sofia and Nacka
forum, see Figure 14. The volume might differ between Emme and Visum and if so
these differences are important to be aware of. The question to be answered is if a
change in the transit line volumes will affect the total volume at a station. Volumes
at the stations can be significantly smaller than in reality which is due to the cut of
transit lines, i.e. some will walk the whole way due to shorter distances between origin
and destination. This could affect Kungsträdgården station especially since the north
going metro only reaches T-centralen with the current network limitation (in reality
this line continues towards Akalla and Hjulsta). This leads to that a lot of passengers
most likely will choose to walk the short distance instead of using a transit line. The
OD-matrix is limited and the passengers have new destination/origin nodes where the
distance between the new nodes might be less than in the unlimited network. Therefore
the results will not be realistic, however there are settings that can force the passengers
to board under certain circumstances (for example when the walk time is longer than
a specified number of minutes).
The volumes of each transit line will be collected from the simulation output and anal-
ysed in Section 6.2. This gives an indication of how route choices are made in respective
software. Other important output will be the total number of public transport trav-
ellers, the total number of those who choose to walk from origin to destination, and
the number of transfers. These results will contribute to a more comprehensive view
of how travellers choose among the transit line options.
55
4 A CASE STUDY
56
5
Results
All simulation results from the sensitivity analysis and the base/future scenario for
respective software algorithm are presented in this chapter. The important output
for comparing the results will be total transit volume, transit line volumes, passenger
kilometres, in-vehicle times, total walking volume, and the average number of transfers
for the passengers.
The result from simulation of the base scenario model is summarized in Table 23. As
showed the total number of boardings is larger in Visum compared to Emme. This is
due to that the travellers choose routes containing more transfers in Visum and less
people walk the whole way to their destination node. This means that in Visum every
person does an average of 0.4 transits per trip compared to in Emme where the average
is 0.36 transfers per trip.
Table 23: Output from the base scenario simulation in Emme and Visum
Base Transit N.o. N.o. Transfers Passenger Auxiliary Assigned
scenario volume boardings transfers per passenger kilometres transit volume demand
Emme 75 516 102 707 27 191 0.36 301 126 21 197 96 713
Visum 75 957 106 482 30 525 0.40 307 481 20 756 96 713
The volume (number of boarding) on each transit line for the base scenario is showed
in Table 24 and 25. The absolute difference between Emme and Visum can also be seen
in the table along with the relative difference, which states the absolute difference in
relation to the average value of both software results. The number of boardings differ
between almost every transit line, in some cases the volumes are lower and some higher
57
5 RESULTS
but as stated before the total number of boardings are larger in Visum. The result
from Emme is graphically shown in Figure 15, where the purple symbolizes the transit
volumes and red represents the volumes out from the origin nodes to the network
(connector links). The result from Visum can be seen in Figure 16, and the volume is
represented by the blue bars and the walking volume from origin or to destination is
marked in green.
Table 24: Total number of passengers boarding on each type of transit mode in respec-
tive software, base scenario
Boardings Emme Boardings Visum Absolute difference Relative difference
Table 25: Number of passengers boarding on each line in respective software, base
scenario
Line Boardings Emme Boardings Visum Absolute difference Relative difference
58
5.1 Simulation results
59
5 RESULTS
The summarized result from the future scenario simulation can be seen in Table 26.
The total transit volume is the same and the number of boardings is still lower in
Emme but the difference is a little smaller compared to the base scenario. There are 1
624 more boardings in Visum. The total number of transfers are lower in Emme and
every traveller does in average 0.35 transits per trip and in Visum 0.37 transits per
trip.
Table 26: Number of passengers boarding on each line in respective software, future
scenario
Future Transit N.o. N.o. Transfers Passenger Auxiliary Assigned
scenario volume boardings transfers per passenger kilometres transit volume demand
Emme 75 995 102 645 26 650 0.35 304 606 20 718 96 713
Visum 76 222 104 269 28 047 0.37 309 985 20 491 96 713
In Table 27 and 28 the results from the future scenario are presented. Note that the
transit lines are not all the same in both scenarios because some of the transit lines
were changed due to the building of the new metro line. The table states the number
of boardings on each transit line in both Emme and Visum, the absolute difference
and the relative difference. Figure 17 illustrates the results in Emme and Figure 18
demonstrates the final result of the simulation in Visum.
Table 27: Total number of passengers boarding on each type of transit mode in respec-
tive software, future scenario
Boardings Emme Boardings Visum Absolute difference Relative difference
60
5.1 Simulation results
Table 28: Number of passengers boarding on each line in respective software, future
scenario
Line Boardings Emme Boardings Visum Absolute difference Relative difference
61
5 RESULTS
62
5.2 Model analysis results
This section shows the evaluation of how the different software algorithms behave when
adjusting one transit assignment parameter at a time, while keeping the OD-matrix
fixed. At first there will be a table stating the results from experimenting in Emme,
i.e. Table 29. Then the same results for Visum are shown in Table 30.
Table 29: Results from the parameter analysis in Emme
Parameters Test value Total transit volume Number of transfers Transfers/person
The results from analysing the difference between Emme and Visum with respect to
the difference between the original results of transit volumes and number of transfers in
each software. The positive values represents that Visum have that much more transit
volume or number of transfers than Emme, see Table 31.
63
5 RESULTS
Table 31: The difference between Visum and Emme with respect to the difference
between the original results
Parameters Test value Difference in transit volume Difference in transfers
The resulting output using the original parameter values (see list in Table 4) can be
seen in Table 32. There are about 75 500 persons that use the public transport network
which of approximately 27 200 transfers during the route. The number of persons who
chooses to walk the whole way between origin and destination is almost 21 200.
When changing the boarding time, from 5 to 0, 4.5 and 5.5, it leads to the results
seen in Table 33. When there is less time penalty for boarding the transit volume will
increase along with the number of transfers during a route. When the penalty is zero
minutes, most demand choose to ride with the transit lines. If the penalty is set to 4.5
instead, the result leads to that less people travel with transit lines and there are fewer
transfers made. If there is a higher boarding penalty than originally, 5.5 minutes, the
total transit volume and number of transfers will decrease.
Table 33: Parameter analysis results for the boarding time weight in Emme
Boarding Transit Total boarding N.o. Transfers Passenger Auxiliary
time volume boarding transfers per passenger kilometres transit volume
The result from changing the wait time factor is showed below in Table 34. The wait
time factor can be set between 0 and 1, and gives the distribution of travellers arrival
to the transit stops. If the factor is 0.5 there is a uniform distribution of travellers and
lower values can be used if there are long headways or passengers know the timetable.
This means that the travellers wait half the waiting time in average, which in the
assignment is represented by the passengers arriving at the station after half the waiting
64
5.2 Model analysis results
time. A higher value of the wait factor gives less total transit volume and more demand
will decide to walk to the destination instead, which reflects a transport system with
high timetable uncertainties.
Table 34: Parameter analysis results with wait time factor in Emme
Wait Transit Total boarding N.o. Transfers Passenger Auxiliary
time factor volume boarding transfers per passenger kilometres transit volume
The waiting time weight indicates how the travellers perceive the time to wait for a
transit vehicle, see result in Table 35. The original value is 1.5 and when the wait
time is less important, e.g. 1 or 1.4, the transit volume will increase. If it is weighted
with 1 approximately 2 300 more persons will decide to use public transport instead
of walking the whole way. The number of transfers will increase significantly, about 13
200 more. This depends on that more routes with transfers will be attractive due to
that the wait time for transferring is weighted less. When the weight is 1.4 the number
of transfers will decrease with about 200, however the total volume that chooses public
transport increases with more than 800. This depends on that passengers have more
incentive to take a route with only one boarding than to board several times. If the
weight is set to 1.6, more people (290) will choose to walk instead and the number of
transfers decreases with almost 900. When the weight is increased to 2, the number
of transfers decrease with about 3 200 and approximately 1 700 persons less choose to
use public transport.
Table 35: Parameter analysis results with wait time weight in Emme
Wait Transit Total boarding N.o. Transfers Passenger Auxiliary
time weight volume boarding transfers per passenger kilometres transit volume
The auxiliary time weight will affect how many that chooses to walk instead of using
the public transport network. It will also affect the number of transfers due to that
the transfer walk time (all walking inside the network) is included in the auxiliary
time. The result from this parameter analysis are available in Table 36. The original
value for the weight is 2 and when the time is unweighted (compared to in-vehicle
time, i.e. weight equals 1) significantly more demand will walk instead of riding with
a public transit vehicle. It differs with more than 13 000 and the number of transfers
will increase with almost 15 500. If the weight is 1.9, almost the original value, about
445 more persons will choose to walk to the destination and the number of transfers
decreases with 2 030. When the weight is increased to 2.1 the demand that uses public
transport decreases with approximately 1 170 and the number of transfers drops with
about 145.
65
5 RESULTS
Table 36: Parameter analysis results with auxiliary time weight in Emme
Auxiliary Transit Total boarding N.o. Transfers Passenger Auxiliary
time weight volume boarding transfers per passenger kilometres transit volume
A parameter translation has been made between Emme and Visum so the sensitivity
analyses are done with the same conditions. The result with these original values can
be seen in Table 37.
75 957 106 482 30 525 0.40 307 481 307 481 96 713
The boarding time in Emme correspond to the boarding time penalty PuT in Visum
and the assignment parameter is changed from values between 0-5.5 see Table 38 for
the results. With a penalty of 0 an increment of the total transit volume of over 4
800 people and more will choose a route that includes transfers, almost 47 000 more
than the original parameter result. If the boarding penalty changes to 4.5 there is an
increment, however not as large as without penalty. Approximately 800 more will use
public transport instead of walking and the number of transfers will almost be 1 300
more. If the boarding time instead was set to 5.5 minutes, the total number that use
the public transport will decrease with 550 passengers and the number of transfers will
be 1 800 less.
Table 38: Parameter analysis results with boarding time penalty in Visum
Boarding Transit Total boarding N.o. Transfers Passenger Auxiliary
time penalty volume boarding transfers per passenger kilometres transit volume
In Visum the wait time factor is split into two parts, formula for OWT (origin wait
time) and for TWT (transfer wait time), and they are both set to 0.5 in the original
translation. Both parameters have the same values in each test so it should reflect the
same changes as in Emme, see result in Table 39 below. Low values leads to that more
passengers travel with the transit lines, for example the penalty 0.01 gives around 6 600
more passengers and nearly 29 000 more transfers between two vehicles. If the value
is changed to 0.4, the result will still be that more chooses to use public transport,
there will be approximately 1 100 more. The number of transfers will also be around
1 100 more than the original result. If the values instead are 0.6, the total number of
passengers travelling with public transport will be about 1 100 less and the number of
transfers will decrease with nearly 950 compared to the original settings.
66
5.2 Model analysis results
Table 39: Parameter analysis results with the formula for origin and transfer wait time
in Visum
Formula for Transit Total boarding N.o. Transfers Passenger Auxiliary
OWT,TWT volume boarding transfers per passenger kilometres transit volume
The wait time weight in Emme corresponds to the factors for both OWT and TWT,
which originally is set to 1.5. The result after shifting these values can be seen in
Table 40. If the new factor is set to one, which means the time is unweighted, compared
to in-vehicle time, about 2 000 persons will switch from walking to using the public
transport system. There will be about 14 000 more transfers and when the factor is 1.4
there are 200 more transfers compared to the original result. The total transit volume
when the value is 1.4 has increased with more than 740 compared to the decrease of
around 440 when the factor was 1.6. Higher values of these factors will lead to large
differences, especially regarding the number of transfers.
Table 40: Parameter analysis results with factor for origin and transfer wait time in
Visum
Factor for Transit Total boarding N.o. Transfers Passenger Auxiliary
OWT,TWT volume boarding transfers per passenger kilometres transit volume
The auxiliary time is divided into three parts in Visum; Access time (ACT), Egress
time (EGT) and Walk time (WT). The original value in Emme is 2 and therefore
each of part of the auxiliary time in Emme is also 2. In the analysis all three parts
were switched to the same value in each simulation, in Table 41 the results are stated.
Unweighted times, factors equal to 1, gives decreased total transit volumes, about 12
800, and the number of transfers will also be less (approximately 17 100). If the factors
are set to 1.9, the difference will not be as high compared to the original result, about
560 persons chooses to walk instead and the transfers differs with 1 670. If the factor
is set to a higher value, 2.1, 1 090 will switch from walking to taking public transport
instead. People will also choose routes with more transfers (1 642 more) when the
factor is 2.1.
Table 41: Parameter analysis results with factor for access, egress and walk time in
Visum
Factor for Transit Total boarding N.o. Transfers Passenger Auxiliary
ACT,EGT,WT volume boarding transfers per passenger kilometres transit volume
There is a factor in Visum that does not exist in Emme, i.e. a penalty for transfer-
ring. The penalty is set in minutes and will be multiplied with the number of transfers
67
5 RESULTS
made on a trip. With 1 minute in penalty it does not lead to any significant difference
regarding the total transit volume (only 144 more), although it affects the number of
transfers among the routes, about 2 600 less transfers. A penalty of 15 minutes gives
a large difference when it comes to the total volume travelling with public transport,
which decreases with 245 persons. A 5 minute penalty gives a decrease of approxi-
mately 12 500 less transfers and 15 minutes leads to more than 25 300 less transfers.
Therefore this factor should be thoroughly chosen and evaluated before using it in
traffic simulation studies. The results from the mentioned experiments can be seen in
Table 42.
Table 42: Parameter analysis results with factor for number of transfers (NTR) in
Visum
Factor Transit Total boarding N.o. Transfers Passenger Auxiliary
for NTR volume boarding transfers per passenger kilometres transit volume
68
5.2 Model analysis results
By analysing the line itineraries and changing the times in Visum so that they almost
are the same, see Table 43 for the calibrated line run times, it can be determined that
the differences does not depend on the difference in line run times. This is due to the
fact that the difference between the total number of boardings in respective software
still are significantly high.
Table 43: The new run times in Visum compared to the original run times in Emme
Line Emme line run time New Visum line run time Absolute Difference
69
5 RESULTS
Table 44 shows the number of boardings on each line and the absolute difference be-
tween the new run times in Visum and the base scenario in Emme. The total absolute
difference is 3753 boardings more or less.
Table 44: Line boardings with the new run times in Visum and the original run times
in Emme
Line Visum boardings, new run times EMME boardings, original run times Absolute difference
70
5.2 Model analysis results
In Table 45 the boarding difference between the new run times and the original run
times in Visum can be seen. When comparing the original run time results and the new
run time results in Visum the number of boardings only differ with 22. This comparison
states that the calibration of line run times did not lead to any significant improvement.
Therefore the run times could not be the reason why the main differences between the
number of boardings occur.
Table 45: Line boardings with the new and the original run times in Visum
Line Visum boardings, new run times Visum boardings, original run times Absolute difference
71
5 RESULTS
When assigning 100 trips between five different zone pairs the resulting mean in-
vehicle time and number of boardings for respective software can be seen in Ta-
ble 46 and 47.
Table 46: Mean in-vehicle time (minutes) for the five tests with 100 demand
From area To area Emme Visum
From these mean in-vehicle times one can see that Visum have lower, or equal, values
than Emme. This might depend on how the different programs calculates and rounds
numbers in time format.
Table 47: Number of boardings per line for the five tests with 100 demand
From area To area Line Boardings Emme Boardings Visum
The number of boardings are exactly the same for all experiments and lines, which
implies that the algorithms work in the same way in this specific case.
However, one of the lines shows a difference between the algorithms when assigning
only 100 demand from Gullmarsplan to the most eastern part of Nacka municipality in
the future scenario. The results are presented in a diagram, Figure 19, by the number
of boardings on each used line.
72
5.2 Model analysis results
According to these results one can see that the Visum passengers choose itineraries
with more transfers involved than the passengers in Emme. Visum also have three
more lines that passengers choose to travel with. The total number of boardings
will not be the same in both software since Visum have more boardings due to the
assignment algorithm and parameter settings.
The node analysis was made in both Emme and Visum and the results are presented
separately to be compared later in Section 6.3.
All results from analysing the mentioned nodes in Emme can be seen in Table 48 and
49 in respective scenario.
The boarding volumes at T-Centralen will increase (+382) and this depends on that
the number of transfer boardings are higher in the base scenario compared to the future
scenario. The number of passengers alighting at T-centralen will instead decrease (-1
329) in the future scenario. The flow of passengers who passes through the station on
a transit vehicle is basically unchanged.
Kungsträdgården will have more passengers boarding after the opening of the new
metro line and also more total alighting passengers. The largest increase is the number
of through passages in the future scenario, 5 447 more compared to 0 in the base
scenario.
No travellers are passing through Slussen in any of the scenarios. This is due to the cut
of the transit lines and there are no possibilities to just pass Slussen in this network,
73
5 RESULTS
passengers can only board or alight a transit line at this node. The total number of
passengers who alight a transit line (including transfer) at the station will decrease
with 733 in the future scenario. 34 passengers find other itineraries to reach their
final destination and 26 passengers will not start their route at Slussen. The number
that boards a vehicle at Slussen decrease with 1 611 passengers compared to the base
scenario.
The Sofia station does not have any volumes in the base scenario since there are no
transit lines available until the future scenario. Then the new metro will pass Sofia
which will of course increase the volumes and that affects boarding, alighting and
passing travellers.
Nacka Forum will have a larger number of total boarding passengers and alighting
passengers in the future scenario (+1 072 and +402) compared to the base scenario.
The sum of travellers who passes the station onboard of a vehicle will decrease with
about 2 000 passengers. This is due to the fact that the transit lines are designed
differently in the two scenarios. Most transit lines that passes Nacka Forum in the
base scenario have this station as an end station in the future scenario.
Table 48: Node results from the base scenario in Emme
Boarding passengers Passengers through Alighting passengers
Station Initial Transfers Total Final Transfer Total
All results from analysing the mentioned nodes in Visum can be seen in Table 50 and
51 in respective scenario.
At T-centralen the total number of boarding passengers will decrease with 2 522 pas-
sengers from the base to the future scenario. This depends mostly on the number of
passengers that transfers at T-centralen are less after the new metro line is built. The
number decrease with 2 285 passengers compared to before. More passengers will just
pass the node without stopping; approximately 3 300 more travellers will pass the node
in a vehicle. Almost 4 050 people less will alight at the node and it mostly depends on
that less (-3 001) people will alight for a transfer.
Kungsträdgården will have more total boarding and total alighting passengers after
the new metro line is open. However, the largest difference between the scenarios is
the number of passengers that only passes the station (+4 778).
74
5.2 Model analysis results
Slussen will have less people boarding (-764) and alighting (-747) at the stop. The
decrease is caused by that less people will transfer at the node.
Sofia will of course have larger amount of travellers in every category because in the
future scenario there will actually be a transit line passing the node. 3 703 passengers
will pass the node without stopping and totally 1 139 people will board and 699 people
will alight.
Nacka Forum will loose boarding passengers and that can depend on the transit lines
are adjusted when the new metro line is open. 181 people less will start their journey
at Nacka Forum but 54 more will do a transfer boarding. Fewer people will pass the
stop (-1 350) after the new metro is open and the number of alighting will decrease
with 395 passenger. This also depends on that fewer passengers will transfer at the
station.
75
5 RESULTS
Table 52: The result from using the standard transit assignment in Emme 4 and
extended transit assignment (without any additional settings)
Emme version N.o. boardings Passenger kilometres
The result from using the flow distribution at origins choice can be seen in Table 53.
The logit function scale is varied between 0 and 1, where 0 represents an equal dis-
tribution between the connector links and 1 represents all travellers using the best
connector.
Table 53: Result from using the option flow distribution at origins
Scale, distribution at origins N.o. boardings Passenger kilometres
The result from using the flow distribution at regular nodes with auxiliary transit
choices can be seen in Table 54. The scale is the same for the distribution at origins
for all test (0.5) while the scale for the regular nodes are varied. The distribution in
this case is between the amount of travellers that waits for another attractive public
transport line or uses an efficient auxiliary route instead of staying on board the current
vehicle. The logit function scale is varied between 0 and 1, where 0 represents an equal
distribution between the connector links and 1 represents that all travellers use the
best option.
Table 54: Result from using the option flow distribution at regular nodes with auxiliary
transit choices
Logit setting Scale, regular nodes N.o. boardings Passenger kilometres
Logit distribution between auxiliary and wait 1 107 499 306 265
Logit distribution between auxiliary and wait 0.75 107 870 306 827
Optimal strategy (all passengers chooses the same line/aux. path) 0.5 107 079 305 503
Logit distribution between auxiliary and wait 0.5 108 228 307 968
Logit distribution between auxiliary and wait 0.25 109 149 309 909
Logit distribution between auxiliary and wait 0 95 495 276 552
The result from using the flow distribution between transit lines can be seen in Table 55.
Both the scale for the distribution at origins and regular nodes is constant 0.5. The
settings regarding logit distribution between auxiliary and waiting will still be used,
i.e. the optimal strategy is not used in this case in order to obtain some variation in
the passengers choice of route.
76
5.3 Emme 4, extended transit assignment, results
Table 55: The result when using the additional setting to use flow distribution between
transit lines
Flow distribution between transit lines N.o. boardings Passenger kilometres
77
6
Analysis
This chapter concludes the comparisons of the software algorithms from the simulation
results, software sensitivity, and node analysis.
Table 56: Comparison between Emme and Visum results from the base scenario
N.o. public N.o. passengers
transport N.o. N.o. N.o. transfers Passenger that walks Total n.o.
Base scenario passengers boardings transfers per passenger kilometres the entire way travellers
Emme 75 516 102 707 27 191 0.36 301 126 21 197 96 713
Visum 75 957 106 482 30 525 0.40 307 481 20 756 96 713
Difference 441 3 775 3 334 0.04 6 355 -441 0
Emme relative
Visum 0.6 % 3.7 % 12.3 % 11.1 % 2.1 % - 2.1 % 0
Table 57: Comparison between Emme and Visum results from the future scenario
N.o. public N.o. passengers
transport N.o. N.o. N.o. transfers Passenger that walks Total n.o.
Future scenario passengers boardings transfers per passenger kilometres the entire way travellers
Emme 75 995 102 645 26 650 0.35 304 606 20 718 96 713
Visum 76 222 104 269 28 047 0.37 309 985 20 491 96 713
Difference 227 1 624 1 397 0.02 5 379 -227 0
Visum relative
Emme 0.3 % 1.6 % 5.2 % 5.7 % 1.8 % - 1.1 % 0
Future- relative base
scenario in Emme 0.6 % 0.1 % 1.9 % 2.8 % 1.2 % - 2.3 % 0
Future- relative base
scenario in Visum 0.3 % 2.1 % 8.1 % 7.5 % 0.8 % - 1.3 % 0
79
6 ANALYSIS
Table 58: Comparison of the results obtained from Emme with the Trafikverket pa-
rameter values and from Visum with the Trafikförvaltning parameter values
N.o. public N.o. passengers
transport N.o. N.o. N.o. transfers Passenger that walks Total n.o.
Base scenario passengers boardings transfers per passenger kilometres the entire way travellers
Emme,
Trafikverket 75 516 102 707 27 191 0.36 301 126 21 197 96 713
Visum,
Trafikförvaltningen 74 306 96 785 22 479 0.30 291 708 22 407 96 713
Difference 1 210 5 921 4 712 0.06 9 418 -1 210 0
Emme relative
Visum 2% 6% 17% 16% 3% -6% 0
There will be variations regarding the transit lines since some of them have been
edited between the scenarios. These lines will not be of interest when comparing the
scenarios within the same software. This is due to that the boarding volumes can change
drastically from the base to the future scenario because of the editing. Therefore they
are not included in Table 59. However, when comparing the two software algorithms
these edited transit lines are of high importance.
Table 59: Absolute boarding difference between base and future scenario in respective
software
Line Visum Emme
The transit line 53t loose passenger in the future scenario in comparison to before the
new metro line is open. The difference between the scenarios in Visum is a little higher
than in Emme but the difference is almost negligible. Travellers take the new metro
line from Sofia instead of taking the bus and could be a contributing factor why there
are less passengers travelling with line 53t.
The transit line 74r (Mariatorget to Sickla köpkvarter) also have fewer passengers in
the future scenario in both Visum and Emme. The differences are more significance in
Emme where it changes from 2 410 passengers to 921 and in Visum from 2 622 to 1
416 passengers. This is also the case for line 74t which is the opposite direction (Sickla
köpkvarter to Mariatorget), 2 789 passengers less in Visum and 3 115 in Emme. In the
future scenario more people take the metro line to Slussen or T-centralen and transit to
the new blue line to get to their destination/origin in Nacka municipality/Södermalm
instead of taking the bus line 74t.
80
6.1 Comparison of the simulation results
The light rail line 22t, from Solna centrum to Sickla udde, will have more travellers
in the future scenario and in Emme the total increase is 1 106 compared to Visum,
2 088. In this scenario a lot of the buses stop at Nacka Forum and only have routes
within Nacka municipality. So in order to travel from the other areas in Stockholm to
Nacka municipality the choices are not as many as in the base scenario and this can
be a reason for the increased number of passengers on line 22t.
The red metro line, north direction, have an increase in the future scenario with 2 010
passengers and in south direction a small increase of 262 people in Visum. Emme also
have more passengers in the future on the red line in the north direction, 2 205, and
226 more in the south direction compared to the base scenario. The increase in north
direction could be an effect of that more people choose the red line and later change
to blue line in order to reach Nacka Forum faster.
The green metro line in the north direction loses passengers when the new metro line is
open in both Emme and Visum. The decrease is larger in Emme where the difference
is 2 263 people and in Visum 1 760 persons. The loss of passengers could be an effect
of that the travellers might have their origin node closer to the new metro line and
chose to use this instead when travelling north. In Figure 20 the boarding differences
from Table 59 can be seen in a diagram.
Figure 20: Diagram over boarding difference between base and future scenario in re-
spective software
A comparison was also made between the two software algorithms regarding the same
scenario. In Table 28 the boarding volumes for every transit line is showed. Six lines
with the most significant differences have been summarized in Table 60.
Table 60: Boarding difference between Emme and Visum in respective scenario
Line Base scenario Future scenario
The bus line 55t (Hjorthagen to Sofia) have more passengers in Visum (+660) in the
base scenario. Also in the future scenario Visum have more travellers on the bus line,
522 more passengers. It is generally the same regarding line 401r where there are 751
more passengers choosing that line in Visum compared to Emme in the base scenario.
In the future scenario the difference is just 104 more travellers in Visum.
81
6 ANALYSIS
Before opening the metro the light rail 22t, between Solna centrum and Sickla udde,
have 172 passengers in Emme and 83 in Visum so the difference is in total 89 travellers.
After opening the metro the total difference is 1 071.
Regarding the new blue metro line there are no passengers in the base scenario. In
the future scenario there are 3 322 who uses the new metro in the south direction in
Emme and 2 009 people in Visum. The line in opposite direction has 4 894 passenger
in Emme and 3 747 in Visum. So in general there are more travellers choosing the new
blue metro line in Emme.
On the green line, in the north direction, there are 276 more passengers in Visum
compared to Emme in the base scenario. In the future scenario Visum still have more
passengers, 779 more than Emme.
In Figure 21 the boarding differences from Table 60 can be seen in a diagram.
Figure 21: Diagram over boarding difference between Emme and Visum in respective
scenario
The following section visualises the results from Section 5.2.1 with graphs from both
Emme and Visum with respect to the different assignment parameters and output
variables.
The boarding time affects the total travel volume in both Emme and Visum and how
much the result differs when changing the time is very similar in both software pro-
grams, see Figure 22. The total volume increases a bit more in Emme when the
82
6.2 Software sensitivity
boarding time is 4.5, about 55 more will travel with public transport. When the time
instead is 5.5 the volume will decrease more in Visum, 145 less people than in Emme.
The number of transfers will be affected more in Visum. When the boarding time is
4.5 there will be about 800 more transits in Visum and when the value is 5.5 the result
is about 120 transfers less than in Emme.
(a) Transit volume as a function of the board-(b) Number of boardings as a function of the
ing time weight boarding time weight
(c) Number of transfers as a function of the(d) Average number of transfers per passenger
boarding time weight as a function of the boarding time weight
(e) Passenger kilometres as a function of the (f) Auxiliary transit volume as a function of
boarding time weight the boarding time weight
Figure 22: Graphs comparing the result from the parameter analysis of boarding time
weight in each software
83
6 ANALYSIS
Regarding the waiting time factor Emme have a lower public transport volume and
number of transfers when the factor is very low, 0.01. When the factor is 0.4 there is
less difference between the two programs, but when it comes to the number of transits
it decreases in Emme than in Visum. When the factor is 0.6 the results are similar
in both software products regarding the total transit volume, however regarding the
number of transfers it will increase with 1 855 more in Emme than in Visum. All
graphs can be seen in Figure 23.
(a) Transit volume as a function of the wait(b) Number of boardings as a function of the
time factor wait time factor
(c) Number of transfers as a function of the(d) Average number of transfers per passenger
wait time factor as a function of the wait time factor
(e) Passenger kilometres as a function of the (f) Auxiliary transit volume as a function of
wait time factor the wait time factor
Figure 23: Graphs comparing the result from the parameter analysis of wait time factor
in each software
84
6.2 Software sensitivity
When it comes to the wait time weight (factor for OWT and TWT) both software
programs are about equally sensitive to value changes. The most relevant simulation
results are shown in Figure 24. The most significant difference occurs when the pa-
rameter value is 2 (0.5 higher than originally), where Visum obtains a decrease of 2
330 persons that does not choose public transport in comparison to Emme where the
difference from the original result is around 1 650.
(a) Transit volume as a function of the wait(b) Number of boardings as a function of the
time weight wait time weight
(c) Number of transfers as a function of the(d) Average number of transfers per passenger
wait time weight as a function of the wait time weight
(e) Passenger kilometres as a function of the (f) Auxiliary transit volume as a function of
wait time weight the wait time weight
Figure 24: Graphs comparing the result from the parameter analysis of wait time
weight in each software
85
6 ANALYSIS
Both Emme and Visum receives decreased transit volumes and number of transfers with
lower auxiliary weights, as can be seen in Figure 25, this depends on that the travellers
does not have as high resistance towards walking compared to when the original settings
were used. When the weight is set to 2.1 there are much more transfers in Visum than
in Emme, both compared to each other and compared to the original result. In Emme
the transfers only increase with 400 compared to the original and in Visum there are
about 1 640 more transfers. Figure 25e shows that the passenger kilometres in Visum
increases faster than in Emme when the auxiliary time weight is between one and
two. That is, in Visum passengers are more likely to use public transport. However,
the most frequently used parameter value for the auxiliary time weight is 2, so the
difference between the programs will not be significant.
86
6.2 Software sensitivity
(a) Transit volume as a function of the auxil-(b) Number of boardings as a function of the
iary time weight auxiliary time weight
(c) Number of transfers as a function of the(d) Average number of transfers per passenger
auxiliary time weight as a function of the auxiliary time weight
(e) Passenger kilometres as a function of the (f) Auxiliary transit volume as a function of
auxiliary time weight the auxiliary time weight
Figure 25: Graphs comparing the result from the parameter analysis of auxiliary time
weight in each software
87
6 ANALYSIS
In Visum there are an extra penalty for the number of transfers made, which does not
exist in Emme. With this parameter one can adjust the number of transfers, boardings,
and passenger kilometres. The total amount of transit volume and auxiliary volume
will not be significantly affected by adding a transfer penalty as can be seen in the
graphs below, Figure 26. Note that the Emme value is fixed in all graphs and is shown
as a straight line in order to compare it with Visum.
(a) Transit volume as a function of NTR (num-(b) Number of boardings as a function of NTR
ber of transfers) (number of transfers)
(c) Number of transfers as a function of NTR(d) Average number of transfers per passenger
(number of transfers) as a function of NTR (number of transfers)
(e) Passenger kilometres as a function of NTR (f) Auxiliary transit volume as a function of
(number of transfers) NTR (number of transfers)
Figure 26: Graphs comparing the result from the parameter analysis of n.o. transfer
penalty in each software
The results conclude that all examined output values will be equal if NTR is increased
to a value between 1 and 2.
88
6.2 Software sensitivity
The run times in Table 43 are similar for both Emme and Visum, and the small
differences can depend on rounding of times in the programs. The total number of
boardings, in Table 44, are still higher in Visum and that does not occur due to the
differences of the travel times at the specific transit lines. This can be seen in Table 45
which states that the change of run times in Visum does not result in any significant
differences. The conclusion of this model test is that Visum have more boardings than
Emme and changing the run times only have a small effect on the boardings on each
transit line. This also implies that the strategy of how to choose a route is calculated
in a different way in Visum compared to Emme.
The mean in-vehicle time indicates if people in general are spending more time riding
vehicles due to longer route choices in aspect of time or length. If more people walk
the distance between origin and destination the mean in-vehicle time will be less than
if more used the existing transit lines. Table 46 shows the mean in-vehicle times
stated for Emme and Visum and below in Table 61 the absolute difference is shown in
minutes. As stated in the table there are only small differences between the software
programs.
Table 61: Absolute difference between the mean in-vehicle time from the analysis with
100 demand
Area (from) Area (to) Absolute difference
The OD-matrix generates 100 trips from an origin node to a specific end destination.
How the 100 passengers travel from these defined nodes are shown in Table 47. The
volumes will split up exactly the same in both Emme and Visum and that is why
the mean in-vehicle times are almost the same. The difference be caused by that the
rounding of time can differ between Visum and Emme. Although, one test led to a
difference in the number of boardings. This was the test with 100 demand between
Gullmarsplan and the eastern part of Nacka, see Figure 19. In general there are more
total boardings in Visum, which indicates that those passengers transfer more compared
to the passengers in Emme. As showed in the figure Visum travellers use 11 different
transit lines compared to Emme, where 8 transit lines are used. This fact, along with
the results from the full matrix, indicates that Visum tend to split the flows on routes
containing more transfers than Emme does.
89
6 ANALYSIS
Figure 27: Comparisons between the base and future scenario in Emme and Visum at
T-centralen with respect to the number of passengers boarding, alighting or passing
through the station.
90
6.3 Comparison of the node analysis
At Slussen there are more alighting passengers in Visum but less boardings in compar-
ison to Emme. The station at Nacka Forum will have some more passengers in total
except regarding the ones alighting to reach their final destination. The biggest differ-
ence is that there are more through passing passengers in Emme (+750) and more who
boards after transferring (+143). A complete comparison between the two scenarios
and software results is displayed in Figure 28.
Figure 28: Comparisons between the base and future scenario in Emme and Visum
at Slussen with respect to the number of passengers boarding, alighting or passing
through the station.
91
6 ANALYSIS
Regarding the future scenario there are more differences concerning all the stations. In
Sofia there are more through passing in Emme, almost 1 600 more than in Visum. In
T-centralen the number of final alighting is larger in Visum (+1 379) and the number
alighting to do a transfer is higher in Emme (+603). A complete comparison between
the two scenarios and software results is displayed in Figure 29. The base scenario is
not represented in this diagram since Sofia station does not exist before building the
metro, i.e. in the base scenario.
Figure 29: Comparisons between the base and future scenario in Emme and Visum at
Sofia with respect to the number of passengers boarding, alighting or passing through
the station.
92
6.3 Comparison of the node analysis
In Kungsträdgården there are more passengers generally in Emme and the largest dif-
ference is the number that does their final alighting, 878 more compared to Visum.
Also the volume that passes the station on a vehicle is larger, about 670 more pas-
sengers. In Slussen the volumes are overall higher in Visum and mostly because of
more transfer boarding (+660) and transfer alighting (+598) in comparison to Emme.
A complete comparison between the two scenarios and software results is displayed in
Figure 30.
Figure 30: Comparisons between the base and future scenario in Emme and Visum
at Kungsträdgården with respect to the number of passengers boarding, alighting or
passing through the station.
93
6 ANALYSIS
Nacka Forum have a greater number of transfer boarding in Emme (+1 161) but the
number of through passing passengers is higher in Visum (+97). A complete compar-
ison between the two scenarios and software results is displayed in Figure 31.
Figure 31: Comparisons between the base and future scenario in Emme and Visum at
Nacka Forum (bus stop) with respect to the number of passengers boarding, alighting
or passing through the station.
These results might be the most important with respect to the case study, since the
metro end-station will be located at Nacka Forum. In the figure one can see that more
travellers will chose the metro in Emme, while travellers go by highway buses in Visum
instead. In this specific case the results show that an investment in a new metro will
be of greater use in Emme than in Visum.
94
6.3 Comparison of the node analysis
A comparison between the Emme and Visum results at each node regarding the base
scenario can be seen in Figure 32.
95
6 ANALYSIS
Figure 33 shows a comparison between the Emme and Visum results at respective node
from the future scenario.
96
6.4 Comparison of the algorithms for public transport assignment
As mentioned in the previous section no logit split is used in the Emme standard transit
assignment. However it is implemented in the latest version of Emme where extended
transit assignment complements the basic standard assignment. The other settings
that could be used in the extended assignment will be analysed in this section.
97
6 ANALYSIS
There should be the same results in Emme 4 with standard assignments as in Emme
4 with extended assignment and optimal strategy settings. However, Table 52 shows a
difference. This could be explained by that the extended transit assignment calculates
more exact and therefore less boardings were obtained for the extended version of the
optimal strategy. The standard assignment in Emme 4 gives exactly the same number
of boardings as in Emme 3, which proves that the same strategy algorithm is used (can
also be confirmed in the manual of Emme 4, [26]).
When using the scale distribution at origins the higher scale value gives more demand
on the shortest connector, the highest value will result in the optimal strategy (i.e.
all demand on the best connector). When the scale is set to zero all the connectors
from the origins will have the same demand and in this case it will result in larger
number of total boarding in the system, see Table 53. When the scale is set to 1 the
boardings will be more similar to the result from using the optimal strategy. To obtain
a more realistic model one should calibrate the scale value so that the results resembles
a measured traffic behaviour for that specific situation.
Table 54 shows the results from varying the scale factor for flow distribution with
auxiliary transit choices. This implies that less travellers board transit lines when the
scale factor is lower, i.e. the flow is distributed equally on all available auxiliary transit
choices as well as transit lines. With the scale set to 1 the system behaves like the
optimal strategy, this means that the passengers choose only the best option and that
leads to more boardings on transit lines. A realistic scale factor should be somewhere
between the highest and the lowest value.
In Table 55 the results from using flow distribution based on frequency and the distri-
bution based on both frequency and transit time to destination are shown. When using
the latter option, faster lines with lower frequency will be more attractive compared to
if only the frequency is taken into account. When the scale is constant the distribution
based on both frequency and transit time gives a lower amount of total boardings (106
529) compared to only frequency based (108 228). However, when studying the total
passenger kilometres the difference is only 1 341 kilometres. This means that in general
every passenger travels 2.85 kilometres compared to 2.88 kilometre.
An important result is that the frequency and transit time assignment generate similar
number of boardings and passenger kilometres as in Visum (base scenario). The number
of boardings in Visum is 106 482 and 106 529 in Emme 4 with the last mentioned
assignment settings. Passenger kilometres in Visum is 307 481 and 306 627 in Emme
4.
98
7
Conclusions and future work
This chapter will conclude the comparisons between the algorithms in respective soft-
ware. The objective to be answered was What are the differences with modelling a
public transport network in Emme and in Visum, based on that the passengers only
have information about the travel times and the line frequency, and why does the dif-
ferences occur? The following list answers the main objective in this master thesis as
well as states some important conclusions.
• The differences are:
– That the public transport algorithms take different parameters in more or
less consideration
∗ Visum first focuses on shortest total travel time and then considers
other lines with respect to maximum waiting time → lines with long
headways and or travel times are excluded
∗ Emme first focuses on shortest travel time and then considers the total
travel time for other lines with half the waiting time → less routes are
determined to be attractive than in Visum
– Rounds off the values in the origin/destination matrices
– Standard parameter settings differs between software (at least in Sweden)
• It is important to be aware of the differences in order to conduct a more correct
simulation and knowing how to interpret the results
• However, it is more important for the results to choose the right parameter values
than the right software
The algorithms generate differences in the route choices, i.e.: transit line volumes, the
number of boardings, total transit/auxiliary volume, and passenger kilometres. In the
current situation, there is no scientific explanation as to why they have chosen the
parameters contained in Trafikverket’s models. The same applies to the parameters
of Visum, which seems to have been chosen based on Emme parameters or traveller
behaviour surveys. The difference between those parameter values is that Visum has a
waiting factor equal to 1 and waiting weights equal to 2 and only a penalty for the num-
ber of transfers made instead of boarding penalty as in Emme optimal strategy.
99
7 CONCLUSIONS AND FUTURE WORK
The examples mentioned in Chapter 3 and the analysis with 100 demand stated in
Chapter 5 shows that the software results sometimes might be the same. This could
depend on the network construction or that differences propagate with larger amount
of transit lines and assigned demand. It could also depend on that the line times are
prioritized differently, according to the results seen in brief in Table 62 below.
Emme takes the shortest line travel time into account. Visum takes the shortest total travel time compared to
the faster lines combined maximum travel time into account.
This represents a situation where the travellers
have to wait the maximum waiting time.
Emme compares the examined lines travel time Visum compares the examined lines travel time
with the faster (in terms of which has the shortest with the faster (in terms of which has the shortest
travel time) lines combined total travel time. total travel time) lines combined total travel time.
Thus, if one were to board the examined line Thus, if one were to board the examined line
instead of waiting for one of the faster lines. instead of waiting for one of the faster lines.
Emme includes the wait time in the flow Visum includes the wait time in both the flow
distribution and in the second attractivity test. distribution, the first and second attractivity test.
Emme can exclude lines that have a little Visum can exclude lines with a long headway, and if the
longer travel time than another line. headway is reduced very little this line may become attractive.
This might exclude those lines that depart more frequently This is not very realistic and might exclude
and logically some travellers should choose that line. faster lines with less frequent departures.
To summarize the conclusions with respect to the studied project, Emme optimal
strategy might overestimate the new metro lines effect or perhaps Visum algorithm
will underestimate it. Either way it is interesting to see that there is such a significant
difference between the two assignment algorithms.
An alternative algorithm that is available in Emme 4 was mentioned and studied pre-
viously in this thesis, called extended transit assignment. In this extended version of
the standard transit assignment one can add extra costs for boarding, in-vehicle time,
and auxiliary time. It also contains a flow distribution to and from origin/destination
nodes, so called logit distribution. The percentage that chooses a particular route
could be based on line frequency (headway) or a combination of the frequency and
travel time. Regarding the first option the travellers are unaware of the line departure
time and with the second option the travellers have knowledge about the timetable,
which is more realistic. This could be managed in Visum as well and as stated previ-
ously in the report the differences between the examined results does mostly depend
on the parameter settings. For example the auxiliary time weight and wait time factor
affects the results more than the wait time weight and boarding penalty. One can
also adjust the extra penalty for transfers in Visum in order to recreate the results
from Emme optimal strategy. Another option is to change the spread factor in Emme,
since increased waiting times can reproduce the results from Visum as well. There-
fore the choice between which software to use is mainly based on choice of preference
and the knowledge of how the software algorithms work along with correct parameter
adjustments are more important.
In future important infrastructure projects we suggest that both "extended transit
assignment" in Emme and "headway based assignment" in Visum is used. These are
frequency based assignments with some similar functionalities. Further work on this
topic could also be to calibrate the assignment parameters for both Emme and Visum,
at least for one or a few projects so that it could be determined if the parameter values
in each software are realistic in the examined traffic system. One might also continue
100
the comparison with the other assignment algorithms available and determine the most
realistic assignment for different transit systems. Our recommendation is to examine
the significance of assignment parameters even further and to decide suiting values for
Sweden today. We also suggest that traffic analysts focus more on calibration of these
parameters in the beginning of the calibration process.
101
References
[1] Software manual from INRO. EMME/2 User’s Manual Software Release 9. Mon-
tréal, Canada, April 1999.
[2] Software manual from PTV Group. PTV Visum 13 manual. Karlsruhe, Germany,
2013.
[3] TSS-Transport Simulation Systems. Aimsun - traffic modelling without bound-
aries. Website, http://www.aimsun.com/wp/?page_id=21, May 2014.
[4] Caliper Corporation. Transmodeler traffic simulation software. Website, http:
//www.caliper.com/transmodeler/Simulation.htm, May 2014.
[5] Gunnar Johansson. Kollektivtrafikanternas resvägsval I Stockholms innerstad - en
jämförelse mellan verkligt och modellberäknat resvägsval. Report from Stockholms
Läns Landsting Trafiknämnden, Stockholm, 1984.
[6] Luis G. Willumsen Juan de D. Ortuzar. Modelling transport. Book published by
John Wiley & Sons Ltd, 4th edition, 2011.
[7] Christer Hydén m.fl. Bengt Holmberg. Trafik i samhället, grunder för planering
och utformning. Book published by Studentlitteratur Lund, 1996.
[8] Report from WSP Analys&Strategi. Trafikprognoser - en introduktion för den
nyfikne. 2007.
[9] National Cooperative Highway Research Program (NCHRP). Report 716: Travel
Demand Forecasting: Parameters and Techniques. Transportation Research Board
2012 Executive Committee, 2012.
[10] California department of Transportation. What is modeling? Website, http:
//www.dot.ca.gov/hq/tsip/otfa/microsim/whatis_modeling.html, Septem-
ber 2012.
[11] Gunnar Lind. Strategisk utvärdering av intelligenta transportsystem. Report from
Kungliga Tekniska Hgskolan, Trafik- och transportplanering, Stockholm, 1998.
[12] Carl Hägerwall Stein. Prognosverktyg för trafikflöden - en jämförelse av emme/2
och visum. Master’s thesis, Master thesis from Lunds Tekniska Högskola, Lunds
Universitet, 2007.
103
[13] Wahba M Parveen M, Shalaby A. G-emme/2: Automatic calibration tool of the
emme/2 transit assignment using genetic algorithms. J Transp Eng, 133:549–555,
2007.
[14] Fung WCS. Calibration and validation of transit network assignment models.
Master’s thesis, Master thesis from University of Hong Kong, 2005.
[15] Horowitz AJ. Subjective value of time in bus transit travel. Transportation,
10:149–164, 1981.
[16] Wardman M. Public transport values of time. Transp Policy, 11:363–377, 2004.
[17] Wardman M Abrantes PAL. Meta-analysis of uk values of travel time: An update.
Transp Res A, 45:1–17, 2011.
[18] Fonzone A Hemdan SMH Shimamoto H Bell MGH Kurauchi F, Schmcker J-D.
Estimating weights of times and transfers for hyperpath travelers. J Transp Res
Board, 2284:89–99, 2012.
[19] Clas Rydergren. Comparison of headway-based public transport models: Numer-
ical experiments for stockholm. Public Transp, 5:177–191, 2013.
[20] Isabelle Constantin Michael Florian. A note on logit choices in strategy transit
assignment. EURO J Transp Logist, 1:29–46, 2012.
[21] INRO. Emme/4 - a better way to model. Website, http://www.inro.ca/pdf/
emmebrochure.pdf, 2014.
[22] Heinz Spiess Michael Florian. Optimal strategies: A new assignment model for
transit networks. Report from Transpn. Res.-B journal, 23B:83–102, 1989.
[23] Molde universitet i Norge Odd Larsen. Kollektivtrafikassignment i visum og
emme/3. Report from AB Storstockholms Lokaltrafik, December 2010.
[24] Christian Nilsson. Emme/2 training, advanced transit assignment. Course mate-
rial, November 2002.
[25] Matti Pursula Hannu Kangas, Annu Korhonen. Back to the basics of emme/2
transit assignment. In Report from the 6th European EMME/2 users group con-
ference 1997, 1997.
[26] Software manual from INRO. EMME Prompt Manual, Version 4.1. Montréal,
Canada, April 2014.
[27] M. Florian M. Cepeda, R. Cominetti. A frequency-based assignment model for
congested transit networks with strict capacity constraints: characterization and
computation of equilibria. Transportation Research Part B: Methodological, 40
Issue 6:437–459, 2006.
[28] Stockholms Läns Landsting Trafikförvaltningen. Förstudie tunnelbana till nacka.
Underhandsrapport till statens förhandlingspersoner, 2013.
104
Appendix
Table 63: Difference in the number of boardings between base and future scenario in
Visum
Absolute boarding difference Relative boarding difference
Line between base and future scenario between base and future scenario
43r 4 0.00
43t 241 0.03
53r 826 0.15
53t 1 503 0.26
55r 323 0.15
55t 204 0.06
59ra 45 0.13
59rb 1 0.01
59ta 10 0.21
59tb 0 0.00
66r 7 0.02
66t 77 0.29
74r 1 206 0.30
74t 2 789 0.32
76r 221 0.08
401r 841 0.53
401t 27 0.02
402r 20 0.02
402t 162 0.14
403r 314 0.53
403t 3 0.01
409r 638 0.88
410r 0 0.00
410t 760 0.86
411t
411ra
411rb
413r 394 0.44
413t
414r
417r
417t
420t 122 0.22
422r 417 0.49
425t 163 0.11
430r 96 0.33
430ta 43 1.00
430tb
431t 419 0.05
434r 96 0.33
435r 107 0.33
442t
443r 88 1.00
443t 380 0.47
444r 255 0.94
445t
448t 32 0.34
449t 36 0.15
471r 96 0.18
471t 229 0.25
474r 464 0.30
474t
821r 87 0.29
821t 56 0.33
840r 141 0.47
840t 227 0.44
22r 678 0.52
22t 1 106 0.87
25r 20 0.06
25t 115 0.08
26r 37 0.06
26t 162 0.09
Blue r 302 1.00
Blue t 475 0.95
New metro line, Blue r
New metro line, Blue t
Red r 262 0.01
Red t 2 010 0.07
Green r 623 0.02
Green t 1 760 0.04
All lines 21 720 0.10
Table 64: Difference in the number of boardings between base and future scenario in
Emme
Absolute boarding difference Relative boarding difference
Line between base and future scenario between base and future scenario
43r 6 0.00
43t 230 0.02
53ra 882 0.18
53ta 1 495 0.22
55r 680 0.25
55ta 66 0.03
59ra 59 0.22
59rb 0 0.00
59ta 11 0.33
59tb 0 0.00
66r 14 0.10
66t 48 0.25
74ra 1 489 0.45
74ta 3 115 0.39
76ra 422 0.18
401ra 194 0.27
401ta 119 0.13
402rb 20 0.02
402tb 47 0.04
403ra 73 0.27
403ta 19 0.13
409r 612 0.82
410r 0 0.00
410t 646 0.78
411t
411ra
411rc
413ra 139 0.09
413ta
414rb
417r
417t
420tb 212 0.43
422ra 374 0.30
425Xt 45 0.03
430ra 30 0.58
430tF 0 0.00
430ta
431t 673 0.09
434ra 30 0.58
435r 35 0.64
442ta
443r 78 1.00
443t 159 1.00
444r 40 0.13
445ta
448t 57 0.61
449t 30 0.12
471r 45 0.04
471t 436 0.42
474r 176 0.65
474t
821r 133 0.61
821t 10 0.08
840r 39 0.25
840t 14 0.04
22ra 833 0.56
22ta 2 088 0.86
25ra 4 0.01
25ta 236 0.14
26ra 0 0.00
26ta 156 0.09
Blue r 175 0.34
Blue t 472 0.95
New metro line, Blue r
New metro line, Blue t
Red r 226 0.01
Red t 2 205 0.07
Green r 436 0.02
Green t 2 263 0.05
All lines 22 096 0.11