Professional Documents
Culture Documents
Hartl
October 2013
Universitt Stuttgart
Institut fr Straen- und Verkehrswesen
Lehrstuhl fr Verkehrsplanung und Verkehrsleittechnik
Abstract
Abstract
A task of traffic engineers is to investigate the impact of traffic demand on past,
present, and future transport networks while considering social, ecological, and
economic issues. The challenge in transport planning is to find the right balance
between all aspects. To solve this optimization problem, methods like transit
assignment models have been developed to support the traffic engineer to analyze the
current deficiencies and design better public transport networks. Depending on the
purpose of planning, the requirements differ among the transit assignment models
according to the type of public transport system modeled, supply and demand
representation, level of details, input and output values, reliability and effort. Each
model has strengths and weaknesses, and suggests specific assumptions about the
information provided to the travelers. Consequently, the results of the models vary.
The aim of this thesis is to compare route choice in the macroscopic schedule-based
assignment in VISUM and microscopic simulation-based assignment in BusMezzo, and
thus giving transport planners a good understanding of the model characteristics by
explaining the underlying modeling principles, comparing route choice behavior, and
evaluating the assignment results.
The model comparison done in this thesis brings light to
how the effect of overcrowded vehicles is represented in both models,
how passengers are distributed in the network due to capacity restrictions and
how different degrees of information are affect assignment results.
The challenge of comparing two transit assignment models, which are structured quite
differently, is to use an appropriate network. The network example needs to be as
simple as possible but still covering all relevant phenomena. Before comparing the
models, it is necessary to define the initial conditions: what is actually comparable and
which level of similarities are achievable? The main focus lies not on simplifying the
models as much as possible in order to reproduce the same results. Rather, it is more
important to ensure that the same starting condition is used to evaluate the output data
and point out the difference of the models. Transport supply of the schedule-based
assignment in VISUM is modeled deterministically based on the assumption that all
passengers have the knowledge of a reliable timetable. In BusMezzo, however,
transport supply is not deterministic, but based on a stochastic simulation-based
model. To make the models in some way comparable BusMezzo is forced to have a
deterministic transport supply. This is done by assuming that all agents have real-time
information of the entire network. This accommodates for the knowledge of the reliable
timetable used in the schedule-based transit assignment model in VISUM.
To analyze the differences of the two models, first the deterministic state with the same
degree of information is compared. Then, different sets of scenarios and subsets are
VuV 2013
Abstract
VuV 2013
Abstract
Furthermore, increasing the demand, decreasing the vehicle capacity or reducing the
frequency in order to enforce capacity constraints in BusMezzo significantly raises the
average total travel time because of denied boarding. The schedule-based transit
assignment model covers crowding effects by a using crowding function, but the effect
is marginal on the passengers travel time, because the schedule-based transit
assignment model optimizes the total impedance value to a stochastic equilibrium.
The comparison between the BusMezzo and the schedule-based transit assignment
model clearly shows the strengths and shortcomings of both model classes.
Furthermore, it contributes to the understanding of the basic fundamentals of each
model structure and highlights the right of existence. The appropriate application of the
models strongly depends on the scope of work. The decision which model fits best to
the assignment needs to be considered by the traffic planner and his/her experience.
The comparison also shows that more scientific work needs to be done to compromise
the shortcoming (e.g. crowding function and the limitation of congestion, overestimation
to high-frequented lines, and travel time calculations) to an adequate level of
acceptance to approximate the real behavior of travelers.
VuV 2013
Zusammenfassung
Zusammenfassung
Die Aufgabe eines Verkehrsingenieurs ist den Einfluss von vergangenen,
gegenwrtigen
und
zuknftigen
Ereignissen
auf
das
Verhalten
von
Verkehrsteilnehmern zu untersuchen. Die Herausforderung fr die zielorientierte
Lsungsfindung liegt in der Bercksichtigung von sozialen, konomischen und
kologischen Gesichtspunkten. Um die Lsungsfindung zu erleichtern, sind Methoden
wie Verkehrsumlegungsmodelle entwickelt worden. Abhngig vom Planungszweck
unterscheiden sich die Anforderungen an das Modell erheblich in Bezug auf den Typ,
die Darstellung von Angebot und Nachfrage, Detailierungsgrad, Eingangs- und
Ausgangsgren, Zuverlssigkeit und Performance. Jedes Modell besitzt Strken und
Schwchen und trifft unterschiedliche Annahmen ber die Informiertheit, die dem
Reisenden zu Verfgung gestellt wird. In Folge dessen unterscheiden sich die
Ergebnisse.
In der Masterarbeit wird die Routenwahl des mikroskopischen, simulationsbasierten
Verkehrsmodell BusMezzo mit der makroskopischen, fahrplanfeinen Umlegung,
implementiert in der Software VISUM, verglichen. Das Ziel der Arbeit ist dem
Verkehrsplaner ein Verstndnis darber zu geben wie sich die zwei Modelle im
direkten Vergleich Verhalten, wo ihre Schwchen und Strken liegen und wie die
Defizite kompensiert werden knnten.
Der Modelvergleich der Arbeit behandelt vordergrndig
wie der Effekt von berfllten Verkehrsmittel in beiden Modellen dargestellt wird,
wie Reisende im Netz unter Beachtung von Kapazittsbeschrnkungen verteilt
werden und
wie die Informiertheit von Reisenden die Umlegungsergebnisse beeinflussen.
Eine Herausforderung beim Vergleich von zwei unterschiedlich strukturierten Modellen
liegt in der Verwendung eines adquaten Beispielnetzes. Das Netz muss so einfach
wie mglich sein, dennoch aber alle untersuchungsrelevanten Phnomene abdecken.
Bevor die Modelle verglichen werden knnen, ist es notwendig die Ausgangssituation
zu definieren, was verglichen werden kann und welche Gemeinsamkeiten sich die
Modelle teilen. Der Anspruch liegt nicht darin, die Modelle soweit zu vereinfachen bis
sie die gleichen Ergebnisse reproduzieren. Vielmehr geht es darum, die gleichen
Bedingungen zu schaffen, um die Modelle vergleichen zu knnen.
Die Angebotsseite der fahrplanfeinen Umlegung ist deterministisch modelliert unter der
Annahme, dass alle Reisende mit zuverlssiger Fahrplaninformation ihre
Verbindungswahl treffen. Das steht im Gegensatz zum simulationsbasierten,
stochastischen Modell BusMezzo. Um die Modelle vergleichbar zu machen, mssen in
BusMezzo auf der Angebotsseite die zuflligen Einflsse eliminiert werden. Wenn sich
BusMezzo deterministisch verhlt, mit gleichzeitiger netzweiter Echtzeitinformation fr
VuV 2013
Zusammenfassung
alle Reisenden, fhrt das zu zuverlssiger Fahrplankenntnis und erfllt die Ansprche
an die Vergleichbarkeit.
Der Vergleich der Modelle beginnt auf der deterministischen Ebene mit der gleichen
Bereitstellung von Information. Anschlieend werden unterschiedliche Sets und
Szenarios entwickelt, um u.a. den Einfluss von Kapazittseinschrnkungen und dem
Einfluss von Informiertheit zu untersuchen.
Kapazittsengpsse in Form von berfllten Fahrzeugen werden in BusMezzo durch
eine absolute Beschrnkung der Platzanzahl erreicht. Im Gegensatz dazu steht der
Ansatz der fahrplanfeinen Umlegung. Es beschreibt das Gefhl des Unbehagens durch
berfllung mittels einer Kapazittsbeschrnkungsfunktion. Dabei verringert sich die
Attraktivitt der Verbindung je voller das Verkehrsmittel ist. Die Attraktivitt, auch
Widerstand genannt, hngt unter anderem vom Auslastungsgrad multipliziert mit der
Dauer des Fahrplanfahrtelements ab. Diese Art der Implementierung ist nicht fhig
Kapazittsbeschrnkungen mit dem richtigen Ma abzubilden, da Reisende
grundstzlich immer in der Lage sind einzusteigen. Das liegt in der Abbildung der
Nachfrage der fahrplanfeinen Umlegung. Deswegen ist es wichtig die Funktion der
Kapazittsbeschrnkung anzupassen, damit berfllte Verbindungen nicht weiter
genutzt werden. Diese Eigenschaft ist zum derzeitigen Entwicklungsstand nicht
adquat umgesetzt.
Ein wesentlicher Unterschied zwischen den Modellen besteht in der Verteilung der
Nachfrage auf das Netz. Die fahrplanfeine Umlegung sucht erst nach mglichen
Verbindungen und filtert anschlieend die Verbindungen heraus, die am
wahrscheinlichsten unter realen Bedingungen gewhlt werden. Der Menge an
gefilterten Verbindungen wird ein Widerstand mittels einer Widerstandsfunktion
zugeordnet. Ausgehend vom Widerstand der Verbindung wird die Nachfrage mittels
eines wahrscheinlichkeitstheoretischen Verteilungsmodell als Einzelentscheidung auf
das Netz modelliert. BusMezzo berechnet keine Verbindungen, sondern filtert alle
sinnvollen Wege als Input fr den situationsangepassten Entscheidungsprozess. Jeder
Reisende trifft entlang seines Weges eine Vielzahl von Entscheidungen. Die am Ende
entstandene Verbindung ist nicht das Ergebnis einer Verbindungswahl, sondern vieler
Einzelentscheidung, angepasst an die Situation im Netz. Auf den ersten Blick ist das
ein wesentlicher Vorteil gegenber der statischen fahrplanfeinen Umlegung in VISUM.
Es erlaubt u.a. die Bereitstellung von Echtzeitinformation und die dynamische
Fahrzeugkoordinierung. In Fllen bei denen eine Linie mit einem geringem Takt einer
Linie mit hoher Taktfolge gegenbergestellt wird, berschtzt BusMezzo die Anzahl der
Reisenden zu Gunsten der Linie mit hoher Taktfolge. Im Modell whlt ein Reisender
seine (Teil-)Verbindung indem er eine Zufallszahl zieht und diese in Beziehung zur
Wahrscheinlichkeit der Alternativen setzt. Da alle verfgbaren Verbindungen eine
Wahrscheinlichkeit besitzen, besteht immer die Chance, dass Reisende die weniger
attraktive Verbindung whlen. Wenn diese Entscheidungen oft hintereinander getroffen
werden muss, erhht sich die Anzahl derjenigen, die sich fr die unattraktive,
VuV 2013
Zusammenfassung
hochfrequentierte Linie entscheiden. Diese Schwche sollte mit einer sinnvollen an die
Situation angepassten Filterungsregel kompensiert werden.
Wird die Nachfrage erhht, die Fahrzeugkapazitt verringert oder die Frequenz der
Linie reduziert, entstehen Kapazittsengpsse. Diese Engpsse wirken sich in
BusMezzo direkt auf die Reisezeit aus, da Reisende nicht in der Lage sind in das
gewnschte Fahrzeug einzusteigen und somit auf das nchste Fahrzeug warten oder
ihre Verbindungswahl berdenken mssen. Die fahrplanfeine Umlegung deckt
berfllungseffekte mit einer Kapazittsbeschrnkungsfunktion ab. Die Auswirkungen
auf die Reisezeit von Passagieren sind jedoch marginal, da das Umlegungsverfahren
den Widerstand in Netz in einem iterativen Prozess zu einem stochastischen
Gleichgewicht optimiert.
Der Vergleich zwischen BusMezzo und der fahrplanfeinen Umlegung zeigt die Vor- und
Nachteile beider Modelle in Bezug auf das verwendete Beispielnetz. Zudem trgt der
Vergleich und die Erklrung der grundlegenden implementierten Theorien zum
Verstndnis des jeweiligen Modells und ihrer Daseinsberechtigung bei. Die geeignete
Anwendung des jeweiligen Modells hngt stark vom jeweiligen Einsatzzweck ab und
muss jeweils vom Verkehrsplaner auf Grund seiner Erfahrung entschieden werden.
Der Vergleich zeigt zudem, dass mehr wissenschaftliche Arbeit ntig ist, um die
Schwachstellen der Modelle (z.B. Kapazittsbeschrnkungsfunktion zur Einschrnkung
der berfllten Verbindungen, berschtzung von hochfrequentierten Linien und
Reisezeit Berechnung) auf ein adquates Ma zu reduzieren, um das Verhalten von
Verkehrsteilnehmern bestmglich abzubilden.
VuV 2013
Selbstndigkeitserklrung
Selbstndigkeitserklrung
Hiermit erklre ich, dass ich die vorliegende Masterarbeit eigenstndig verfasst habe
und keine anderen Hilfestellungen oder Quellen als die angegebenen in Anspruch
genommen habe.
Insbesondere habe ich keinen bezahlten Dienst mit der Anfertigung der gesamten
Arbeit oder Teilen der Arbeit beauftragt.
Maximilian Hartl
VuV 2013
Glossary
Glossary
ADC
APC
ATTT
AVL
CONNECTION
Destination
FB
Frequency-Based
GTC
HDWY
Headway
INTEGER
IVT
In-Vehicle Time
LINK
LOAD
MNL
NI
Network Indicator
NrTransfer
Number of Transfers.
NODE
Origin
OD-PAIR
PrT
Private Transport
PuT
Public Transport
SB
Schedule-Based
SCENARIO
SET
STOP
TAM
TRAVELER
TRIP
TRIP SEGMENT
VoT
Value-of-Time
WalkT
Walking Time
VuV 2013
Glossary
BM
BusMezzo
RTI
Real-Time Information
WaitT(BM)
PJT
SB-TAM
VISUM
WaitT(V)
VuV 2013
10
Contents
Contents
1
Introduction
13
1.1
Motivation
13
1.2
Research Goals
14
1.3
Outline of Work
14
Survey
16
2.1
16
2.2
General Information
19
2.3
Estimation Process
20
22
3.1
22
3.2
25
3.3
26
3.4
Level of Information
28
3.5
Capacity Constraints
29
3.6
30
33
4.1
Mezzo
33
4.2
BusMezzo
34
4.2.1
Object Framework
34
4.2.2
Simulation Flow
35
4.2.3
Implemented Models
37
4.2.4
40
4.2.5
Real-Time Information
49
51
5.1
Connection Search
51
5.2
Pre-Selection
52
5.3
53
5.4
Connection Choice
53
VuV 2013
11
Contents
5.5
6
Crowding Functions
54
Model Comparison
58
6.1
Classification
59
6.2
61
6.3
62
6.4
65
6.5
Example Network
65
6.6
70
6.7
74
6.8
85
6.8.1
89
6.9
91
6.10
95
Conclusion
98
References
101
List of Tables
104
10 List of Figures
105
Appendix
107
VuV 2013
12
Introduction
1 Introduction
1.1
Motivation
VuV 2013
13
Introduction
1.2
Research Goals
The objective of the thesis is to facilitate the planners comprehension of the model
characteristics by comparing, explaining, and evaluating the route choice of the
microscopic simulation-based transit assignment model BusMezzo (BM) and the
macroscopic schedule-based transit assignment model (SB-TAM) implemented in the
software framework of VISUM.
1.3
Outline of Work
This work starts by presenting a survey in chapter 2 Survey on the estimation of the
value-of-time (VoT) for a transfer between two transit lines, as well as the willingness of
passengers to accept longer travel times when traveling in less crowded vehicles
depending on the ratio of volume to capacity. The survey is used as introductory part to
specify some of the fundamental relations between the real world and the simplified
implementations in transit assignment models.
To evaluate the models BusMezzo and the schedule-based transit assignment model
in VISUM, first the fundamentals of modern traffic simulation are outlined in chapter 3
Traffic Modeling Fundamentals. Therefore, the different model types according to the
level of aggregation (Micro-, Meso, Macroscopic) and time relation (static vs. dynamic)
are classified followed by a short description of the modeling principles for private and
public assignment models. Furthermore, the impact of information and capacity
restrictions are described. Chapter 3 closes with the definition of the deterministic and
stochastic user equilibrium.
Chapter 4 Simulation-Based Transit Assignment Model explains the principal model
structure of BusMezzo. It concentrates on the subjects of simulation flow, implemented
models, and dynamic path choice models. The latter describes in detail the choice-set
generation process followed by the path choice decision process as well as the
evaluation of alternative paths and the actual path decision. The chapter closes with a
description of real-time information (RTI) in BusMezzo.
Chapter 5 presents the principle model structure of the Schedule-Based Transit
Assignment Model. It begins with the description of the connection search, followed by
the filtering process of all reasonable connections and the calculation of the
connections impedance. Furthermore, the connection choice and the distribution of
travelers are explained. The chapter closes with an analysis of the impact of the
additionally provided capacity restriction function.
Chapter 6 Model Comparison forms the core of the work. It describes the model
classification, travel behavior aspects considered in the model, the comparable level of
the two models, the information degree provided in each model and necessary
simplifications. Additionally, it describes the travel time correlation between VISUM and
VuV 2013
14
Introduction
VuV 2013
15
Survey
2 Survey
The originally idea of the survey was to estimate the coefficients of the utility function,
used in the schedule-based transit assignment model and BusMezzo, in the analysis in
chapter 6. Unfortunately, it was not possible, within the limited time of the thesis, to
analyze the survey before implementing the network and running the assignments.
Therefore, the survey is used as introductory part to specify some of the fundamental
relations between the real world and the simplified implementations in transit
assignment models.
Many implemented theories in transit assignment models are derived by observing the
natural behavior of travelers. The observation is transferred into a mathematical
approach to simulate and, especially, to forecast travelers behavior for planning
purposes. One of the major parameters, besides travel time, which influences the route
choice in public transport, is the transfer rate. Since transit assignment models are not
able to reflect all influencing parameters in a one-to-one correlation, parameters are
transferred into impedance. The impedance is mainly represented through the unit
time. This means that all influencing parameters with or without a correlation to time
are transferred to a value-of-time. This is also true for the number of transfers. Since
transferring has no direct relation to time, surveys try to estimate the value-of-time
which expresses to what amount travelers would accept to travel with a more time
consuming connection instead of transferring once. This kind of survey is called stated
preference survey (see (Hicks & Turner, 1999) for details). The method of stated
preference tries to derive the value-of-time by providing a choice of several discrete
options to the respondents. The aim of the survey is to define the coefficient of the
parameter transfer rate for public transport within a travel time shorter than one hour.
Another focus of the survey is to allocate the importance of overcrowded vehicles. It
follows the same principles but analyses the dependency of the value-of-time
depending on the ratio between volume and capacity.
This chapter first presents the structure of the survey. Secondly, the general
information (Gender, Age etc.) of the respondents is analyzed. Finally, the estimation
process is explained and the results are presented.
2.1
The survey is designed in the framework of the web-platform SurveyMonkey and was
conducted in German. The link to participate in this survey was open to public access
for about ten weeks and started in July 2013. The link was available on the homepage
of the Department for Transport Planning and Traffic Engineering of the Institute for
Road and Transport Science, University of Stuttgart and was also passed to the
authors personal mailing list. In total 243, people responded to the survey. About 90%
VuV 2013
16
Survey
of the participants answered all questions. The survey itself was structured into three
main parts, as seen in Table 1.
Table 1
Survey Structure
Response
Rate
a
b
98 %
92 %
87 %
It was interesting to observe that with the progress of the survey, the response rate
decreased. This is derivable by the motivation of the respondents to finish the survey
along the process of answering the monotonous questions. A complete list of the
translated queries and the corresponding answers are given in Appendix A.
Since it was unforeseeable who would actually participate in the survey, the choice
situations of the stated preference experiment are constructed in a way that everybody
is able to answer them without any additional knowledge. This is done with the best of
authors knowledge to avoid that people cancel the survey before completing all choice
situations but, even more important, that people understand and answer the question
correctly. To make it easy for the survey participants to grasp the context of the
decision situation, the survey is equipped with sketches, pictures and explanatory text
passages given in surveys screenshots in Figure 1. Most of the time, transferring is a
regular part of a connection (except direct connections) and therefore most people are
familiar with the personal correlated meaning of it. More difficult to capture is the
parameter congestion and what it means to travel in a crowded vehicle. Especially, the
abstract degree of volume to capacity ratio, explained in chapter 2.3 is hard to imagine.
Therefore, pictures are provided to illustrated different degrees of crowded respectively
overcrowded public transport systems
Since the number of questions in a survey is limited to a for the participant acceptable
number, the range of travel time is within one hour. It represents the regular travel time
for inner city OD-pairs. The travel time values of both connections are chosen such that
the statistical experimental design is most likely to captures all representative travel
times and correlations. To exclude the propagation of the same question order, the
questions in part b and c are given to each respondent randomly. The provided answer
to choose both connections is considered in the analysis as half an answer for each
connection. This is derived by the question type. To force the participants to give an
answer, most questions are carried out as a single select answer. But the possibility to
choose both connections as a third choice is also provided. The assumption: If a
participant would accept both connections but needs to decide which one he/she
chooses, the distribution is equally. That is the reason why the answer for both
connections can be spitted into half an answer for connections.
VuV 2013
17
Survey
Figure 1
VuV 2013
Screenshots Survey
18
Survey
2.2
General Information
Due to the fact that most of the respondents are from the authors family environment
or related to the environment of the University of Stuttgart, the responding group is
characterized as young educated people with an affinity to use public transport as a
standard transport mode, but with equally distributed income. Therefore, the amount of
respondents cannot be seen as a representative cross-section of society. The
distribution between female and male is almost equally represented. This is deducible
by analyzing Figure 2 and Figure 3. However, the respondent group is very familiar
with the properties of public transport, hence they are able to rate the queries about the
connection choices properly.
Figure 2
Figure 3
VuV 2013
19
Survey
2.3
Estimation Process
)]
Where:
Query
Impedance of connection for query
The amount of respondents chosen connection
VuV 2013
20
Survey
higher the congestion rate becomes. Therefore, the following impedance function is
used.
The coefficient
vehicle over all queries. This implementation assumes a linear relationship for the
volume to capacity ratio up to one hour travel time regardless of the in-vehicle time
(IVT). A more appropriate way to define congestion is to take into account the travel
time (PTV VISUM 12.5 Fundamentals, 2012). By estimating the value-of-time for the
same travel time correlation (blue dots) with different degrees of overcrowding, Figure
4 shows a linear relation (trend line). This is comprehensible by the subjective
perception. The longer the travel time and the higher the crowding level, the higher the
discomfort of the connection. Ergo, the connection becomes less attractive and
travelers shift to connections with more travel time and less travelers on board. The
linear correlation is in line with the presented results of (Pownall, Prior, & Segal, 2008)
at the 21st European Transport Conference 2008. The linear crowding function to
capture the effect of congestion is considered in the transport planning software VISUM
and will be discussed in chapter 5.
Figure 4
VuV 2013
Value-of-Time Congestion
21
3.1
Skim categories:
VuV 2013
22
Most transit assignment models are classified into three major steps as listed below. To
fulfill stable convergence conditions, some of the steps need an iterative procedure.
Search process:
Choice process:
To simulate the movements of travelers along their route, traffic flow models use
different levels of aggregation. They are classified into classes: microscopic,
mesoscopic and macroscopic, according to the level of detail and aggregation.
VuV 2013
23
Macroscopic Models
According to (Papageorgious, 1997) the macroscopic transit assignment model
describes the transition to the continuum theory. Probably the most famous
macroscopic transit model was developed by Lighthill-Whitham and has its origin in the
scientific research field of hydromechanics.
Microscopic Models
Another extreme, according to the level of detail, is the microscopic traffic model.
Vehicles are represented individually and the behavior of each vehicle depends on the
interaction with other vehicles. Additionally, vehicles subject to braking and
acceleration processes, as well as to the characteristics of the transport network (e.g.
light-signal system, right of way rules, lane assignment). Furthermore, the human factor
is considered by the models cognitive and reactive capability. Since some of the
components are subjected to stochastic processes the assignment needs to be
repeated until the results present an adequate mean situation of the network.
Mesoscopic Models
Mesoscopic models are a combination of macroscopic and microscopic modeling
approaches. That means that skim categories of the network are used but vehicles are
simulated individually, however, their second-by-second movement is not modeled.
Figure 5
The characteristics of private (PrT) and public transport (PuT) differ significantly.
Therefore, it is necessary to specify individual assignment models separately in order
to simulate the model characteristics properly. Note that the separation of the models is
VuV 2013
24
required, but the interaction (e.g. bus lines are usually on regular streets and flow with
the surrounding traffic) should not be neglected. The classification and explanation in
chapter 3.2 and 3.3 are taken from (Friedrich, 2012)
3.2
A private transport (PrT) model can be described in the major steps. Firstly, a route
search is performed which finds the choice-set of all alternative routes a traveler
considers on his way from his origin to his destination. Secondly, the route choice is
selected, in which the traveler chooses one of the alternative routes in the choice-set
according to the routes utility. And lastly, the traffic flow through the network, in which
vehicles are processed along their chosen routes and interact with each other.
VuV 2013
25
Capacity-Depending
Model
Link i
Macroscopic Flow
Model
Link i+1
Link i
Microscopic Flow
Model
Link i+1
i-1
i+1
Individual Vehicles
unified
Figure 6
In macroscopic models, the traffic flow is considered as continuous flow like a fluid
through a pipe. The velocity
is derived from the traffic density . The density is
defined as the number of vehicles within a path interval of the length
without
explicitly modeling lanes or vehicles. The correlation between velocity and density is
presented in the fundamental chart.
The most detailed representation of traffic flow is given in a microscopic traffic
assignment model. The complexity reaches from simple models like the cellular
automate (Nagel & Schreckenberg, 1992) to complex psycho-physical vehicle-following
models (Wiedemann, 1974).
3.3
One of the major differences between private and public transport is the time
dependency. Public transport participants are not able to decide freely when to depart
at their origin, because they depend on the schedule times of the transport system, e.g.
VuV 2013
26
bus or train. Therefore, the decision process is extended from a spatial route to a time
dependent connection.
VuV 2013
27
3.4
Level of Information
Depending on the specific model and the aggregation, different levels of timetable
information are provided to public transport travelers. Therefore, it is necessary to
distinguish between two types of information:
VuV 2013
28
3.5
Capacity Constraints
Traditional assignment models assume that travel time and costs are the main
attributes influencing travelers decisions. Empirical studies prove that passengers, in
reality, consider several qualitative aspects, which impair or improve the experience of
travelling (Tirachini, Hensher, & Rose, 2013). In the case of public transport, this
includes the number of travelers sharing one bus or train. The relevance of these
qualities becomes more important in developing and developed economies, since the
income of the population increases over time (Tirachini, Hensher, & Rose, 2013).
Consequently, public transport travelers are more likely to attach more value to the
service quality and comfort features (Tirachini, Hensher, & Rose, 2013). The disregard
VuV 2013
29
for capacity limitations is an unsatisfactorily simplification which does not reflect the
reality in highly loaded public transport systems (PTV VISUM 12.5 Fundamentals,
2012). Capacity limitations can affect travelers decision process in different ways:
3.6
The terms stochastic and deterministic will be widely used in chapter 6 to define the
comparable level of the two models, as well as for the determination of the input, output
and data characteristics. To obtain a fundamental comprehension of these
expressions, the following example will be used to clarify the underlying principles.
The example network shown in Figure 7 is simply structured. The demand of 1000
vehicles requests to travel between the origin and destination. The network provides
two alternatives; one with short travel time but less capacity and the other one with
longer travel time but more capacity. Depending on the equilibriums objective function
the demand will be distributed differently to route 1 and 2. Therefore, the deterministic
(DUE) and stochastic user equilibrium (SUE) as well as the system optimum (SO) will
be highlighted and described.
Equilibrium methods are widely spread in every-day planning to estimate the
distribution of traffic flows in transport networks. To optimize the objective function it is
necessary to distinguish between two deterministic approaches, also known as
Wardrops principles (Wardrop, 1952).
VuV 2013
30
routes hold higher or equal impedance and everybody chooses the best route. In the
example networks chart, the two blue solid lines present the impedance (travel time)
for route 1 and 2 according to the actual volume. The equilibrium state is reached if the
(
)
(
) is fulfilled.
condition
The approach is based on the principle of the individual trying to maximize the personal
utility. For practical use however, the assumption that travelers are provided with full
information is questionable, because not each traveler can be continuously served with
information or acts without personal preferences (Boden & Treiber, 2009). These
weaknesses are compensated in the stochastic user equilibrium with variables to
estimate spontaneity and individuality and lack of information.
VuV 2013
31
Demand
CR-Function according to BPR:
(
)
(
Dij = 1000
,
Route 1:
O
Route 2:
Figure 7
VuV 2013
Route 1
t0,1 = 8
Cap1 = 500
Route 2
t0,2 = 16
Cap2 = 800
32
4.1
Mezzo
Most of the existing transit models are time-based assignment models. The core of the
simulation is the progress from one to the next time step while each equally scaled time
step calculates the changes and updates the network status. In contrast, Mezzo is an
event-based traffic simulation tool which progresses from one to the next event. The
model specifies which changes are classified as events and orders them into an event
list. Events are called as they appear in the event stack (Oded & Tomer, 2008).
Vehicles are simulated individually, but lanes are not explicitly presented. The link
structure is divided into a running and a queuing part. The running part is not affected
by the downstream capacity limit and describes the earliest link exit time. The travel
time on the link depends on the ratio between loads and link capacity. The queuing part
imitates the delay process if capacity is exceeded and characterizes the process of
vehicles queuing in a single lane waiting to exit the link. Queue servers determine the
capacity limitation on turning movements. Turning movements are modeled
stochastically to regulate delays. Vehicles are randomly generated following a negative
exponential distribution by time-dependent OD-pair flow matrices according to a prespecified vehicle mix. The route choice follows a multinomial Logit model and might be
influenced by the information degree provided to the vehicle.
VuV 2013
33
4.2
BusMezzo
By extending the model Mezzo to simulate the interaction between private and public
transport, the modularized object-orientated framework helps to implement the dynamic
transit operation and assignment model BusMezzo.
4.2.1
Object Framework
Additional classes like bus types, bus vehicles, bus lines, bus routes, bus trips and bus
stops are implemented and define the characteristics of the objects shown in Figure 8.
The subclass vehicle type inherits its characteristics from the object bus type and on
the other hand the object bus vehicle is described by the bus type, bus route, and bus
trip. Each bus trip is assigned to a bus line and a bus route. The bus route is specified
with an ID and an ordered sequence of links. Bus lines initialize the subclass of the
object actions, which defines general procedures in the simulation, determine the
scheduled trips, and the list of stops. The deposited timetable is used as a reference
point for each bus trip to adjust the actual travel time to the timetable with the help of
holding strategies to absorb delays. Stops are allocated to links with the characteristic
assumptions about the spatial position, dwell time, and waiting time of individual
travelers represented by an agent. Each time a bus arrives, the dwell time function
calculates the time the bus needs to spend in the station until all agents are alighted
and have boarded the vehicle, and summarizes the waiting time of agents which are
forced to wait because of denied boarding.
Figure 8
VuV 2013
34
4.2.2
Simulation Flow
At the beginning of the simulation all objects are initialized. By initializing the objects,
some of them register an event and save them in the event stack. Most of the events
result in a new sequence of results. Aside from the introduced objects, it is necessary
to implement several new event types to properly represent the transit simulation
model BusMezzo. A general overview of the simulation flow is given in Figure 9.
When the simulation starts, BusMezzo reads the bus line list and generates individual
trips with the corresponding objects bus lines, bus routes and bus types, and registers
the events in the event stack. If the vehicle has not been introduced to the system yet
(first trip on its trip chain), it generates a bus vehicle object and assigns it to the
required bus type. After that, the vehicle enters the first link on the line route. Once a
bus enters a link as sequence of its trip, it checks whether there is a stop and if the bus
services it. If no stop is located on the link, BusMezzo calculates the link travel time
depending on the current traffic conditions. In this case, there is no difference between
vehicles in Mezzo and busses in BusMezzo, because both objects are running on the
same network and are treated as agents with different attributes (e.g. seat capacity,
length). If a stop is on the link, BusMezzo calculates the travel time to the stop, and
books an event for entering the stop. When the bus enters the stop the dwell time is
calculated and the model checks if the bus is subjected to any control strategies (e.g.
coordination of the departure time to a predefined timetable).
The implementation of holding control strategies in the main loop requires additional
steps to execute the control logic and to determine the appropriate action. For
example, if a bus enters a stop, and holding strategies are activated, the control
strategy checks for how long the bus needs to be held in the station to minimize
accumulated delays according to the timetable and the actual temporal position of the
bus.
The outputs of the queries determine if the process books an event for the stop exit
time. Exiting a stop is similar to entering a link, the model checks if there are any
further stops downstream on the link, calculates the travel time for the link section
based on the traffic conditions and the loop starts over. By reaching the end of its route
BusMezzo checks if any additional trips are assigned to the vehicle. If yes, then the trip
process is activated and progressed through the system (trip chaining). If this is the last
trip of the line, the vehicle terminates.
On the output level, the simulation lists the collected data on stop level for each
individual traveler or bus. The main outputs of BusMezzo are line ID, trip ID, vehicle ID,
stop ID, traveler ID, early and late arrivals, dwell times, boarding and alighting
passengers, occupancy, denied boardings, selected paths, and travel times between
stops. On a larger scale of aggregation, e.g. at trip level, line or OD-stop level, it
presents a summarized list of the chosen paths or line loads.
VuV 2013
35
Figure 9
VuV 2013
36
Number of Replications
BusMezzo, according to its definition, is a stochastic simulation-based transit
assignment model. Therefore, it is necessary to run several simulations in order to find
a meaningful average and to evaluate each execution. Each run of a simulation is a
single shot of the current situation, also called within-day learning. That means that
there is no interdisciplinary exchange between the simulations. In fact, no learning
process takes place as it is performed in a day-to-day learning process. At the current
state, this feature is under development by (Gkioulou, 2013). This would also enable
access to simulate travelers behavior by shifting from overcrowded to less crowded
vehicles depending on travelers practical experience.
To receive statistically verified results, several simulations are needed. To quantify the
number of simulations, the following formula can be used (Dowling, Skabardonis, &
Vassili, 2004).
( )
( )
( )
Where:
( )
( )
( )
initial simulations
initial simulations
4.2.3
Implemented Models
BusMezzo requires a detailed representation of its basic attributes to describe the main
elements such as travelers arrival and alighting process, dwell time, travel time, and
trip chaining.
VuV 2013
37
Dwell Time
The implemented travel time calculation consists of two parts. One is the riding time
between stops, which depends on the vehicle density and the dwell time. The dwell
time describes the process at the stop from the start of opening the doors, travelers
boarding and alighting until the transit vehicle closes the door and leaves the stop to
enter the link again. The dwell time function implemented in BusMezzo is based on the
Transit Capacity and Quality of Service Manual (Kittelson & Associates, KFH Group,
Parsons Brinkkerhoff Quade & Douglass, & Hunter-Zawarski, 2003). This approach
distinguishes between boarding and alighting separately for each door. The door with
the longest service time is crucial for the dwell time. If the bus stop is placed in-lane or
uses a bus bay, the delay time is captured by the used function because more time is
needed to re-join the traffic flow. The dwell time function is given by:
(
VuV 2013
38
Where:
is the dwell time for line at stop on trip
is the required service time for the front respectively rear door. It
depends on the total number of travelers boarding and alighting
and the crowding level on the bus
indicates if the bus top is in-lane or a bus bay
describes the physical space at the bus stop (e.g. 20 meters)
are parameters to specify the dwell time function
describes the error term for unpredictable events
To support the implemented holding strategies, the departure time is given by the
following formula for line at stop on trip :
(
Where:
Departure Time
Actual Arrival Time
Dwell Time
Departure Time as results of the holding strategy
Travel Time
The driving time between stops is the major part of a transit trip. Levinson (Levinson,
1983) estimated in his research that 9 to 26 percent of the total travel time is
contributed to dwell time and 12 to 26 percent of the time is spend in traffic delays. The
variables affecting the riding time in urban busses are subjected to deceleration and
acceleration processes due to the high density of stops every few hundred meters.
Depending on the independency of the public transport system (bus on links floating
with PrT vs. trains on independent rail tracks) the travel time reliability increases while
the service variability decreases. According to several researchers, the travel time of
busses and the arrival process tends to follow a right skewed distribution (Strathman,
et al., 1999), (Dessouky M. , Hall, Nowroozi, & Maurikas, 1999). BusMezzo keeps the
flexibility according to the estimated distribution and provides several functions like
normal, lognormal, Gumbel and gamma distribution to describe the travel time
variability with its characteristic runs.
Trip chaining
Besides the published timetables which show the service frequency to public transport
users, an additional schedule exists. It is known as a driving roster and is used by the
VuV 2013
39
operating company to manage the fleet and driver coordination. Each transit vehicle
and driver needs to fulfill a certain workload within a working period. From the time a
transit vehicle leaves the depot until it returns for servicing, it makes several trips.
Therefore, it is required to simulate trip chains. Mezzo usually generates and eliminates
vehicles between OD-pairs. But transit vehicles have different characteristics than
private vehicles. That is why busses should only be eliminated when they reach the
destination of their last trip. If busses would not be affected by unreliability and were
always on time, trip chains would be unnecessary. Since traffic is subjected to
stochastic processes, the feature of trip chains is needed to simulate the propagation of
delays and layover times adequately. Layover times have the goal to buffer delay to
avoid delay propagations along the route. Layover times may be spread along the line,
at the end of the line or a combination of the two (TCRP, 2003). Alternatively, these
recovery times are necessary for servicing (e.g. refuel) and breaks for drivers. Thus,
there is always some recovery time needed. The actual departure time of a trip from
the origin bus stop (dispatching time) is calculated as the maximum between the
schedule departure time and the arrival time of the bus from the previous trip at the
origin stop of the following trip plus a minimum recovery time plus a lognormal
distributed error term describing the stochastic departure delays. The actual
dispatching time is given by:
(
Where:
Actual Departure Time for trip
by bus
by bus
from the previous trip
4.2.4
The transit path choice model approach in BusMezzo is a two stage choice process
shown in Figure 10. The first stage is determined by the base of the deterministic
network configuration (timetable and walking distances), the static path set for each
given OD-pair and uses it as input for the dynamic path choice model. Each generated
agent takes successive decisions along its path which is triggered by events. Each
alternative is evaluated by agents preferences and expectations. The expectations
depend on prior knowledge and the accessibility to real-time information. Agents
choice execution relies on capacity restriction.
VuV 2013
40
The separation allows determining a general set of paths as a pre-assignment step and
hence this step needs to be executed only once. It optimizes the performance in terms
of computing time, but it is no essential requirement for the principle work flow of the
model. Therefore, this step could be implemented either statically or dynamically.
Figure 10
4.2.4.1
Choice-Set-Generation
The spatial choice-set generation is the basis for the dynamic path choice model. In
terms of route choice, the path choice is not trivial and aims to find all reasonable paths
for given OD-pairs. Because the path set is used as input for the path choice, the
model needs to also find paths for OD-pairs with no demand since they might become
attractive alternatives under some circumstances during the dynamic assignment. The
elimination process of paths is an optimization problem between dismissing irrelevant
paths and keeping the majority of the travelers used paths. By referring to the
elimination criterion of the schedule-based assignment model in chapter 5, both models
face the same problem in this step of the assignment but the number of paths in
VISUM will be the same or even more likely smaller compared to BusMezzo. The
filtering rules in BusMezzo are looser than in VISUM, because agents are facing the
decision of alternative connections dynamically. Note that the path set in VISUM is a
subset of BusMezzos path set. The principles in this assignment step are the same for
both models. Figure 11 shows a general overview of the choice-set generation
process. The dashed line in the figure marks the steps until the models assume the
same approaches. After this, different settings regarding the strictness of the filtering
process can be applied.
VuV 2013
41
Figure 11
Path Generator
First, the path generator calculates all direct paths, which do not involve a transfer stop.
Indirect paths are constructed by a recursive search method for given OD-pairs (Tan,
Tong, Wong, & Xu, 2007) illustrated in Figure 12. Therefore, a suitable data structure is
constructed to examine paths between the origin stop and all other stops. This is done
by increasing the number of transfers and considering all stops which are reachable
within the next connection link to the destination stop. At each stage, an additional stop
is used as an intermediate origin to restart the search loop. This method expands a
search tree to the depth of stops. The width of the tree depends on the reachable
stops within one stop level.
VuV 2013
42
Figure 12
Path Generator according to (Cats O. , 2011) and (Tan, Tong, Wong, &
Xu, 2007)
To limit the depth and width of the search tree and to filter out unreasonable paths, the
method is subjected to constraints (similar to the branch & bound method mentioned in
chapter 5). The maximum number of transfers is a pre-defined criterion to limit the
depth of the search tree. Additionally, paths are subjected to two logical constraints:
No loop:
Filtering
The applied recursive search method may generate a large variety of paths but
previous studies suggest that the path set of the choice-set generation model should
be reduced by non-compensatory rules (Recker & Golob, 1979), (Cantillo & Ortuzar,
2005) before passing the path set to the path choice phase. The aim of the filtering
rules is to emulate behavioral constraints and to assess each path independently from
alternative paths in the path set. The rules are given by:
VuV 2013
43
No opposing lines:
Dominancy rules
Merging
Checking the provided paths individually, they fulfill all the filtering criteria but might
have the same spatial distribution to a certain percentage. Therefore, these paths can
be merged together in terms of a hyperpath (Spiess & Florian, 1988). It helps to avoid
the potential of overlapping paths which might cause bias in the choice model.
Alternatives which share the same sequence of lines but differ in their intermediate
stops can be merged as long as the route between the intermediate stop is the same.
The same rule is used to merge paths which share the same sequence of stops and
vary the lines. The merging process leads to the master set.
Dynamic Filtering
The dynamic filtering process uses the master set each time a traveler takes a decision
to examine the filtering rules to exclude paths which are unreasonable under timedepending conditions. Considered are the following criteria:
Availability:
Worthwhile to wait:
VuV 2013
44
between different user groups. The path set is the basis for the dynamic path decision
process which will be explained in the following.
4.2.4.2
Path Choice
The dynamic path decision process includes the connection, boarding, and alighting
decision.
Connection Decision
The first decision a traveler faces is the connection decision. He/she chooses at which
stop he/she might start his trip by evaluating all the path alternatives from the
selectable stops to the destination. Usually it is a non-transit (e.g. walking) connection
to the next stop. When travelers alight at a stop, a connections decision also takes
place. The connection decision is limited to the accessibility of stops.
Boarding Decision
Once a traveler decides to wait at a stop, each arriving trip triggers a binary boarding
decision. The traveler applies the decision rules to balance between staying at the stop
or boarding the approaching vehicle. The actual boarding process might be subject to
capacity constraints. The boarding rule becomes important in the analysis in chapter 6.
Alighting Decision
Once the traveler is on board, he/she faces an alighting decision each time the event
trigger is pulled. Therefore, it is necessary to distinguish between two extreme
modeling approaches. Essential for the process is the exact timing of the alighting
decision. Either the decision is based on the set of all potential alighting stops as soon
as the traveler boards the vehicle or the decision event is triggered each time the
vehicle approaches a stop. Before taking a decision, the alighting process needs to
identify all relevant stops of the path set. However, it is assumed that travelers take an
intermediate approach. Hence travelers immediately take a non-adaptive decision upon
boarding. Once on board, they may reconsider the decision during the ride due to the
changing conditions.
4.2.4.3
Independent of the actual decision (connection, boarding, alighting), the traveler needs
to evaluate the provided alternatives. The path evaluation spans an evaluation tree with
different levels of aggregated choice alternatives. Each choice can be presented as a
set of actions with different path alternatives. The presented sketch in Figure 13 is an
example of one possible decision tree. Actually, the decision tree varies between stop
decision (connection and alighting decision) and a binary decision between stay and
move (boarding and repetitive alighting decision).
VuV 2013
45
Figure 13
for action
Where
is a scaling factor.
4.2.4.4
All the presented sub rules of the path choice model lead to the process of making a
path decision. In terms of the dynamic path choice model, the output of a travelers
path choice is an adaptive path choice depending on the travelers progress in the
network. Because the decision progress is a sequence of single decisions, the
implemented dynamic path choice model is in contrasting to static assignment models.
Travelers in static assignment models consider a path choice as a single decision for
the whole path. Travelers in BusMezzo do not choose between paths, instead they
choose between alternative actions which correspond to certain path stages. The
adaptive decision process is shown in Figure 14.
VuV 2013
46
The path decision choice starts with the random generation process of travelers
according to the given OD demand matrix with the expected arrival rates. Each traveler
updates his/her perceptions with the accessible information or prior knowledge. Each
traveler first takes the connection decision and thus decides where to initialize his/her
trip or to stay at the current location in case their origin is a transit stop. While travelers
wait for the arriving vehicle their expectations might be updated with provided
information at the stop (RTI or elapsed waiting time) which deviates from their
expectations and leads to a reconsideration of the connection search. Each time a
vehicle approaches, travelers are confronted with a binary boarding decision, whether
the traveler boards or stays at the current stop to wait for another vehicle. Once on
board the traveler updates his/her perceptions on downstream alternatives by the
provided information. While on board, the traveler may receive new information which
deviates from his/her expectations and leads to reconsidering the alighting decision.
Each time the vehicle approaches a stop, the model checks which of the travelers
needs to alight at the current stop. A new connection decision is performed, once the
traveler alights. While exiting the vehicle, the traveler decides between the alternatives
of waiting for another line, walking to another transit stop or walking to his/her
destination. If the traveler has reached his/her destination, the trip is terminated.
Otherwise, the current stop is considered as a new origin stop and the path choice
decision process is recalled until the trip chain is completed.
VuV 2013
47
Figure 14
VuV 2013
48
4.2.5
Real-Time Information
The large enrichment for public transport system users (e.g. time savings) as well as
for transport operators (e.g. automated vehicle location (AVL) Data for fleet
management) by collecting and providing real-time information has been studied
extensively (Lappin & Bottom, 2001). Therefore, the public transport systems are wellequipped with communication and information technologies. The collected data was
first used by transit operators to structure the fleet management and react faster and
more precisely in case of service disruption or fleet coordination. In a later step,
information is provided to transit users to increase the service level and to minimize the
variability of unreliability. On both sides of a public transport system, operators and
passengers profit by real-time information. There is no limitation on the scope of realtime information, e.g. to travel times or expected arrival times. The information degree
might extend to provide crowding conditions, expected and unexpected service
disruptions or for planning purposes with the aim to rise the service quality. To
dynamically imitate the impacts of real-time information using BusMezzo to predict
precise information for further downstream conditions, the presented framework in
Figure 15 is implemented into the model. Inputs are represented in parallelograms,
ovals show the data processing, rectangles express models, and outputs are described
by rounded rectangles. The core of the real-time information process is within the
dashed frame. The three modules within the frame represent the transit operation on
the supply side and traffic dynamics (in blue), the generation process for real-time
information (in purple) and on the demand side traveler decision (in orange).
Figure 15
VuV 2013
49
VuV 2013
50
5.1
Connection Search
For each origin zone a search tree of suitable connections to all stops in the network is
generated. This ensures that not only the best, but all adequately good connections are
found for the origin zone. The resulting path tree allows a highly differentiated
distribution of the traffic demand. To evaluate the quality of a connection, a search
impedance is used. The search impedance is calculated to:
VuV 2013
51
The parameters are journey time (JT), number of transfers (NrTransfer), transport
system specific impedance (TSys_Imp) and vehicle journey specific impedance. For
the analysis in chapter 6 only the journey time and number of transfers are considered.
Dominance
A pair-by-pair comparison is important to identify redundant connections. If a
connection is with no regard better than any another connection within the same time
series, the connection is dominated by others and removed from the choice-set. A
connection c is dominated by connection c if
c is within the same time series of c
NrTransfer (c) NrTransfer (C)
SearchImp(c) SearchImp(c)
There must be a real inequality in at least one of the previous criteria
Bounding
Independent of the temporal distribution of a connection the following rules are
implemented to exclude connections which differ too much in one or several criteria
from the optimum:
SearchImp(c) > minimum SearchImp(C) factor + constant
Journey time(C) > minimum journey time(C) factor + constant
NrTransfer(C) > minimum NrTransfer(C) + constant.
The connection search ensures that all logical connections are found according to the
settings.
5.2
Pre-Selection
The pre-selection step values and compares all found connections. Only reasonable
connections are provided to passengers connection choice. To identify unreasonable
connections, the following exclusion criteria are consecutively executed:
SearchImp(c) > minimum SearchImp(c) factor + constant
(or no limitations; just branch & bound)
Journey time(C) > minimum Journey time(C) factor + constant
(unless the connection is optimal with respect to the number of transfers)
NrTransfer(C) > minimum NrTransfer(C) + constant
(unless the connection is optimal with respect to the journey time)
VuV 2013
52
5.3
5.4
Connection Choice
The connection choice uses the impedance of a connection within a time slice in
respect to the impedance of all other connections to calculate the distribution of traffic
demand on the given choice-set. The percentage of the demand, which is distributed
to the connections, is calculated with the help of a choice model, according to the
following formula:
(
)
(
VISUM provides several choice models (e.g. Logit, Kirchhoff, BoxCox, Lohse (PTV
VISUM 12.5 Fundamentals, 2012)). In the case of the model analyzes in chapter 6, the
Logit model is selected.
VuV 2013
53
Logit Model
In the Logit model, the percentage of the demand distribution to a connection is
calculated as follows:
The Logit model evaluates the differences of the connection impedance. There is no
difference for the model if the travel time of two routes is 10 and 15 or 100 and 105.
The distribution on both alternatives would be the same. The sensitivity of passengers
towards increased impedance is described by the parameter and is set to
. This
ensures a relatively strict reaction to small changes in impedance and assumes the
same condition than BusMezzo.
5.5
Crowding Functions
The default settings in the schedule-based transit assignment model do not consider
the fact of overcrowded connections. That means that demand is distributed
independently from the impedance of the connection and the volume to capacity ratio.
However, if the ratio of volume to capacity is considered in the impedance function,
connections which seem unattractive considering their perceived journey time might
become attractive connections. Thereby, the set of possible connections might change
and grow. The capacity restriction in the schedule-based assignment aims to simulate
discomforts for passengers due to crowded or overcrowded vehicles. The approach
approximates the fact that passengers exceeding the capacity need to use different
connections. However, there is no strict limitation of capacity if the vehicle capacity is
saturated. The general sequence of the procedure is shown in Figure 16.
Independent from the volumes distributed to each connection, the default connection
process and the pre-selection are carried out in the first step of the procedure.
Followed by an optional connection search which calculates the connection search a
second time. In order to do this a pre-assignment distributes loads to the pre-selection
to ensure that connections, which could be dominated by other connections from the
first connection search, are in the choice-set. Optionally, the first and the second
selections are merged. This forms the choice-set for the transit assignment process.
The connection impedance is calculated independent from the parameter volume to
capacity ratio. Step six initializes the actual iteration process. To describe the
impedance because of crowding, VISUM uses the expression Vol/Cap ratiodependent impedance. It is added to the connection impedance from the previous
step. The Vol/Cap ratio-dependent impedance changes through the iteration process
depending on the volume to capacity ratio. A smoothing process (e.g. MSA (PTV
VISUM 12.5 Fundamentals, 2012)) prevents oscillations. It increases the probability
VuV 2013
54
that the iteration process converges. If the termination conditions are verified, the
iteration procedure terminates.
Start
Default connection search and pre-selection indenpendent from volumes
Optionally, the connection search is calculated a second time to take the
calculated volumes into account. This ensures that connections which could be
dominated by other connections are in the choice-set.
Optionally, merging both selection searches
Connection impedance is calculated independently from the volume to
capacity ratio
Calcualtion of the impedance based on the volume/capacity ratio and total
impedance of connection.
Smoothing process to prevent oscillations
Measurements to describe the distance from the balanced state.
no
If the termination
conditions are verified the
procedure
will be cancelled.
yes
End
Figure 16
The linear and the SSB function are presented in the following as a pre-explainable
step for the analysis in chapter 6. The German Railway (DB) functions is not used
during the investigations and not further discussed in this thesis.
VuV 2013
55
Linear
The linear crowding function defines the impedance for a vehicle journey item as
follows:
(
and
{
Parameter
and
are user-specific inputs. The parameter
calculated from the derivation at the position and . The conditions are:
(
and
are
)
(
)
)
VuV 2013
(
(
)
)
56
Figure 17
VuV 2013
57
Model Comparison
6 Model Comparison
To obtain a better practical knowledge about the effectiveness of the simulation-based
and schedule-based transit assignment model approach, a designed example network
will be presented and used for the comparison. The network needs to be as simple as
possible but still covering all relevant phenomena. On the one hand, the example
needs to consist of the characteristics of each single model. On the other hand, a
certain degree of complexity is necessary to reflect the specific assumptions of each
model.
The presented network is loosely based on the research network of (Spiess & Florian,
1988). The author is not aware of a more appropriate network to fulfill the needs for the
comparison. Therefore, the network will be used to elaborate the details of the different
model approaches. The example network will be elaborately described and presented
in Chapter 6.5.
To use the presented theoretical background information in Chapters 4 and 5, a
moderate software framework is needed. A representative for the macroscopic transit
assignment model PTV VISUM 12.5 is used to cover the schedule-based transit
assignment model. The microscopic part is implemented in the simulation-based
software BusMezzo. Both models are carried out in the object orientated programming
language C++. In the following, the software names are representatively used for the
assignment model. To make this point clear so that it does not cause any
misunderstandings, the allocation is considered bellow. Additionally, the term traveler is
used as general expression for somebody who travels in the network. If the
characteristic of a traveler is related to VISUM, the expression passenger, and
respectively to BusMezzo the term agent is used.
(
(
)
)
VuV 2013
58
Model Comparison
6.1
Classification
Another way of presenting the introduced models from the previous chapters is to take
a look at the scale of the supply and demand side. Table 2 illustrated the classification.
Table 2
Supply/Demand
Micro
Macro
Micro
BusMezzo
Macro
SB-TAM
FB-TAM
VuV 2013
59
Model Comparison
Table 3
Modeled Factors
SB-TAM
BusMezzo
Planning Interval
Type of Equilibrium (Results)
Choice model
Deterministic
Stochastic
Joint car and PT assignment
Traffic dynamics representation
Accordion effects - Delays
Service Disruptions
Supply
Holding Strategy to compromise delays
reliability
capacity restriction (Queuing Process)
Dwell time
Vehicle types
Vehicle scheduling/Fleet assignment
Generation Process
Demand
representation OD-level with connectors (walking links)
Random taste variation
Demand
Within-day learning (adaptive path choice)
modeling
Day-to-day learning
Vehicles
Level of
representations Travelers
Cost
In-vehicle time
Access time
Egress time
Walk Time
Access stop wait time
Path evaluation Transfer wait time
components Number of transfers
Number of operator changes
Departure time choice (hidden waiting
time)
Mode specific constant
Stop and line preferences
Discomfort due to Crowding
Frequency knowledge
Timetable knowledge
Degree of
Information
Real-Time Information
Group dependent information degree
L
Sto
Det
x
S-M
Sto
Sto
(x)
x
x
Meso
x
x
x
x
x
x
x
x
Sto
x
x
x
*
I
I
Macro
x
Det
x
x
I
F
x
x
x
x
x
(x)
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
Sto Stochastic
Det Deterministic
I Individually
F - as Flows
* - in development (BM)
Meso Mesoscopic
Macro Macroscopic
S,M,L - Short, Middle, Long
VuV 2013
60
Model Comparison
6.2
Supply/Demand
Stochastic
Deterministic
Stochastic
BusMezzo
SB-TAM
Deterministic
(DUE)
To make the models comparable, it is necessary to have the same starting conditions.
Because it is not possible to manipulate VISUM to behave stochastically on the supply
side, the only possible adjustment is to force BusMezzo to behave deterministically.
This is done by eliminating all the stochastic processes e.g. the vehicle travel time,
boarding and alighting processes, and the dwell time at stops. At the bottom line, the
reliability is set to 100%.
Another major difference is the information degree provided to travelers. BusMezzo
features four different levels of information which are elaborately described in
Chapter 4 and shows some similarities to the information level of the headway-based
assignment model condensed in Chapter 3. In contrast, the SB-TAM considers only full
information. The expression full information might be not the proper description in this
context. Full information includes also knowledge about monetary costs. In this
comparison the effect of costs is excluded, thus the expression full information, is set
equal to the usage of timetable knowledge. Timetable knowledge aligns more
adequately to the information levels provided in BusMezzo. By observing the
comparable level, the information degree in BusMezzo needs to be set to network realtime information, as is shown in Table 5.
VuV 2013
61
Model Comparison
Table 5
Deterministic
BusMezzo
VISUM
No RTI
Platform/
Stop RTI
Station/
Cluster RTI
Network RTI
Timetable Knowledge
By forcing BusMezzo to be deterministic on the supply side and using network realtime information for all agents equals in reliable timetable knowledge as is defined in
VISUM for passengers. These definitions form the starting conditions for the
comparison level between the models. Table 6 gives an aggregated overview of the
comparable level of information and supply characteristics. The classification will
change among the investigations.
Table 6
Supply
Modell
BusMezzo
Deterministic
VISUM
6.3
Information
Network RTI
Timetable Knowledge
Since several of the implemented theories in BusMezzo and VISUM are different, some
simplifications are needed to get the same or at least the closest possible starting
conditions to make the models comparable and examinable. One of the differences is
the parameter variety in the utility function. VISUM provides a large range to specify the
utility function as is presented in Chapter 4. BusMezzo provides the parameter IVT,
WalkT, NrTransfer and WaitT(BM) in the utility function. In fact, the number of
parameters varies and needs to be reduced to the same quantity. Hence the amount
will be reduced to the main unit used in BusMezzo. All other provided parameters in
VISUM are set to zero and have no influence on the assignment results. The definition
of the selected parameters is the same in both models except for the waiting time. The
parameter waiting time is too fundamentally important for the assignment results
impact that it cannot be left out. For this reason a means needs to be found to
compromise the parameter differences.
VuV 2013
62
Model Comparison
area represents the time before an agent has been generated relating to BusMezzo
and, in terms of VISUM, the time slice before passengers experience their desired
departure time.
Figure 18
VISUM defines demand as a total number of passengers who are in the network from
the beginning of the assignment time interval. The time component is considered by
the definition of time series and the weight for each time slice. The more the weight of
one time series, the greater is the percentage of passengers amount from the total
demand. Within one time slice, also called desired departure time, there is no
difference in impedance. This means in consequence that passengers are able to
choose trips which departure earlier as their desired departure time. Of course this
comes along with an increase of impedance for connections departing out of the time
slice and they might become less attractive. This is not the case in BusMezzo, because
each individual agent is generated and before an agent has not been generated into
the system, he/she cannot board any trip. From the moment a specific agent is
introduced to the system and decides to wait at the originating stop to board the
arriving vehicle, the waiting time starts to count and might influence his/her decision
during the decision process through his/her journey. Consequently, to merge the
parameters, the coefficient early temporal utility of a connection in VISUM needs to be
set to a sufficiently high value for the impedance to be high enough so that passengers
do not take a trip which departures before the desired departure time. This involves the
fact that agents are not able to board before being generated. The coefficient late
temporal utility of a connection is apparently equal to the waiting time at the origin stop
in BusMezzo within the time series tolerance. The time series tolerance specifies the
different amount of agents at the introduction point compared with the consistent
number of passengers. As already mentioned, BusMezzo introduces agents randomly
and uniformly distributed within the generation time. Because the generation follows a
VuV 2013
63
Model Comparison
distribution, it cannot be guaranteed that during a specific time slice, e.g. one minute,
the exact same number is introduced compared to adjacent time slices of the
generation time process. Overall, the number of agents introduced to the system
balances to the deterministic demand level in VISUM, thus some differences in waiting
time might occur. The smaller the time series, the smaller the impact of the time slice
tolerance. Therefore, the length of the time series has an essential influence on the
calculation of the impedance. Hence it needs to be small enough to fit the assumptions
of the agent generation process in BusMezzo. By comparing the network results in a
pre-working step it has been shown that setting the time series to a length of one
minute is adequate. Smaller time series have marginal effects on the assignment
results, but the calculation time increases disproportionately.
This simplification does not represent the exact same phenomenon, but it is a
moderate way to implement it. With the parameter definition and the coefficient setting,
the starting conditions are as similar as they can possibly get and the results are
comparable. The summarized utility function with its characteristic parameters and
coefficients for BusMezzo and VISUM might look something like those presented
below. The parameters and their coefficients do not change during the analyses. For
the sake of completeness, the utility function used in VISUM also lists the additional
capacity constraints term. BusMezzo does not provide such a term, because the
capacity constraint effects are covered by the boarding or better the denied boarding
process.
(
( )
( )
(
)
(
( )
Where:
[min]
999
( )
Of course, there is motivation to discuss the values of the coefficients. They are loosely
based on the common accepted values named in the literature e.g. (Wardman, 2001),
(ITP, VWI, 2006) and the presented survey in chapter 2. For analyzing the two model
approaches it is sufficient.
Because BusMezzo uses the multinomial Logit model to calculate the probability of an
alternative, the default choice model in VISUM is adjusted to it. A full overview of the
used setting in VISUM is listed in appendix B.
VuV 2013
64
Model Comparison
Passengers in VISUM have full timetable knowledge. Therefore, the access to the
origin stop is always coordinated to the first departing trip, which means that the origin
wait time is constantly zero. VISUM provides the option to approximate the adapted
origin wait time but it is a constant supplement value calculated with the input of the
service frequency. The origin waiting time has the same value for all connections of an
OD-Pair (PTV VISUM 12.5 Fundamentals, 2012). According to the network settings,
the service frequency varies respectively among OD-Pairs. Therefore, it is more
suitable to use the temporal utility of a connection which is defined above to approach
similar starting conditions. The alternative to consider the origin wait time is not as
similar to BusMezzos behavior, but it would include the origin wait time in the travel
time. This fact needs to be retained referring to the travel time correlation described in
the following chapter and the results analysis.
6.4
In the following lines the correlation between the travel time for a given OD-Pair and
line or the sequence of chosen lines for both models will be discussed. Referring to the
general form of the utility function and the simplifications which are necessary to
archive similar starting conditions, VISUM does not provide origin waiting time. In
contrast, BusMezzo considers origin waiting time along with the transfer waiting time in
the combined parameter waiting time. The set up for the impedance in VISUM
considers the fact of desired and actual departure time by increasing the path
impedance but this is not implied in the total travel time. The total travel time is defined
as the length of journey time. Based on this there is a relation between the total travel
time in VISUM and BusMezzo which is defined in the following expression.
( )
6.5
Example Network
The investigation network is shown in a general form in Figure 19. It consists of five
stops where travelers may start, transfer and/or determine their journey. The network
also provides five different lines with different headways, travel times, and vehicle
capacities. One two-minute walking link exists between stops 2 and 5. Transferring
platforms at stop two or three implies a walking time (WalkT) of additionally two
minutes. The complete timetables of line 1,2,3,4 and 5 are respectively listed in
VuV 2013
65
Model Comparison
appendix C. The vehicle capacity of Line 1, 2, 3 and 5 is 50 travelers, while Line 4 has
a vehicle capacity of only 25 travelers. In this network, there is no difference between
seats and total capacity.
In this network, only demand of public transit is considered. The interaction with private
transit vehicles is ignored. This is a major simplification and reduction of the model
scope especially for BusMezzo. But on the other hand, this allows getting a closer
comprehension of the operation method respectively.
Figure 19
The simulation time is generously determined from 7.00 to 10.00. Within the range of
simulation time, the generation time is fitted to the length of one hour starting from 7.30
to 8.30. The demand is generated during this period and equally distributed either
through the whole period (BM) or in equal one minute time slices (V). The large pre and
post run ensures that all travelers arrive at their destination within the simulation time
and are included into the result set. The earliest trip departs at 7.24 (pink line) and the
latest at 9.02 (green line). The number of trips overlap the traveler generation time.
Table 7 shows an overview of the line characteristics.
Table 7
Network characteristics
Nr
Name
Line
segment
Frequency
-1
[h ]
Hdwy
10
[min]
IVT
[min]
Stop(i,j)
(1,2)
Line
Pink
Green
Line
Yellow
(1,4)
Generation
Time
Capacity
[pax]
50
500
50
500
50
200
(2,3)
Line
Vehicle
Capacity
[pax]
6
10
15
(2,3)
25
4
(3,4)
Line
Blue
(3,4)
20
10
25
500
Line
Brown
(5,4)
15
15
50
750
VuV 2013
66
Model Comparison
Network Indicator
In addition to the vehicle capacity, the generation time capacity (GTC) is calculated by
multiplying the length of the generation time with the frequency and the vehicle
capacity of the line. Connecting this value with the amount of generated travelers gives
a rough indicator of how busy the network is.
In a later step, the network indicator value is used during the investigation to define the
capacity levels. Exceeding the ratio value over one (or 100%), more demand is created
than the generation time capacity is theoretically able to carry and agents, for example,
would need to wait at a stop until they are able to board a trip which departs after the
generation period. This will be discussed and explained in detail in the following
chapters.
The introduced network has been developed in the framework of VISUM. Therefore, it
features some VISUM specific characteristics and it is not possible to transfer the
network topology directly to the technical implementation in BusMezzo. The
implementation differences are described in the following lines. It also gives information
as to how and why the models use such a way of realization. By specifying the model
characteristics, the explanation and definitions refer to (Cats O. , 2011) and (PTV
VISUM 12.5 Fundamentals, 2012). For detailed explanations reference is made to
(Cats O. , 2012), (Burghout, 2011) and (PTV VISUM 12.5 User Manual, 2012).
To implement demand into the system VISUM uses zones. Each zone characterizes
the origin and/or the destination of movements. That means each journey starts at a
zone and terminates in another zone. Zones form the connection between the demand
and the transport supply (network). Each zone is connected to the network with
connectors and corresponds to an access or egress route. Therefore, a connector
forms either the origin or the destination connector. By connecting zones with the
network, passengers are able to enter and to exit the network and demand can be
assigned to it. The network itself basically exists for the purpose of representing nodes,
links and stops with their characteristic attributes. Nodes are defined as locations
where roads intersect, links starting or terminating, and turnings are specified. A link is
thereby defined between two nodes. A stop is used to let passengers board, transfer or
alight a public transit vehicle and can either be located on a node or on a link. Hence
zones are connected to nodes with a stop point in the represented network. All five
lines running in the network are assigned to a specific line route with a determined
sequence of stop points and a line route profile. The line route profile describes the
departure, transfer, and arrival times at each stop and defines the travel time between
stops. Because the line route profile output, like travel time, is independent from the
topologys attributes of the network, it is possible to assign different travel times
VuV 2013
67
Model Comparison
between the same stops to different line routes even though they run on the same link.
Assigning different travel times for individual line routes to the same link simplifies the
network, because no separate nodes and links are needed for each individual line
route. VISUM provides the possibility of calculating the travel time on the link attributes,
but it is not mandatory. Each individual vehicle journey is implemented into the model
as a pre-assignment step. The element of a vehicle journey between stops is called
vehicle journey item. Within one vehicle journey item, the calculated attributes stay
constant. For example, passengers are only able to board, transfer or exit a vehicle at
a stop point and not during the run.
Figure 21
VuV 2013
68
Model Comparison
VuV 2013
69
Model Comparison
stop is defined by the dwell time function. To reduce the impact of the additional
introduced links to connect origin nodes with stops (56,109), stops with destination
nodes (1211) and links to introduce stops (12,67,1415,912,34), the speed density
function is set to a sufficiently high value. On all other links, the combination of velocity
and length reflects the default travel times listed in Table 7.
By taking a comparable look at the generalized network representation between Figure
20 and Figure 21, it is obvious that the structure seems simpler in VISUM. Two other
facts which boost the network topology are the stop definition and the different ways of
describing the travel time. VISUM provides a separate interface to enter the time
profiles in a pre-assignment step. BusMezzo abdicates to define the time profiles to
simulate reliability and calculates the values from the link attributes and the assigned
speed density function.
6.6
Figure 22
Set Overview
Figure 22 shows the classification of the comparable level and different sets of
comparison. Each set might consist of several scenarios. The whole table of sets and
their scenarios is listed in Table 8. The lowest level represents the comparable level
how it is defined in chapter 6.2 with a deterministic behavior on the supply side and
network real-time information provided to agents in BusMezzo. By using this limitation
on BusMezzos side the models are comparable. The influence of capacity restrictions
has a major focus in the thesis. Therefore, different types of capacity constraints are
VuV 2013
70
Model Comparison
VuV 2013
71
Model Comparison
Table 8
Set
Scenario Overview
Type
Model
A
A
A
A
A
A
A
A
A
Capacity
Restriction
UVC
UVC
UVC
LVC
LVC
LVC
LVC
LVC
UVC/LVC
Crowding
Function
M
M
MC
M
M
M
MC
MC
SC
BM
V
BM/V
BM
V
V
BM/V
BM/V
BM
A
A
UVC/LVC
UVC/LVC
B
B
B
B
B
B
B
B
UVC
UVC
UVC
LVC
LVC
LVC
LVC
LVC
SC
V
SC
V
Overview_Set_A(Lin)
Overview_Set_A(SBB)
M
BM
M
V
MC
BM/V
M
BM
M
V
M
V
MC
BM/V
MC
BM/V
B
B
B
UVC/LVC
UVC/LVC
UVC/LVC
B
B
Lin
SBB
Lin
SBB
Lin
SBB
Scenario Name
A_U(BM)
A_U(V)
A_U(MC)
A_C(BM)
A_C(V(Lin))
A_C(V(SBB))
A_C(MC(Lin))
A_C(MC(SBB))
A_BM(SC)
A_V(Lin(SC))
A_V(SBB(SC))
B_U(BM)
B_U(V)
B_U(MC)
B_C(BM)
B_C(V(Lin))
B_C(V(SBB))
B_C(MC(Lin))
B_C(MC(SBB))
B_BM(SC)
B_V(Lin(SC))
B_V(SBB(SC))
LVC
LVC
SC
BM
SC
V
Lin
SC
V
SBB
Overview_Set_B(Lin)
Overview_Set_B(SBB)
M
BM
M
V
Lin
LVC
BM
C_NetRTI(BM)
LVC
LVC
BM
Lin
B_D()_BM
B_D()_V
C_TTK(V)
A_C(V(Lin))
C_NoRTI(BM)
Overview_Set_C
D
D
VVC
VVC
M
M
BM
V
D
D
VFC
VFC
M
M
BM
V
VuV 2013
Lin
D_VVC()_BM
D_VVC()_V
Lin
D_VFC()_BM
D_VFC()_V
72
Model Comparison
Where:
UVC
LVC
VVC
VFC
NoRTI
NetRTI
TTK
M
MC
SC
By comparing the models, basically the loads on line segments, the average total travel
time, and the number of transfers is considered to deviate and to explain the difference
in the models approaches.
Table 9 shows the schematic proceeding of the load line segment comparison. The Set
in the upper left cell indicates the set name. The inner four scenarios declare the output
results calculated by BusMezzo or VISUM either with or without capacity restraints.
Next to the actual load are also the generation time capacity and the ratio listed. The
linear crowding function is used as a default function. Additionally, the SBB function is
used and presented on a separate page. The column model comparison compares the
results among the models. The row scenario comparison draws a comparison among
the results of one model under different scenarios (unlimited vs. limited capacity).
Table 9
Scenarios of Set A
Set
BusMezzo
VISUM
Model Comparison
Unlimited Capacity
S_U(BM)
S_U(V)
S_U(MC)
Limited Capacity
S_C(BM)
S_C(V(Lin))
S_C(MC(Lin))
S_BM(SC)
S_V(Lin(SC))
Header
Scenario Comparison
The evaluation of the results is carried out with the measure of the total value, the
relative value, and the GEH. The GEH is defined as follow and attaches the total and
relative value (UK Department of Transport, 1996).
Where:
Load model M
Load model C
VuV 2013
73
Model Comparison
6.7
Set A represents the general framework mentioned in chapter 6.5 and is called the
base case. The aim is to get a first comprehension of the model behavior. Table 10
shows the demand injected to the system as travelers per hour. Because the
generation time is exactly one hour, the demand representation may be used as total
number of travelers introduced to the system.
Table 10
OD-Pairs
Stop 1
Stop 2
Stop 3
Stop 4
Stop 5
Stop 1
300
60
300
Stop 2
360
Stop 3
240
Stop 4
Stop 5
360
The average total travel time correlation can only be used if capacity is not considered.
In most of the application purposes, capacity plays a role in the assignment. The
question may also rise as to why this rule is not executed for all comparable cases. The
comparison of travel time is in this context very difficult, and therefore the usage of
travel time correlation is an exception and cannot be used in general. When it comes to
capacity restriction, travel times are not comparable due to the definition and what it
VuV 2013
74
Model Comparison
represents respectively. Thats why the travel time calculated with BusMezzo is written
down in its original form.
Table 11
Set A ATTT
[min]
Unlimited Capacity
BusMezzo
16,8
Limited Capacity
16,8
VISUM
13,3
13,5 (Lin)
14,6 (SBB)
Two facts are conspicuous in Table 11. One is the invariable travel time in BusMezzo,
independent from the usage of capacity. To show how busy the network is, the average
occupancy of the network is calculated to a network index of 66% by putting the
demand into ratio with the generation time capacity. This shows that the volume to
capacity ratio of the network is simply too low to cause significant congestion effects.
Considering the fact that the results are aggregated with a five percentage error term
estimating travel time, the values have an oscillation of five percent, too. Therefore, the
limitation of the capacity has a slight impact, but it is within the tolerated deviation. The
other peculiarity is the major difference between the linear and the SBB function in
VISUM. Using the linear function, the calculated travel time represents the minor
impact of the capacity restriction. Due to the steep rise of impedance, the SBB function
influences the travel time much more than the linear function with a smoother
impedance increase. Minor changes in the route choice causes major penalties.
Therefore, less attractive connections referring to long in-vehicle times become more
attractive, because the ratio of loads to capacity is lower. In total, the travel time
increases.
Transfer Rate
The transfer rate shown in Table 12 is also used for pointing out the work flow of the
models. Obviously, the impact on transfers in BusMezzo is the same as on travel time.
The network indicator is simply too low to cause major congestion effects, thus agents
are not denied of boarding and the amount of transfers stays even. In contrast, the
impact in VISUM is much higher. Considering the unlimited vehicle capacity, the
transfer rate between BusMezzo and VISUM is ten time less often.
Table 12
[%]
Unlimited Capacity
BusMezzo
5,8
Limited Capacity
5,8
VuV 2013
VISUM
0,58
0,43 (Lin)
0,18 (SBB)
75
Model Comparison
To understand why the transfer rates are so different, it is necessary to take a close
look at the choice-set and the choice-set generation model with its filtering rules. For
further detailed information refer to (Cats O. , 2011) and (PTV VISUM 12.5
Fundamentals, 2012).
VuV 2013
76
Model Comparison
An agent is forced to take a decision process whenever the event trigger is pulled. In
the graph below one agent has to decide between six alternatives. Each alternative has
a certain probability. The probability is an output value depending on the utility and the
choice model. The decision process itself is carried out by drawing a random number
between zero and one. The greater the probability of an alternative, the higher the
probability the alternative is chosen. A random number, for example, is calculated to
0.65 the agent takes alternative 4. VISUM only provides a minimum of connections with
a transfer compared to BusMezzo. Hence agents use more unattractive connections
with respect to the number of transfers. This explains the value difference by the power
of ten.
Figure 23
VuV 2013
77
Model Comparison
difference is obvious between lines pink, yellow, blue and brown between stops 2, 3, 4
and 5, 4. The reason why this occurs will be discussed in the following lines.
The difference between the lines pink and yellow between stops 2 and 3 is partly
explained by considering the fact of the choice-set. In addition only passengers from
stop 1 to 3 take the pink line directly to their destination and all other passengers exit at
stop 2. Except for the limited capacity case, for passengers from other OD-Pairs use
the pink line. On the other hand, BusMezzos agents use the pink line more often in the
section. This is caused by the generation process of agents and triggering the event
trigger.
Passengers coordinate their arrival time to the departure time of the trip. Using the pink
line automatically implies a transfer a stop three. This makes the connection
unattractive because the additional impedance of five minutes per transfer is relatively
high compared to the short travel times in the network. Most of the passengers use the
fast yellow line to reach their destination.
In contrast, agents are randomly generated and do not coordinate their arrival time with
the departure time of a certain trip. In terms of distribution choice model the cleverest
decision might be to wait at stop 2 until the fast yellow line approaches for the
destination stop 4. Because the frequency of the yellow line is low compared to the
pink line agents are forced to take the decision process every single time a vehicle
approaches. This phenomenon can be explained more easily by taking a look at the
next line segment between stop 3 and 4 because the option of walking is excluded.
Triggering Event
Lets pretend only demand from stop 3 to 4 is considered. 240 travelers have to decide
between a line with less in-vehicle time and low frequency or a line with more in-vehicle
time and high frequency. According to the definition in chapter 4 the event trigger is
pulled by taking a connection, boarding or alighting decision. A randomly generated
agent is introduced to the system and faces the decision process at stop 3. Therefore,
a connection decision needs to take place. Because the generation process is not
coordinated to any departure time, the agent has only one option and decides to wait at
stop 3. Whenever the next trip of a line arrives the agent faces a boarding decision
(boarding vs. waiting). While taking the decision, he/she is aware of the utility of all
alternatives because network real-time information is provided. The agent choses the
actual alternative by picking a randomly generated number and sets it in correlation to
the probability of the alternative. This is described in Figure 23. This means the greater
the utility of an alternative, the higher the probability that the alternative is chosen.
Every time the event trigger is pulled, the decision process starts again.
VuV 2013
78
Model Comparison
Table 13
VISUM
Yellow Line
31
128
Blue Line
210
112
240
=240
Taking a look at stop 3 where two lines with contrary line characteristic departure, the
event trigger has a major impact on the results. The presented loads in Table 13 are
the results of a separate assignment. It should be noted that the sum of the demand in
BusMezzo is not exactly 240 travelers. This comes from the stochastic generation
process and the calculated average of a sufficient high number of simulations runs.
Only demand from stop 3 to 4 is considered. The difference is significantly high. It
seems VISUM distributes the loads more balanced over all provided connections while
BusMezzos agents prefer the blue line with a long travel time and high frequency. To
understand how this works, let us take a look at the timetable in Figure 24. Decision
Figure 24
Timetable of the Yellow and Blue Line: BusMezzo (left), VISUM (right)
The yellow lines headway is 15 minutes with a travel time of 4 minutes. In contrast, the
blue line departures every 3 minutes but the in-vehicle time is 10 minutes. VISUM
dominates some of the blue trips (marked light blue in the figure, right), because the invehicle time is greater, the departure is earlier and the arrival is later. These trips are
removed from the choice-set. Hence they do not carry any loads. In BusMezzo these
trips are part of the choice-set. Each agent chooses with a certain probability whether
he/she wants to board the alternative or not. In some cases an agent might be
generated very close to the departure time hence the waiting time is almost zero and
one of the dominated trips becomes attractive and the agent boards the vehicle. But
this does not explain why the loads among the lines is significantly different. Therefore,
the process of the event trigger needs to be clarified.
VuV 2013
79
Model Comparison
Since the event trigger is launched in a sequence of three minutes and in addition
every 15 minutes according to the lines approaches. Agents face a decision process
quiet often. In the given situation agents decide between boarding or waiting until the
next vehicle departures. This leads to a chain of decisions and the multiplication of the
decision probabilities.
Figure 25 is taken into consideration to illustrate the process by a theoretical example.
The decision probability between two alternatives is estimated at 50% (e.g. flipping a
coin). On the first decision level, 50% decide whether to board or to stay at the stop.
Whenever the next vehicle approaches, the likelihood of the decision process stays the
same but only 50% of the demand faces the second level. Hence, the number of
deciders reduces by the multiplication of the probability to 25%. If you think about the
example of flipping a coin, the probability to get head or tail x times in a row is
.
Figure 25
In our case line blue induces a high number of decisions and agents tend to board the
slow high-frequented blue line, because of the fact that this line causes a lot of decision
processes and agents board the blue line although it could be smarter to wait until the
fast yellow line approaches.
In summary, a high-frequented line causes more decisions thus having a major impact
on the travelers distribution. It is also questionable how realistic this might be, because
the implementation allows agents to deny the first arriving vehicle of a line but allows
boarding the second arriving vehicle of the same line. Hardly any user of public
transport systems would do this. Either they will board the first arriving vehicle of a line
or they wait until the next arriving vehicle of another line departures. This is the reason
why the model comparison between BusMezzo and VISUM for the unlimited capacity
scenario delivers dissimilar results especially for the relations stop 2, 4 and 5, 4.
VuV 2013
80
Model Comparison
Limited Capacity
The explanation, given in the previous sections, gives a detailed exploration of the
basic behavioral characteristic specifying BusMezzo and VISUM. These characteristics
are truly independent from the limitation of the capacity. Therefore, the limitation of
capacity will be discussed on the following pages. As already mentioned in chapter 3.5,
BusMezzo and VISUM describe different phenomena by considering congestion
effects. BusMezzo takes into account the actual limitation of capacity. This means if a
vehicle has reached the capacity limit nobody is able to board and they have to wait at
the stop. If no alternative is coming, the agent does not reach his destination. In
contrast, VISUM describes the discomfort due to crowding. Because it is just a felling
passengers are in principal always able to board independently from the degree of the
volume to capacity ratio. Of course, densely occupied vehicles become less attractive
but if no alternative is provided passengers will still board. The used crowding functions
are characterized in chapter 5.5.
Comparing the scenarios unlimited and limited capacity with the model BusMezzo, it is
obvious that the impact of the capacity restriction is marginal. This can be derived by
the network business. The network volume to capacity ratio is only 70 % of the
theoretical amount the network is able to carry during the generation time process. The
quantity of generated agents is simply too low that no denied boarding process takes
place. The scenario comparison would be more interesting if demand was increased.
The in-vehicle time in VISUM has a major impact on the distribution to the connection
choice. Therefore, passengers are attracted by those lines. If discomfort is considered
passengers try to avoid crowded vehicles and take into account the in-vehicle time of
the trip segment. The impact of the linear function is relatively low to crowdedness.
Hence a lot more passengers still take e.g. the fast yellow line than the generating time
capacity is able to carry. What is questionable, however, is the fact the assignment
allows passengers to board highly overcrowded vehicles. There is no actual limitation
to the total number of passengers fitting into one vehicle. The number of passengers
boarding a vehicle in reality might fluctuate highly depending on the situation. But in
fact, at a certain degree simply no more passengers are able to board. BusMezzo
considers this fact with a clear limitation in the boarding process. The closest notion of
a travelers behavior in reality might first consider the fact of discomfort, and if a certain
level of volume to capacity ratio has been reached, a strict refusal of boarding
processes takes place.
It is not astonishing that the impact on the comparison between the unlimited and
limited scenario using VISUM (independent from the crowding function) is much higher
than it is running the assignment with BusMezzo. Obviously the limitation on the
vehicle capacity has no direct influence on the capability of boarding. Enough capacity
is provided to the travelers that they can take their desired trip. But if crowding is
considered using VISUM, the core of the assignment changes from a stochastic
VuV 2013
81
Model Comparison
distribution to an iterative learning process. After all realistic connections have been
determined (depending on the operators selected settings and filtering rules) and used
as provided choice-set for each connection, the individual impedance is calculated.
Depending on the desired departure time, the attractiveness varies among passengers.
The distribution model allocates the demand to the connection. The higher the
impedance, the lower the percentage of passengers using the connection. Because the
discomfort cannot be described before an assignment, the assignment needs to run
iteratively to estimate the additional impedance. This process can be described as a
learning process. The learning process ends when the defined terminating conditions
are fulfilled. Therefore, the network responds more sensitively to capacity limitations in
respect to feelings of discomfort from crowding.
The default-used linear crowding function in VISUM provokes a slight tendency to more
similarity compared to BusMezzos output in the unlimited and limited scenario
(A_U(MC) vs. A_C(MC(Lin))). To catch the effect more appropriately of a stricter
punishment by exceeding the vehicle capacity, the SSB function with its characteristic
progression is used in addition, hence the model response more sensitive
(A_V(Lin(SC)) vs A_V(SBB(SC))). On the other hand, comparing the output results the
models are more similar except on the line load segment between stop 2, 3 on the pink
line.
In summary it should be remembered that both models assume different approaches to
estimate the behavior of travelers. Aside from the developmental purpose from a short
to middle planning tool (BusMezzo) to a middle to long planning tool (VISUM) implies
different assumptions. Some assumptions are similar and some vary. The average total
travel time represents different things according to the model definition. Only in some
exceptions the travel time can be derived and be comparable. The decision process
also differs from a single decision (VISUM) to a sequence of decisions (BusMezzo)
along the path. This arouses in BusMezzo an overestimation to high-frequented lines.
Otherwise the limitation of capacity is implemented opposingly. VISUM describes the
discomfort due to crowding and has no strict limitation to the vehicle capacity. It tries to
cover the fact by adding additional impedance to overcrowded connections. The
provided functions only work for a realistic range of volume to capacity ratio. When the
network business increases, BusMezzos approach of individual agents covers
overcrowdedness with denied boarding processes, in turn increasing travel time.
Chapter 6.8 deals with the same initial situation but the demand level is increased by
40%. A stepwise increase of the demand representation analyzes and presents
chapter 6.8.1 in an aggregated form.
VuV 2013
82
Model Comparison
Figure 26
VuV 2013
Overview Set_A(Lin)
83
Model Comparison
Figure 27
VuV 2013
Overview Set_A(SBB)
84
Model Comparison
6.8
The demand increase is the results of an estimation process to the point where denied
boarding takes place but the network is still able to handle the amount of generated
travelers mostly within the generation time. Table 14 specifies the increased demand
representation by 40%. The network index is calculated to 93%.
Table 14
OD-Pairs
Stop 2
Stop 3
Stop 4
Stop 5
Stop 1
420
84
420
Stop 2
504
Stop 3
336
Stop 4
Stop 5
504
Like expected, the average total travel time listed in Table 15 increases with the
increase of the demand presentation. But the increase is higher in BusMezzo than in
VISUM. This can be derived from the fact how travel time is defined in each model.
Each denied boarding process implies an additional waiting time and in consequence a
higher travel time. The busier the network, the higher the probability that agents are not
able to board their chosen trip and have to wait at the stop until the next vehicle
approaches. The travel time increases dramatically with the number of denied
boardings. In VISUM the business of the network implies a balanced distribution over
all provided connections in the network. The aim is to shift passengers from highly
utilized to low frequented connections, so that in total the sum of impedance is minimal.
This principal is also known as user defined equilibrium. To achieve this balance some
passengers need to switch from attractive to less attractive connections until the ratio
of loads to capacity multiplied with the travel time of the trip segments balances.
Table 15
Set B ATTT
[min]
BusMezzo
Unlimited Capacity
16,8
Limited Capacity
17,5
VISUM
13,3
13,8 (Lin)
14,4 (SBB)
The unlimited capacity scenario with the increased demand mirrors the results from
set A (A_U(BM); A_U(V) vs. B_U(BM); B_U(V)). The loads are respectively distributed
with the same proportion, but the loads are 40% higher. VISUM considers an increase
of exactly 40%, BusMezzo an increase of approximately 40% within the limit of error
tolerance.
VuV 2013
85
Model Comparison
Because the network is quiet busy, the impact of the scenario comparison (B_BM(SC),
B_V(Lin(SC)); B_V(SBB(SC))) increases from the consequence explained in
chapter 6.7. The effect is more observable using the model VISUM than BusMezzo and
even more clearly using the SBB than the linear crowding function. The results of the
line loads segments are listed in Figure 28 and Figure 29.
In spite of the strict punishment, based on SBB characteristics, the function is not
capable of avoiding overcrowded vehicles exceeding the vehicle capacity with the
default settings mentioned in chapter 5.5. This may arise theoretically from the low
coefficient in the SBB function which implies only a duplication of the trip segment
travel time. For example, the travel time of the trip segments of the yellow line are
relatively short compared to the remaining impedance factors and other lines trip
segments travel time. Therefore, a duplication of travel time does not raise the
impedance level to a high sufficient level apparently. By raising the value of the
coefficient , it should strictly punish connections which exceed the vehicle capacity.
Consequently, a limit is set to board vehicles over the volume to capacity ratio.
By raising the coefficient , it seems like VISUM distributes loads step-by-step to lines
in correlation to the value of . If the value of exceeds a certain value, the results are
non-comprehensible and illogical. Within the scope of the master thesis, it is not
possible to investigate the characteristic of the SBB functions behavior, which leaves
room for additional scientific work.
It should be kept in mind that the provided crowding functions in VISUM have not been
developed to imitate the process of denied boarding. Instead it describes the
discomfort due to crowding. The degree of volume to capacity ratio has a quantifying
range. It would be desirable to implement a function which describes the discomfort of
crowding to a certain level of volume to capacity ratio. After exceeding this level, the
function imitates the behavior of denied boarding with additional high sufficient
impedance. Care should be taken that the iterative process, with such a function,
converges to obtain results.
VuV 2013
86
Model Comparison
Figure 28
VuV 2013
Overview Set_B(Lin)
87
Model Comparison
Figure 29
VuV 2013
Overview Set_B(SBB)
88
Model Comparison
6.8.1
Set B uses the same default settings as set B except that the demand presentation
varies by the parameter . The parameter simply multiplies the demand presentation
from the base case scenario (Set A) listed in Table 10. The impact of the demand
increase is taken into consideration by taking a look at the aggregated results of the
average total travel time and the number of transfer.
While demand level raises the travel time, each individual agent experienced with
BusMezzo increases with an over proportional correlation Figure 31. This is caused
mainly by the number of denied boardings. The higher the demand increase, the higher
the number of denied boardings, and the higher the average total waiting time which
leads to a higher average total travel time. On the other hand, the impact on the travel
time in VISUM is marginal. Because VISUM does not simulate travelers as individual
passengers and travelers are presented in terms of traffic flows, no actual boarding
process takes place. The origin waiting time is coordinated to the connection hence the
start waiting time is always zero. In addition the waiting time between connections does
not increase by the increase of demand itself, it increases because passengers are
shifted by the choice distribution model to slower connections until the network
balances to the optimum of a minimum of impedance. The choice-set of connections
does not change during the assignment and is considered as static. Therefore, it is not
remarkable that the impact is very low. The in-vehicle time goes twice as input into the
impedance function which will be optimized in the network. One is the parameter invehicle time and the other is the in-vehicle time for each trip segment multiplied with
the output of the crowding function. The major input for the crowding function is the
volume to capacity ratio for each trip segment. Therefore, the in-vehicle time is an
important influence factor in the distribution process. This causes a major difference in
the model structure. In BusMezzo, the in-vehicle time and the travel time is only an
output of the model result. The distribution is the result of the decision process along
the decision path of each individual agent. Of course, the decision itself is influenced by
the parameter of the impedance function and each agent tries to optimize its decision
process. But the system does not optimize the network distribution to a superior aim.
VISUM uses the provided in-vehicle time for each trip segment and the complete
network information (e.g. departure, arrival, transfer time etc.) to optimize passengers
distribution in terms of a network consideration. The travel time can be seen as output
from the internal in-vehicle time distribution optimization.
It is necessary to distinguish between the linear and SBB function while looking at the
results. Taking a look at the linear function, a slight increase of the travel time is
noticeable as impact of the demand increase. The transfer rate still stays at the same
level. Referring to the characteristics of the linear function and its decent punishment to
overcrowded connections, the behavior mirrors in Figure 31. On the contrary, the SBB
function behaves slightly different. Apparently the increased demand level causes an
insignificantly reduction of the travel time. The transfer rate behaves similarly to the
VuV 2013
89
Model Comparison
linear crowding function. Only the amount of transfers is constantly lower. This arises
from the strict punishment of connections in the iterative distribution process. Since the
transfer cost is relatively high, correlated to the low travel time, connections including
one transfer becomes even less attractive. The characterization of the SBB function is
not within the scope of the master thesis and leaves room for further scientific
research.
Figure 30
Figure 31
VuV 2013
90
Model Comparison
One needs to bear in mind that the travel times are incomparable, because the origin
waiting time is considered in BusMezzo and left out in VISUM. Therefore, only the
quality of the runs is able to be judged. It seems logical that the number of transfers is
higher in BusMezzo than in VISUM. BusMezzo, on the one side, provides a larger
choice-set hence the choice possibility is larger and the number of decisions for each
agents leads to a greater amount of transfer choices. On the other hand, VISUM
reduces the choice-set to logical connections. The logical filtering rule excludes all
connection including a transfer except for the OD-pair 1 to 4. Most of the passengers
from stop 1 to stop 2 take the attractive direct green line and hardly anybody in the
network chooses the alternative of transferring. Except for the alternatives previously
mentioned, the filtering rule eliminates all connections including a transfer. In
combination with the fact that the provided alternative, including a transfer, hold little
probability, explains the general low rate of transfers. The actual transfer rate fluctuates
a little bit by increasing the demand level due to the quantity of passengers choosing
the alternative including one transfer.
6.9
Capacity constraints may arise from different impacts. The most common case is
probably the increase of the demand presentation, representative for the peak hour,
which has been discussed in the previous chapter. Complementary, the reduction of
the vehicle capacity provokes capacity constraints as well as lowering the frequency of
a line. To analyze the impact of the vehicle capacity and the frequency, the generation
time capacity is lowered by 10% steps.
()
The reduction of the frequency is an approach problem since the value of the
frequency needs to be an integer (relevant for Line 3,4,5). Therefore, the process is
carried out to the best of knowledge to find the optimum between reducing the
frequency and fitting the target of 10% steps. This causes unusual headways. The first
trip departures at the same time as the first run from the base case. All other trips
depend on the estimated frequency on the headway. In the case of vehicle capacity
reduction, the error to the exact value is within half a capacity unit (e.g. seat). This is
relevant for line 4 (90%, 70%). The capacity unit is rounded down to the next integer.
A complete table of the used framework is listed in Table 16 and Table 17. Table 17
shows in the column for one reduction step the selected frequency (Sel. Freq.) which
calculates with the vehicle capacity, the selected GTC (Sel. GTC) and the theoretically
targeted GTC (100% GTC). The results for the travel time and transfer rate are
presented in Figure 32, Figure 33, Figure 34 and Figure 35. The frequency reduction
might have a fluctuating impact on the results. This depends explicitly on the exact
coordination of the trips towards each other. For example, reducing the frequency
might cause a better coordination of a connection with less waiting time.
VuV 2013
91
Model Comparison
Table 16
Line
Frequency
[h -1]
100%
50
90%
45
80%
40
70%
35
60%
30
2
3
50
50
45
45
40
40
35
35
30
30
10
4
4
5
25
50
22
45
20
40
17
35
15
30
20
15
Frequency Variation
Frequency Reduction [h -1]
650
675
160
400
400
600
600
350
350
150
140
350
350
500
525
6
6
2
12
9
100% GTC
450
150
350
Sel. GTC
750
450
400
350
Sel. Freq.
750
180
400
7
7
3
14
10
100% GTC
500
200
400
60%
Sel. GTC
500
450
400
Sel. Freq.
200
450
8
8
3
16
12
100% GTC
200
450
70%
Sel. GTC
500
450
Sel. Freq.
500
9
9
4
18
13
100% GTC
500
80%
Sel. GTC
500
Sel. Freq.
10
10
4
20
15
100% GTC
Sel. Freq.
1
2
3
4
5
90%
Sel. GTC
Line
100%
300
300
300
300
100
120
300
300
450
450
VehCap [pax]
Table 17
10
50
50
50
25
50
As far as the models have been explained in the previous chapters and the reader
obtained a basic understanding of the behavior of the models, it is expected that the
average total travel time in BusMezzo increases with the reduction of the hourly
capacity. Compared with the results of VISUM it presents a qualitative similar picture to
set B. Differences are noticeable in the smoothness of the run and the travel time
using the SBB function. Attention needs to be paid to the scale range, too. The graph
of set B shows only the range from
(100% -250%) demand increase. The
variation of the vehicle capacity and the reduction of the frequency are limited to the
range from
(60% - 100%) of the generation time capacity. The results of
set B and set D cannot be equalized since the proportion of the influence is different.
The influence of increasing the demand by 10% is lower than reducing the frequency or
the vehicle capacity by the same amount. Reducing the vehicle capacity and increasing
the demand level leads in the end to the same results, but the step size is not
comparable. Therefore, the sets need to be analyzed separately and only qualitative
conclusions can be drawn.
VuV 2013
92
Model Comparison
Figure 32
Figure 33
The transfer rate shows partly a different behavior. Taking a look at the very similar
case of capacity reduction and increasing the demand, the runs show qualitatively
related runs. The linear function mirrors about the same transfer rate in both sets. The
same is valid for the transfer rate using BusMezzo. The only exception in this
comparison is the SBB function. Apparently, there is a tendency to the same number of
transfers while reducing the frequency in each model. All runs seem to aim at the same
level of transfers. But this cannot be proved by looking at the chart and remains an
assumption.
VuV 2013
93
Model Comparison
Figure 34
Figure 35
In summary it should be kept in mind that increasing the demand level and decreasing
the vehicle capacity has the same qualitative impact on the assignment results.
Reducing the frequency influences the runs of the travel time and leads to counterrotating processes between BusMezzo and VISUM.
VuV 2013
94
Model Comparison
RTI Impact
BusMezzo No RTI
BusMezzo Network RTI
VISUM(Lin) Timetable Knowledge
VuV 2013
ATTT [min]
17,16
16,81
13,50
95
Model Comparison
The discussed impacts are additionally visible in distribution of the line loads given in
Figure 37. The sketches from the row network real-time information and VISUMs
timetable knowledge are already known from set A (limited capacity scenario). Taking
into consideration the scenario and model comparison, it shows the influence of the
information degree. The results are more extreme. Segments with high-frequented
lines are more loaded and vice versa. The mentioned model differences which are the
reason for the differences become more effective and cause load shiftings. Over all,
BusMezzo is more similar to VISUM if network real-time information is provided.
Another way of looking at the consequences is to consider the distribution of agents
travel time and the corresponding frequency, also known as histogram. Figure 36
shows the results. The values of the relative frequency stacks are higher for low travel
time classes providing network real-time information. In contrast, the number of agents
with longer travel time is higher if no real-time information is accessible. This implies
the shifting from short to long travel time if network or no information is provided.
Figure 36
VuV 2013
96
Model Comparison
Figure 37
VuV 2013
Overview Set C
97
Conclusion
7 Conclusion
The aim of the thesis was to facilitate the planners comprehension of the model
characteristics by comparing, explaining, and evaluating the route choice of the
microscopic simulation-based transit assignment model BusMezzo and the
macroscopic schedule-based transit assignment model implement in the software
framework of Visum.
The survey in the context of the thesis was used to estimate the value-of-time for
transferring between transit lines and riding in an overcrowded public transport vehicle.
It was found, that travelers would accept to ride approximately 7 min longer in favor of
not having to transfer transit lines. This value is in line with the values of scientific
literature and confirms its validation.
The second part of the survey dealt with the impact of overcrowded vehicles. The
analysis of the respondents answer shows a linear correlation of travel time.
Different types of congestion, information degree, level of aggregation, and types of
assignment models are presented as preparatory steps to explain the foundations of
the microscopic simulation-based transit assignment model BusMezzo and the
macroscopic schedule-based transit assignment model in VISUM.
The challenge of comparing two transit assignment models, which are structured
differently, is to use an appropriate sample network. The network example needs to be
as simple as possible but still needs to cover all relevant phenomena. On the one
hand, the examples need to include the characteristics of each single model. On the
other hand, a certain degree of complexity is necessary to reflect the specific
assumptions of each model. Before comparing the models, it is required to define the
initial conditions: what is actually comparable and which similarities are achievable?
The main goal is not to simplify the models as much as possible in order to reproduce
the same results. Rather, it is more important to ensure that the same starting condition
is used to evaluate the output data and point out the differences in the models. The
best possible and suitable simplification is used as a compromise.
On a larger scale the level of comparison has to be defined. Considering the supply
side of the schedule-based assignment, it is a deterministic model with reliable
timetable information. This means all trips are on time, no delay is propagated and the
boarding or alighting process does not cause any delay in departure time. This is in
contrast to the stochastically simulation-based model implemented in BusMezzo. To
make the models in some way comparable, BusMezzo is forced to behave
deterministic on the supply side. Forcing BusMezzo to be deterministic and using
network real-time information for all agents equals in reliable timetable knowledge, how
it is defined in the schedule-based transit assignment model for passengers.
VuV 2013
98
Conclusion
VuV 2013
99
Conclusion
number and sets it in correlation to the probability of the alternative. Since all provided
alternatives hold a certain probability, there is a chance that travelers choose this less
attractive option. If this decision needs to be faced several times in a row, the amount
of travelers increases, taking a less attractive line in favor to a high-frequented line.
This shortcoming needs to be compensated by a logical, dynamically adaptive filtering
rule.
By increasing the demand, decreasing the vehicle capacity or reducing the frequency
in order to cause capacity constraints, the average total travel time in BusMezzo raises
significantly because of denied boarding. The schedule-based transit assignment
model covers crowding effects by using additional impedance functions depending on
the ratio between loads and capacity multiplied with the travel time for each vehicle
journey item and distributes the loads to the choice-set, respectively. But the effect is
marginal on passengers travel time because the schedule-based transit assignment
model optimizes the total impedance value to a stochastic user equilibrium.
The comparison between the microscopic simulation-based transit assignment model
BusMezzo and the macroscopic schedule-based transit assignment model in VISUM
clearly showed the strengths and shortcomings of both model classes. Furthermore, it
contributed to the understanding of the basic fundamentals of each model structure
and highlights the right of existence. The appropriate application of the models strongly
depends on the scope of work. The decision which model fits best to the assignment
needs to be considered by the traffic planner and his/her experience. It also showed
that more scientific work needs to be done to compromise the shortcoming (e.g.
crowding function and the limitation of congestion, overestimation to high-frequented
lines, and travel time calculations) to an adequate level of acceptance to approximate
the real behavior.
VuV 2013
100
References
8 References
Centre
VuV 2013
101
References
VuV 2013
102
References
Spiess, H., & Florian, M. (1988). Optimal Strategies: A New Assignment Model for
Transit Networks. Centre de recherche sur les transports, Universit de
Montral.
Strathman, J., Dueker, K., Kimpel, T., Gerhart, R., Turner, K., Taylor, P., et al. (1999).
Automated bus dispatching, operations control and service reliability.
Transportation Research Record 1666, pp. 28-36, Washington DC.
Sweet, R. (1997). An aggregated measure of travel utility. Transportation Research
Part B, 31 (5), 403-416.
Tan, M., Tong, C., Wong, S., & Xu, J. (2007). An algorithm for finding reasonable paths
in transit networks. Journal of Advanced Transportation, 41 (3), 285-305.
TCRP. (2003). Transit capacity and quality of service manual (TCQSM) 2nd edition.
TCRP Report 100, Transportation Research Board, Washington DC.
Tirachini, A., Hensher, D., & Rose, J. (2013). Crowding in public transport systems:
Effects on users, operation and implications for the estimation of demand.
Transportation Research Part A P. 36-52.
UK Department of Transport, H. A. (1996). The Design Manual for Roads & Bridges
(DMRB). (http://www.standardsforhighways.co.uk/dmrb/).
Wardman, M. (2001). Public Transport Values of Time. Leeds, UK: Institute for
Transport Studies.
Wardrop, J. (1952). Some theoretical aspects of road traffic research. Proceedings of
the Institution of Cicil Engineers.
Wiedemann, R. (1974). Simulation des Verkehrsflusses. Schriftenreihe des Instituts fr
Verkehrswesen, Heft 8, Universitt (TH) Karlsruhe.
VuV 2013
103
List of Tables
9 List of Tables
Table 1
Table 2
Table 3
Table 4
Table 5
Table 6
Table 7
Table 8
Table 9
Table 10
Table 11
Table 12
Table 13
Table 14
Table 15
Table 16
Table 17
Table 18
Table 19
Table 20
Table 21
Table 22
Table 23
Table 24
Table 25
Table 26
Table 27
Table 28
VuV 2013
104
List of Figures
10 List of Figures
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6
Figure 7
Figure 8
Figure 9
Figure 10
Figure 11
Figure 12
Figure 13
Figure 14
Figure 15
Figure 16
Figure 17
Figure 18
Figure 19
Figure 20
Figure 21
Figure 22
Figure 23
Figure 24
Figure 25
Figure 26
Figure 27
Figure 28
Figure 29
Figure 30
VuV 2013
105
List of Figures
Figure 31
Figure 32
Figure 33
Figure 34
Figure 35
Figure 36
Figure 37
VuV 2013
106
Appendix
Appendix
Appendix A
Appendix B
Appendix C
VuV 2013
Survey
Used Parameter in VISUM
Timetable Details of the Example Network
108
112
114
107
Appendix
Appendix A
Table 19
Survey
Survey Part a (General Information)
1*. Gender
Female
Male
2*. How old are you?
6-18
19-23
24-30
30-60
Over 60
3*. Your highest level of education
Pupil/Trainee
School Degree
University Degree
Doctoral/Postdoctoral
4. Your monthly income [] (if applicable)
[%]
46,3
53,7
[%]
[%]
5,0
33,9
56,6
4,5
[%]
Respondents
112
130
Respondents
5
54
118
60
5
Respondents
12
82
137
11
Respondents
0 - 100
101 - 200
201 300
301 500
501 1000
1001 2000
2001 3000
3001 and more
4,7
4,2
7,5
17,9
17,5
16,0
13,2
18,9
10
9
16
38
37
34
28
40
[%]
Respondents
51,7
13,6
34,7
125
33
84
[%]
Respondents
16,1
47,9
28,1
26,4
34,7
25,6
35,5
0,4
39
116
68
64
84
62
86
1
VuV 2013
2,1
22,3
48,8
24,8
2,1
108
Appendix
Table 20
7*.
[%]
Respondents
48,7
9,2
42,1
111
21
96
8*.
[%]
Respondents
18,4
48,7
32,9
42
111
75
9*.
[%]
Respondents
21,9
44,7
33,3
50
102
76
[%]
Respondents
37,3
22,8
39,9
85
52
91
[%]
Respondents
6,1
84,2
9,6
14
192
22
[%]
Respondents
11,4
65,8
22,8
26
150
52
[%]
Respondents
18,0
47,8
34,2
41
109
78
[%]
Respondents
0,9
95,2
3,9
2
217
9
[%]
Respondents
1,3
95,2
3,5
3
217
8
[%]
Respondents
16,2
59,2
24,6
37
135
56
VuV 2013
109
Appendix
[%]
Respondents
1,3
97,4
1,3
3
222
3
[%]
Respondents
24,1
25,0
50,9
53
55
112
[%]
Respondents
37,7
10,5
51,8
83
23
114
[%]
Respondents
6,4
82,3
11,4
14
181
25
[%]
Respondents
14,5
57,3
28,2
32
126
62
[%]
Respondents
24,5
30,9
44,5
54
68
98
[%]
Respondents
3,2
89,5
7,3
7
197
16
[%]
Respondents
5,5
84,1
10,5
12
185
23
[%]
Respondents
12,7
57,7
29,5
28
127
65
[%]
Respondents
1,4
95,5
3,2
3
210
7
[%]
Respondents
1,8
94,1
4,1
4
207
9
VuV 2013
110
Appendix
Table 21
threefourths
full
Almost
completely
full
[%]
R.
[%]
R.
[%]
R.
4,2
70,9
24,9
9
151
53
19,2
52,6
28,2
41
112
60
43,2
33,3
23,5
92
71
50
[%]
R.
[%]
R.
[%]
R.
2,3
82,2
15,5
5
175
33
16,9
65,3
17,8
36
139
38
40,8
31,9
27,2
87
68
58
[%]
R.
[%]
R.
[%]
R.
3,3
82,2
14,6
7
175
31
17,4
64,3
18,3
37
137
39
50,2
29,6
20,2
107
63
43
[%]
R.
[%]
R.
[%]
R.
2,4
88,2
9,4
5
187
20
17,9
66,0
16,0
38
140
34
46,2
34,4
19,3
98
73
41
[%]
R.
[%]
R.
[%]
R.
0,5
93,9
5,7
1
199
12
16,5
71,2
12,3
35
151
26
41,0
41,0
17,9
87
87
38
[%]
R.
[%]
R.
[%]
R.
0,9
96,7
2,4
2
204
5
6,6
87,2
6,2
14
184
13
26,5
62,6
10,9
56
132
23
[%]
R.
[%]
R.
[%]
R.
0,5
98,1
1,4
1
208
3
14,2
81,1
4,7
30
172
10
41,0
50,5
8,5
87
107
18
VuV 2013
111
Appendix
Appendix B
Table 22
VuV 2013
112
Appendix
Table 23
VuV 2013
113
Appendix
Appendix C
Table 24
Timetable of Line 1
Stop 1
Table 25
Stop 2
Timetable of Line 2
Stop 3
Stop 1
Stop 4
Dep.
Arr.
Dep.
Arr.
Dep.
Arr.
Trip1
7.24
7.31
7.33
7.39
Trip1
7.32
7.57
Trip2
7.30
7.37
7.39
7.45
Trip2
7.38
8.03
Trip3
7.36
7.43
7.45
7.51
Trip3
7.44
8.09
Trip4
7.42
7.49
7.51
7.57
Trip4
7.50
8.15
Trip5
7.48
7.55
7.57
8.03
Trip5
7.56
8.21
Trip6
7.54
8.01
8.03
8.09
Trip6
8.02
8.27
Trip7
8.00
8.07
8.09
8.15
Trip7
8.08
8.33
Trip8
8.06
8.13
8.15
8.21
Trip8
8.14
8.39
Trip9
8.12
8.19
8.21
8.27
Trip9
8.20
8.45
Trip10
8.18
8.25
8.27
8.33
Trip10
8.26
8.51
Trip11
8.24
8.31
8.33
8.39
Trip11
8.32
8.57
Trip12
8.30
8.37
8.39
8.45
Trip12
8.38
9.03
Trip13
8.36
8.43
8.45
8.51
Trip13
8.44
9.09
Trip14
8.42
8.49
8.51
8.57
Trip14
8.50
9.15
Trip15
8.48
8.55
8.57
9.03
Trip15
8.56
9.21
Trip16
8.54
9.01
9.03
9.09
Trip16
9.02
9.27
Trip17
9.00
9.07
9.09
9.15
Table 26
Timetable of Line 3
Stop 2
Stop 3
Stop 4
Dep.
Arr.
Dep.
Arr.
Trip1
7.26
7.30
7.32
7.36
Trip2
7.41
7.45
7.47
7.51
Trip3
7.56
8.00
8.02
8.06
Trip4
8.11
8.15
8.17
8.21
Trip5
8.26
8.30
8.32
8.36
Trip6
8.41
8.45
8.47
8.51
Trip7
8.56
9.00
9.02
9.06
VuV 2013
114
Appendix
Table 27
Timetable of Line 4
Table 28
Timetable of Line 5
Stop 3
Stop 4
Stop 5
Stop 4
Dep.
Arr.
Dep.
Arr.
Trip1
7.30
7.40
Trip1
7.32
7.47
Trip2
7.33
7.43
Trip2
7.36
7.51
Trip3
7.36
7.46
Trip3
7.40
7.55
Trip4
7.39
7.49
Trip4
7.44
7.59
Trip5
7.42
7.52
Trip5
7.48
8.03
Trip6
7.45
7.55
Trip6
7.52
8.07
Trip7
7.48
7.58
Trip7
7.56
8.11
Trip8
7.51
8.01
Trip8
8.00
8.15
Trip9
7.54
8.04
Trip9
8.04
8.19
Trip10
7.57
8.07
Trip10
8.08
8.23
Trip11
8.00
8.10
Trip11
8.12
8.27
Trip12
8.03
8.13
Trip12
8.16
8.31
Trip13
8.06
8.16
Trip13
8.20
8.35
Trip14
8.09
8.19
Trip14
8.24
8.39
Trip15
8.12
8.22
Trip15
8.28
8.43
Trip16
8.15
8.25
Trip16
8.32
8.47
Trip17
8.18
8.28
Trip17
8.36
8.51
Trip18
8.21
8.31
Trip18
8.40
8.55
Trip19
8.24
8.34
Trip19
8.44
8.59
Trip20
8.27
8.37
Trip20
8.48
9.03
Trip21
8.30
8.40
Trip21
8.52
9.07
Trip22
8.33
8.43
Trip22
8.56
9.11
Trip23
8.36
8.46
Trip23
9.00
9.15
Trip24
8.39
8.49
Trip25
8.42
8.52
Trip26
8.45
8.55
Trip27
8.48
8.58
Trip28
8.51
9.01
Trip29
8.54
9.04
Trip30
8.57
9.07
Trip31
9.00
9.10
VuV 2013
115