Vous êtes sur la page 1sur 91

Traffic Flow on Escalators and Moving Walkways:

Quantifying and Modeling Pedestrian Behavior in a Continuously Moving System

Peter D. Kauffmann

Thesis submitted to the faculty of the


Virginia Polytechnic Institute and State University
in partial fulfillment of the requirements for the degree of

Master of Science
In
Civil Engineering

Dr. Shinya Kikuchi


Dr. Antoine Hobeika
Dr. Ralph Buehler

February 4, 2011
Falls Church, Virginia

Keywords: escalator, moving walkway, cellular automata, pedestrian, capacity

Copyright © 2011 by Peter D. Kauffmann


Traffic Flow on Escalators and Moving Walkways:
Quantifying and Modeling Pedestrian Behavior in a Continuously Moving System

By
Peter D. Kauffmann

ABSTRACT

Because of perceived deficiencies in the state of the practice of designing escalators and moving
walkways, a microsimulation-based model of pedestrian behavior in these moving belt systems
was created. In addition to implementing walking and stair climbing capabilities from existing
pedestrian flow literature, the model utilized following behavior and lane change decision logic
taken form studies performed in the field of automotive traffic flow theory. By combining
research from these two normally independent fields with moving belt operational
characteristics, a solid framework for the simulation was created.

The model was then validated by comparing its operation to real world behaviors and
performance metrics found in the literature in order to verify that the simulation matched the
choices made by actual pedestrians. Once this crucial function had been completed, the model
could finally be used in its original purpose of determining the capacity of a belt under region-
specific input parameters. This paper also discusses other applications for which the model is
suitable, including performing sensitivity analysis of both existing and proposed belt systems,
analyzing the impacts of operational rule sets on the performance of escalators and moving
walkways, and analyzing the effect of queue growth on the storage area needed for pedestrians in
an ambulatory facility. Through the use of this model and the logic contained within it,
engineers and planners will be able to gain a more accurate understanding of pedestrian flow on
moving belts. The result of this increased understanding will be more effective and more
efficient transportation systems.
Acknowledgements

I would like to first thank my advisor, Dr. Shinya Kikuchi, for his continued support and
encouragement throughout my graduate career. His courses have opened my eyes to the
underlying issues that face the field of transportation engineering in this country and around the
world, and outside of class he has provided me with the opportunity to travel abroad and gain an
international perspective on engineering issues. Without his advice on this project, I would
likely never have found such a unique and engaging thesis topic, and his comments have helped
it become a much stronger document.

I would also like to single out the members of my thesis committee for their invaluable support
in this process. It wasn’t always easy to schedule meetings of the three committee members –
especially considering they work out of three separate Virginia Tech campuses – but because of
their flexibility, we made it work. I thank Dr. Antoine Hobeika for advising me on the “how” of
the model, ensuring that its operation was based in sound logic, and I thank Dr. Ralph Buehler
for making sure that I remain mindful of the “why” of the project, thereby keeping the purpose
and goal of the model in the forefront of my thoughts.

Also at Virginia Tech, I would like to thank the faculty of the Transportation and Infrastructure
Systems group. I would like to especially call out the professors who assigned significant term
projects. Although they seemed overwhelming at the time, I now understand that including large
projects in the TISE coursework helps to prepare students for larger research undertakings, so
thank you Dr. Montasir Abbas and Dr. Hesham Rakha, as well as Dr. Hobeika and Dr. Kikuchi.

My continued gratitude goes out to the Via Department of Civil and Environmental Engineering
staff. Lindy Cranwell, Merry-Gayle Moeller, Val Dymond, Kara Lattimer, Donna Sanzenbach,
Jeny Beausoliel, and the rest of the gang – you all were always there for me when I needed
something, and for that, I will always be indebted to you. A big thanks goes out to former staff
members Shelia Collins and Vickie Mouras as well.

Furthermore, I would like to thank Dr. Randy Dymond and Prof. Jeff Connor for giving me a
way to support myself financially during the time I should have been working on my thesis.
Teaching Measurements was a lot of fun, and I learned a lot about what it is like to be on the
other side of the assignment dropbox. Your continued advice and support really aided me in my
development as a student and as a teacher. Additional thanks goes out to Scott Miller and Kyle
White, the two best student teaching partners I could have asked for.

Acknowledgement is also owed to my roommates Greg Swieter, Brad Shapiro, and Jeremy
Henry, as well as my friends Kevin Finelli, Mallory Brangan, and Elaine Huffman for motivating
me to go traveling while working on my thesis. Traveling through big airports and using public
transit systems played a large role in forming the approach I took to writing this document.

Finally, I’d like to thank my parents for never once asking impatient questions about when I was
going to finish up my graduate study and get a real job, and I’d like to thank my girlfriend for
moving to France for a year so I finally had an excuse to sit down and get to work on this project.

You all are the best.

iii
Table of
o Conten
nts

Chapterr 1 – Introdu
uction to Mo oving Belts....................
. .................................................................. 1
1.1 Prevalence of Escalatorrs and Moving Walkwayys .............................................................. 1
1.2 History and
d Developmeent of Movin ng Belt Systtems........................................................... 2
1.3 Design of Pedestrian
P In
nterface Areaas .................................................................................. 3
1.3.1
1 Necessarry Simplificaations ................................................................................................ 4
1.3.2
2 Flaws in Belt Specifiication ............................................................................................... 4
1.3.2.1 Deteermining Cap pacity ................................................................................................ 4
1.3.2.2 Acco ounting for Regional
R Diffferences in Pedestrian S Streams ................................. 5
1.3.2.3 Effects of Crush h Loading........................................................................................... 5
1.4 Proposed Approach
A to Belt Capacitty Analysis ................................................................... 5
1.4.1
1 Model Operation
O ............................................................................................................... 5
1.4.2
2 Source Data
D ....................................................................................................................... 6
1.4.2.1 Incluusion of Auttomotive Choice Behavi or .............................................................. 6
1.4.2.2 Senssitivity Analysis Capabillities .............................................................................. 6
Chapterr 2 – Literature Review w ......................................................................................................... 7
2.1 Pedestrian Behavior.............................................................................................................. 7
2.1.1
1 Unrestriccted Conditio on...................................................................................................... 7
2.1.1.1 Trav vel Characterristics ................................................................................................ 7
2.1.1.2 Avaiilable Spacee ......................................................................................................... 8
2 Complicaating Factors ........................................................................................................ 9
2.1.2
2.1.2.1 Stairr Climbing ........................................................................................................... 9
2.1.2.2 Elev vation Chang ge..................................................................................................... 10
2.1.2.3 Botttleneck Effeccts ................................................................................................... 11
2.1.2.4 Follo owing Behav vior ................................................................................................. 13
2.1.2.5 Passsing Choice......................................................................................................... 13
2.2 Traffic Flow
w Theory ........................................................................................................... 14
2.2.1
1 Followin ng Behavior ........................................................................................................ 14
2.2.1.1 Exissting Derived d Equation Models M ........................................................................ 15
2.2.1.2 Rulee-Based Mod dels ................................................................................................. 15
2.2.2
2 Passing Choice
C ................................................................................................................ 16
2.3 Belt Characcteristics ............................................................................................................ 17
2.3.1
1 Geometric Specificattions ................................................................................................ 17
2.3.2
2 Operational Parameters ................................................................................................... 18
2.4 Capacity Analysis
A ............................................................................................................... 19
2.4.1
1 Empiricaal Capacitiess ....................................................................................................... 19
2.4.2
2 Level of Service Dettermination ...................................................................................... 19
2.5 Modeling Framework
F ......................................................................................................... 20
2.5.1
1 Simulatio on Approach h ...................................................................................................... 20
2.5.1.1 Cellu ular Automaata (CA) Fraamework ...................................................................... 20
2.5.1.2 CA Approach
A to
o Pedestrian Flow Modelling ......................................................... 20
2.5.1.3 Critiicism of CA Approach....................................................................................... 21
2.5.1.4 Calib bration of Model
M Param meters............................................................................ 21
2 Modeling
2.5.2 g Language ........................................................................................................ 22

iv
Peter D. Kauffmann Tablee of Contentts v

Chapterr 3 – Model Developmen


D nt .................................................................................................... 23
3.1 Initializatio
on ....................................................................................................................... 23
3.1.1
1 Belt Chaaracteristics ......................................................................................................... 24
3.1.1.1 Belt Parameters ........................................................................................................ 24
3.1.1.2 Belt Rules ................................................................................................................ 25
3.1.1.3 Draw w Simulation n Environmeent ............................................................................... 25
3.1.2
2 Initial Peedestrian Staate .................................................................................................... 25
3.1.2.1 Adju ust for Consttraints.............................................................................................. 26
3.1.2.2 Com mpute Variab bles .................................................................................................. 26
3.1.2.3 Pedeestrian Param meter Selectiion .............................................................................. 26
3.1.2.4 Popu ulating the Belt B .................................................................................................. 27
3.2 Operation ............................................................................................................................ 27
3.2.1
1 Inducing g Movement ........................................................................................................ 28
3.2.2
2 Boundary y Operationss ...................................................................................................... 29
3.2.2.1 Arriv vals .................................................................................................................... 29
3.2.2.2 Departures................................................................................................................ 30
3.2.2.3 Endiing Conditio ons ................................................................................................... 30
3.2.3
3 Implemeenting Pedesttrian Interaction ............................................................................. 30
3.2.3.1 Follo owing Behav vior ................................................................................................. 31
3.2.3.2 Passsing Behavio or ..................................................................................................... 32
3.2.4
4 Accountiing for Comp plicating Factors ........................................................................... 34
3.2.4.1 Pedeestrian Limittations ............................................................................................. 34
3.2.4.2 Botttleneck Facto ors ................................................................................................... 34
3.2.4.3 Belt Parameters ........................................................................................................ 35
3.3 Data Collecction ................................................................................................................... 35
3.3.1
1 Real-Tim me Outputs.......................................................................................................... 36
3.3.2
2 Time-Space Diagram ms .................................................................................................... 36
3.3.3
3 Capacity y Analysis ........................................................................................................... 37
Chapterr 4 – Resultss and Discusssion ............................................................................................... 39
4.1 Model Valiidation................................................................................................................ 39
4.1.1
1 Operation ........................................................................................................................ 39
4.1.1.1 Follo owing Behav vior ................................................................................................. 40
4.1.1.2 Passsing Choice......................................................................................................... 41
4.1.1.3 Floo or-Belt Interfface Effects ....................
. ................................................................ 42
4.1.2
2 System Outputs
O ............................................................................................................... 43
4.1.2.1 Com mparison to Empirical
E Caapacities ...................................................................... 43
4.1.2.1.1 Capacity Scenario 1 ........................................................................................ 44
4.1.2.1.2 Capacity Scenario 2 ........................................................................................ 44
4.1.2.1.3 Capacity Scenario 3 ........................................................................................ 44
4.1.2.2 Creaation of Time-Space Diaagrams ......................................................................... 44
4.2 Potential Applications
A ........................................................................................................ 46
1 Sensitivity Analysis of Belt Systems ............................................................................. 46
4.2.1
2 Analysis of Proposed
4.2.2 d Rule Impleementation .................................................................. 47
4.2.2.1 Pedeestrian Behaviors as “Ru ules” ............................................................................ 47
4.2.2.2 Stan nd-Only and Walk-Only Restrictionss ............................................................... 48
4.2.2.2.1 Rule Scenaario 1 (Exam mple Scenarioo 2 in Sectioon B.4)................................. 48
Peter D. Kauffmann Tablee of Contentts vi

4.2.2.2.2 Rule Scenaario 2 (Exam mple Scenarioo 4 in Sectioon B.4)................................. 48


4.2.2.2.3 Rule Scenaario 3 (Exam mple Scenarioo 5 in Sectioon B.4)................................. 48
4.2.3
3 Platform Sizing ............................................................................................................... 49
4.3 Suggestionns for Furtherr Research ...................................................................................... 50
4.3.1
1 Belt Perfformance in Terms of Deelay ............................................................................. 50
4.3.2
2 Microsim mulation of Bottleneck
B Effects
E .......................................................................... 51
4.3.3
3 Effect off Impatience on Pedestriaan Aggressivveness ..................................................... 52
Chapterr 5 – Conclu
usions................................................................................................................... 53
5.1 Theoreticall Approach anda Modelin ng Strategy .................................................................. 53
5.2 Comparison to Empiriccal Pedestriaan Behavior and Capacitty Data................................. 54
5.3 Real-Worldd Applications of the Mo odel ............................................................................. 55
5.4 Summary ofo Contributiions of the Project P ......................................................................... 56
Referencces ........................................................................................................................................... 57
Appendiix A – NetLogo Model Code .............................................................................................. 60
;; A.1 Create Glob bal Variablees ..................................................................................................... 60
;; A.2 Create Turttle-Specific Variables ........................................................................................ 60
;; A.3 Create Pedestrian Classses ................................................................................................... 61
;; A.4 Define Exaample Scenarios ................................................................................................. 61
;; A.5 Select Pedeestrians for NetLogo
N Tim me-Space Diiagram ..................................................... 65
;; A.6 SETUP rou utine ................................................................................................................... 66
;; A.7 Adjust for Constraints
C module
m ........................................................................................... 66
;; A.8 Compute Variables
V mo odule ................................................................................................ 66
;; A.9 Draw Simu ulation Envirronment mod dule............................................................................. 67
;; A.10
0 Populate Belt module ......................................................................................................... 67
;; A.11
1 File Outputt module............................................................................................................. 69
;; A.12
2 GO routinee .......................................................................................................................... 70
;; A.13
3 Arrivals moodule .................................................................................................................. 70
;; A.14
4 Move Queu ue module .......................................................................................................... 72
;; A.15
5 Move Belt module .............................................................................................................. 73
;; A.16
6 Lane Chang ge module .......................................................................................................... 74
;; A.17
7 Departures module ............................................................................................................. 76
;; A.18
8 Create Plotts module ........................................................................................................... 77
Appendiix B – Modeel Operation n Instructio ons ............................................................................... 79
B.1 Version 1.1
1 ......................................................................................................................... 79
B.2 Operation ............................................................................................................................ 79
B.3 Explanationn......................................................................................................................... 79
B.4 Example Sccenarios ............................................................................................................. 80
B.5 Error Codees ........................................................................................................................ 81
B.6 Future Impprovements ......................................................................................................... 82
B.7 Related Moodels ................................................................................................................... 82
B.8 Revision History
H ................................................................................................................. 83
B.9 Credits and
d Referencess ....................................................................................................... 84
Table of Figures

Figure 1.1: Typical Escalator Bank, Dulles International Airport .................................................. 2


Figure 1.2: Typical Moving Walkway Installation, Dulles International Airport .......................... 3
Figure 2.1: Facial Ellipse Effect in the Downwards (Left) and Upwards (Right) Directions ...... 11
Figure 2.2: Escalator with Glass Handrail Balustrades, Dulles International Airport .................. 11
Figure 2.3: Moving Walkway with Lane Markings, Minneapolis-St. Paul International Airport 12
Figure 2.4: Standard Escalator Dimensions .................................................................................. 18
Figure 3.1: Flowchart of “SETUP” Routine ................................................................................. 23
Figure 3.2: Model Control Board in NetLogo .............................................................................. 24
Figure 3.3: Simulation Environment as Rendered in NetLogo..................................................... 25
Figure 3.4: Pedestrian Populated Simulation Environment as Rendered in NetLogo .................. 27
Figure 3.5: Flowchart of "GO" Routine ........................................................................................ 28
Figure 3.6: Flowchart of “Move Belt” Subroutine ....................................................................... 31
Figure 3.7: Flowchart of “Lane Change” Subroutine ................................................................... 32
Figure 3.8: Flowchart of Lane Check Algorithms ........................................................................ 33
Figure 3.9: Plot Showing Queue Growth under Steady-State Conditions at Capacity................. 38
Figure 4.1: NetLogo Simulation Showing Following Behavior ................................................... 40
Figure 4.2: NetLogo Simulation Showing Passing Behavior and Gap Acceptance ..................... 41
Figure 4.3: NetLogo Simulation Showing the Floor-Belt Interface ............................................. 42
Figure 4.4: Time-Space Diagram Showing Following and Lane Change Behaviors ................... 45
Figure 4.5: Excerpt of NetLogo Interface Showing Queue Simulation and Tabulation .............. 50
Figure 5.1: Comparison of Escalator Capacities from the Model and from the Literature .......... 53

vii
Chapter 1 – Introduction to Moving Belts
In the early days of urban transportation, achieving mobility was not difficult. Travelers could,
given enough time, walk to whatever destination they needed to reach. If necessary, a horse or
carriage might be used. However, as human settlements have become more and more dispersed,
it has become necessary to develop new technologies to speed up travel within urban areas. In
contemporary society, cities require tremendous transportation networks to handle the needs of
their citizens. However, even with this reliance on mechanized modes like automobiles, buses,
and trains, the need to accommodate pedestrians remains paramount.

In many public facilities like transit centers, airports, and even shopping centers, conglomeration
of modes have occurred in order to gain the efficiencies of scale. Because of this, these
structures have become so large that their designers cannot reasonably expect the average user to
travel across the facility by means of walking alone, at least not in a timely manner. To this end,
engineers and inventors have spent the last century developing innovations that can speed up
pedestrians and facilitate movement through urban interface areas.

1.1 Prevalence of Escalators and Moving Walkways


Today, escalators and moving walkways have become an integral part of the urban aesthetic.
Travelers have come to expect the presence of these enabling devices in most if not all
significant public facilities. Escalators have become commonplace wherever elevation change is
present, both to allow pedestrians to traverse a longer distance than normal even when carrying
luggage and also to keep the flow of travelers at a stable and high rate even in the presence of
vertical obstacles. Similarly, moving walkways have seen widespread implementation in
dispersed facilities like airports, which are spread out by necessity, or adjacent transit stations,
which may be connected to facilitate transfers between lines.

Both devices provide several key benefits. First, they serve to reduce overall travel time across a
facility. At the same time, moving belt systems also increase both the horizontal and vertical
distances that pedestrians are able to traverse by reducing the level of physical effort that must be
expended. Finally, and perhaps most importantly in the context of a high-traffic area like a
transit station, the presence of an escalator or moving walkway improves the overall flow
through an otherwise constricted area, thereby reducing the area required in the facility to
accommodate a given number of people. By conveying all passengers along at a constant rate,
the belt also creates some minimum speed at which all riders must travel. The conveyance of the
belt both serves to increase throughput while at the same time decreases the speed differential
through the constriction, therefore reducing conflicts.

Despite the benefits that escalators and moving walkways provide, their numbers are limited by
virtue of the limited number of public facilities where their usage is required. While it may be
true that in high capacity, low-rise facilities there are few alternatives to escalators for
movement, in the vast majority of hospitals, office buildings, and apartment buildings this
purpose is fulfilled by the elevator. To this end, while the 30,000 escalators in operation in the
United States (Slaughter, 2004) may seem like a large quantity, this number pales in comparison
to the 700,000 elevators in operation (Hession). However, while there may have been over
twenty times as many elevators, they only carried slightly more passengers. In the US and
Canada, there are 325 million elevator passengers per day compared to 245 million escalators

1
Peter D. Kauffmann Chap
pter 1: – Intrroduction to Moving Beltts 2

passengeers per day (Hession). Therefore, it can be sseen that despite their rrelatively lim
mited
number, escalators an
nd moving walks
w on aveerage carry a much greatter load per uunit.

1.2 History
H and
d Developm
ment of Mo
oving Belt Systems
The deveelopment off moving belt technolog gy was spreaad out over several decades, and inndeed
still continues to thiss day. The first
f patent for
f such techhnology on rrecord was ffiled in Auguust of
1859 by Nathan Am mes for an innvention he called “revoolving stairss.” This devvice, while ccrude
and imm mensely dan ngerous by modern stan ndards, wass the first tto include a progressioon of
horizontaal step-like surfaces
s – caalled “cleatss” – for stannding (Amess, 1859). Eaarlier models had
instead consisted
c off a simple belt with regularly
r sppaced woodeen slats to provide traaction
(Lampug gnani, Hartwwig, Simmen,, & Imorde, 1994).

Howeverr, the first modern


m versiion of the escalator as iit is known today was ppremiered bby the
Otis Eleevator Corpo oration at the
t Paris Exposition
E inn 1900. P Practically aall contempporary
escalatorrs follow thee model laid out by Otis. Safety feaatures that w
were introducced in this m model
include side
s balustraades for suppport. How wever, the mmost notewoorthy innovaation of thee Otis
escalatorr was the innclusion of a number off flat steps aat the entry and exit pooints to faciilitate
loading and unloadiing (O'Neilll, 1974). By B adding in space foor two – aand later thrree –
consecutive cleats to o leave theiir step treadds in a horizzontal posittion, the esccalator allow
ws its
passengeers enough tiime to establish themsellves on a sinngle step beffore the belt transitions tto the
inclined regime of steps.
s A ty
ypical escalaator bank iss shown in Figure 1.1. This partiicular
installatio
on is configuured for two
o-way operattion, with tw
wo parallel KKONE escalaators operatiing in
each direection, separaated by a by
ypass staircasse.

Figu
ure 1.1: Typical Escalato
or Bank, Dullles Internatiional Airporrt
Peter D. Kauffmann Chap
pter 1: – Intrroduction to Moving Beltts 3

With tim
me, further deevelopments in escalatorr technology have emergged, from hanndrails that m move
along wiith the steps to interlocking cleat designs thaat significanntly reduce the risk off foot
entrapmeent (Lampug gnani, Hartwwig, Simmen,, & Imorde, 1994). How wever, it is innteresting too note
that throuughout the last century escalator teechnology w was developeed much eaarlier than thhat of
moving walkways.
w Although both
b types of
o moving b elt systems are now coomparable too one
another, it is intriguinng that the complicated,
c , torque-intennsive, and fr
frankly dangeerous system m that
makes up p an escalato or was developed in advaance of the vvastly simpler moving w walkway. Inn light
of the faact that it is only with th he relatively
y recent advvent of largee indoor faciilities that suuch a
device iss required, this lag begiins to make sense. A ttypical moving walkwaay installatioon for
two-way operation iss shown in Figure
F 1.2.

Figure 1.2: Typical Movving Walkwa


ay Installatioon, Dulles Innternationall Airport

1.3 Design
D of Pedestrian
P Interface Areas
When deesigning pub blic facilitiess, one aspectt of importannce to considder is that oof pedestriann flow
within thhe facility. Even the mostm attractiv
ve and functiional structuure will be uunsuccessful if it
proves to o be too diffficult for its users to traaverse. Althhough a movving belt willl not necesssarily
transformm a structuree that is disspersed horizzontally or vertically innto somethinng with a tiightly
linked an nd effective floorplan, thet inclusion n of these syystems can go a long w way in bringging a
spread-ou ut structure closer togeth her. Furtherrmore, even in a situatioon where a m moving belt iis not
needed because
b of distance,
d theiir inclusion can help to promote ridder comfort and convennience
(O'Neill, 1974). Witthout the presence of an escalator, thhe area takenn up by verticcal transporttation
facilities would be prohibitively y large, no matter wheether the allternative traansportationn was
provided d by stairs or elevators.

To this end, architeects and eng gineers musst be sure tto consider the dynamiics of pedesstrian
circulatioon when dev veloping theese structures. Interfacee areas generrally includee facilities w where
there is a change of mode
m or vehiicle, a characteristic perhhaps best exxemplified byy a transporttation
station liike a train station or airrport. Howeever, a shoppping compllex, office bbuilding, or other
large pub blic structurre could alsso count in this categoory because of the factt that users must
transitionn from the mode
m of theirr arrival to whatever
w actiivity may bee occurring aat the endpoint of
their trip
p. The impo ortant behavvior to note is the preseence of pedeestrian circuulation withiin the
facility, since
s this acction must be
b accounted d for in the design phasse in order tto assure thaat the
final faciility design will
w have suffficient pedeestrian conveeyance capaccity.
Peter D. Kauffmann Chap
pter 1: – Intrroduction to Moving Beltts 4

1.3.1 Necessary
N Simplificati
S ions
Althoughh the capabillity exists to
o simulate thhe movemennt of crowds throughout a facility, oonly a
small perrcentage of interface
i areeas are subjeected to this level of anaalysis. Typiically, simullation
is only performed
p in
n facilities where
w crush loading
l condditions may occur that w will create uunsafe
ns for users. However, even in this situation, it is more ccommon for designers too use
condition
architectu
ural standard
ds to determmine the widtth of a corri dor or the nnumber of exxit doors reqquired
based on the projecteed flow of peeople along that
t particul ar path.

With reggards to escaalators and moving


m belts, design is typically peerformed byy determininng the
normal and
a peak load ding of userss who desiree to travel allong a particcular pedestrrian corridor. For
instance, a designer may know that t during rush
r hour theere may be some number of peoplee who
desire to
o exit an un nderground subway staation. Throough the use of publishhed architecctural
standardss or projecteed loading cu
urves, they can
c determinne how manyy escalators will be needded to
handle th
he outbound flow. Unfo ortunately, th
here are a nuumber of flaw ws and impeerfections thaat are
present in
n this approaach.

1.3.2 Flaws in Bellt Specifica ation


In the scenario descrribed above, a designer wants to dettermine whaat configurattion of escallators
is necesssary for subw
way installattion. The cu
urrent state oof the practicce would invvolve conduucting
research in a publishhed manual or
o manufactu urer’s guidellines to deterrmine the caapacity of a ggiven
escalatorr configuratiion and then n determinin
ng the quanntity and sppecifications needed to meet
demand. However, this meth hod may prrove to be unacceptably inaccuraate under ceertain
conditionns.

1.3.2.1 Determining
D Capacity
C
At preseent, the capaacity of an escalator or moving w walkway is determined by lookingg at a
manufactturer publicaation and finnding the “p
practical cappacity” baseed on the sppeed of operration
permitted
d by that loccality (ThysssenKrupp Elevator,
E 20004). Unforttunately, thee methods beehind
this meth
hod are someewhat oversiimplified.

To deterrmine the prractical capaacity, the manufacturer


m first compuutes the theooretical trannsport
capacity,, which is eq
qual to the number
n of trreads per houur times thee number of people eachh step
can supp
port. In conv ventional insstallations with
w a speed of 0.5 m/s, a tread depthh of 0.4 m, aand a
tread with capable off holding twoo passengerss, this value is 9000 passsengers per hhour:

1 2
0..5 ⁄ 2.5 9000
0.4

Up to thiis point, thee logic behin


nd capacity determinatioon for escalaators is sounnd. However, in
order to convert thiis theoreticaal capacity to practicall capacity, tthe standardd approach is to
account for
f loading inefficiencies, thereby reeducing the ccapacity dow
wn from this theoretical llevel.
While this approach is backed up p by observation and byy research, thhe specific m
method by wwhich
the practtical capacitty is determ
mined is thro
ough a multiiplicative addjustment faactor, usuallyy 0.8
(ThyssennKrupp Elev vator, 2004).. By using what in moost applicatioons is a connstant adjusttment
Peter D. Kauffmann Chap
pter 1: – Intrroduction to Moving Beltts 5

factor, th
he impacts off numerous complicating
c g factors aree effectively ignored, froom loading ddelays
to luggag ge to the added
a space required by y walkers aas opposed to standers. Althoughh this
problem is not uniquue to movingg belt facilities as even hhighway capaacity calculaations contaiin the
usage off constant adjustment factors,
f this does not m mean that thhe current situation muust be
accepted.

1.3.2.2 Accounting forr Regional Diifferences in Pedestrian


P Sttreams
While it is importantt to include thet impacts of these com mplicating faactors, it is eequally impoortant
to realizee that the dessires and chaaracteristics shown by ussers in differrent regions are not the ssame.
That is, the
t speed and d aggressiveeness parameeters that aree seen in New w York Cityy will be diffferent
from those in Newpo ort News, juust as the beehavior of a rider on ann escalator inn a transit sttation
during peeak hours will
w be more rushed than n a passengeer at a shoppping mall w will on a Tueesday
afternoon n. It is theorized that iff pedestrian behavior
b couuld be changged within tthe context oof the
capacity model, the results
r wouldd be applicab ble over a w
wider range oof sensitivityy scenarios.

1.3.2.3 Effects
Ef of Crussh Loading
As mentioned abovee, simulation n of pedestrrian flows iss sometimess used in siituations whhere a
sudden innflow of users can raisee safety con ncerns, eitheer through thhe potential for tramplinng or
crushing or because of the risk of overflow wing a limiteed loading aarea. An addditional practical
on of pedesstrian flow simulation
applicatio s can
c be foundd in the studdy of emerggency evacuuation
scenarioss. To this en
nd, it is sugggested that microsimula
m ation of movving walkwaays may sim milarly
be able to
o account for the intricaccies present under these conditions.

Under crrush loading conditions, microsimulaation would,, for instancee, allow the engineer to track
the growwth of a queu ue on a subw m. If the quueue reacheed some crittical length, there
way platform
may be potential
p for members off the crowd to fall onto the tracks, sso the designner would ennsure
that this condition
c woould not occcur given pro
ojected train unloading rrates. Howevver, an addittional
benefit ofo the use of microsim mulation in determiningg belt capaacity would be that vaarious
parameteers such as thhe ones discuussed above might be immplemented in the modell.

1.4 Proposed
P Approach
A to
t Belt Cap
pacity Anallysis
Because of the perceeived drawb backs presen nt in the currrent state off the practicce in movingg belt
capacity analysis, th
his paper prroposes the developmennt of a micrrosimulationn frameworkk that
describess and quantiffies pedestriaan flow on moving
m belt surfaces.

1.4.1 Model
M Operration
This mod del would bee capable off taking in a range of rellevant inputt parameters, from pedesstrian
desires an
nd choices to belt speciffications and
d operationall characteristtics. This innformation w
would
be used to define th he simulatio
on environm ment, at whicch point a llogical fram mework and rules
would bee used to go overn the intteraction of all participaants within the system. In this waay, all
automataa within the model will exhibit behaaviors basedd on their ow wn desires w within the coontext
of the overall
o systtem rules. Through the t interacttion of eveery entity, tthe system--wide
characterristics may be tabulateed over tim me to determ mine relevaant operationnal measurees of
effectivenness.
Peter D. Kauffmann Chap
pter 1: – Intrroduction to Moving Beltts 6

1.4.2 Source Data a


In order to adequateely define th he rules thatt govern eacch entity’s bbehaviors, a literature reeview
will be undertaken
u in
n order to quuantify and codify
c the chhoices madee by a personn in an activity as
mundanee as walking g or stair climbing. As will be seenn below, beecause of thee relative laack of
research conducted on o the speciific topics of
o following and passingg behavior w while climbbing a
moving belt,
b this infformation will
w be pulled from a diiverse array of fields. W While pedesstrian
behaviors like stair climbing
c andd walking sp peeds can b e taken direectly from appplicable stuudies,
additionaal informatioon will be taaken from reesearch condducted on waalkers withinn bottleneckks and
applied to
o the constriictions that are
a present in n an escalatoor.

1.4.2.1 In
nclusion of Au
utomotive Ch
hoice Behavio
or
Furtherm
more, to accoount for the thought pro ocess that ressults in folloowing behavviors and paassing
choice, information originally developed
d to model highhway drivinng will be inntegrated intto the
model. Although itt may seem m improper to use autoomotive folllowing and passing rulles to
determin
ne the behav vior of pedesstrians, the means by w which these human factors decisionns are
made aree actually verry similar by
y virtue of th
he fact that inn both casess it is a persoon who is maaking
the choicce. Additionnally, since these autommotive rules aare relativelly well fundded, there exxists a
much larrger and morre substantial basis of wo ork on whichh to base thiss section of tthe model.

1.4.2.2 Seensitivity Ana


alysis Capabiilities
Through the use of research fro om a varietyy of unconv entional souurces in a novel approaach to
moving belt
b analysiss, these unad ddressed facctors may bee included iin the transpportation anaalysis
and decission-making g process. Furthermore,
F it will be ppossible to vvary the inpuut factors sliightly
to determ
mine how seensitive the solution is to slight chhanges in anny number oof characteriistics,
from inpput stream volumes
v to belt speed to traveler characteristtics. In thiss way, projected
changes in passengerr mix or opeerational rules can be innvestigated aand accounteed for in adddition
to increases in passen
nger volumees.
Chapte
er 2 – Lite
erature Re
eview
The statee of the praactice of dessigning mov ving belt faccilities has sseveral key shortcominggs, as
were disccussed in thhe previous section. In order to recctify these iissues, this rreport propooses a
new microsimulation n framework k of pedestrrian behavioor to gain a clearer undeerstanding oof the
actual caapacity of a belt system. This mod del will incluude key elem ments of peddestrian behhavior
while at the same tim me accountiing for the variabilities
v in pedestriaan mix and aaggressiveneess as
well as anny operation
nal rules or restrictions
r th
hat may be iin place on thhe facility.

Howeverr, because th he literature indicates no


o projects thhat have attemmpted this aapproach, at least
not with the same co onditions andd constraintss that are preesent on a m
moving surfaace operatingg in a
narrow bottleneck,
b th
he theoreticaal frameworkk that drivess the model hhad to be deeveloped from m the
ground up.
u In ordeer to accuraately model the behaviiors present within thiss sort of system,
informatiion needed to be pullled from a number off sources, fr from pedestrrian behavior to
automotive traffic floow theory. Additional references
r coovering the operation off a belt systeem as
t practice of determin
well as the ning capacity and level of service in such a syystem have been
pulled ass well. Finaally, an inveestigation off varying moodeling pracctices was aalso conductted to
find a suiitable meanss of creating the desired simulation.

2.1 Pedestrian
P Behavior
The foun ndation of th
he model mu ust come from m the basic bbehaviors exxhibited by ppedestrians uusing
the faciliity. Significcant literatu
ure exists deescribing thee capabilitiees of pedestrrians in termms of
their trav
vel characterristics and ho
ow people teend to use thhe space avaiilable withinn a system. From
this basee, the modell can be streengthened by b the inclussion of variious compliccating factorrs, as
these connstraints willl cause the entities
e preseent in the moodel to be reestrained as tthey would iin the
real worlld by impediiments like stairs
s and enttry bottleneccks.

2.1.1 Unrestricted
U d Condition n
At the most
m fundameental level, pedestrian behaviorb is ddefined by hhow a persoon moves in open
space. There
T are a number
n of characteristiccs that pedesstrians exhibbit both whille moving aand in
terms of how they filll space.

2.1.1.1 Travel
T Characcteristics
Escalatorrs and movin ng walkway ys have been n designed too fit the sizee and physiccal capabilitiies of
locomotion that hum mans exhibit. Therefore, this model m must be baseed on these ssizes. At resst, the
average adult
a male can
c be appro oximated by an ellipse w with dimensioons of 18” bby 24” (0.455m by
0.6m). Itt is importan
nt to note thaat this design
n ellipse is a conservativve size selectted to accounnt for
a numberr of factors, including laarger bodies,, luggage annd other perssonal articless, and body sway
while waalking (Lee, 2005).

At the saame time, th


he average person is cappable of susttaining a maaximum walkking pace of 240
feet per minute (O'NNeill, 1974),, or about 1.2
1 meters pper second. Any pace above this sspeed
would innvolve both feet leavinng the groun nd simultaneeously and would best be classifieed as
running. In maintaaining this level
l ort, the aveerage stride length is 11.58m for m
of effo males,
compared d to 1.32m for
f females (Sutherland, Olshen, Biiden, & Wyaatt, 1988). It is importaant to
note thatt this measu
urement is giiven in term
ms of stride length and is therefore measured aas the

7
Peter D. Kauffmann Chapter 2:: Literature R
Review 8

distance between step ps of the sam


me foot. Th he distance bbetween eachh individual step is know
wn as
the step length
l and iss equal to on
ne-half of thee stride lengtth.

As will be
b seen later,, belt systemms have been n made in succh a way thaat these param meters are wwithin
the physiical capacityy of the system. In this way, escaalator steps aare deep enoough (0.4 m meters
deep) forr riders (on average
a 0.3 meters deepp) to occupy consecutivee steps, just as two riderrs can
stand sid
de by side ono a moving g walkway. Similarly, tthe operatioon of belts is restrained by a
number of o factors in
ncluding safeety and the ability of huumans to haandle the intterface between a
moving surface
s and a stationary one. Becau use of these llimits, the sppeed at whicch a belt opeerates
(no greatter than 180 feet per min nute) is held
d below the mmaximum unnassisted waalking speedd (240
feet per minute),
m therreby allowinng users to overcome thee motion of tthe belt and avoid a poteential
hazard iff necessary.

2.1.1.2 Available Spacce


Accordin ng to the literature, therre are four basic
b elemeents involvedd in the dynnamics of ccrowd
motion: time,
t space,, energy, andd informatioon. Time coovers the w way in whichh pedestrian flow
peaks an nd ebbs acro
oss a period of time, spaace deals wiith how the crowd utiliizes the avaiilable
area, eneergy indicatees the amounnt of effort exerted
e by tthe individuaals and how
w this inform mation
travels th
hrough the crowd in the form of sho ockwaves, annd informatiion means thhe way the ccrowd
becomes aware of in nstructions or
o hazards (Fruin,
( 19844). In this last case, infformation caan be
transmittted through formal mean ns like a pub
blic addresss system or a dynamic m message boaard or
informallly through conversation or rumor, th he latter of w
which can cau
ause a signifiicant safety tthreat
if the cro
owd begins too converge on
o a single constricted
c p oint.

Althoughh all elemeents of crow wd dynamiccs are impoortant in thhe context of designinng an
ambulatoory facility, the factor th
hat impacts pedestrian
p bbehavior most on the beelt itself is thhat of
space. AtA the mostt basic leveel, space utiilization is ggoverned byy size of a person andd any
belongings they may y happen to o have withh them. Hoowever, anoother importtant restriction is
formed by
b the social conventionss that exist which
w drive ppeople to avvoid body coontact with oothers
(Lee, 200
05).

As beforre, the size of


o a human body
b can bee representedd by an ellippse. At a baare minimumm, the
body takkes up an ellipse with an
n area of 1.5 5 square feett (Fruin, 19884). Howevver, this size does
not inclu
ude the spacing for sway y and other factors thatt was outlineed by Lee. Rather, thiss area
allows a person no control
c overr their own movement,
m aas with this level of spaacing they aare in
ontact with others. Frruin states that “at appproximatelyy 3 square feet per peerson,
direct co
involuntaary touching g and brush
hing againstt others willl occur, a ppsychological thresholdd that
should generally
g bee avoided inn most pub blic situationns. Below w 2 square feet per peerson,
potentially dangerou us crowd forrces and psy ychological stresses maay begin to develop” (F Fruin,
1984).

In this co
ondition, sho ockwave pro opagation will
w be directt and unrestrricted, leading to dangeer if a
crowd in this condition is rigidlyy confined byy architecturral features liike corridor walls, doorw
ways,
guardraills, or even esscalators. An
A additionall danger com mes from thee mechanicall conveyancee that
is inherenntly present in escalatorrs only serves to reducee the individdual’s controol over theirr own
movemen nt as well (F
Fruin, 1984).
Peter D. Kauffmann Chapter 2:: Literature R
Review 9

In order for pedestriaan movemen nt and choicce to occur, ddensity musst be reducedd. As occuppancy
decreasess, the amounnt of space available
a per person will see a corressponding inccrease. At a level
of 5 squaare feet per person,
p pedeestrians are able
a to standd stationary without toucching each oother.
Fruin sugggests that th
his level is the minimum m “in most nnormal waitinng situationss,” listing seeveral
exampless including standing on nboard an escalator.
e A
At 10 squarre feet per person, waalking
becomes possible on n the part of the pedestriians, provideed this moveement is donne in cooperration
with otheers. Finally, at 20 squaree feet per person free moovement is ppossible (Fruuin, 1984).

Howeverr, in additionn to the bodyy ellipse and


d the buffer zzone that suurrounds it, aanother impoortant
factor off spacing thaat should bee consideredd is the “paccing zone”, that is, the area requireed for
stepping forward wh hile walking (Lee, 2005). It is this l evel of spaccing that raisses the amouunt of
area requuired per peerson to highh levels in the
t Fruin esttimates. Hoowever, stilll further spaace is
required to detect annd respond too the terrain
n and conditiions ahead, w which Lee ccalls the “sennsory
zone”. Although
A this area does not
n need to be b kept entirrely free of oother pedesttrians for thee user
to make sound deciisions, it neeeds to be of o relativelyy low densitty for them m to satisfacttorily
respond tot obstacles ahead.

2.1.2 Complicatin
C ng Factors
The speeed at which a pedestrian is capable of o moving iss determinedd based on how they perrceive
the surroounding enviironment. The T informaation collecteed in this sccan of the im mmediate arrea is
coupled with that in ndividual’s personal
p chaaracteristics as well as tthe characterristics of thee trip
they are making – th hat is, the triip’s purposee, the presennce of luggagge, and theirr familiarityy with
the routee they are tak
king. The in nfrastructuree present, inccluding the llevel of elevvation change and
presence of shelter oro climate control
c has a factor in ttheir speed choice, as ddo environm mental
factors liike the weath
her or even thet attractiveness of the environmennt in terms oof the presennce of
shops or the possibiliity of crime (Hoogendoo orn & Daam men, 2005).

When ap pplied to the case of a mo


oving belt sy
ystem, severral of these eelements are of concern iin the
context of
o a modeling situation n. In addition to the oobvious impedement cauused by stepps or
elevation
n change, ussers will alsso experiencce delay from m factors liike bottleneck effects oor the
presence of other ridders. Thesee factors willl serve to hhinder the baaseline, unreestricted waalking
parameteers describedd above.

2.1.2.1 Sttair Climbing


g
The speeed at which a person iss capable off climbing oor descendinng a set of sstairs will allmost
invariablly be less thaan their speeed if they weere travelingg along levell ground. A study conduucted
by Fujiyaama and Tyller was desig gned to deterrmine the avverage climbbing speed off different cllasses
of pedesstrians, grou uped by agee. They fo ound horizonntal speeds of 0.44 to 0.76 m/s w while
ascending and 0.47 to 0.87 m/s while
w descen
nding. Whil e this range is fairly widde, the variaability
could noot be tied too any variab ble other thaan the preferrences of inndividual ussers (Fujiyam ma &
Tyler, 20
004)

Additionnally, the study found thaat there is no


ot a significcant differencce in walkinng speed bettween
the old and young cllasses, nor iss there a sign
nificant diffe
ference preseent when thee user is clim
mbing
at a “no ormal speed d”. Howeveer, when assked to asccend and deescend the stairs quickkly, a
differencce emerged at the 95% significancce level or aabove. Thee researcherrs concludedd that
above a certain poin nt, leg streng
gth does no ot benefit thee walking taask. Youngger users maay be
Peter D. Kauffmann Chapter 2:: Literature R
Review 10

capable of
o traveling faster
f if desiired, but a “n
normal” clim
mbing speed does not reaach the maxiimum
speed of elderly perssons and therrefore the av verage speedd for all userrs under norrmal conditioons is
similar reegardless of age or gend
der (Fujiyamaa & Tyler, 22004).

Other stuudies have fo


ound largelyy similar speeeds when addjusted to giive the horizzontal, unasssisted
pedestriaan speed. A study of th he London Undergroun
U nd gave a vaalue of 1.40 m/s for waalkers
onboard descending escalators accross all useer classes, whhich when thhe escalator speed is facctored
out and the
t diagonall speed is co onverted to a horizontal one gives a speed of 0.55 m/s (Davvis &
Dutta, 20002). Similaarly, a study
y on the Waashington Meetro shows a stair-climbbing speed oof 0.8
m/s, which corresponds to a speeed of 0.69 m/s (Sutthaw awassuntorn,, 2010). Ussing this rannge of
values, it is possiblee to determiine a range of walkingg speeds on stairs to goo along withh the
previouslly defined sp
pread of unreestrained waalking speedds.

Another important faactor to conssider when analyzing


a staair climbing motion is thhe fact that ttravel
along a staircase
s is restricted by the constraiints of the stteps themsellves. That iss, the dimennsions
of the staair riser and tread surfacces will limitt the numberr of possiblee locations fo
for a user to place
their feett while ascen nding or descending. Th herefore, thee user will fiind themselvves limited a pace
where thee average steep length eq quals the depth of the staiir tread (Leee, 2005).

2.1.2.2 Elevation
E Change
It is not just
j the presence of stairrs that createes an impediiment to peddestrian motiion; the elevvation
change ittself can pro
ove a detrim
ment to walkeers. First, immpedance iss observed as a climber gains
elevationn through the strain of physical
p exeertion. The maximum sspeed a peddestrian desirres is
seen to decrease
d slig
ghtly as their climb proogresses. HHowever, anoother imporrtant detrimeent to
forward progress
p com
mes from thee impact thatt elevation c hange can hhave on sightt lines.

Various studies havee shown an interesting relationship


r between the jam density observed on a
staircase and the direection of trav
vel along thaat facility. SSpecifically, the spacingg between ussers is
observed d to increase when travell is in the upphill directionn. This behhavior was deetermined too be a
result off a “facial ob
bscuring”. Facial obscu uring occurss when theree is a lack oof forward sspace
availablee in front of a user, and specifically in front of their face. When traveeling in the uuphill
direction
n, a user stan
nding directlyy in front off the rider inn question w
will – by virtuue of their hhigher
standing position – block
b all of the rider’s view since tthe rider wiill be staringg directly at their
shoulderss or back. This
T violation n of the forw ward facial eellipse can cause hesitatiion or discom mfort
on the part of the riider and cau uses greater open spacee to exist w when travelinng uphill. IIn the
downhilll direction, higher
h densiities are obsserved becauuse the oppposite effect occurs: useers in
front of the
t rider in question
q are standing on a lower stepp, and thereffore the riderr can see oveer the
top of thee forward usser’s head, giiving the illu
usion of greaater forwardd space (Lee, 2005).

This faciial ellipse co


oncept is illu
ustrated in Figure
F 2.1, bbelow. It iss interestingg to note thaat this
figure haas been creaated through h a modificcation of thee standard ccopyright-frree United S States
Department of Tran nsportation escalator pictogram, aas developedd by the fformer Ameerican
Institute of Graphic Arts (Geism mar, Chwast,, de Harak, Lees, & Viggnelli, 2011). In this fiigure,
the standdard solitary escalator rider has been n copied andd the facial ellipse highlighted in yeellow
to show the
t conflict that
t arises when
w travelin
ng in the upw
wards directiion.
Peter D. Kauffmann Chapter 2:: Literature R
Review 11

Figurre 2.1: Facia


al Ellipse Eff
ffect in the Downwards (L
(Left) and Uppwards (Righ
ght) Directionns

2.1.2.3 Bottleneck
B Efffects
Another feature of moving
m belt facilities
f thatt must be acccounted for in the propoosed model is the
constrictiing effect thhat the balusstrades – thee railing struuctures that are required by code oon all
contempo orary escalaators and mo oving walkw ways – have on the pedeestrian streaam. An escaalator
and its glass
g balustraades are shoown as Figurre 2.2, show wing the connstricting, coorridor-like eeffect
that is caaused by the handrails.

Fortunateely, there ex
xists a sizeab
ble body of research
r on fflow throughh bottleneckks. Most nottably,
Serge Hooogendoorn runs a lab at a the Delft University
U oof Technologgy where voolunteers aree sent
through experimentaally designeed constrictiions to prodduce microsscopic pedeestrian flow data
(Daamenn & Hoogend doorn, 2003)).

Fig
gure 2.2: Escalator with
h Glass Hand
drail Balustrrades, Dullees Internationnal Airport
Peter D. Kauffmann Chapter 2:: Literature R
Review 12

Insights gained from m these praactical expeeriments inddicate that w while shoullder width alone
indicates that it maay be possib ble for two pedestrianss to operatee side by sside in a naarrow
d as pedestriians attempt to maintainn separation from
bottlenecck, the lateraal space thatt is required
the side walls as well as each other preveents this froom occurringg. Becausee of the factt that
pedestriaans within thhe bottleneck k both wantt to maintainn their forwaard sight linnes and achieve a
satisfacto
ory level of horizontal spacing,
s a behavior thatt Hoogendooorn calls “ziipping” emeerges.
Zipping patterns
p are characterizeed by each su ubsequent usser taking upp a position to one side oor the
other wiithin the bo ottleneck, much
m he teeth of a zipper. In this wayy, the strides of
like th
consecutive users do o not interfeere with each h other, an individual ppedestrian iss able to ideentify
forward obstacles,
o an
nd lateral spaacing is mainntained (Hooogendoorn S S. , 2004).

In a folllowing expeeriment, it was


w determiined that thee overlap ppresent betw ween the parrallel,
zipped laayers of pedeestrians link
ks the flow of
o each streaam together, despite the fact that theey are
in differeent “lanes”. Because a pedestrian
p in
n one layer ttakes up morre than half of the corriddor, it
becomes difficult forr another useer in the adjaacent layer too pass, thereeby making tthe speed off each
stream generally
g con
nstant. Add ditionally, th
he shoulder sway exhibbited by a peedestrian as they
walk mak kes passing nearly impo ossible. To this end, thhe study conncludes that the capacityy of a
bottlenecck is stepwisse, based onn the width ofo the corriddor and how w many layerrs or lanes iit can
support (Hoogendoor
( rn & Daameen, 2005).

Since Hooogendoorn’’s experimen nts generallyy take placee in a meter--wide bottleeneck – the same
width as a standard escalator
e or moving
m walkkway – it is believed thaat his findinggs can be appplied
to the prroposed mod del. Howev ver, the expperimental p edestrians ddesire to maaintain separration
from the walls as weell as each other,
o unlikee the stationary pedestriians on an escalator whoo can
choose to
o “hug” the railing,
r whicch allows oth her users to pass in a paarallel layer. This behavvior is
supportedd by the incclusion of phhysical walkk/stand lane markings foound in ambbulatory faciilities
throughoout the United States, as seen in Figuure 1.2 and e specially in Figure 2.3, bbelow.

Figure 2.3: Moving


g Walkway with arkings, Minnneapolis-St.. Paul Internnational Airpport
w Lane Ma
Peter D. Kauffmann Chapter 2:: Literature R
Review 13

A subseqquent study detailing


d thee difference between
b singgle-file flow
w in a narrow
w corridor annd the
flow of pedestrians
p in a wider corridor confirms the ““zip effect” observed bby Hoogenddoorn.
Howeverr, this study goes one sttep further in that it obsserved this bbehavior on escalators iin the
London Undergroun nd. Addition nally, that report
r foundd the presencce of a “cappacity drop”” that
occurs when
w a constrrained corrid dor system like
l an escallator goes frrom normal flow to satuurated
flow, as this transitiion causes th he packing structure off the pedestrrians to adjuust to handlle the
higher in
nflow. It is the
t transition n itself that causes a tem
mporary but significant drop in flow w, the
presence of which caan significan ntly disrupt the
t operationn of a system m. Becausee of this dropp, the
study pro
oposes meterring users in n such a way y that the peaak inflow dooes not exceeed this maxiimum
sustainab
ble flow rathher than the capacity of the system since this w would incur tthe capacity drop
(Cepolina & Tyler, 2005).
2

Within thhe context of


o a transit sy
ystem, the function
f of m
metering couuld be approoximated thrrough
progressiive platform
m design. Sp pecifically, by positioninng the escalaator in a wayy that prevennts all
those useers disembarrking a train from reachiing the escallator at the same point, tthe peak inflow is
spread ouut to keep th
he escalator operating
o in a sustainabl e manner.

2.1.2.4 Following
Fo Beh
havior
Again, Hoogendoorn
H n’s research provides an answer to aanother criticcal impedim ment to pedesstrian
motion: the presencee of other users
u within
n the bottlenneck. Usingg experimenntally determ mined
pedestriaan headway observations
o s, a composiite headway distributionn model was developed. This
model is called a co omposite mo odel becausee it contains two sets off behavior eqquations, onne for
those whho are unconstrained and d another for those who aare constrainned by the faact that they must
follow annother user. Above som me free headwway, the model states, itt can be saidd that a pedesstrian
is not connstrained by
y any other pedestrians
p present
p aheaad. This dessired empty zzone is uniqque to
each peddestrian and assumed to be exponenttially distribbuted as a ressult of user choice as w well as
individuaal variations in step size and frequenncy (Hoogenndoorn & Daaamen, 2005).

Howeverr, the equatiions contain ned within the


t composite headwayy distributioon model reequire
calibratio
on based on n field data to fully opeerate the moodel. Furthhermore, thee data providdes a
distributiion of headw ways for the overall syystem, whichh would be difficult too implementt in a
microsimmulation-baseed model like the one proposed inn this paper. Additionaally, basing each
simulated d user’s actiions on the goal
g of main
ntaining a sppecific headw
way is believved to not m
model
real-worlld conditions as well as a model in which the uuser varies ttheir speed ddirectly baseed on
their obseerved headwway.

2.1.2.5 Passing Choicee


There dooes not exist much informmation about how a walkker decides to pass anotther in free sspace.
Blue andd Adler pressent a logicaal method fo or determiniing when a pedestrian ddesires to chhange
lanes to walk
w aroundd a slower user,
u but thiss method is bbased on paassing in opeen space (Bllue &
Adler, 1998). Therrefore, this technique is unsuitablle the purpose of moddeling pedesstrian
behavior in a constraained system
m like what ex xists in a botttleneck or ccorridor.

In a corrridor scenarrio, pedestriian behaviorr tends to fform inform mal but rigiddly defined lanes
similar to
o those seen on a highwaay. While passing
p behaavior may noot be formallly codified, m
many
moving belt
b installations operatin ng in publicc transit systeems show ann informal trend where those
users whho choose to stand on thee belt stand to the outsidde lane, leavving the insiide lane avaiilable
Peter D. Kauffmann Chapter 2:: Literature R
Review 14

to those who choosee to walk. In I fact, Daviis and Duttaa report that the Londonn Undergrouund is
perhaps unique
u in th
hat the transsit authority has installeed signage inn several staations instruucting
users to keep to the left – the ou utside lane in
i Commonw wealth natioons – unlesss actively paassing
someone. In this waay, both stan nders and sloower walkerrs are kept oout of the paassing lane uunless
occupanccy reaches th he point wheere both lanees must be fi filled (Davis & Dutta, 20002). Becauuse of
the simillarities that exist betweeen passing g behavior oon a highwaay and passsing behavior of
pedestriaans within a constrictingg corridor, thhis project w
will look towwards developments that have
been mad de in traffic flow theory in order to model
m these parameters.

Howeverr, further un nderstandingg can still bee gained wiithin the fielld of pedesttrian passingg. A
subsequeent paper by the Blue an nd Adler proposes that ppeak of lane changes willl occur wheere an
occupanccy of betweeen 20 and 40%4 is obseerved. This occupancy is defined aas the numbber of
“cells” occupied, whhere each esccalator or mo oving walkw way tread witth depth of 00.4 meters iss split
into two standing positions, yield meters deep bby 0.5 meterss wide. Blue and
ding a cell siize of 0.4 m
Adler repport that theese occupanncies “are prresumed to bbe the speed-flow-denssity combinaations
that havee the most volatile
v dynaamics”, meaning that thhis range is w where the best gains maay be
had fromm a lane chaange (Blue & Adler, 20 001). Below w this densitty there willl be enoughh free
space too allow users to travell unimpeded d, thereby keeping lanne changes to a minim mum.
Converseely, above this
t density the efficieency of a laane change is greatly reduced beccause
acceptable gaps will be hard to fiind in the adj
djacent lane.

While th he formation n of lanes hash the poten ntial to resuult in a losss of capacityy because oof the
inefficien
ncy that commes from sub boptimal paccking of rideers on the beelt (Daamen & Hoogenddoorn,
2003), thhe tendency of users to o form laness provides sseveral beneefits. In moost non-satuurated
conditionns, the outside lane willl exhibit den nse standingg, either eveery step or oon alternate steps
dependin ng on wheth her the belt is i traveling uphill or doownhill, whiile the insidde lane will carry
walkers traveling att a higher effective sp peed (Sutthhawassuntornn, 2010). Because off this
operationn, the benefiits of lane foormation maay be realize d in terms oof user satisffaction, increeased
speeds inn lower floww periods forr those who o desire to wwalk quicklyy, and also inn that it proovides
order in an otherwise chaotic sy ystem. To address
a the ppotential forr capacity too be increaseed by
prohibitinng passing behavior,
b thee proposed model
m will inncorporate a number of rules that goovern
belt operration so thatt various rulee-based scen
narios may bbe tested.

2.2 Traffic
T Flow
w Theory
Because of the similarities betw ween pedesstrian flow w within a coorridor and the operatioon of
vehicles along a roaadway, techn niques relateed to traffic flow theoryy were revieewed to form m the
basis of movement within
w the model.
m Althhough Hooggendoorn andd Bovy statte that “vehiicular
mulation mod
flow sim deling approaches are geenerally not applicable tto pedestriann flow modeeling”
(Hoogendoorn & Bov vy, 2003), th
hey fail to prrovide justifi
fication of thhis assertion. For this prooject,
it is stro
ongly believved that thee constraintts present oon a movinng belt surfface throughh the
bottlenecck-like balustrades and the social conventionn of lane foormation maake movingg belt
systems very
v much analogous
a to automotive behaviors, aas will be seeen below.

2.2.1 Following Behavior


B
Because of the level of attentio on that has been paid tto traffic floow theory over the pastt few
decades, there existss a sizeable body of woork includingg numerous mathematiccal approachhes to
Peter D. Kauffmann Chapter 2:: Literature R
Review 15

the situaation of car-following behavior.


b Each of theese models considers sslightly diffferent
variabless and utilizess a differentt weighting scheme
s in an attempt too turn measuurable param
meters
about thee vehicle streeam into a prredictive fraamework for following bbehavior.

2.2.1.1 Existing
Ex Derivved Equation Models
A numbeer of these models
m can bee found in th
he literature.. One of thee first vehicuular car-folloowing
models to be develop ped was created by engineers from the Generall Motors Coorporation arround
fifty yeaars ago. Th he GM mo odels are a set of stim mulus-responnse models of car-folloowing
behavior, where thee stimulus comes from changes in the headwaay present bbetween thee two
vehicles in conjunction of the speeed of the leead vehicle, w while the ressponse of thee model is shhown
in the reesulting acceeleration or deceleration n of the folllowing vehiccle. Thus, the extent oof the
response of the follow wing vehiclee is directly related to thhe speed of th
the lead vehiicle and inveersely
related to
o the headwaay present beetween the two,
t implyinng that as spaacing increaases the actioons of
the lead vehicle
v havee a diminisheed effect on the
t speed off the followinng one (Mayy, 1990).

Since thee development of the GMM models, seeveral other m models havee come alongg. Next cam me the
Greenshiields model which attem mpted to rellate macrosccopic trafficc characterisstics to the sspeed
exhibitedd by a following vehiclle (Rakha & Crowther, 2002). Suubsequent m models like G Gipps
(Rakha, Pecker, & Cybis,
C 2007
7), Pipes, annd Van Aerdde (Rakha & Crowtherr, 2002) havve all
become more
m advancced, accountting for vario
ous other pootentially useeful parametters in an atttempt
to accuraately model the
t decisionss a driver maakes as a ressult of the raaw data inpuuts that they ccould
potentially receive.

2.2.1.2 Rule-Based Mo
odels
Howeverr, within thee context off the pedestrrian analoguue, it is perhhaps more ddesirable to have
followingg behavior governed by b a rule-b based methood instead of a compllicated seriees of
equationss. While carr-following behavior
b is easily
e broke n down to itts operationaal componennts, in
that a driver can exppress their desired
d speedd through thhe analog innput of the aaccelerator ppedal,
pedestriaan following
g is a far moore organic and
a innate aability. Hum mans are harrd-wired to leave
some ammount of spacing and maaintain a con nstant headw way to thosee around theem without rreally
having too give it much thought, and in orderr to program m these goalss and desiress into a com
mplete
modeling g frameworkk they must be
b distilled in
nto a straighhtforward yeet comprehennsive rule sett.

To this end,
e a rule-b
based model of car-follo owing was fofound in the literature ass a componeent of
the TRan nsportation ANalysis
A SIIMulation System, or T TRANSIMS.. TRANSIM MS bills itseelf as
“the nextt generation planning/simmulation mo odel” becausse of the waay it implem ments advancces in
computinng technolog gy and data availability
a to
t perform tr
travel demannd analysis aand forecastiing at
the netwoork level whhile modeling pedestrian n, automobilee, and transiit operationss at the indivvidual
level thro
ough microssimulation (Hobeika & Gu, 2004). In this waay, TRANSIMS gives aall the
benefits of the conveentional fouur-step modeel of transpoortation plannning while at the same time
accountinng for microo-level behav
viors like rou
ute planningg and regionaal congestionn.

The beneefit of TRAN NSIMS to th he creation of the propossed model coomes from itts microsimuulator
module. Because TRANSIMS
T is designed d to handlee large netw works, the ddevelopers oof the
system wanted
w to enssure that the software wo ould not be bbogged dow wn by compleex calculatioons of
desired acceleration
a or other facttors as is useed in the equuation-basedd models desscribed abovve. In
order to combat
c this issue, they decided
d to im
mplement a coarse simuulation approoach using a cell-
Peter D. Kauffmann Chapter 2:: Literature R
Review 16

based neetwork, desccribed below


w, along witth rule-base d behavior algorithms (Hobeika & Gu,
2004).

The resulting algorithhms are baseed on two baasic rules. F


First, a user wants to acccelerate wheerever
possible, up to their desired max
ximum speed d. Second, ddeceleration will occur oonly if necesssary,
albeit witth some low
w probability of random deceleration
d to mimic reeal-world behhaviors. Beccause
of the simplicity
s and
a elegance of these rules, the only inputt variables required inn the
microsimmulator are the
t present speed of th he subject veehicle, its m maximum atttainable speeed –
whether limited by the vehicle or by a sp peed limit, aand the forw ward headw way (Los Allamos
National Laboratory,, 2006).

Using thhese parametters, at each h time step every


e vehiclle on the network compputes the forrward
gap aheaad and makes decisionss based on the relationnship betweeen its currennt speed and the
forward gap.
g If the speed is greeater than thee gap, it slows down to a speed enssuring that it will
not violaate the gap in
i the next time
t step no o matter at w what speed tthe lead vehhicle is traveeling.
Otherwisse, and if th he current speed
s is lesss than the maximum speed, it w will acceleratte by
whateverr acceleration factor the vehicle is capable
c of acchieving. Inn each resultting case, theere is
some low w probabilitty – usually y 5% - thatt the vehiclee will slow w by some iincrement foor no
particularr reason othher than to mimic
m distraaction and otther actual bbehaviors, wwhile at the same
time giviing the systeem a bit of randomness
r to keep it ffrom settlingg into an unrrealistic situuation
(Los Alamos Nationaal Laboratorry, 2006).

2.2.2 Passing
P Cho oice
When it comes
c to mo odeling passsing choice, human
h behaavior is basedd on just thaat – choice. Lane
change behavior
b is a discrete, binary decisio
on process b ased on logiic and rules rather than some
continuou us numericaal calculationn. Thereforee, a rule-bas ed approachh to lane chaanges and paassing
is the on
nly practical approach to o use in the proposed mmodel, and beecause of thhe clearly deefined
analogy between
b lanes on a highhway and thee lane-like laayers observved in pedesstrian flow w within
bottleneccks it seemss that trafficc flow theorry is once again uniquuely suited tto this partiicular
problem area. Forttunately, TR RANSIMS was w created with a set of rules too govern paassing
behavior in addition to the previo ously discusssed followinng algorithm
m.

The TRA ANSIMS miicrosimulato or contains twwo differentt sets of lanne change allgorithms, oone to
govern thhe passing of
o slow-moviing vehicless and one to ensure that a vehicle is able to be iin the
appropriaate lane to make
m a requiired turn alo
ong their rouute. The prooposed modeel will utilizze the
first algo
orithm, whichh contains several condiitions that m
must all be m
met for a lanee change to ooccur
(Los Alamos Nationaal Laboratorry, 2006).

The first condition th hat must be met is that the gap forrward in the current lane is less thaan the
distance the vehicle would traveel in the nex xt time step. If this prooves true, thhe vehicle bbegins
scanning g the adjacennt lane with the
t second condition,
c thhat the gap fo
forward in thhat lane is grreater
than the gap forward d in the currrent lane. There
T is no ssense changging lanes iff the forwardd gap
there is leess than whaat exists at present.
p Nexxt, the microssimulator chhecks to deteermine if adjacent
lane’s gaap is greater than or equ ual to the currrent speed oof the vehiclle. If and onnly if all of these
criteria are met, the model
m makess a final checck to ensure that any lanne change that occurs wiill not
cut off a vehicle behhind in the ad djacent lane by checkingg that the avvailable gap backwards iin the
adjacent lane is at leaast as large as
a the maxim mum speed tthat any vehhicle on the nnetwork couuld be
Peter D. Kauffmann Chapter 2:: Literature R
Review 17

traveling
g. This is too ensure that no matter what the caapabilities of that adjaceent back vehhicle,
they willl not be traveeling so fastt as to collid
de with the ssubject vehiccle at the next time step (Los
Alamos National
N Labboratory, 200 06).

In additio
on to the TRRANSIMS-d defined proccedures for cchanging lannes, the propposed model will
also inclu
ude addition
nal decision logic that will
w serve to drive sloweer users backk into the ouutside
lane wheen faced withh a faster waalker approaaching from behind. In this way, thhe users will keep
right wheenever possiible in order to ensure th
hat the passinng lane is keppt free from
m impediments.

2.3 Belt
B Characcteristics
Now thatt sufficient background
b exists to deffine the behaavior of the pedestrians within a mooving
belt systeem, the systtem itself neeeds to be constructed.
c Fortunatelyy, the publicc safety conncerns
that impaact the operaation of escaalators and moving
m walkkways meanns that there exists a sizzeable
amount of o documen ntation and regulation
r th
hat applies to the desiggn and capaabilities of these
facilities.

2.3.1 Geometric
G Specificatio
S ons
First off,, the physicaal layout off a moving belt
b system is standardized to faciliitate ease off use,
ergonom mics, and pub blic safety. The individ dual tread ddimensions aare mandateed such that only
specific step widthss of 24”, 32 2”, and 40”” (up to 1.00 meters) arre allowed, with signifficant
restrictions present on
o the situattions where narrower trreads may be used. Thee rise and ruun of
each step p are constraained to be 7-7/8" (0.2mm) and 15-33/4" (0.4m), respectivelyy, accordingg to a
manufactturer (Schin ndler Escalattors, 2010). These speecifications are confirm med by Ameerican
Society of
o Mechanical Engineerss’ Safety Cod de for Elevators and Esccalators (Doonoghue, 19881).

At presennt, the mostt common trread width on o escalatorss and movinng walkwayss is 40 inchees, or
approximmately one meter.
m The reeason for thee prevalencee of this partticular tread width is thaat it is
the minimmum width that can ad dequately suupport two inndependent travel laness for pedestrrians.
The reasoon that wideer belts are allmost never seen in pracctice is that a wider belt would encouurage
the formaation of a thiird lane in th
he center of the belt. Thhis behavior could provee to be dangeerous,
as those pedestrians traveling in n the center lane wouldd not be able to reach a handrail dduring
d conditions, so belt widtths are kept to
saturated t one meterr to prevent tthis third layyer of pedesttrians
from form
ming (O'Neiill, 1974).

Additionnal specificattions also ex xist that are not as pertinnent to the ooperation off the model. One
such specification states that thee requiremen nt that three flat steps exist on eachh end on low w-rise
escalatorrs of less thann 10 meters with at leastt four steps m
mandated onn taller installlations. Annother
specificaation mandattes that the level of in nclination off an escalattor belt is nnot to exceeed 30
degrees (Welch,
( et all., 2009). Th
his parameteer was used to convert ddiagonal disttances and sppeeds
into horizzontal ones.

Other sou urces provid


de alternate values
v for the specified aangle of incllination. AS
SME requires that
escalatorrs not exceedd 30 degrees with an alllowable maargin of erroor of plus orr minus 1 deegree,
meaning that contraactors can technically “cheat”
“ upw
wards in steeepness as llong as theyy are
confidentt of their in
nstallation precision (Do onoghue, 19981). Europpean specifiications alloow an
incline of
o 30 degreees but permiit installation ns up to 35 degrees in the case off low belt sppeeds
(Bangash h & Bangash h, 2007).
Peter D. Kauffmann Chapter 2:: Literature R
Review 18

For moviing walkway y installation


ns, the comm
monly acceppted requirem
ment is that the incline oof the
belt shou
uld not exceeed 12 percennt (Bangash & Bangash, 2007).

A diagraam illustratin
ng these speccifications iss shown as F
Figure 2.4. This figure has been crreated
by annottating a pictture of a KOONE escalattor found att Dulles Inteernational AAirport. Notte the
yellow and
a black markings,
m wh
hich are sug ggested by thhe ASME SSafety Codee to highlighht the
edges of each tread for
f safety (Donoghue, 19 981).

Figuree 2.4: Standa


ard Escalatoor Dimensioons

2.3.2 Operational
O l Paramete ers
An addittional area ofo concern fo or governmeental regulattors is the sppeed at whicch escalatorrs and
moving walkways
w arre allowed too operate. The most typiical requirem ment on the ppart of muniicipal
codes is that the speeed of an escalator should d not exceedd 90-100 feeet per minutee, or 0.44 too 0.51
meters peer second (W Welch, et al., 2009). Ho owever, the AASME codee provides thhe opportunitty for
localitiess to approvee operation at a higher speed proovided that it is deemeed safe by local
regulatorry officials, although th his provisionn is rarely taken advanntage of (D Donoghue, 1981).
Addition nally, variouss transit ageencies such as
a the Londoon Undergroound have cconducted sttudies
with escalator belt speeds
s of up p to 180 feeet per minuute, or 0.91 meters per second (O'N Neill,
1974), allthough this is well in excess
e of thee typical opeeration speeed within thaat system off 0.72
meters per second. Generally, moving wallkways are ppermitted too operate upp to 180 feeet per
minute, sos long as their degree off inclination
n is kept low (Donoghue,, 1981).

In the O’Neill
O studyy that descrribed Londo on’s experim ments, an asssessment oof the througghput
possible under vary ying operatiion speeds was also iincluded. It was determined thaat an
nal speed off 145 feet peer minute (0..74 meters pper second) w
operation would proviide the maxiimum
flow of passengers
p through
t a moving
m belt system, wheether an esccalator or a m
moving walkkway
Peter D. Kauffmann Chapter 2:: Literature R
Review 19

(O'Neill, 1974). Thee stated reasson for this occurrence


o ddeals with thhe interface between thee belt
and the stationary
s flo
oor surface on
o either end d. While a hhigh belt speeed allows fofor faster speeed of
movemen nt, higher sppeeds cause trouble
t for users
u when thhey have to board or dissembark from m the
belt. Noot only is it hazardous to change from fr a consttant platform
m to a moviing one andd vice
versa, co
oncerns are also
a raised by b the poten ntial for danggerous pileuups to occur if a movingg belt
continuess to convey passengers
p into
i a crowed d area.

2.4 Capacity
C An
nalysis
As was previously
p deescribed in Section
S 1.3.2
2.1, determinning the capaacity of a mooving belt syystem
is at pressent a very approximate
a endeavor. By combiniing an idealiized occupaancy with thee belt
speed an nd then adju usting by somme arbitrary y occupancyy factor, the present statte of the praactice
fails to account
a for a number of o key factorrs while devveloping a vvalue that iss not sensitiive to
changes ini pedestrian n stream chaaracteristics.

2.4.1 Empirical
E Caapacities
Measured d flow ratees for escalators generaally give caapacities inn the vicinitty of 4000--6000
passengeers per hour. This range of values wasw found accross a varietty of sourcess, including 5400
from the manufacturrer publicatio on shown prreviously (T ThyssenKruppp Elevator, 2004), a sliightly
higher esstimate of 6400
6 passen
ngers per ho our from Fruuin based onn occupancyy statistics alone
(Goodmaan, 1992), an nd an empirrically defineed range of about 4100 to 5400 passsengers perr hour
based on
n the combin nation of treead width, beelt speed, annd incline present on thhat particularr unit
(Turner, 1998). O’N Neill further confirms th hese values, albeit on a 448” escalatoor which is a size
no longeer permitted by legal co odes (O'Neilll, 1974). S Subsequent sstudies havee gone furthher in
analyzing
g the capacitty of each laayer of an esscalator to giive the capaccity of both the standingg lane
and the walking
w lanee under a rep presentative scenario off choice behhavior (Davis & Dutta, 22002)
(Sutthaw
wassuntorn, 2010).
2

2.4.2 Level of Service Determ mination


As with most transp portation facilities, the level of seervice of a moving bellt system caan be
determin ned by comp puting its vollume-to-capacity ratio aand occupanncy rate. Thhese factors aallow
the enginneer or plannner to determ
mine what am mount of thee facility’s caapacity is beeing utilized. For
stairs, escalators, and
d moving waalkways thiss approach iss the only reeal way of m measuring seervice
quality since
s in all but the mosst extreme peak
p loadingg conditionss delay is nnot an appliicable
service metric.
m

In terms of volume-to-capacity (v/c) ratio, on ne analysis pperformed oon the Washiington Metroo was
developeed such that stations whiich possesseed one or moore facilitiess with a v/c ratio at or aabove
0.75 were deemed too need enhanncement. A v/c v ratio bettween 0.5 annd 0.75 flaggged that statiion as
needing a more detaiiled study off its pedestriian facilitiess (WMATA,, Access andd Capacity S Study,
2008).

To give a more descrriptive valuee of service quality


q of itss pedestrian facilities, W
Washington MMetro
publishedd a guide att the same time
t as the previous stuudy that dettailed the m methods by wwhich
level-of-sservice grad
des could be assigned to these facilitiies. The detterminationss were basedd on a
desired space
s per peedestrian – in
i effect, thee inverse off occupancy – as well aas the sustainable
flow per unit width ono the facilitty. The ratin
ngs ranged ffrom LOS A A, where a suufficient areea per
Peter D. Kauffmann Chapter 2:: Literature R
Review 20

pedestriaan exists freeely select speeed and to paass slower-m moving userss all the wayy to LOS F w
where
“complette breakdow wn in pedestrrian flow witth any stopppages” is thee norm. Usiing these criiteria,
the accepptable servicce level for pedestrians
p on was deemed to be
o stairs andd moving bellt facilities w
LOS C, defined
d as a space per pedestrian
p off 10 to 15 sqquare feet. W
While speedds are observved to
be slightlly restricted due to the innability of fast
fa walkers tto pass otherr slower-mooving pedestrrians,
this behaavior is posssible within the norms of o gap accepptance in thhe adjacent llane. At LO OS C,
standing riders on beltb systemss start to caause noticeaable conflictts (WMATA A, Site Plannning
Manual, 2008).

2.5 Modeling
M Framework
F k
With thee collection of informattion regarding pedestriaan behaviorr and dynam mics compileed, it
became time to dettermine how w best to construct thee simulationn. The goaal in selectiing a
simulatio
on approach and softwaare was to find f an effecctive framewwork that w would providde the
desired simulation
s caapabilities. First, the siimulation neeeded to havve the abilityy to create, alter,
and track
k the current and desired d states of a multitude
m off individual eentities as thhey move thrrough
the simullation space. Once the entities
e weree inserted innto the simullation space,, a predefineed set
of rules and
a behavio oral parameteers would govern the innteraction annd progressioon of all enttities.
This sim
mulation spacce itself wass another keey capabilityy, in that thee selected m model frameework
would neeed to posseess the abilityy to display the current condition oof the modell at any instaant in
time to en
nsure its opeeration and adherence
a to
o real-world operations.

2.5.1 Simulation Approach


A
Accordin ng to Hoogendoorn and Bovy, theree are three m model types that can be used to desscribe
flow-based systems like those found
f in tran
nsportation engineeringg problems. At the brooadest
level, maacroscopic modeling
m usses system-wwide param meters like aaverage flow w and densiity to
describe the conditioon of the syystem. Narrrowing in soomewhat, m mesoscopic m models like those
found in gas flow models do nott distinguish h between thhe individuall players in a system, innstead
choosingg to use vellocity distrib butions to simulate
s theeir behaviorr; but they still providde for
interactio
on between the entitiess in the sysstem throughh rule-basedd approachees. At the most
detailed level
l are miccrosimulatio
ons like the one
o sought foor use in thiss project thaat seek to desscribe
the behavvior of an en
ntire system through thee choices andd interactionns of those entities it conntains
(Hoogendoorn & Bov vy, 2000).

2.5.1.1 Cellular Autom


mata (CA) Fra
amework
In the coourse of investigating TRANSIMS
T and its rulees of driver interaction, a similarityy was
noted beetween the approach
a uttilized by thhe TRANSIM MS microsiimulator andd the constrraints
imposed by the treead spacing g on an esscalator. T TRANSIMS S uses a cooarse simullation
methodology known n as cellular automata, oro CA. CA divides each link foundd on the nettwork
umber of reg
into a nu gularly spaceed cells, each
h of which i s the size off one system
m user. Eachh user
is progreessed throug
gh the netwo ork using a simple set oof rules, witth the qualiffication that each
movemen nt must be some
s integer multiple ofo the cell siize such thaat the occupaancy of eachh cell
remains one
o (Hobeik ka & Gu, 200 04).

2.5.1.2 CA
A Approach to
t Pedestrian
n Flow Modeling
Althoughh cellular auttomata simu
ulations are generally
g moore commonn in other fiellds like com
mputer
science or
o automotiv ve behavior, they have seen
s some immplementatioon in pedesttrian simulattions.
Peter D. Kauffmann Chapter 2:: Literature R
Review 21

Blue and
d Adler constructed a series of mo w a through one-
odels that deescribed peddestrian flow
dimensio
onal free spaace using CAA principles (Blue & Addler, 1998) (Blue & Adller, 2001). T These
models contained
c ru overned the interactionns between pedestrians,, however, those
ules that go
approach
hes were deemed to be in nferior to thee previously defined TRAANSIMS ruule sets.

Howeverr, the Blue and a Adler experiments


e provide som me invaluabble guidancee in applyingg CA
characterristics to thee pedestrian
n paradigm. Most notaably, the coontinuous sppeed distribuutions
observed d in pedestriaan behaviorss must be redduced to a seeries of discrete speeds, all in increm
ments
of one ceell per secon t configuraation of an escalator lim
nd. Fortunaately, since the mits the locaations
where it is possible for
f a user to place their feet
f to the trread surfacess, it was deccided to makke the
cell size in the produ
uced model equal
e to the depth
d of a treead, 0.4 metters.

After a range
r of waalking speeds was consttructed usingg this setup,, a problem was discovvered.
Using a full
f tread as the cell sizee does not alllow enoughh variation inn the possiblle walking sppeeds
to provid
de suitable seensitivity in walking speeed. Consequ quently, it waas decided too adjust the sspeed
choice opptions to hallf-cell incremments per seecond, givingg twice as m many options for the waalking
speed off the system users. As it i turns out, this change allows for tthe inclusionn of a previously
omitted characteristic
c c of human behavior,
b steep straddlingg. This pracctice occurs wwhen a user has a
foot on one
o step and their other foot f on an addjacent step,, effectively causing thee user to occuupy a
position midway betw ween two stteps. With th he inclusionn of followinng behavior, spacing bettween
users willl be maintaained even in n the face of
o a rider strraddling stepps, with the added beneefit of
additionaal positions present
p in the available speed
s profilee.

2.5.1.3 Criticism of CA
A Approach
Again, Hoogendoorn
H n offers som
me criticism of a propossed aspect off this projecct. He and B Bovy
assert thaat CA modeels’ “oversim mplified or incomplete
i bbehavioural rules prohibbit application to
complex situations where
w the miicroscopic behaviour of pedestrians is importannt” (Hoogenddoorn
& Bovy, 2003). In this project,, it is believ ved that the cconstrained,, corridor-baased environnment
haracteristic of escalators and movin
that is ch ng walkwayss provides a suitably rigid frameworrk for
a CA mo odel to work k. The pressence of lanees and balusstrades meanns that theree is little neeed to
worry ab bout the trajeectory and tu
urning behavvior of the peedestrians wiithin the sysstem, which is the
primary concern of Hoogendoo orn and Bov vy. Additiionally, the natural fit between reegular
spacing ofo the escalaator treads and
a cellular framework oof the CA m model lends credibility tto the
approach h taken in thiis paper.

2.5.1.4 Calibration off Model Param


meters
Howeverr, there remaain several important
i faactors to connsider duringg the construuction of the CA
model. OneO paper on o implemen nting car-folllowing proccedures disccusses issuess with calibrrating
the modeel and the im
mportance off preventing a behavior tthey refer too as “particlee hopping.” This
conditionn exists wheere platoons of lane channgers form iin somethingg of a “tailgating dance”” that
is a mannifestation off a “cooperaative ping-pong effect” that stems from all useers followinng the
exact samme decision--making proccess. Thereffore, if theree is a series oof users following each other
closely and
a the criteria for a lanne change arre met with all of them,, it is possibble to see seeveral
pedestriaans make lane chang ges back and a forth iin rapid suuccession ((Rickert, N Nagel,
Schreckeenberg, & Latour,
L 1996
6). In orderr to reduce the occurreence of this sort of eveent, it
becomes necessary to o add in com
mplicating faactors like deelays betweeen lane changes or even some
Peter D. Kauffmann Chapter 2:: Literature R
Review 22

small, ranndomized prrobability th


hat a user willl decide nott to change llanes even w
when it may be in
their bestt interest to do
d so.

2.5.2 Modeling
M Laanguage
In order to providee the capabiilities needeed by the pproposed simulation, a program ccalled
NetLogo was selecteed to constrruct the mod del. NetLoggo is at oncce a program ming languaage, a
compilerr, and a sim mulation sofftware. Baased on thee storied Loogo program mming langguage,
NetLogo is designed d to providee an object-ooriented appproach to prroblem solviing. By creeating
individuaal simulated d entities – called “tu urtles” – eaach turtle ccan be giveen ownershiip of
nteract with other turtlees and its ennvironment. For
characterristic variablles that it can use to in
instance, a turtle cou
uld be design nated a “com
mmuter” withh some user-defined dessired speed tthat it
wants to attain whilee traveling allong a movinng belt. Bassed on these variables annd the rules uunder
which the simulation n operates, thhe system as a whole exhhibits behavviors approacching those ffound
in the reaal world. Gllobal variablles can then be used to eextract systeem-wide parrameters fromm the
model (WWilensky & Tisue,
T 2004)).

In additiion to the benefits


b of the object-o oriented fram mework with individuaal user “turrtles”,
NetLogo provides an nother imporrtant benefit in that it proovides the caapability to ccreate a grapphical
display of
o the simulation enviro onment. Thrrough this ssimulation w window, the programmeer can
see the model
m in actiion and observe the pro ogression of turtles throuugh the systtem. Howevver, a
more imp portant beneefit of this feeature is thatt it allows thhe programm
mer to verify that the model is
behavingg as intended and that the t operation does indeeed approachh real-worldd behavior. It is
through verificationss such as th his that the calibration of applicable model paarameters caan be
performeed, as was deescribed in Section
S 2.5.1.4, above.

Finally, although NetLogo


N allo
ows for pro ogramming tto occur w within a CA frameworkk, the
software itself is wriitten to allow
w for actionns to occur iin a continuoous manner.. By assignning a
continuou us, decimal speed to eacch user ratheer than a dis crete value tthat is a mulltiple of the tread
spacing, additional points
p may be created in the speed pprofile. In thhis way, userrs may be abble to
travel a “half step”, putting theem astride tw wo consecuutive treads as is characcteristic of m many
transit ussers, as was described
d in Section 2.5.1.2, above.
Chapte
er 3 – Mod
del Develo
opment
The literaature review
w revealed a plethora of pertinent
p staatistics and pprinciples. E
Even thoughh little
direct ressearch existss on the specific topic of
o determininng the capaccity of a mooving belt faacility
through microsimula
m ation, by pulling the necessary inform mation fromm research coonducted in other
fields ennough data can be colllected to co onstruct the proposed m model. In this sectionn, the
developmment of the moving
m belt model
m will be
b describedd including aall necessaryy assumptionns and
the logicaal frameworrk contained therein.

In order to make thee resulting prrogram easieer to visualize and compprehend, thee model wass split
into two primary secctions. Firsst, there exists a SETUP UP routine too construct aand initializze the
model an nd its compo onent variabbles. Depen nding on the selections m made by thee user, the m model
can startt either emppty or popullated with pedestrians
p bbased on the inflow thaat was speccified.
After beiing properlyy initialized, the GO rou utine is whatt provides thhe actual funnctionality oof the
model. Finally,
F seveeral additionaal protocols have been e nacted to faccilitate data collection.

3.1 In
nitializatio
on
In order to properly initialize the model,, a SETUP routine waas developeed. This sset of
instructio
ons providess all the info
ormation thatt NetLogo nneeds to adeqquately definne the simullation
environmment and to configure
c itss initial pedeestrian state. The compoonents of the SETUP rooutine
are show
wn in Figure 3.1,
3 below.

Figuree 3.1: Flowcchart of “SETUP” Routiine


SETUP runs
r through
h several setts of processses in the coourse of initiializing the ssimulation sspace,
but becauuse of how it functions each
e iteration
n of SETUP P by definitioon must cleaar out any setttings
that existted previously. Thereforre, SETUP onlyo needs too be run oncce before thee operation oof the
model caan begin with h the GO com mmand.

Each of the
t moduless contained within
w SETUUP has a speecific function in the logical progreession
of model initializatiion. The fiirst three steeps process the user-deefined inputt data and m make
adjustmeents as neceessary wherre conflicts may exist, while the following steps definee the
simulatioon space. In
n this section
n SETUP willl be brokenn down basedd on the twoo processes w which
produce observable results – th he ones thatt define the simulation space – wiith the proccesses
involved in the prevvious three modules
m divided into thheir role in ddefining eithher the simuulated
belt or in g the system with pedesttrians. In orrder to proviide a better understanding of
n populating

23
Peter D. Kauffmann Chapter 3: Model
M Deveelopment 24

the inputts available within the model, the control boarrd as renderred in NetL Logo is show wn as
Figure 3..2, below. In this figuree, the green boxes
b repressent user inpputs, the tann fields are ooutput
dials, an
nd the blue buttons represent actions, includinng the aforeementioned SETUP andd GO
routines.

Figuree 3.2: Model Control Boaard in NetLoogo

3.1.1 Belt
B Charactteristics
Because of the rangee of capabilitties affordedd by the moddel, the proceess of defininng the simullation
environmment can req quire a fair amount of input on thhe part of thhe user. A Although exaample
scenarioss are provideed to give a series of dem
monstrationss to outside oobservers, foor someone uusing
the modeel for researcch purposes there is a great
g deal off flexibility iin its functioonality. In eeither
case, the model requ uires informaation about the
t belt as w well as the rrules that exxist on boardd it in
order to create
c the initial simulatiion environm
ment.

3.1.1.1 Belt
B Parameteers
The paraameters that define a moving
m belt system gen erally are thhose that peertain to its size,
overall operation,
o an
nd type of beelt that is being modeledd. To this ennd, the user is asked to input
the lengtth of the beelt in meters as well ass the speed of its operration in meeters per seccond.
Howeverr, because the t model is i being run n under thee principles of cellularr automata (CA)
operationn all of thesee values musst be converrted the cell--based units that will bee used througghout
the modeel in the “Co ompute Variaables” modu ule. This proocess requirees convertingg all lengths from
meters innto cells by dividing by y the cell side length, w which as preeviously statted is 0.4 m meters
along thee direction of travel. Sinnce this distaance was choosen based oon the standaard tread deppth of
an escalaator step, these units are called “treadds” throughoout the modeel.

The userr is additionally able to specify the type of bellt to be usedd in the moddel. The opptions
availablee are walkwaay, up escalaator, and dow wn escalatorr. While esccalators are divided baseed on
directionn of travel due
d to the hesitation
h ob
bserved in ppedestrians ffrom obscurring of the ffacial
ellipse, walkways
w aree not dividedd in this mannner becausee their inclinnation is keptt to a low ennough
level thatt the facial ellipse is not significantly
y violated.
Peter D. Kauffmann Chapter 3: Model
M Deveelopment 25

An additional capabiility of the model


m is the ability to siimulate behaavior on a sttaircase – albbeit a
narrow staircase
s – by creating an
a escalator--based simullation with a belt speedd of zero. Inn this
situation,, code contained in the “Adjust
“ for Constraints”
C ” module auttomatically eensures that there
are no peedestrians seet to be statiionary, as th
here would bbe in the casse of an esccalator or mooving
belt. It is this typ pe of verificcation that is containedd within thhe “Adjust” module soo that
impracticcal scenarioss are automaatically flaggged or adjustted.

3.1.1.2 Belt
B Rules
Several parameters
p are
a also avaiilable to the user to speecify rules thhat could be implementeed on
the belt. One such h rule includ des the optiion to speciify walk-onlly or stand--only restricctions
onboard the simulateed escalator or moving walkway.
w Thhis rule wass provided inn order to test the
effect of a behaviorall prohibition
n on the overrall capacity of the systeem. The userr can also diisable
lane chaanging funcctionality within the model m if the experimeental scenarrio calls forr the
independdent operatio
on of the beltt’s “lanes.”

An additiional conditiion that can be altered on


o board the belt is in reggards to the facial ellipsee that
impacts downhill
d esccalators and
d staircases. A slider is present on the simulatiion control bboard
(Figure 3.2)
3 that allo ows the useer to specifyy how manyy pedestrianss will choosse to observve the
facial elllipse princip
ple. The remaining useers will ignoore the desiire to leave extra free sspace
between them and the pedestrian n immediately in front.

3.1.1.3 Draw
D Simulation Environm
ment
Once all of this info ormation hass been colleected and prrocessed, thee model’s frramework caan be
constructted. Commaands are sen nt to NetLog go that definne the size oof each cell, or “patch,”” as it
calls them
m, and using g the user-deefined belt leength the proogram createes a graphic representatiion of
a belt surrface as show
wn in Figuree 3.3. In this figure, thee blue grid reepresents thee lane- and ttread-
based cell configuraation, compleete with shaading designned to makee the individdual cells visible.
Along thhe length of the
t belt, 40 distinct
d cellss are visible,, which at 0..4 meters perr tread repreesents
a horizonntal belt lenggth of 16 meters.
m Otheer purely aessthetic featuures present in the simullation
environmment are the black cells, representing g handrails, and the lighht gray cellss on either end of
the belt showing
s wheere arrival qu
ueuing and departure
d disspersion occurs.

Fig
gure 3.3: Sim
mulation Envvironment ass Rendered iin NetLogo

3.1.2 In nitial Pedestrian State e


Once thee belt has been drawn within
w the simulation
s sspace, the innitial pedestrrian state caan be
established. This process involv ves tabulatin
ng and proceessing the innput parametters that relaate to
pedestriaan quantity and
a behavio or. Followinng this step, the user ccan choose tto have NettLogo
automatically populaate the belt with
w a baselin ne scenario.
Peter D. Kauffmann Chapter 3: Model
M Deveelopment 26

3.1.2.1 Adjust for Con


nstraints
Using thee user-input data, the first thing that must be donne is to veriffy that the daata meets wiith all
logical constraints. As discusseed before, th his step invoolves performming adjustm ment of the input
data to en nsure that an
ny rules inclu
uded in the model
m are noot broken byy user inputss. An exampple of
this situaation is seen if the user sets
s the belt speed to zerro, which w would imply tthat a staircaase is
being sim mulated. Sin nce a staircaase cannot convey
c its paassengers along at speedd, the model will
automatically upgrade any ped destrians deesignated ass “standers”” to the “w walker” cateegory,
ensuring that the mod del continuees to operate smoothly annd efficientlyy.

3.1.2.2 Compute Varia


ables
Once thee adjustmen nts have beeen performed to ensuree that all ruule sets havee been folloowed,
pedestriaan variables may be computed fro om the raw w input dataa. This innvolves not only
convertinng walking and climbin ng speed uniits from metters per secoond to treadds per seconnd, as
discussedd above; it also includees calculatin
ng convertinng any relevvant percentaages from wwhole
numbers to their corrresponding fractions.
f

her adjustmeent is requireed in the callculation of the actual nu


Still furth number of peedestrians inn each
of the thhree pedestriian classes using the user-defined total inflow w rate and thhe percentagge of
turtles th
hat is are assiigned to eachh class. Class options arre describedd in Section 33.1.2.3, beloow. It
is importtant to note that
t there arre only two pedestrian
p m
mix sliders ppresent in thee model’s coontrol
panel (Fiigure 3.2). The
T first slid der controls the percentaage of the tootal inflow thhat is made up of
the walker class whiile the secon nd shows wh hat percentaage of the reemaining peedestrians – those
who are not standin ng – will bee defined ass fast walkeers, a class known as ccommuters. The
remainin ng percentag ge left on this second sllider represeents those w who are norm mal walkerss as a
fraction of
o those who o are not assiigned to the stander classs.

3.1.2.3 Pedestrian Parameter Seleection


For this model,
m threee classes of pedestrians
p have
h been eestablished too account foor different llevels
of user agggressivenesss. At the loowest end off this scale, ““standers,” aas their namee implies, chhoose
to stand still
s on the belt.
b Howev ver, a speed slider
s can stiill be seen foor these pedeestrians in F
Figure
3.2 sincee these userss still requirre some speeed that theyy observe w while walkinng along the gray
patches that
t constitutte the loadin ng and unloaading areas. In the simuulation envirronment, stannders
are show
wn as black wedges
w pointting in the diirection of trravel.

Next are the “walkerrs,” which poossess user-d


defined leveels of climbinng speed, waalking speedd, and
required free space behind
b in th
he opposite lane in ordeer to make a lane channge. Walkerrs are
representted by the co
olor green.

Finally, at
a the highesst level of agggressivenesss is the classs of pedestrrians known as “commuuters.”
This nam me was not chosen
c to refflect any lev
vel of judgmment on transsit system commuters except
perhaps tot show thatt they have thet most reasson to be huurried in theiir travels thrrough ambullatory
facilities like transitt stations. Consequentlly, commuteers are expected to be set to the most
aggressiv ve, speed-baased tendencies, meaning g higher leveels of walkinng speed and climbing sspeed
along wiith a lower level of spaace required to completee a mergingg operation. Commuterrs are
displayed d as red wed
dges.
Peter D. Kauffmann Chapter 3: Model
M Deveelopment 27

3.1.2.4 Populating thee Belt


Using the previously y establishedd pedestrian class param meters, the ppedestrian mmix present iin the
simulatio
on environm ment can be fully defined. On the ccontrol boarrd (Figure 3.2) there exxists a
hat causes the simulation
switch th n to automatiically populaate the belt w
with pedestrrians if desireed by
the user. When thiss module is activated, the t model crreates pedesstrians of eaach class thaat are
randomlyy distributedd along the leength of the belt. The qquantity of ppedestrians tthat are geneerated
is based on a Poisson n distributio
on centered around
a the eexpected vallue of pedesstrians that wwould
be presennt on the beelt under thaat given beltt length andd operation speed. Thee lane in whhich a
given peddestrian is geenerated is determined
d based
b on othher user-seleccted lane asssignment opttions.

Each ped destrian witthin the system is creaated with innitial speed, merging, aand facial ellipse
parameteers appropriaate to its classs. Additionnally, each ppedestrian is set to point to the right since
this is deefined as the forward dirrection in thee model as wwell as set too the approprriate lane. A
At the
same tim me, each ped destrian is initialized
i with
w several other counnt variables to keep tracck of
parameteers such as itts presently desired
d speed and the tim
me since it laast changed lanes.

An exammple of the reesulting simuulation environment is sshown below w as Figure 3.4. Note thhat in
this moddel the forwaard directionn will alwayys be to thee right, in thhat pedestriaans will enteer the
simulatio
on from the left and pro ogress forwarrd along thee belt until tthey exit on the right annd are
removed from the environment
e t and counteed. It can also be seeen that undeer the condiitions
imposed when this figure was generated, all a walkers – the blackk wedge shaapes – are aat the
bottom of
o the simullation. How wever, when n the directtion of travvel is considdered it beccomes
apparent that this meeans they aree traveling in
i the right llane and aree pointing foorwards alonng the
model. Walkers
W (greeen) and com
mmuters (redd) can likewiise be seen inn the left lanne.

Fiigure 3.4: Peedestrian Po


opulated Sim
mulation Envvironment ass Rendered inn NetLogo

3.2 Operation
O
Followin ng the complletion of the SETUP rou utine, the moodel is ready to run. In oorder to begiin the
simulatioon process, the GO ro outine must be selectedd. Two opptions exist to executee this
procedurre. The userr can either select the “GGo Once” buutton to proogress the model throughh one
“tick,” th
he name giveen by NetLo mulation timee step whichh in this moddel is one seccond,
ogo to a sim
or they can
c select th he “GO” buttton. GO is what NetLoogo calls a “forever buttton,” meaniing it
will conttinue to run n until the user
u stops th
he simulatioon, an errorr occurs, or the simulattion’s
ending crriteria are met.

Figure 3..5 shows a flowchart


f of all the modu ules containned within thhe GO routinne. At the eend of
the chartt, a diamondd block can be
b seen. In flowchart nnotation, ovaals representt start and ennding
command ds, rectanglles represen
nt actions or o modules, and rectanngles with bbars on thee end
representt subroutines. Howeverr, the diamo ond block reepresents a decision strructure, usuaally a
true or false
fa decision n. In this case,
c the deccision presennt in the GOO routine exxists to see iif the
routine’s ending cond ditions have been met by y the end of a given simuulation secoond.
Peter D. Kauffmann Chapter 3: Model
M Deveelopment 28

Figu
ure 3.5: Flow
wchart of "G
GO" Routinee

3.2.1 Innducing Mo ovement


In develo
oping the mo
odel, the firsst goal was to
t induce moovement. Inn early versioons of the m model,
before th
he implemen ntation of following
fo annd passing rrules, pedesttrians present in the syystem
simply walked
w unim
mpeded along g their originnal trajectorry at their orriginal speedd. This behhavior
caused pedestrians
p to pass rightt through oth her, slower moving peddestrians beccause the loogical
framework that causes a real-wo orld human to t slow dow wn upon enccountering annother pedesstrian
had not yet
y been prog grammed.

Howeverr, these early steps of model


m devellopment werre necessaryy in order tto verify thaat the
underling
g frameworkk of the model function ned satisfacttorily beforee more advaanced subrouutines
were addded. While pedestrian
p characteristiccs like desireed speed hadd been accouunted for thee belt
populatio
on module (SSection 3.1.2
2.4), the facct that the m
model involvees pedestriann movementt on a
moving belt
b surface meant that thet speed off the belt muust be includded in addition to the rellative
walking speed.

The proccess of com mputing the total speed d simply invvolves addinng the relattive speed tthat a
pedestriaan achieves through
t climmbing to thee belt speed that exists iin the simullation. This total
speed rep presents of a pedestrian n on the bellt as seen byy outside obbserver, wheereas the rellative
speed is just
j that – reelative to thee movement of the belt – so this speeed is as if it w
were observeed by
someone else standin ng stationaryy on the belt..

With all the passeng


gers on the belt
b traveling g at their apppropriate tootal speed, thhere remains one
more factor that must be accountted for beforre additional process modules can bee implementeed on
Peter D. Kauffmann Chapter 3: Model
M Deveelopment 29

the system
m. Although it may seeem as though h all pedestriians created in Section 33.1.2.4 contaain all
the necesssary parameeters and beh
havioral charracteristics ffor model opperation, in rreality all thaat has
been programmed in nto them thuus far is theirr desired sp eed. An addditional variable needs to be
included in the frammework of every
e pedesttrian that reepresents their current speed, sincee any
obstruction – like a slower walk ker on the beelt – will byy rule cause a conflict bbetween the user-
defined speed
s of thee pedestrian
n and the am mount of disstance that the pedestriian is capabble of
advancinng.

In this way,
w even if an entity on
n the belt iss slowed dowwn for somee reason, theey will still have
enough “knowledge”
“ ” to know whhat speed thhey should acccelerate bacck up to oncce the obstruuction
is passed
d. However, since at th his point in the
t model development following bbehavior (Seection
3.2.3.1) has
h not been n implementted, the curreent speed vaariable will be initially set to the deesired
speed vallue.

With the above logicc in place, model


m develoopment couldd finally proogress to addditional moddules.
The speeed logic conntained with hin this secction has beeen includedd in the dettailed moveement
procedurres contained
d in Section 3.2.3.1.

3.2.2 Boundary
B Operations
O
Once mo ovement of the simulatted pedestriaans was achhieved, it became timee to implem ment a
number of boundary y operationss to deal with
w the inteerface betweeen the beltt and the siimple
walking environmen nt. As show wn in Figurre 3.5, the mmodules thaat represent these condiitions
involve pedestrians
p entering
e thee system and
d arriving att the belt, ppedestrians ddeparting thee belt
and bein
ng removed from the sim mulation env
vironment, aand finally tthe procedurres by whicch the
GO routiine itself is teerminated.

3.2.2.1 Arrivals
In order to assure coontinued operation of th he model inn successive time steps, new pedesttrians
must be introduced tot the simulaation space. This functiion is perforrmed by thee arrivals moodule.
In this module,
m pedesstrians are generated at the
t entry endd of the belt and assigneed parameterrs in a
process very
v similar to the belt population
p procedure
p deescribed in S Section 3.1.22.4, above. Each
entity is assigned a heading and d lane basedd on its pedeestrian classs along withh a desired sspeed
based onn its walking
g capabilitiess and an initial speed eqqual to this ddesired speedd. Dependinng on
i that simulation secon
the queue situation in nd and whethher the userr has instruccted the moddel to
ignore laane assignmment under certain
c unbaalanced connditions, thee pedestrian may insteaad be
assigned to a randomm lane.

As beforre, the quanttity of pedestrians with


hin each classs is determ
mined by thee total numbber of
pedestriaans within that
t class th
hat will be generated within a oone-hour sim mulation, ass was
calculated in Section
n 3.1.2.2. Sppecifically, in each secoond the nummber of geneerated pedesttrians
within a given class is defined by a Poisson--distributed random num mber centereed around a vvalue
equal to the number of one-hourr arrivals divvided by 36000 seconds. In practice, the sum of these
one-second Poisson-rrandom arriv vals value neearly alwayss sums up too the originall one-hour arrrival
quantity, as would bee expected of
o a sum of Poisson-distr
P ributed randoom numbers.
Peter D. Kauffmann Chapter 3: Model
M Deveelopment 30

Once gennerated, the pedestrians are adjusted


d backwards to ensure thhat they are iin the back oof the
queue, should one exist. Thesse pedestriaans will lateer be goverrned by thee “Move Quueue”
procedurres discussed
d in Section 3.2.3.1.
3

3.2.2.2 Departures
D
Just as peedestrians must
m be insertted into the simulation eenvironmentt, they must bbe removed from
the simullation enviroonment to ap pproximate real-world
r bbehavior. Thhe departurees module chhecks
the belt system for any pedestrrians that haave progresssed past thee end of thee belt treadss and
removes them from the model space using NetLogo’s DIE comm mand. As thhey are remooved,
several global
g count variables coorresponding g to the varioous pedestriaan classes tabbulate the exxiting
pedestriaans. If this module
m is in
nspected (coomplete moddel code is ccontained inn Appendix A A), it
can be seeen that this module also o contains so
ome computaation of systtem-wide vaariables for uuse in
the modeel’s data colllection operaations.

3.2.2.3 Ending
E Condittions
The simu ulation can be
b terminated d in one of several
s wayss. Many of tthese methodds were obseerved
during th he model development
d t, as any error
e messaage dealing with the eexecution of the
simulatioon’s calculattions will haalt the GO routine. Foor instance, if the modeel comes uppon a
section of
o unexecuttable code or o has to perform
p an undefined m mathematicaal operationn like
dividing by zero, then the simulaation will sto op. Howeveer, the more ccommon proocedure by w which
the modeel exits its iteerative execu
ution proceddure is contaained within the diamondd block preseent in
the GO routine
r flow
wchart (Figu ure 3.5). Th his module represents tthe logical ddecision bettween
continuinng the modeel for another iteration ofo one simulaation secondd or stoppingg the GO rooutine
after thatt particular cycle. The decision maade at this ppoint in the routine is bbased on thee exit
conditionns that goverrn the modell.

There aree two exit co


onditions forr the GO rou utine in this model. Thee first condittion checks tto see
if the user-defined maximum
m qu
ueue has beeen reached,, which wouuld mean thhat the queue has
grown to o be so lonng that it is approachin ng the edgess of the simmulation envvironment. This
conditionn is known as
a queue oveerflow, and if i the queue were to reacch the edge of the simullation
window it could cau use NetLogo to crash. The T second ccondition chhecks that thee elapsed tim me of
the simuulation, as measured
m by the cumulaative numbeer of ticks thhat have been tallied by the
model, does
d not exceeed the user--specified simulation lenngth. As can
an be seen inn Figure 3.5, each
iteration of the GO routine increases the ticck counter bby one throuugh the incllusion of a T TICK
module. Therefore,, it can be said that GO G will perfform the saame set of pprocedures eevery
simulatedd second unttil the exit co
onditions aree reached.

3.2.3 Im mplementiing Pedestrrian Interaction


Although h great proggress had beeen made up to this poinnt in the devvelopment oof the modeel, the
configuraation still diid nothing too prevent peedestrians frrom simply walking thrrough each oother.
Clearly, rules of ped destrian interraction weree needed beffore the moddel could bee checked aggainst
empirical capacity vaalues. To peerform this function,
fu the following aand passing rrules for highhway
ystems found
traffic sy d during thee literature review
r (Secttion 2.2) woould need too be implemented
within thhe context off this pedestrrian model.
Peter D. Kauffmann Chapter 3: Model
M Deveelopment 31

3.2.3.1 Following
Fo Beh
havior
Followinng rules werre implemen nted first. As
A can be seeen in the G
GO routine fflowchart (F Figure
3.5), there are two modules that require pedestrians
p tto move, sppecifically ““Move Belt”” and
“Move Queue.”
Q Ass their namees imply, Move
M Queue involves thhe progressioon of pedesttrians
waiting to
t board the belt surface while Movee Belt goverrns the moveement of peddestrians whho are
have alreeady boardedd the belt. The generall procedures involved inn the two following behhavior
modules can be seen in Figure 3.6.

Figure 3.6: Flowcharrt of “Move Belt” Subrooutine


It is important to noote that the procedures
p shown
s in Fiigure 3.6 rellate specifically to the M
Move
Belt sub broutine and d that slighht differences exist beetween Movve Belt andd Move Queue.
Specificaally, since Move
M Queue applies in th he horizontaal waiting sppace before entering thee belt,
pedestriaans are not subject to fatigue
f as th
hey would bbe when clim mbing a staairs so the iinitial
“distancee traveled for
f speed drrop” decisio on and resuulting “Slow w with Distaance” subrooutine
implemen ntation (Section 3.2.4.1)) is ignored.
Peter D. Kauffmann Chapter 3: Model
M Deveelopment 32

Although h all other logic remain ns in place, since Movee Queue occcurs by definition withiin the
context of
o a queuing situation in most cases the subsequeent decisionns will consisstently yieldd little
to no heaadway and consequently y no ability to
o accelerate . Therefore,, forward progression wwill by
and largee stem fromm gaps that open as ped destrians aree able to booard the beltt at the inteerface
between the queue an nd the belt.

An additiional factor that must bee considered at the interfface betweenn the queue aand the belt is the
change that
t pedestrrians experieence betweeen their deesired walkinng speed aand their deesired
climbingg speed. Oncce a pedestriian crosses onto
o the beltt, the model marks that ppedestrian aas “on
belt” which triggers several actio ons. First, the
t pedestriaan’s desiredd speed is chhanged from their
maximum m walking sp peed to theirr maximum climbing speeed. In the case of the sstander classs, this
means thhat their desiired speed iss instantaneoously set to zero. For oother classess, this meanss that
their maaximum desiired speed will w decreasse; howeverr, since thiss is speed iis relative too the
movemen nt of the beelt their totall speed willl likely increease. This increase in average speeed is
what keeeps a queue from develo oping at the base of the belt in unsaaturated connditions. Annother
function of the “on belt”
b markerr is that it teells the moddel that that particular ppedestrian iss now
subject to
o the Move Belt
B and Lan ne Change su ubroutines raather than M Move Queue procedures.

3.2.3.2 Passing Behavvior


After thee implementaation of follo owing rules, the operatioon of the moodel was obsserved to deggrade
significanntly from the previous itteration of th
he model. HHowever, thiis drop in peerformance oof the
simulated d belt system
m was necessary to apprroximate reaal-world behaavior, and itt in fact servved to
bring thee observed throughput
t of
o the modeel’s simulateed escalatorss and movinng belts closser to
empirically determin ned capacitiees from the various
v studi es found in Section 2.4.1 of the literrature
review.

Howeverr, one final step


s still rem
mained, and that was thee implementtation of lane changing rrules.
Although h it is comm mon practicce on escalaators and m moving walkkways that those remaaining
stationary
y relative to
o the belt sh hould stay in
i the outsiide lane witth those peddestrians acttively
walking or climbing along the belt sticking to the insidee, in reality people do nnot always addhere
to these guidelines.
g Therefore, thhe “Lane Chhange” subrooutine was ddeveloped inn order to proovide
a structure to pedesttrian passingg within the simulation. The logicaal steps conttained withiin the
Lane Chaange modulee are shown in Figure 3.7 7.

Figure 3.7: Flowchart of “Lane Chhange” Subrroutine


Peter D. Kauffmann Chapter 3: Model
M Deveelopment 33

The Lanee Change su ubroutine beg gins by calcu ulating the oopen space aaround each pedestrian iin the
system, both
b ahead of
o and behind d the pedestrrian in both the current and adjacennt lane. Nextt, one
of two prrocedures is run to deterrmine if a lan ne change iss warranted, depending oon which lanne the
pedestriaan is currentlly traveling in. These prrocedures arre shown in Figure 3.8. Also includded in
this figure are the three
t sets off rules that are used bby the lane change check algorithm ms to
determinne which ped destrians willl benefit froom a lane chhange. Firstt, the pedestrrians are scaanned
to ensuree that they are
a in the app propriate lanne and that tthey have not recently cchanged intoo that
lane. Seecond, pedesstrians eligib ble for a righht lane channge are checcked to see if they are bbeing
“tailgated
d” by a faster pedestrian n who is folllowing clossely behind. If this is tthe case, theey are
immediattely flagged d for a lan ne change. Finally, pparameters for the rem maining eliigible
pedestriaans are passeed through four
f tests based off of thhe TRANSIM MS lane chaange rules. If all
four condditions in thiis set are met, the pedesttrian is flaggged for a lanee change.

Figure 3.8:
3 Flowcha
art of Lane C
Check Algoriithms
Peter D. Kauffmann Chapter 3: Model
M Deveelopment 34

After thee model has evaluated all a pedestriaans and flaggged the ones who woulld benefit frrom a
lane chaange, all peedestrians whow qualified for a sshift are m moved to thhe adjacent lane
simultaneeously. Th he reason th hat the lanee change ccheck algoriithms are pperformed oon all
pedestriaans on the belt
b before any pedestrrians are mooved is so that no peddestrians gaiin an
advantagge in their deecision proceess as a resu
ult of their pposition on thhe belt or anny other arbitrary
parameteer. Thereforre, all pedesttrians make their
t decisioons based onn the gaps arround them aat the
same insttant in time,, thereby enssuring that an
a open gap for a walkerr will not be filled by a ffaster
moving commuter
c wed to make a decision, for instance. In this wayy, the
beefore the waalker is allow
GO routiine sequencee and protoco ols (Figure 3.5)
3 are mainntained.

3.2.4 Accounting
A for Compliicating Facttors
Even after implemen nting pedesttrian interacction protocools in the mmodel, there still remainned a
number of
o other com mplicating faactors that haad previouslly been idenntified in the literature reeview
(Section 2.1.2) that needed
n to be addressed.

3.2.4.1 Pedestrian Lim


mitations
The firstt complicatiing factor to
t be impleemented in the model came from m the pedesttrians
themselv ves. Human ns naturally will tire unnder the straain of prolonnged physiccal exertion, so a
module wasw added to o the GO rou utine to causse the simulaated pedestrrians to exhibbit the sympptoms
of tiredness. This beehavior is manifested
m in
n the model in a reductioon of the deesired speed for a
given peedestrian oncce they hav ve traveled a certain di stance. In this model,, the “Slow with
Distance” routine (sshown in Fiigure 3.5) performs
p thiss function. When enabbled, the rooutine
causes peedestrians to
o slow by onne tread per second afteer traveling 220 meters hhorizontally aalong
the belt and another tread per second forr every addiitional 10 m meters, as ssuggested byy the
literaturee.

Howeverr, in order to implemeent this module a distiinction needded to be m made betweeen a


pedestriaan’s initial desired
d speed
d and the sp peed they deesire at a givven momentt in time. Inn this
way, thee “speed dessired now” variable
v waas initially s et equal to the “speed desired” byy that
pedestriaan, but after 20
2 meters th he equality was
w altered sso that the sppeed desiredd now was innstead
equal to the speed desired
d minuus one tread d per secondd. After 300 meters, thhis became sspeed
desired minus
m two trreads per seecond, and so s on. Thiss process waas implemennted to makke the
subroutinne more intu uitive and eassier to program.

3.2.4.2 Bottleneck
B Facctors
or shown in the literature, especiallyy in the Hooogendoorn exxperiments, is the
Another critical facto
importannce of modeeling the imppediment to o travel that exists at a bottleneck. In many oof his
corridor experimentss, Hoogendo oorn observ ved a noticeeable delay in pedestriaan travel whhen a
substantiial volume of
o pedestrianns was funn neled down through a bottleneck ((Hoogendooorn &
Daamen, 2005). Theerefore, it is imperative that this moodel should eexhibit a com mparable levvel of
hesitation
n from bothh the bottlen
neck and thee interface tthat exists bbetween stattionary floorr and
moving belt.
b

Howeverr, when com mparing the operation off the model and the ressulting capaccities determ mined
therein, it
i appears th ulation does indeed apprroximate thee queuing annd delay inccurred
hat this simu
by travellers as they traverse the system baseed on the caapacities shoown in the liiterature (Seection
2.4.1). It is believ ved that thiss similarity has two ccauses. First, althoughh Hoogendooorn’s
Peter D. Kauffmann Chapter 3: Model
M Deveelopment 35

experimeent featured pedestrians


p converging ono the bottleeneck from nnumerous directions andd only
merging at the entryy to the botttleneck, peddestrians andd especially transit commmuters exhiibit a
common set of courrtesies to on ne another ini their norm mal routine, and one oof these is qqueue
disciplinee. Transit commuters have been observed tto form disttinct queuess at the base of
escalatorrs under cru
ush loading conditions while attem mpting to exxit transit sttations (Davvis &
Dutta, 20002), much like the behaavior exhibiteed by the peedestrians in this model.

Furtherm
more, the delaay caused by y pedestrian hesitation aat the interfaace between a stationary floor
and the moving
m belt is approximaated quite well
w by a com mpletely inaddvertent featture of the m
model.
Since thee transition of
o each pedeestrian between the Movve Queue annd the Movee Belt procedures
causes thhem to verify y that a suitaable gap exiists on the beelt ahead off them beforee boarding, some
pedestriaans are delayyed ever so slightly
s in th
hat they mayy have to wait a secondd before findding a
gap that fits their neeeds. This occasional
o delay
d very cllosely approoximates thee behavior thhat is
observedd when a diveerse stream of o pedestrianns attempts tto board a mmoving belt ffacility.

3.2.4.3 Belt
B Parameteers
Still another complicating facto or that has been
b implemmented in thhe model deeals with thee belt
parameteers themselves, as was im mplemented all the way back in the model initiaalization phaase of
developmment. It is important
i to
o recall that the literaturre review reevealed that facial obsccuring
(Section 2.1.2.2) can n cause discoomfort to sttair climberss or pedestriians travelinng upwards oon an
escalatorr if they follo
ow too closeely behind th
he passengerr in front of tthem. In ordder to accounnt for
this tendeency in the model, somee amount off space mustt be reservedd in front off pedestrianss who
choose too observe thee facial ellip
pse concept.

Since thee user is ablee to define whether


w a belt is travelinng in the upwwards directiion as well aas the
percentag ge of pedestrrians who chhoose to obsserve the faccial ellipse (F
Figure 3.2), tthese prefereences
can be immplemented in the modeel. First, exttra code wass added in thhe Populate B Belt and Arrrivals
modules to determin ne which ped destrians – ifi any – wouuld be randoomly selecteed to observve the
facial elllipse through a combination of ran ndom probabbilities and the belt typpe present in the
simulatioon. If these conditions were
w satisfieed, the pedeestrian wouldd be assigneed a facial ellipse
size of onne tread. If not,
n the faciaal ellipse varriable wouldd be set to zeero.

The Mov ve Belt routine was then n modified to


o include an additional ddistance equual to the vallue of
the faciall ellipse variiable to each
h computed headway
h term
m present inn the logical decision treee. In
this way,, a pedestriaan who obserrves the faciial ellipse cooncept will bbegin to slow
w earlier in order
to preserrve one tread d of spacing between theemselves annd the forwaard pedestriaan when travveling
on an uph hill facility.

3.3 Data
D Collecction
With thee model perfforming in a way that accurately
a poortrays realiistic behavioor, steps muust be
taken to ensure thatt analyses can
c be perfo ormed on siimulations thhat are condducted using the
model. This
T means that data caan be collectted followinng a simulattion run as wwell as at reegular
intervals during the run. To do this, sev veral outputts were connstructed aloong with seeveral
subroutinnes that perm
mit the user to
t expand the functionaliity of the moodel as a whhole.
Peter D. Kauffmann Chapter 3: Model
M Deveelopment 36

3.3.1 Real-Time
R Outputs
O
As was seen
s in the NetLogo-rend
N dered contro Figure 3.2, seeveral real-time outputs were
ol board in F
placed on the contro ol panel forr easy viewiing. In adddition to dissplays showiing tallies oof the
number of o pedestrian ns who had entered and d exited the system throough that insstant in timee, the
results of other com mputations contained wiithin the “C Create Plots”” module off the GO rooutine
(Figure 3.5)
3 were shown. Theese calculateed parameteers includedd the currennt and maxiimum
observedd queue leng gths, as welll as the dennsity and innstantaneouss flow rate exhibited byy the
model. A running av verage flow rate was alsso displayedd once the m model had beeen given tim
me to
stabilize into a steaddy condition, which wass defined as occurring aafter the beltt had a channce to
clear itseelf of station
nary pedestrrians five tim
mes. The llast real-tim me display “ddial” showed the
number of o simulated seconds thaat had passed d since the sttart of the sim
mulation.

Beside thhe array of dials,


d two ploots were also included tto allow the user to trackk the progreession
of these and other variables
v thrroughout thee simulationn period. Inn addition tto the previously
mentioneed density, flow,
f queue,, and pedesttrian count vvariables desscribed abovve, a secondd plot
was created to show the averagee speed of eaach class of pedestrians for all mem mbers presenntly in
the system
m. This plot was includ ded to show the
t level of iimpedance ththat the walkker and comm muter
classes experienced
e through a comparison n between thhe user-defifined definedd speed andd the
observedd average speeed.

3.3.2 Time-Space
T e Diagrams
Another method to observe
o the progressionn of users thhrough the ssystem and to determinne the
level of impedance experienced d as a resultt of obstrucctions like th
the bottlenecck, the floorr-belt
interface, and slowerr pedestrianss is through a time-spacce diagram. This type oof plot show ws the
trajectory
y of a singlle entity within the sysstem, with tthe elapsed time on the x-axis and the
distance traveled dispplayed alongg the y-axis. Because o f this setup, the speed oof the traveleer can
be determmined at a gllance by observing the sllope of the pproduced linne at any poinnt in time.

Procedurres were impplemented in


n the model that
t permit thhe user to trrack up to thrree pedestriaans at
any one time in a reeal-time tim
me-space diaggram. The model is caapable of ranndomly seleecting
destrians; however, the user
three ped u also hass the option oof overridinng these randdom selections by
enabling the “Select Ped” subro outine and ch
hoosing up to three of their own too track in a third
plot.

Unfortun nately, the data


d contain
ned within thet NetLogoo produced plots is traapped withinn the
model, making
m this information n impossiblee to export in this form mat. Additiionally, the plots
themselvves are difficcult to interprret, especially over a lonng time interrval. To oveercome this iissue,
a “File Output”
O subro outine was created
c that gives
g the useer the optionn to export aall relevant m
model
data to ann external teext file for an
nalysis or pllotting in anoother prograam. If activaated, this extternal
file will report the simulation sy ystem inputss for the purrposes of reepeatability. Additionallly, at
one-second intervals NetLogo will w print the results of thhe various pperformance variables ass well
as the po osition, speeed, and desiired speed of o every pe destrian witthin the system. Usingg this
informatiion, all the plots
p produceed in NetLog go can be repproduced ouutside of the program.
Peter D. Kauffmann Chapter 3: Model
M Deveelopment 37

3.3.3 Capacity
C An
nalysis
Finally, since
s one off the major pu urposes of th
his model w was to determmine the capaacity of a mooving
belt systeem under diffferent inputt stream charracteristics, a final subrooutine was addded to alloww the
user to conduct
c a “CCapacity Teest” of the entire
e systemm. When tthis module is activatedd, the
simulatioon automaticcally takes precautions
p to
t ensure thhat queue ovverflow cannnot occur. T To do
this, the model
m monittors the queu ue length to see if it apprroaches the mmaximum quueue. If it ppasses
the warn ning value – by default, 75 pedestrians in eitherr queue lanee – the systeem automatiically
removes the extra members
m of the
t queue. However,
H b efore it does this it tallies the offennding
queue members
m into
o a global “queue
“ remooved” count variable. This quantitty of pedesttrians
removed from the qu ueue is addeed to the qu ueue length iin all displayays and plotss featuring qqueue
length.

The reaso on for ensurring that the model will never


n experiience queue overflow whhen perform ming a
capacity test is so thaat the inflow
w rate can inittially be set to an arbitraary high valuue on the firsst test
run undeer a new set of parameteers. Even with w the infloow set to a high value, the belt willl still
only be able
a to passs a certain number
n of peedestrians nno matter hoow many aree waiting inn line.
Therefore, although the inflow is well in excesse of thhe observed flow exitingg the queuee, this
observed d flow could d then be useed as the inflow in the second iteraation of the simulation. The
outflow isi continuallly used as th he inflow inn subsequennt iterations of the capaccity test unttil the
system reaches a po oint where th he queue is observed too reach steaady-state connditions. A At this
point, thee inflow andd the outfloww will be equual and the syystem will bbe at its maxiimum sustainnable
capacity.

The capaacity level deetermined ussing this procedure shouuld be the maaximum flow w where the input
and output volumes match,
m meanning that thee queue shouuld act in a rrelatively staable mannerr. An
example of this queu ue behavior can be seen n in Figure 3.9, below. This plot is the resultt of a
simulatioon performed d on an uphiill belt with 50% of peddestrians observing the fa facial ellipse at an
inflow off 4300 pedesstrians per hoour, which was
w the valuee computed by the Capaacity Test moodule
to be thee maximum sustainable flowf under these
t condittions. Althoough the queeue length caan be
seen to rise
r and fall with the sto ochastic natuure of arrivaals under thee model, it remains relattively
stable at around 20 people
p in botth lanes, witth times of llonger queuees and periods with no qqueue
at all. The
T importan nt behavior to note in th he plot, howwever, is thaat the queuee length doees not
grow con nstantly overr the course of the entiree hour mode led in this siimulation, wwhich would have
been the case in an oversaturated d simulation environmennt.
Peter D. Kauffmann Chapter 3: Model Development 38

40
35
30
Queue Length

25
20
15
10
5
0
0 300 600 900 1200 1500 1800 2100 2400 2700 3000 3300 3600
Elapsed Time (seconds)

Figure 3.9: Plot Showing Queue Growth under Steady-State Conditions at Capacity
Chapte
er 4 – Results and Discussio
D on
With thee model fully y developed d and in wo orking order,, it could finnally be utiilized to sim
mulate
pedestriaan behavior in
i moving beelt systems. As will be seen below,, the model hhas the capaability
to answer various ressearch questiions proposeed at the begginning of thhe project wiithin a controolled,
repeatablle frameworrk. Followin ng a series of
o observatioons designedd to validate the perform mance
of the mo n terms of itts representaation of pedeestrian choicce behaviorss as well as in its
odel, both in
ability to
o match empirically determined capaacity values uunder the saame input paarameters, seeveral
additionaal application
ns were testeed.

While the purpose off the model, as it was orriginally connceived was to compute the capacityy of a
moving belt
b system under
u region
nally determmined input pparameters, tthe model w will be seen bbelow
to be cap
pable of cond ducting seveeral other annalyses. Thee applicationns investigatted in this seection
include innvestigating
g the sensitiv
vity of a prop
posed belt soolution to chhanges in its inputs, analyyzing
belt systeems under proposed
p opeerational rulee sets, and ddetermining the requiredd platform siize to
handle qu ueuing under crush loading conditions.

4.1 Model
M Valid
dation
In order to verify thaat the modell created durring this prooject accurattely represennts human chhoice
behavior and the reall-world cond ditions obserrved in a mooving belt syystem, a seriies of tests hhad to
be conduucted. These validation n proceduress were impleemented to ccheck that tthe diverse sset of
behavioral source daata as well as the proccedures and rule sets ppulled from a wide rangge of
transporttation researcch indeed prrovided a suiitable basis ffor the simullation.

There weere two steps involved in i this validaation processs. First, thee operation oof the model was
assessed through a number
n of visual
v meanss to ensure that the sim mulated pedeestrians obseerved
practical behaviors ini terms of following behavior,
b paassing choicce, and in thhe level of delay
present at
a the floor--belt interfaace as a ressult of bottleeneck and bboarding efffects. Afteer the
operation
nal practicess had been verified,
v the model’s
m outp
tput performmance characcteristics couuld be
compared d to the capaacity that waas observed under empirrical studiess found in thhe literature w
when
run under the same in nput parameeters.

4.1.1 Operation
O
Based onn informationn and data collected during field stuudies that weere found in the literature, the
observed
d operationaal behavior of the mod del could aadequately bbe compared to empiriically
ned human behavior with
determin h the goal off verifying thhe model’s ooperation. T
The data thaat was
gleaned from the literature wass not limiteed to numerrical results; rather, maany examples of
diagrammmatic and photographic
p c examples of pedestriaan behaviorr were contained withinn the
sources found
f during
g the literaturre review.

In this way,
w queuingg regimes annd pedestriann flow formations couldd be observaationally verrified.
The visuual manner byb which thiis validation
n process waas conductedd occurred inn much the same
way as thhe following g and passin
ng behaviorss, which weere verified tthrough an inspection oof the
processedd by which lane changiing was execcuted and h eadways weere maintainned in the faace of
obstructin
ng pedestriaans.

39
Peter D. Kauffmann Chapter
C 4: Reesults and D
Discussion 40

4.1.1.1 Following
Fo Beh
havior
Althoughh no video evidence
e wass found in th he literaturee, following behavior is common ennough
in everyd
day life for this verificaation step to
o be perform med based oon personal experience. The
main validation step that needed d to be condducted was too ensure thaat the prograammed folloowing
logic mattched up witth that whichh was found d in the TRA ANSIMS liteerature and thhen subsequuently
programmmed into thee model and then to veriffy that normal followingg behavior occcurred.

First, thee model waas observed d in operatiion under a wide rangge of input parameterss and
operationnal rule scheemes to ensuure that the following
f rulles were exeecuted propeerly. Fortunaately,
the objecct-oriented nature
n of NeetLogo mean ns that thesee rules were graphicallyy displayed iin the
simulatioon environmment, much as a they are seen
s in Figuure 4.1, beloow. This figgure providees the
opportunnity to validaate that thesee TRANSIM MS rule sets are observeed by the moodel as well as to
visually verify
v that th
hey match reeal-world con nditions.

Fig
gure 4.1: NettLogo Simullation Showiing Followinng Behavior
For instaance, in the simulation
s second
s showwn in Figure 4.1 a platooon of comm muters, in redd, can
be seen forming
f in th
he left lane – at the top of
o this figuree – behind a walker, in ggreen, becauuse of
the lowerr climbing sppeed assigneed to the waalker. Since it is known that the direection of travvel in
the simullation is from
m left to rig
ght across thee screen, com mmuters jusst entering thhis platoon aare at
the leftm
most edge of the
t figure, with
w the lead ding walker ccloser to the right of the figure. It can be
seen thatt those commuters just entering thee platoon arre observed to leave a llarger gap – two
treads – between theemselves and d the commu uter in frontt than those commuters further alonng the
belt, whoo only leave one tread.

The reason for this behavior


b is because
b the rearwards ccommuters aare still travveling at a hhigher
nd therefore the TRANS
speed, an SIMS logic is causing thhem to leavve a larger gap forward. The
commuteers farther ahead
a along the belt haave had to sslow down because off the walkerr, and
because of
o this drop in speed onlly need to leave one emppty tread to aaccount for aany sudden ddrops
in speed they may en ncounter. This
T behaviorr is similar tto how autom mobile driveers reserve llarger
forward headways
h att higher speeds to accouunt for the nnecessary reeaction and ddecelerationn time
required in that situattion.

It is also interesting to note that in the simullation shownn in Figure 44.1 the belt direction waas set
to downh hill, meaning
g that the faccial ellipse was
w not a faactor. Thesee gaps observved in this ffigure
exist soleely to leave decelerationn room in th he case of a sudden stoop on the paart of the leaading
pedestriaan. The faciial ellipse cooncept becom mes importaant when useers are tighttly packed oonto a
belt in saaturated conditions, as thhose pedestrrians who chhoose to obsserve the faccial ellipse eeffect
will be coompelled to leave an em mpty tread in front of them
mselves for personal com mfort.

In the co
ourse of condducting this evaluation process,
p an additional effect was obbserved wheen the
Slow witth Distance module wass activated by b the user. Because oof the meanss by which users
slow dowwn in the reeal world as a result of fatigue, thee effective thhroughput of which a bbelt is
Peter D. Kauffmann Chapter
C 4: Reesults and D
Discussion 41

capable decreases
d ass its length in
ncreases. This is becauuse a longer belt will seee its riders tiire as
they proogress forwaard along its length, causing
c a ccorrespondinng drop in average rellative
pedestriaan speed tow wards the un nloading endd. Above a certain floow rate, thiss speed dropp will
cause a shockwave to form and d travel bacckwards dowwn the lengtth of the beelt. Beyondd this
shockwav ve front are users who area constrain ned by the ddecelerated ppedestrians. Therefore, even
though walkers
w and commuters
c d not slow until they nnear the end of the belt, other pedesttrians
do
g further bacck will still experience delays as a platoon of slow movving riders fforms
traveling
behind thhese fatiguedd pedestrianss.

4.1.1.2 Passing Choicee


Because of the level of personal preference present in m making lane change deciisions, more rigid
verificatiion of thesee procedures was requiired. How wever, becauuse of the llack of literrature
regardingg passing behavior
b onn stairwayss and escaalators as w well as wiithin constrained
passagewways like mo oving belt systems
s in general,
g the primary foccus of this vvalidation w
was in
ensuring that the TR RANSIMS passing rules were follow wed. To thiis end, a simmilar processs was
undertaken as was seen in thee validation of followinng behaviorr in Sectionn 4.1.1.1, abbove.
Specificaally, the behhavior of th
he model un nder a widee variety off operationall conditionss was
visually inspected
i fo
or conformityy. An exammple of the mmodel that hhighlights paassing behavvior is
shown inn Figure 4.2.

Figure
F 4.2: NetLogo
N Sim
mulation Showing Passinng Behavior and Gap Accceptance
In this fiigure, the left travel lane at the top of the figurre is limitedd by a walkeer (green) wwho is
completin ng a pass off a platoon of
o standers (black
( wedgges) in the riight lane. H
However, beccause
walkers are defined d with a slo ower climbiing speed tthan commuuters (red) are, a queuue of
commuteers has form med behind the leading g walker. Their desirred speed inndicates thaat the
commuteers want to go faster bu ut cannot make
m a pass to the righht because oof the platooon of
standers in that lane. In this scen nario, passin ng decisionss can be undderstood throough the pressence
of the thrree commuteers present in n the right laane, who willl now be insspected one at a time.

The com mmuter locateed farthest back


b on the belt – that iis, closest too the left of the figure – was
observedd to make a change
c into the left lanee in next simmulation secoond. This beehavior is loogical
because this
t pedestriian meets thee minimum eligibility ruules of beingg on the belt and havingg free
space open to the lefft in addition n to all fourr of the TRA ANSIMS lanne change crriteria outlinned in
Figure 3..8. The firstt of these criiteria, a lack
k of free spacce in the currrent lane, caan be seen inn that
there is only
o one open cell in frront of this pedestrian.
p Secondly, tthere can be seen to be more
free spacce available in the oppo osite lane. Next,
N trafficc in the adjaacent lane is observed to be
moving fast
f enough to permit an n improvem ment in speedd in the left lane. Finallly, adequatee free
space is available beehind in thee left lane to o allow a m merge while satisfying thhe preferredd free
space reqquired by a commuter of o three free treads undeer this operaational scenaario. Since all of
these tests are passeed, a lane ch hange will be b conductedd at the apppropriate poiint in the ovverall
structure of the GO routine.
r
Peter D. Kauffmann Chapter
C 4: Reesults and D
Discussion 42

Next, thee middle com


mmuter in thhe right (botttom) lane off Figure 4.2 iis inspected. This pedesstrian
was obseerved to havve recently changed lanees to the righht in an attemmpt to pass the commuter in
front of him in the left lane sinnce that commmuter’s sppeed was lim mited by thee leading waalker.
Howeverr, as happen ns in the reall world, he did
d not lookk far enoughh ahead befoore merging right
and was therefore blocked
b by the
t platoon of standers before he ccould pass tthe source oof his
slowdowwn.

Finally, the
t forwardm most commu uter in the rig
ght lane of F
Figure 4.2 has been stucck at that possition
on since he entered the belt nearly twenty simuulation secoonds ago. Inn that
in the staander platoo
timeframme, no accepttable gap hass yet presentted itself in tthe adjacentt lane to merrge left, and so he
has beenn trapped in the slow lan ne for the en ntire journeyy. This behhavior indicaates an addittional
factor thaat merits furrther researcch (Section 4.3), that off increasing aggressivenness exhibiteed by
users as the amount of delay theey have incu urred increaases. Driverrs and pedesstrians in thee real
world have been obseerved to chaange their behavior in thee face of proolonged obsttruction, so ffuture
study cou uld examinee this effect on moving belt systemss to see if ppassenger preference chaanges
with timee.

4.1.1.3 Floor-Belt
Fl Inteerface Effectss
The finall area of model operation that must be inspectedd for real-woorld functionnality is the point
of interfaace between
n the station
nary floor an nd the moviing belt surfface. This area is show wn in
Figure 4..3, below. As
A has been discussed previously,
p thhere are twoo significant impacts thaat this
location in
i the modell has on pedeestrian behav
vior.

Figurre 4.3: NetLogo Simulattion Showingg the Floor-B


Belt Interfacce
First, thee hesitation observed
o in pedestrians
p when
w it commes time for them to trannsfer betweeen the
floor and d the belt is directly observed in th his model inn how pedesstrians use thheir own intternal
decision process to judge
j when an acceptin ng a gap preesent in fronnt of them tto board the belt.
Although h this behaviior is based on a combiination of peedestrian observance of the facial ellipse
and the TRANSIMS
T S following rule
r set, in practice
p it seeems to clossely model thhe behaviorrs and
decisionss made by acctual pedestrrians using reeal belt systeems.

The seco ond cause of delay at the floor-belt in nterface resuults from the fact that thee entry to a bbelt is
by definiition a bottlleneck in thaat it represeents a constrriction fromm free movem ment on a trransit
platform to the narro ow belt. Th he belt itself provides rrigid boundaaries to the movement of its
riders thrrough the veertical balustrrades that ho
old the systeem’s handraiils in positioon. In typicaal belt
systems, these balusttrades are usually separaated by a disttance of one meter.

In experiiments by Hoogendoorn and validateed by otherss, bottleneckks like the onnes at the enttry to
a belt weere observed to have a siignificant efffect in limitiing the rate aat which peddestrians cann pass
through a constrictedd system, as seen in Secction 2.1.2.3 . However,, as was disccussed in Seection
3.2.4.2, it is believeed that the set of info ormal rules that governns transit paassenger queeuing
behavior provides a framework that t is simillar to that ussed in the quueuing regimme of this mmodel.
Peter D. Kauffmann Chapter
C 4: Reesults and D
Discussion 43

Since traansit passen


ngers typicallly queue in n a relativelly linear faashion at thee entrance tto an
escalatorr or moving walkway
w (D
Davis & Duttta, 2002), thee effects of tthe bottlenecck are minim
mized
since thee pedestrianss only havee to constricct from the narrow queuue into the belt. This is as
opposed to constricting into a corridor from m a wide areea of free sppace as was measured iin the
Dutch exxperiment (HHoogendoorn n & Daamen n, 2005).

One areaa of future research (Section 4.3) proposed a fter the couurse of this project invvolves
developinng a modulee for implemmentation in the
t model thhat incorporaates pedestriian wayfindiing in
crowd sccenarios. Th
his module could
c then bee used to moodel pedestriian behaviorr in mergingg onto
the belt by
b way of thee bottleneck
k created by the
t balustraddes.

4.1.2 System Outp puts


Followinng the verification of the model’s operation
o in the previouus section, itt became tim me to
assess the validity off the various outputs produced by thee simulationn. Even withh the operatiion of
the modeel having beeen successfu ully checked d against both real-worlld behavior aand experim mental
observatiions found in n the literatu
ure, if the ou
utputs produuced by the mmodel do noot similarly m match
their exp
perimental co ounterparts then
t the model itself coould not be llabeled as a success. Too this
end, this section willl investigatee the accuraccy of the moodel’s compputed capacitty values rellative
to those determined in empiricall studies. Attention
A willl also be paiid to the moodel’s outputt data
routine and especially y to how thee resulting daata can be ussed to investtigate pedesttrian trajectoories.

4.1.2.1 Comparison to
o Empirical Capacities
C
Perhaps the most im mportant valiidation that needed to bbe performedd before thee model couuld be
approvedd was that capacity. Ass was mentioned beforee, the modell was createdd to simulatte the
behavior of pedestrians on a moving
m belt system,
s withh one of thee major endd goals beinng the
determinnation of thatt belt’s capaacity under a user-defineed set of inpuut parameterrs. Howeveer, for
this goal to be satisfaactorily met,, the belt cap
pacities deteermined by tthe model thhrough simullation
had to bee checked aggainst empiriical sources found in thee literature.

Even thoough one off the stated goals


g of the model was to separatee the practice of movingg belt
design frrom the exissting methods of utiliziing capacityy curves andd empirical data sourcess, the
accuracy
y of the simu
ulation comppared to real--world instaallations musst be ensuredd. It is onlyy after
the propoosed model has been prroperly veriffied that a mmathematicall basis can bbe brought tto the
process by
b which mo oving belts are
a selected and
a specifiedd for use in aambulatory ffacilities.

In order to use the model


m to calcculate capacity through simulation, the Capacityy Test subrooutine
(Section 3.3.3) had to o be activateed. With thee capacity teest switch sett to “on,” the model couuld be
initialized with the input
i parameeters set forrth in the litterature. Whhile the studdies found iin the
literaturee review werre clear abouut the parameeters under w which the beelt was operaated, it was ffound
that seveeral studies provided in nsufficient data
d about iimportant chharacteristics like pedesstrian
speed, ag ggressiveness, and choice behavior to completelyy set the inpputs of the m model as desccribed
in Chapter 3. In theese cases, av verage valuees of normall human exeertion and asssumptions aabout
choice beehavior weree used, someetimes in ad ddition to vaalues from otther studies conducted iin the
same geo ographical reegion. For instance,
i a study
s on clim
mbing speedd conducted at a Londonn-area
university y (Fujiyamaa & Tyler, 20004), describbed in Sectioon 2.1.2.1, w
was used to pprovide addittional
informatiion to a cap pacity studyy conducted on escalatoors in the saame system (Davis & D Dutta,
2002) feaatured in Secction 2.4.1.
Peter D. Kauffmann Chapter
C 4: Reesults and D
Discussion 44

4.1.2.1.1 Capacity Sccenario 1


The first scenario thaat was invesstigated was based on F ruin’s measuurements onn a downhilll belt.
By settin
ng the modell to 20% stannders and 600% of the waalkers being commuters with lane chhange
allowed and no rulee restrictionss, a capacityy of approxximately 66000 pedestriaans per hourr was
observedd on an escaalator operatting at 0.4 m/s.
m This vvalue is veryy close to F
Fruin’s compputed
capacity of 6400 peddestrians per hour (Goodm man, 1992).

Since thee Fruin studdy did not liist the averaage behaviorral parameteers exhibitedd the pedesttrians
during th
he measurem ment period, the input paarameters in this scenarioo were varieed to see how w the
simulatioon’s computed capacity changed. As A it turned out, very siimilar capaccities were ffound
even wheen changing g the speed preferences
p and lane chaanging aggrressiveness pparameters oof the
various pedestrian
p classes. It is believed
b thatt the reason for this is thhat under satuurated condiitions
there is insufficient
i room availaable on a moving belt ssystem for ppassing to occur. Moreeover,
since passsing is next to impossibble in these conditions
c it appears thaat the entire ppedestrian sttream
is limited
d by the preesence of walkers
w sincee the users tthat desire tto move faster are unabble to
achieve these
t speeds. Therefore, it is believed that the ccapacity of th the belt is tieed in large ppart to
the speed d at which the belt op perates, therreby makingg the inform mation gatheering precauutions
describedd above largely unnecesssary.

4.1.2.1.2 Capacity Sccenario 2


A second scenario was then analyzed to determine tthe effects of the faciaal ellipse onn the
observedd capacity. When the model
m was set
s to operatte in an uphhill directionn under the same
pedestriaan parameterrs with 80% of pedestriaans set to obbserve the faacial ellipse,, the result w
was a
capacity of 3800 ped destrians perr hour. Datta presented by Turner iin his sectioon of the Vertical
Transporrtation Hand dbook show wed a lowerr limit of caapacity for escalators ooperating at 0.45
meters per second to o be 4051 pedestrians
p per
p hour (Tuurner, 1998). Althoughh this simullation
lacks the resolution necessary
n to model a bellt speed of 00.45 meters pper second, it is believedd that
the slighttly lower cap
pacity found
d at a slightly
y lower speeed in the sim
mulation mattches the findings
of the em
mpirical studyy.

4.1.2.1.3 Capacity Sccenario 3


A final scenario
s wass simulated with
w the passing modulee disabled too mimic the theoretical study
of lane-sspecific cap pacity condu ucted by Davis
D and D Dutta. Thiis study used mathematical
approachhes to capaccity, which fortunately provided am mple pedesttrian and beelt speed daata to
duplicatee their proceedures in simmulation forrm. The m model compuuted a belt ccapacity of 7200
pedestriaans per hour,, whereas thhe study foun
nd a capacityy of 6600 peedestrians peer hour (Davvis &
Dutta, 20 002). It is believed that the reaso on for this ddiscrepancy is that the theoretical study
assumed a gap of tw wo treads between
b wallking pedesttrians in ordder to accouunt for clim mbing
motion. Conversely,, the simulattion determin ned that at thhis flow ratee the belt woould be conggested
and the users
u would not be ablee to sustain a walking paace and insttead was ablle to separatte the
pedestriaans on the beelt by a gap of
o only one tread.
t

4.1.2.2 Creation of Tim


me-Space Dia
agrams
In order to gain a beetter understtanding of the model’s operation, tthe capabilitty to create time-
space diaagrams was implemented
i d into the model in Sectiion 3.3.2. A
As discussedd in that sectiion, a
time-spacce diagram is a plot thhat shows th with elapsed time
he trajectoryy of a moviing entity w
Peter D. Kauffmann Chapter 4: Results and Discussion 45

plotted on the x-axis and the distance traveled since time zero on the y-axis. This type of plot is
beneficial because the interaction between travelers within a system over can be determined from
a static plot. Additionally, by inspecting the slope of the various traveler trajectories, the speed
of an individual pedestrian can be determined since distance traveled over time elapsed is speed.

Although NetLogo is capable of generating its own plots, these figures lack clarity and are not
suitable for detailed analysis. Therefore, the file output module described in Section 3.3.2 was
employed to export simulation data for use in this analysis. In order to generate a suitable time-
space diagram for analysis, the model was allowed to run for sixty seconds in a low flow
environment of 1200 pedestrians per hour under the input parameters present in Example
Scenario 0, described in Appendix B, Section B.4. The second-by-second pedestrian data
contained in the output file were plotted in Microsoft Excel, and an excerpt from this plot is
featured as Figure 4.4, below.

55
Horizontal Position (no. treads)

50
Stander 3
45 Stander 4
Stander 9
40 Stander 12
Commuter 16
35 Commuter 17
Walker 20
30 Commuter 21

25
35 40 45 50 55 60
Time Elapsed (seconds)

Figure 4.4: Time-Space Diagram Showing Following and Lane Change Behaviors
Figure 4.4 shows the trajectories of eight different simulated travelers observed during a time
window of 25 simulation seconds over a 30-tread (12-meter) length of belt. In this figure, the
pedestrians are individually color coded, with solid lines representing travel in the right lane and
dotted lines indicating travel in the left lane.

The first conclusion that can be drawn from this figure is the distinct speed preferences of each
of the three pedestrian classes. The four standers are all observed in this scenario to have a
constant slope of 1, indicating that they are traveling at one tread per second (0.4 meters per
second), which is the belt speed in Example Scenario 0. Likewise, the orange line representing
Walker 20 can be seen to have a constant slope of two treads per second which is equivalent to
the belt speed plus its climbing speed of 0.4 meters per second. The three commuters present in
this scenario represent a more interesting situation, as their relatively aggressive behavior causes
some of them to come into conflict with the other users in the system. While Commuter 16
exhibits a constant speed of three treads per second through a combination of the belt speed and
its climbing speed of 0.8 meters per second, this is because it is traveling in the left lane for its
Peter D. Kauffmann Chapter
C 4: Reesults and D
Discussion 46

entire jou
urney througgh this sectio
on of the beelt and there fore does noot need to ddecelerate whhen it
passes thhe two walkkers travelinng in the rigght lane. H However, thee other two commuters both
experiencce slowdow wns when th hey come up pon slower travelers inn their curreent lane andd are
compelleed to take acttion to maintain their desired speed, as will be ddiscussed bellow.

Commuter 17 represents perhapss the most in nteresting caase in the timme-space diagram show wed in
Figure 4.4. When it i enters thiss plot at 38 seconds, itt is exhibitinng a speed of one tread per
second, much
m below its desired speed
s of threee treads perr second. TThis is becauuse it is folloowing
two tread ds behind Stander
S 12. At time sttep 41, the dotted line in the figuure indicatess that
Commuter 17 entereed the left laane to pass Stander
S 12, merging bacck right afteer completinng the
pass. For this pedesttrian, a “commpleted pass”” is indicatedd when it haas achieved a rear headw way of
three treaads to satisfy
y the TRAN NSIMS lane change
c ruless discussed iin Section 3.2.3.2. How wever,
once bacck in the rig ght lane Com mmuter 17 quickly enccounters anoother standeer. Ordinariily, it
would immmediately switch
s back left again, but
b under thhe parameterrs of Exampple Scenario 0 all
pedestriaans must waiit five secon nds since theeir last lane change in oorder to be eeligible to chhange
again. At
A time step 51, these fiv ve seconds have
h elapsedd and the Commuter 17 ppasses Standder 9,
remaininng in the left lane for the duration of the time-spaace diagram..

Commuter 21, on thee other hand d, appears to be in confliict with Wallker 20 whenn it enters F Figure
4.4 at tim me step 49. However, this behaviior can be eexplained byy the fact thhat Walker 20 is
travelingg in the leftt lane with Commuter 21 progresssing in thee right lanee, indicatingg that
Commuter 21 is execcuting a passs in the rightt lane. Com mmuter 21 is able to traveel at its full sspeed
of three treads
t per seecond until it begins to encounter thhe effects off Stander 122 at time steep 54.
Howeverr, since Com mmuter 21 haas not chang ged lanes reccently it is im
mmediately able to swittch to
the left laane and resu
ume its full sp
peed, passin
ng Stander 122 and Standeer 9 in the prrocess.

It is in th
his way that time-space diagrams
d maay be created from the mmodel’s outpput data andd used
to investtigate the peerformance ofo the modeel. Throughh the exampples describeed above, a brief
glimpse into
i the logiic and proceedures contaained within the model tto govern foollowing andd lane
change choice
c behavviors can be found. Using observaations made of this and other time-sspace
diagramss, the validattion of the model
m can be brought to a successful conclusion.

4.2 Potential
P Application
A ns
Througho out the proccess of develloping and assessing
a thee performance of the moodel, the priimary
focus waas on its orriginal purpo ose of deterrmining thee capacity oof a belt system througgh an
effective simulation environmen nt. With thiss goal comppleted, other potential appplications oof the
model deetermined du uring the validation process can be e xplored.

4.2.1 Sensitivity Analysis


A off Belt Systemms
The first and perhapss most obvio ous applicatiion of the m
model is to peerform sensitivity analyssis on
proposed d solutions. Already in n Section 4.1.2.1, the ccapacities coomputed by the model were
subjectedd to such a sensitivity analysis by varying thee input paraameters to see what chaanges
resulted in
i the resulting maximu um flows. However,
H in ppractical dessigns it is unnlikely that a belt
specifiedd by a comp petent engineeer will be forced to opperate at cappacity in alll but perhapps the
most extrreme peak lo oading condiitions. Insteead, by perfoorming sensiitivity analyssis on facilitties at
other, lesss critical loaad levels it will
w still be possible
p to seee the belt’ss performancce under a vaariety
Peter D. Kauffmann Chapter
C 4: Reesults and D
Discussion 47

of input parameters. By subjectting both prooposed and existing faccilities to sennsitivity anaalysis,
the engin
neer can gain
n a better und
derstanding of the suitabbility of the iinstallation.

Another reason that analyzing th he capacity ofo saturated systems is oof less importance than those
operatingg under slighhtly lower innflows is thee fact that att capacity, thhe behavior of passengeers on
belt systeems makes thet capacity y of the systtem inherenttly less senssitive to channges in the input
parameteers. At high her flow ratees, pedestrian
ns exhibit hhigher packinng efficienciies, meaningg that
each indiividual pedeestrian has leess room available for m maneuveringg than they oordinarily w would.
At saturaation, the rid
ders are packked so tightlly onto the bbelt that lanne changes aand even waalking
becomes impossible.. In the abssence of thesse confoundding behaviooral factors, the capacityy of a
system iss almost excclusively driven by the belt
b speed. Changes in the input sttream in term ms of
pedestriaan mix, choicce, and clim
mbing capabillities will m make little diffference heree, except perrhaps
in the casse where lugggage carryinng can be appproximated through usee of the faciaal ellipse featture.

In order to
t investigatte the sensitiivity of a mo
oving belt syystem under nominal conditions, useers of
the modeel need only y to vary thee input streamm parameterrs in ways thhat could pootentially bee seen
over the life of the installation and track the t resultingg performannce metrics. This partiicular
applicatio
on would perhaps
p be most
m useful in belts wiith occupanncies rangingg from 20 tto 40
percent of
o cells utilizzed, as this density rang
ge was identtified in the literature ass being the rrange
where paassing and other
o choice behaviors have
h the mosst impact onn the operatiion of the syystem
(Blue & Adler, 2001 1), as describbed in Sectioon 2.1.2.5. Using thesee techniques, a transit aggency
could usee the model to see whatt level of infflow stream change cauuses operatioonal problem ms for
an existin
ng system orr exceeds thee desired perrformance m metrics for a proposed onne.

4.2.2 Analysis
A of Proposed
P Rule
R Implementation
Another feature of th he model thaat can be useed in belt deesign is the iinclusion off rules that caan be
implemen nted on thee simulated belt. Thesse rules werre included to accountt for the typpe of
restrictions that the literature
l inddicated weree put on beltt facilities byy various traansit operatoors or
potentially by the ped destrians theemselves. By activating a rule or coombination oof rules withiin the
simulatioon environm ment, their efffect could beb noted onn the capacitty of the beelt. Additionnally,
through observation
o of the modeel runs and the
t use of tiime-space ddiagram outpputs, the effeect of
proposed d rules on thee performancce of the bellt could be asssessed as wwell.

4.2.2.1 Pedestrian Beh


haviors as “R
Rules”
The mosst rudimentaary rules co ontained wiithin the moodel are the ones thatt affect how w the
simulatioon deals withh pedestrian behavior. Itt can be saidd that model parameters like the optiion to
enable oro disable paassing behav vior could beb considereed a “rule” that a trannsit agency ccould
impose ono its patron ns. Other qu uasi-rules thhat are contaained withinn the model are the abiliity to
cause ped destrians to adhere to sttrict lane asssignment prootocol. Connversely, a bbelt’s riders ccould
instead be
b instructed d to ignore their
t lane assignment if the queue iin their lanee was longerr than
that of th
he adjacent laane by somee quantity deefined by thee user of the model. It iss also possible for
the user to
t instruct th
he pedestrian ns to ignore their assignnment altogeether and chooose their laane at
random anda let passiing behaviorr sort out thee pedestrianss on the beltt itself. A fuurther exampple of
this sort of rule set includes
i thee “walk-left//stand-right”” behavior thhat is observved on escallators
and moviing walks an nd indeed onn highway sy ystems as weell.
Peter D. Kauffmann Chapter
C 4: Reesults and D
Discussion 48

4.2.2.2 Sttand-Only an
nd Walk-Onlyy Restrictions
In this seection, standing and walk king restrictiions will be investigatedd. It was hyppothesized tthat if
a transit agency chose to implem ment a rule on an escallator or movving walkwaay that forceed all
pedestriaans to remain n stationary on the belt, the numberr of pedestrians that the belt was cappable
of handling would increase. A potential usage for tthis scenarioo may be uunder emerggency
evacuatio on protocolss or in the case of severre crush loadding on the part of pedestrians tryiing to
exit a traansit facility
y. If it weree possible fo
or enough ussers to disem mbark from a transit veehicle
onto a platform
p to bring that platform to o its capaciity, dangeroous conditioons might ooccur.
Specificaally, pedestrians at the edge
e of the crowd may be thrown off of the platform andd onto
electrified rails or other
o hazard
dous locatioons as the ccrowd uncoontrollably sshifts due too the
shockwav ve behaviorss present in high-density
h y crowds (Frruin, 1984).

To addreess this reseearch questiion, walk-on nly and stannd-only rulees were impplemented inn the
model. Activating the t stand-on nly restrictio
on would caause all peddestrians to become stannders
whereas activating the
t walk-onlly rule wou uld adjust thhe pedestriann class voluumes in ordder to
bump alll standers into the walkeer class. A series of triaal runs was then conduccted to assesss the
effect off walking annd standing restrictions on the oveerall capacitty of a sam mple belt thrrough
simulatio
on. The scen narios descriibed below can
c be foundd in Section BB.4 of Appeendix B.

4.2.2.2.1 Rule Scenaario 1 (Examp


ple Scenario 2 in Section B.4)
The first simulation that
t was con nducted to adddress the reesearch quesstion was treeated as a coontrol
in that no
o walking orr standing reestrictions were
w enabled on the belt. This scenaario took place on
an uphill escalator with 95% of o passengeers observinng the faciall ellipse using the stanndard
pedestriaan mix, speeeds, and chaaracteristics that
t were ussed in the caapacity scennarios featurred in
Section 4.1.2.1.
4

The reaso on an uphilll escalator was


w used in thist scenarioo was to appproximate thee platform eegress
situation described above.
a It was
w envision ned that thiss scenario took place inn a subterraanean
transit station, and thhe only meaans of escap ping the crow wded platforrm was a baank of escalaators.
Under th his particularr uphill escaalator simulaation, the caapacity of thhe belt was ffound to be 3700
pedestriaans per hour.

4.2.2.2.2 Rule Scenaario 2 (Examp


ple Scenario 4 in Section B.4)
Next, thee scenario was
w repeated except with h a rule impllemented whhere users arre only perm mitted
o the belt. No walking
to stand on g or climbing g is permitteed. Althoughh it is uncerttain how a trransit
agency would
w be ab
ble to enforcce this restriiction, a cappacity of 44000 pedestriaans per hourr was
observedd, indicating a slight impprovement in n throughputut. Based onn observationn of the moddel in
this scenario, it was determined that this imp provement inn capacity wwas a result oof the removval of
imperfections in thee packing of o the belt due d to walkking. As w with any opptimization-bbased
situation,, the removaal of perturb
bations fromm the system m has the efffect of improoving the ovverall
performaance.

4.2.2.2.3 Rule Scenaario 3 (Examp


ple Scenario 5 in Section B.4)
Finally, the
t scenario o was run under
u walk-oonly rules. With all ussers climbinng the escalaator’s
stairs using either waalker or commmuter preferrences, the bbelt was obseerved to exhhibit a capacity of
3800 ped destrians peer hour. Th his slight immprovementt over the bbaseline sceenario was again
believed to be a resullt of the redu
uction in varriability that resulted from the presennce of standeers in
Peter D. Kauffmann Chapter
C 4: Reesults and D
Discussion 49

the origin
nal case. However,
H thee fact remain
ns that the ccapacity wass lower thann that observved in
Rule Sceenario 2 from
m packing in standers alo
one.

4.2.3 Platform
P Sizzing
Based on n the analyssis performed in Section n 4.2.2.2, abbove, the moodel can alsso have a roole in
moving belt system can have
platform sizing for trransit facilitiies. As wass seen in thatt section, a m
the effecct of limitingg the rate att which peddestrians cann exit a conffined area. Since the m model
tracks the growth off the queue at o the belt, aas was seen in Figure 3.9, it can also be
a the base of
used to determine
d thee maximum number of pedestrians
p tthat will be ppresent in thhis location uunder
peak houur crush loadding conditio ons.

In addition to the prreviously diiscussed sceenario wheree transit pattrons are in danger of bbeing
shoved off
o of a transiit platform, another
a situaation where it becomes iimportant too assess the qqueue
length th b facility is when bellts are placeed in sequennce. This soort of
hat exists at a moving belt
situation is found wiith escalatorrs in deep transit stationns and mid-rrise office buuildings andd also
with moving walkw ways in distrributed facillities like aairports. Acccording to the literatuure, if
designerss fail to leave adequatee queuing sp pace aroundd escalators and movingg walks at m major
activity centers,
c the continual drriving action n of the bellt may causee passengerss to be convveyed
into areaas where insu ufficient spaace exists to accommodaate them, esspecially undder emergenncy or
evacuatio on situations. This situ uation preseents a publiic safety haazard since the only waay to
prevent oneself
o from
m being forceed into a pack ked queuingg area is to w
walk backwarrds along thee belt
which would
w requiree the cooperration of all belt passenngers since tthose riders just enterinng the
belt will likely be un
naware of thee situation att the downsttream end (M Mansel, Mennaker, & Harrtnett,
1998).

Therefore, it is of grreat importaance for eng gineers and architects too consider pplatform areaa and
other connditions likee queuing arrea at landinngs betweenn moving beelt facilities in the desiggn of
ambulatoory facilities. Fortunatelly, in additio
on to being able to plot queue behaavior over timme as
was disccussed in Seection 3.3.3, the model is capable of tracking the maximuum queue leength
recorded in a given simulation.
s A screensho ot of the moodel as rendered in NetLLogo is showwn in
Figure 4.5, below. In I this figurre, the instan
ntaneous andd maximum m queue lenggth meters caan be
wards the upp
seen tow per left, with
h the current queue preseent in the moodel shown in the simullation
window near
n the botttom.

As was seen
s in the description of the Capaacity Test mmodule (Secttion 3.3.3), at the maxiimum
throughpput of a beltt facility it can be saidd that a steaady state quueue exists. However, it is
importannt to note thaat a maximum queue len ngth determiined in this wway is only indicative oof one
possible outcome under that arriv val conditionn. In order to determinee the maximmum queue leength
with con nfidence, sim
mulations must
m be run using diffeerent random m number sseeds in ordder to
account for the rand domness thatt is inherentt in stochasttic arrivals. Using thiss queue lenggth in
conjunctiion with thee amount off space requ uired per peedestrian in order to preevent shockkwave
behavior from occurrring (Sectio on 2.1.1.2), the
t total am mount of areaa required too handle queeuing
for a giveen moving belt
b system can be determ mined.
Peter D. Kauffmann Chapter
C 4: Reesults and D
Discussion 50

Fig
gure 4.5: Exxcerpt of NetL
tLogo Interfa
ace Showingg Queue Simuulation and Tabulation
Finally, it
i is also imp
portant to co
onsider the way
w that peddestrian flow w peaks and eebbs over tim me in
an ambullatory facilitty. The connfiguration of
o the entry tto the escalaator or moving walkwayy can
have a great
g impactt on the waay in which pedestrian arrivals aree distributedd. Thereforre, by
designingg a transit platform
p or the
t entry areea of a publlic space in such a wayy that arrivalls are
spread ouut over time due to varyiing walking distances, thhe peak infloow that a mooving belt faacility
must be designed to o handle cann be reduced d, as discusssed in Sectiion 2.1.2.3. Additionally, at
these hig
gher levels of
o inflow, thee growth off a belt’s queeue can and should be ttracked overr time
using this module to o determine how
h much storage
s spacee is requiredd to hold thee queue baseed on
the duration of time that
t oversatu urated condittions are exppected to exiist.

4.3 Suggestion
S s for Furth
her Researrch
During thhe course of this projecct, three add ditional reseaarch questioons presenteed themselvees for
which the model wass incapable of o determiniing a solutioon. These prrinciples aree put forth heere in
the hopess that anotheer, future pro
oject will be able to addrress them.

4.3.1 Belt
B Perform mance in Terms of De elay
The perfo
formance meetrics used by y this modell are almost exclusivelyy designed foor use in cappacity
analysis and in ensuuring that thee fundamenttal characterristics of thiis traffic flow system caan be
monitoreed. To that end,
e the mod del features real-time
r dissplays and fille output prootocols to prresent
informatiion about sttatistics likee instantaneoous flow, sppeed, and ddensity. Thhe maximum m and
average values
v of flo
ow througho out the simullation periodd can be useed to draw cconclusions aabout
the capaacity, while additional displays thaat show queeue length measuremennts can helpp the
engineer to make decisions abou n and sizing of platform
ut the design ms and landinng areas bettween
belts.

Howeverr, another peerformance metric


m that may
m prove uuseful in connducting anaalyses of mooving
belt systtems is thaat of delay.. Delay iss a commoon measure of effectivveness for other
Peter D. Kauffmann Chapter
C 4: Reesults and D
Discussion 51

transporttation facilities, allowing


g a designerr to determiine what levvel of impeddance a userr will
experiencce when passsing through h an intersecction or transit station. T
Therefore, itt is proposedd that
the modeel be adjusted to compu ute the delaay incurred by each peddestrian as tthey traversse the
modeled moving bellt system. This T particulaar value couuld be computed as a difference bettween
the travell time and th
he amount off time that thhey would hhave taken w walking at theeir desired speed.
Separate values could mine delay inncurred on thhe belt and inn the queue..
d be computted to determ

The resuulting delayy measures for the enttire system could provve to be annother impoortant
performaance characteristic for th
he engineer to consider when desiggning an am mbulatory faccility.
The incluusion of thiis performannce metric would
w repreesent a signiificant shift in the apprroach
taken by
y engineers in the desig gn of movin ng belt systeems, as up until this ppoint the priimary
concern in the desig gn of these systems haas been cappacity. This paper hass also previously
proposed
d designing moving
m belt systems to achieve
a a deensity of no more than oone passengeer per
square meter
m – an occcupancy off 40% – in order
o to allow
w pedestriann choice behhavior on thee belt
during normal periods. Accoun nting for ped
destrian delaay in a systtem in addittion to thesee two
measuress of effectiveness could provide a better
b undersstanding of tthe proposedd solutions tto the
problem of pedestrian n locomotion.

Further research
r willl need to beb conducted d before thiis proposal can be impplemented. Most
notably, study is merrited to deterrmine wheth her the compputed delay ppresent on a moving bellt will
even be significant enough
e to track. The qu uestion for tthis study too answer willl be to findd how
much delay is incurrred by a sin ngle pedestriian on a conngested beltt and whetheer this amouunt is
enough to o cause a ch
hange in behaavior. It is entirely
e posssible that peddestrians willl be unconceerned
with a deelay of fifteeen or twentyy seconds co ompared to ttraveling onn an uncongeested belt att their
desired speed so long g as the capaacity exists on
o the belt syystem for them to reach their destinaation.
If furtheer study sho ows to indeeed be the case, then it may be that the incclusion of delay
computattion is not a significant factor
f to connsider in bellt design relaative to the iissues of cappacity
and occuupancy.

4.3.2 Microsimula
M ation of Boottleneck Efffects
Although h the bottleneck effect uttilized in this model apppears to give a good apprroximation oof the
level of constriction
c exhibited in
n actual escallator and mooving walkw way systems,, the fact rem
mains
that this effect is laargely serenndipitous. Therefore,
T iit is propossed that a ffuture project be
undertaken to attemp pt to model thhe bottlenecck effect usinng microsimmulation.

In order to perform this


t analysiss, a continuo ous simulatioon of pedesttrian behavioor would neeed to
be createed. Pedestrrians could be generateed some disstance awayy from the bbelt at randdomly
determinned position ns around thet entry constriction.
c Then, uusing pedesttrian wayfinnding
techniquees, the travelers would make
m their way
w onto the belt. The N NetLogo soft ftware wouldd lend
itself well to this ap
pplication, ass the prograam is capablle of perform
ming the kinnd of continnuous
simulatio
on required here
h in addittion to the cellular
c autom
mata approaach used in tthe model abbove.
Further detail
d regardiing this prop
posal can be found in thee model doccumentation in Section B B.6 of
Appendix x B.

The most notable chhallenge in completing


c this
t researchh is the difficulty in moddeling pedesstrian
wayfinding in crowdd scenarios. Hoogendoo orn’s experimments includded some em mpirical studiies of
crowds funneling
fu theemselves dow
wn into bottllenecks, inclluding somee fascinating spatial trajeectory
Peter D. Kauffmann Chapter
C 4: Reesults and D
Discussion 52

figures showing
s dyn namic queuee formation (Hoogendooorn & Daam men, 2005).. Unfortunaately,
these stud
dies were lim
mited to observationally defining quueue characteeristics and llack definitioon on
the methoods by which the pedesttrians made their
t way intto the queue lanes.

4.3.3 Effect
E of Imppatience on n Pedestria an Aggressiiveness
Observattions made in validatingg the model’ss passing beehavior in Seection 4.1.1..2 lead to annother
suggestio
on for futuree research. It was hypo othesized thaat as the am
mount of timme a pedestriian is
stuck folllowing a slo
ower traveleer ahead of them, the m more impatieent they willl be come. It is
believed that in thiis situation,, impatiencee would m manifest itsellf through more aggreessive
behavior like the relaxation of choice
c param
meters such as the amouunt of free space back thhat is
required to completee a merging g action. Iff the hypothhesis proves to be true, then after some
amount ofo time that a pedestrian n spends traaveling beloow their desiired speed tthe value off their
lane chan
nging accepttance factorss would becoome more lennient.

Howeverr, the inclusion of this function


f intoo the modell would likeely require ooriginal emppirical
study sin
nce minimal information n about the growth
g of pe destrian imppatience withh time was ffound
during th
he literature review
r for th
his project.
Chapte
er 5 – Con
nclusions
The goall of this project was to develop
d a theeoretical fram
mework to m
model pedestrian behaviior on
moving belt
b facilitiess like escalaators and mo oving walkw ways using m
microsimulattion. By draawing
informatiion about peedestrian cap pabilities annd choice beehavior fromm a number of sources iin the
literaturee – includinng studies th hat were orriginally inteended to appply only too the autom motive
environm ment – a co omprehensive and logiccal basis forr the modell was createed. These facts,
practices, and principles were then coded in an objecct-oriented programminng and moddeling
software called NettLogo to creeate a simu ulation envirronment and the contrrols necessaary to
operate itt. Finally, a series of tessts were run to validate tthe performaance of the m
model and too gain
a better understandiing of how it could bee utilized too answer quuestions aboout the capaacity,
sensitivitty, and queuee growth of moving beltt facilities.

5.1 Theoretica
T l Approach
h and Mod
deling Stra
ategy
At the core
c of thiss research was
w the dessire to creaate a modell of pedestrrian behavioor on
continuou usly moving g belts like escalators
e annd moving w walkways. A At present, thhese facilitiees are
designed d and sized using
u capacitty curves thaat are based on empiricaal studies of isolated escaalator
installatio
ons, as wass discussed in Section 1.3.2. Whiile adjustmeent factors aare often used to
convert these
t theorettical capacitiies into pracctical ones, thhe fact remaains that capacity curvess only
give the system’s con ndition undeer a single seet of inflow pparameters. Examples oof theoreticaal and
practical capacity curves from a manufactureer’s design lliterature cann be seen ass the red andd blue
lines, resspectively, in
i Figure 5.1, below. Furthermoree, flow infoormation is only providded at
capacity. As was seen s in Secttion 4.2.1, the
t performaance of a m moving belt system at llower
occupanccy levels is ata least as im
mportant to system
s desiggners as howw it behaves at capacity ggiven
that thesee lower denssities will prrovide a morre desirable state of opeeration in terrms of pedesstrian
choice an nd comfort.

Practical Cap
pacity (Thyssen
nKrupp) Theoretical Capacity (ThyssenKrupp)
Capacities from Emprical Studies Simulated C
Capacities from
m the Model
18000

16000

14000
Flo (ped/hr)

12000

10000
O tp t Flow

8000
Output

6000

4000

2000

0
0.0 0.2
2 0.4 0.6 0.8 1.0
Belt Speed ( m/s)

Figu
ure 5.1: Com
mparison of Escalator
E Ca
apacities fro m the Model and from thhe Literature

53
Peter D. Kauffmann Chapter 5: Conclusions 54

In order to combat these deficiencies in the state of the practice, a model was constructed to
create a microsimulation of pedestrian choice behavior on moving belt facilities. By creating a
simulation of pedestrian behavior, an experimental approach towards moving belt research could
be utilized. In this way, not only could the performance of an escalator or moving walkway
installation be tested at varying levels of occupancy up to its capacity, different inflow conditions
could also be tested to see the effects of changes in the pedestrian stream on belt performance.
Experimental changes to be implemented in the model included input flow, speed distribution,
user aggressiveness, belt characteristics, and even passing and walking restrictions that could be
implemented by an operator like a transit agency.

However, the functionality of the proposed model as a research tool relied on its ability to mimic
real-world behaviors. While ample research existed on pedestrian characteristics like walking
speed and stair-climbing ability as well as moving belt specifications like speed, very little
information could be found about pedestrian choice behavior in confined spaces like those
present on escalators and moving walkways. Most research into the following and passing
choices made by pedestrians dealt with wayfinding and navigation through crowds in ambulatory
facilities like airports and shopping centers, not through what effectively amounted to a walled-in
corridor. To combat this shortcoming in the literature, it was decided that this model would use
findings from another field of transportation engineering as an analogue for the decisions made
by travelers on escalators and moving walkways.

Through the initial literature review as well as everyday observation, it was found that
pedestrians on belts automatically segregate themselves into two layers, one for slower traffic
and one for passing. These parallel layers mimic the lanes found on a typical road, including the
following and passing choice behaviors exhibited by drivers. Therefore, it was determined that
automotive research, specifically the field of car-following behavior, would be the source of the
choice behaviors programmed into the model. Specifically, the behavioral rules found in the
microsimulator module of TRANSIMS were implemented into the model because of the way
that rule-based behavioral models are especially suited to governing pedestrian choice behavior.
Further benefits were gained because of the similarity between the standardized spacing of step
treads on an escalator and the discrete location information found in the cellular automata
simulation framework that comprises the TRANSIMS microsimulator.

5.2 Comparison to Empirical Pedestrian Behavior and Capacity Data


After the model had been successfully programmed in the cellular automata microsimulation
environment, it had to be validated to ensure that it matched closely with real-world behaviors.
In addition to simple visual verification that the programmed behavioral rules were followed,
time-space diagrams were also utilized to create a more permanent record of passing and
following choice. Throughout these processes, it was determined that the simulation accurately
modeled pedestrian behavior in that each of the simulated entities within the system moved in
accordance with observed real-world pedestrian behavior and that the simulated users interacted
in much the same way as their physical counterparts. Furthermore, the microsimulated
pedestrians each possessed internal variables that accurately accounted for their own individual
desires and performance.

However, the basic research questions of this study were answered through an analysis of the
model’s outputs. Data can be collected both on a pedestrian-by-pedestrian basis as well as on a
Peter D. Kauffmann Chapter 5: Conclusions 55

system-wide level for all three fundamental characteristics, specifically speed, flow, and density.
Not only are the instantaneous, average, and maximum values of these characteristics tracked,
other useful parameters like queue length for the system and headway for the users are also
recorded. However, the advantage of a simulation of pedestrian flow over empirical studies is
that changes in the operational state of the system can be modeled. As was stated previously, the
goal of this project was to model both changes in rules implemented on a moving belt system by
the operator as well as variations that may occur in the pedestrian stream entering an escalator or
moving walkway.

To this end, several capacity tests were undertaken to compare the results of the model to
empirically determined capacities found in the literature in order to continue with the validation
of the model as a tool for both research and design. In Section 4.1.2.1, several studies from
design manuals and research studies were simulated using the model, with the resulting capacity
from the model closely matching those found in the literature for a variety of operational
scenarios.

However, the real power of the model comes from its ability to simulate belt operations under a
variety of conditions. The previous state of the practice of belt specification was to rely on
capacity curves, as was previously seen above in Figure 5.1, or empirical studies like those that
were used in the capacity tests, shown as green triangles in the same figure. In contrast, this
model can simulate a limitless number of scenarios over a range of belt speeds and inflow
conditions, resulting in the orange region of output capacities seen in Figure 5.1 that cover the
variety of conditions that can be found in belt installations around the world. Perhaps the most
important aspect of the model is that it can evaluate the performance of a belt in unsaturated
conditions to ensure that it remains at a sufficiently low occupancy to ensure the comfort of its
riders through their ability to make their own behavioral choices, giving the wide range of output
flows seen in the above figure.

5.3 Real-World Applications of the Model


Obviously, a microsimulated model of pedestrian behavior can be used to experimentally predict
the characteristics and performance of the physical system it is meant to represent. It has
previously been seen that capacity of a system can be found under a user-defined set of inflow
parameters. However, there are several additional applications to which the model may be suited
that were discussed in Section 4.2.

The first potential application is not particular dissimilar to the model’s primary purpose of
capacity determination, that of performing sensitivity analyses of belt systems. By assessing the
performance of a proposed system under a range of inflow stream settings, uncertainty in
demand prediction can be accounted for in the design process. However, this same analysis can
be applied to existing systems by varying the input stream and examining the result these
changes have on the operation of a moving belt system in order to determine how long an
existing facility can be expected to operate under nominal conditions before improvement or
replacement becomes necessary.

As was discussed above, the model can determine the capacity of a moving belt facility under the
expected set of input characteristics as well as to test the sensitivity of this sort of solution to
changes in that input. However, other experimental tests can also be run using the simulation
Peter D. Kauffmann Chapter 5: Conclusions 56

framework. One such example described above involves analyzing the performance of an
escalator under a proposed set of operational rules that could potentially be implemented by a
transit authority or other operator, specifically a rule that all users must stand still on the belt was
examined. Under emergency conditions, it may become necessary to direct users to pack more
efficiently onto an escalator in an attempt to increase the overall flow rate leaving a subterranean
transit facility, for instance. An experiment conducted in Section 4.2.2.2 appears to indicate that
this sort of rule has the potential of increasing pedestrian throughput by a significant margin, on
the order of 15 to 20 percent.

Finally, because of its ability to track queue length over the course of a simulation, the model
seems to be suited for use as a tool for platform sizing. Given the stochastic nature of arrivals in
a conventional belt installation or even the ebb and flow seen in transit facilities as trains and
buses arrive and discharge their passenger loads, there exists the potential for significant queue
accumulation at the entry to an escalator or moving walkway. By tracking the length of this
queue over time and accounting for the minimum amount of space required per person in order
to ensure pedestrian comfort and control and to prevent the formation of shockwaves in a crowd,
it becomes possible to determine just how much space is required to safely contain the expected
queue. If this area is supplied in the design of the waiting area at the entry to the moving belt,
whether on a transit platform or simply at a landing between moving walkways, the safety of a
system’s users can be preserved.

5.4 Summary of Contributions of the Project


Based on the results presented in this paper, it seems that the fundamental approach of
quantifying and modeling pedestrian behavior on a moving belt system using microsimulation is
sound. Furthermore, the method by which the model was constructed – using data from a
number of different disciplines within the field of transportation engineering – appears to provide
a satisfactory framework for pedestrian behavior. In addition, it has been shown that the model
created in this project can not only be used to analyze the capacity and fundamental
characteristics of a belt system, it can also fulfill other functions such as providing plots of
pedestrian trajectory and assessing the effect of rule implementation on the operation of the
system under both normal conditions or under evacuation protocols.

In this way, it seems that the model provides a new approach to the design of pedestrian facilities
like escalators and moving walkways. By allowing the user to input the specific parameters
exhibited by pedestrians in that region, the engineer’s ability to accurately predict moving belt
performance is greatly improved. In addition to the benefits to capacity analysis, significant
gains are also observed in how well the growth of a queue and its behavior over time can be
anticipated. Therefore, it is hoped that this model, or at least the approach it takes to quantifying
and modeling pedestrian behavior, is implemented in the field of moving belt specification as a
tool for both research and design.
References

Ames, N. (1859). Patent No. 25076. United States of America.

Bangash, M. H., & Bangash, T. (2007). Lifts, Elevators, Escalators and Moving
Walkways/Travelators. New York: Taylor & Francis.

Blue, V. J., & Adler, J. L. (1998). Emergent fundamental pedestrian flows from cellular
automata microsimulation. Transportation Research Record, 1644, 29-36.

Blue, V. J., & Adler, J. L. (2001). Cellular automata microsimulation for modeling bi-directional
pedestrian walkways. Transportation Research Part B, 293-312.

Cepolina, E., & Tyler, N. (2005). Understanding capacity drop for designing pedestrian
environments. Walk21-VI “Everyday Walking Culture”, The 6th International
Conference on Walking in the 21st Century. Zurich: University College London.

Daamen, W., & Hoogendoorn, S. P. (2003). Experimental Research of Pedestrian Walking


Behavior. Transportation Research Record, 1828, 20-30.

Davis, P., & Dutta, G. (2002). Estimation of Capacity of Escalators in London Underground.
London: London School of Economics and Political Sciences.

Donoghue, E. A. (1981). Handbook on A17.1: Safety Code for Elevators and Escalators. New
York: American Society of Mechanical Engineers.

Fruin, J. J. (1984, May). Crowd Dynamics and Auditorium Managment. Auditorium News.

Fujiyama, T., & Tyler, N. (2004). An Explicit Study on Walking Speeds of Pedestrians on Stairs.
London: Centre for Transport Studies, University College London.

Geismar, T., Chwast, S., de Harak, R., Lees, J., & Vignelli, M. (2011). Symbol Signs. Retrieved
January 1, 2011, from AIGA | the professional association for design:
http://www.aiga.org/content.cfm/symbol-signs

Goodman, L. (1992). Transportation Interface Areas. In J. D. Edwards, & I. o. Engineers (Ed.),


Transportation Planning Handbook (pp. 201-293). Englewood Cliffs, New Jersey:
Prentice-Hall.

Hession, A. (n.d.). NEII On-Line Newsroom. Retrieved October 12, 2010, from National
Elevator Industry, Inc.: http://www.neii.org/presskit/pressmaster.cfm?link=7

Hobeika, A. G., & Gu, Y. (2004). TRANSIMS - the next generation planning/simulation model.
Urban Transport X, 141-151.

Hoogendoorn, S. (2004). Walking behavior in bottlenecks and its implications for capacity. TRB
2004 Annual Meeting CD-ROM. Washington: Transportation Research Board of the
National Academies.

57
Peter D. Kauffmann References 58

Hoogendoorn, S. P., & Daamen, W. (2005). Pedestrian Behavior at Bottlenecks. Transportation


Science, 39(2), 147-159.

Hoogendoorn, S., & Bovy, P. H. (2000). Gas-Kinetic Modeling and Simulation of Pedestrian
Flows. Transportation Research Record, 1710, 28-36.

Hoogendoorn, S., & Bovy, P. H. (2003). Simulation of pedestrian flows by optimal control and
differential games. Optimal Control Applications and Methods, 24(3), 153-172.

Lampugnani, V. M., Hartwig, L., Simmen, J., & Imorde, J. (1994). Vertical: Lift, Escalator,
Paternoster. Berlin: Ernst & Sohn.

Lee, Y.-C. (2005). Pedestrian Walking and Choice Behavior on Stairways and Escalators in
Public Transport Facilities.

Los Alamos National Laboratory. (2006). TRansportation ANalysis SIMulation System


(TRANSIMS) Course Manual: Microsimulator. AECOM.

Mansel, D. M., Menaker, P. J., & Hartnett, G. (1998). High-Capacity Light Rail Transit:
Balancing Stationside and Railside Capacities. Transportation Research Record, 1623,
170-178.

May, A. D. (1990). Traffic Flow Fundamentals. Englewood Cliffs, NJ: Prentice Hall.

O'Neill, R. (1974). Escalators in Rapid Transit Stations. American Society of Civil Engineers
Jornal of the Transportation Engineering Division, 1-12.

Rakha, H., & Crowther, B. (2002). A Comparison of the Greenshields, Pipes, and Van Aerde
Car-Following and Traffic Stream Models. Transportation Research Record, 1802, 248-
262.

Rakha, H., Pecker, C. C., & Cybis, H. B. (2007). Calibration Procedure for Gipps Car-Following
Model. Transportation Research Record, 1999, 115-127.

Rickert, M., Nagel, K., Schreckenberg, M., & Latour, A. (1996). Two lane traffic simulations
using cellular automata. Physica A: Statistical and Theoretical Physics, 231(4), 534-550.

Schindler Escalators. (2010). Schindler 9300 Advanced Edition. Morristown, New Jersey:
Schindler USA.

Slaughter, L. (2004). 108 Congress HR 4995.

Sutherland, D. H., Olshen, R. A., Biden, E. N., & Wyatt, M. P. (1988). The Development of
Mature Walking. Philadelphia: J.B. Lippincott.

Sutthawassuntorn, K. (2010). Study and Analysis of Passenger Movement and Service on


Escalators at Metro Station.
Peter D. Kauffmann References 59

ThyssenKrupp Elevator. (2004). Escalators and Moving Walks planning guide. Memphis,
Tennessee: ThyssenKrupp Americas Business Unit.

Turner, D. L. (1998). Escalators and Moving Walks. In G. R. Strakosch, The Vertical


Transportation Handbook, Third Edition (pp. 210-236). New York: Wiley.

Welch, P., Houseman, H., Kinihan, J., Evans, D., Herndobler, R., Nurnberg, T., et al. (2009).
Heavy Duty Transportation System Escalator Design Guidelines. In APTA Rail Transit
Standards, Volume 5: Fixed Structures (pp. 7.1-7.26). Washington: American Public
Transportation Association.

Wilensky, U., & Tisue, S. (2004). NetLogo: A Simple Environment for Modeling Complexity.
International Conference on Complex Systems. Boston.

WMATA. (2008). Station Site and Access Planning Manual. Washington: Washington
Metropolitan Area Transit Authority.

WMATA. (2008). Washington Metro Station Access and Capacity Study. Washington:
Washington Metropolitan Area Transit Authority.
Appendix A – NetLogo Model Code
;; A.1 Create Global Variables
globals
[
treads
inc_create_walk
inc_create_stand
inc_create_comm
v_belt
v_walk
v_comm
p_stand
p_walk
p_comm
peds_in
peds_in_belt
standers_in
walkers_in
commuters_in
peds_out
standers_out
walkers_out
commuters_out
flow
flow_max
density
n_stand
n_walk
n_comm
o_ellipse
p_ellipse
p_decel
avg_speed_stand
avg_speed_walk
avg_speed_comm
queue
queue_right
queue_left
queue_removed
queue_total
max_queue
time
selected1
selected2
selected3
]

;; A.2 Create Turtle-Specific Variables


turtles-own
[
speed
speed_desired
speed_desired_now
rand_decel
distance_forward
gap_forward

60
Peter D. Kauffmann Appendix A: NetLogo Model Code 61

gap_back
gap_forward_other
gap_back_other
tolerable_gap_back_other
speed_desired_now_back
speed_desired_now_back_other
speed_desired_now_forward
speed_desired_now_forward_other
time_since_lane_change
facial_ellipse
lane
on_belt
]

;; A.3 Create Pedestrian Classes


breed [ standers stander ]
breed [ walkers walker ]
breed [ commuters commuter ]

;; A.4 Define Example Scenarios


to scenario0
ca
set belt_length 24
set Stand_Percent 25
set FastWalker_Percent 60
set beltspeed 0.4
set pedspeed_stander 0.4
set pedspeed_walker 0.4
set pedspeed_commuter 0.8
set climbspeed_walker 0.4
set climbspeed_commuter 0.8
set lanechange_allowed true
set slow_with_distance false
set belt_rules "None"
set belt_type_for_facial_ellipse "Down Escalator"
set obey_ellipse_percent 0
set initially_populate_belt true
set mergespace_walker 0.8
set mergespace_commuter 0.4
set Max_Queue_Diff 20
set ignore_lane_assign true
set inflow 2500
set stopafter_seconds 3600
set decel_probability 3
set min_t_btwn_lanechange 5
set capacity_test false
output-print "Scenario 0 - TNV D 2500 0.4"
set File_Output_TimeSpace false
end

to scenario1
ca
set belt_length 24
set Stand_Percent 50
set FastWalker_Percent 50
set beltspeed 0.4
set pedspeed_stander 0.4
Peter D. Kauffmann Appendix A: NetLogo Model Code 62

set pedspeed_walker 0.4


set pedspeed_commuter 0.8
set climbspeed_walker 0.4
set climbspeed_commuter 0.8
set lanechange_allowed false
set slow_with_distance false
set belt_rules "None"
set belt_type_for_facial_ellipse "Down Escalator"
set obey_ellipse_percent 0
set initially_populate_belt true
set mergespace_walker 0.8
set mergespace_commuter 0.4
set Max_Queue_Diff 20
set ignore_lane_assign false
set inflow 3500
set stopafter_seconds 3600
set decel_probability 3
set min_t_btwn_lanechange 5
set capacity_test false
output-print "Scenario 1 - FNX D 3500 0.4"
set File_Output_TimeSpace false
end

to scenario2
ca
set belt_length 24
set Stand_Percent 50
set FastWalker_Percent 50
set beltspeed 0.4
set pedspeed_stander 0.4
set pedspeed_walker 0.4
set pedspeed_commuter 0.8
set climbspeed_walker 0.4
set climbspeed_commuter 0.8
set lanechange_allowed false
set slow_with_distance false
set belt_rules "None"
set belt_type_for_facial_ellipse "Up Escalator"
set obey_ellipse_percent 95
set initially_populate_belt true
set mergespace_walker 0.8
set mergespace_commuter 0.4
set Max_Queue_Diff 20
set ignore_lane_assign false
set inflow 3500
set stopafter_seconds 3600
set decel_probability 3
set min_t_btwn_lanechange 5
set capacity_test false
output-print "Scenario 2 - FNX U-95 3500 0.4"
set File_Output_TimeSpace false
end

to scenario3
ca
set belt_length 24
set Stand_Percent 50
Peter D. Kauffmann Appendix A: NetLogo Model Code 63

set FastWalker_Percent 50
set beltspeed 0.8
set pedspeed_stander 0.4
set pedspeed_walker 0.4
set pedspeed_commuter 0.8
set climbspeed_walker 0.4
set climbspeed_commuter 0.8
set lanechange_allowed false
set slow_with_distance false
set belt_rules "None"
set belt_type_for_facial_ellipse "Up Escalator"
set obey_ellipse_percent 95
set initially_populate_belt true
set mergespace_walker 0.8
set mergespace_commuter 0.4
set Max_Queue_Diff 20
set ignore_lane_assign false
set inflow 3500
set stopafter_seconds 3600
set decel_probability 3
set min_t_btwn_lanechange 5
set capacity_test false
output-print "Scenario 3 - FNX U-95 3500 0.8"
set File_Output_TimeSpace false
end

to scenario4
ca
set belt_length 24
set Stand_Percent 100
set FastWalker_Percent 50
set beltspeed 0.4
set pedspeed_stander 0.4
set pedspeed_walker 0.4
set pedspeed_commuter 0.8
set climbspeed_walker 0.4
set climbspeed_commuter 0.8
set lanechange_allowed false
set slow_with_distance false
set belt_rules "Stand Only"
set belt_type_for_facial_ellipse "Up Escalator"
set obey_ellipse_percent 95
set initially_populate_belt true
set mergespace_walker 0.8
set mergespace_commuter 0.4
set Max_Queue_Diff 20
set ignore_lane_assign false
set inflow 3500
set stopafter_seconds 3600
set decel_probability 3
set min_t_btwn_lanechange 5
set capacity_test false
output-print "Scenario 4 - FSX U-95 3500 0.4"
set File_Output_TimeSpace false
end

to scenario5
Peter D. Kauffmann Appendix A: NetLogo Model Code 64

ca
set belt_length 24
set Stand_Percent 0
set FastWalker_Percent 50
set beltspeed 0.4
set pedspeed_stander 0.4
set pedspeed_walker 0.4
set pedspeed_commuter 0.8
set climbspeed_walker 0.4
set climbspeed_commuter 0.8
set lanechange_allowed false
set slow_with_distance false
set belt_rules "Walk Only"
set belt_type_for_facial_ellipse "Up Escalator"
set obey_ellipse_percent 95
set initially_populate_belt true
set mergespace_walker 0.8
set mergespace_commuter 0.4
set Max_Queue_Diff 20
set ignore_lane_assign false
set inflow 3500
set stopafter_seconds 3600
set decel_probability 3
set min_t_btwn_lanechange 5
set capacity_test false
output-print "Scenario 5 - FWX U-95 3500 0.4"
set File_Output_TimeSpace false
end

to scenario6
ca
set belt_length 24
set Stand_Percent 50
set FastWalker_Percent 50
set beltspeed 0.4
set pedspeed_stander 0.4
set pedspeed_walker 0.4
set pedspeed_commuter 0.8
set climbspeed_walker 0.4
set climbspeed_commuter 0.8
set lanechange_allowed false
set slow_with_distance true
set belt_rules "None"
set belt_type_for_facial_ellipse "Up Escalator"
set obey_ellipse_percent 0
set initially_populate_belt true
set mergespace_walker 0.8
set mergespace_commuter 0.4
set Max_Queue_Diff 20
set ignore_lane_assign false
set inflow 3500
set stopafter_seconds 3600
set decel_probability 3
set min_t_btwn_lanechange 5
set capacity_test false
output-print "Scenario 6 - FNX U-0 3500 0.4 SLOW"
set File_Output_TimeSpace false
Peter D. Kauffmann Appendix A: NetLogo Model Code 65

end

;; A.5 Select Pedestrians for NetLogo Time-Space Diagram


to select1
if mouse-down?
[
let mx mouse-xcor
let my mouse-ycor
if any? turtles-on patch mx my
[
ask selected1
[
if (breed = standers) [ set color black ]
if (breed = walkers) [ set color lime ]
if (breed = commuters) [ set color red ]
]
set selected1 one-of turtles-on patch mx my
output-type "Selected Ped 1 = " output-print selected1
ask selected1 [ set color orange ]
display
]
]
end

to select2
if mouse-down?
[
let mx mouse-xcor
let my mouse-ycor
if any? turtles-on patch mx my
[
ask selected2
[
if (breed = standers) [ set color black ]
if (breed = walkers) [ set color lime ]
if (breed = commuters) [ set color red ]
]
set selected2 one-of turtles-on patch mx my
output-type "Selected Ped 2 = " output-print selected2
ask selected2 [ set color yellow ]
display
]
]
end

to select3
if mouse-down?
[
let mx mouse-xcor
let my mouse-ycor
if any? turtles-on patch mx my
[
ask selected3
[
if (breed = standers) [ set color black ]
if (breed = walkers) [ set color lime ]
if (breed = commuters) [ set color red ]
Peter D. Kauffmann Appendix A: NetLogo Model Code 66

]
set selected3 one-of turtles-on patch mx my
output-type "Selected Ped 3 = " output-print selected3
ask selected3 [ set color magenta ]
display
]
]
end

;; A.6 SETUP routine


to setup
clear-all
constraints
compute-variables
draw-belt
if (initially_populate_belt = true) [populate-belt]
if (file_output_timespace = true ) [setup-file]
end

;; A.7 Adjust for Constraints module


to constraints
if beltspeed = 0
[ set belt_rules "Walk Only" ]
if belt_rules = "Stand Only"
[
set Stand_Percent 100
set lanechange_allowed false
]
if belt_rules = "Walk Only"
[
;; assume all standers become normal walkers
set FastWalker_Percent ((1 - (Stand_Percent / 100)) * FastWalker_Percent)
set Stand_Percent 0
]
ifelse belt_type_for_facial_ellipse = "Down Escalator"
[
set o_ellipse false
set p_ellipse 0
]
[
set o_ellipse true
set p_ellipse (obey_ellipse_percent / 100)
]
if slow_with_distance = true
[ set belt_type_for_facial_ellipse "Up Escalator" ]
end

;; A.8 Compute Variables module


to compute-variables
set treads (Belt_Length * 2.5) ;; convert meters to number of treads
set v_belt (BeltSpeed / 0.4) ;; convert speeds in m/s to cells/tick
set v_walk (ClimbSpeed_Walker / 0.4)
set v_comm (ClimbSpeed_Commuter / 0.4)
set p_stand (PedSpeed_Stander / 0.4)
set p_walk (PedSpeed_Walker / 0.4)
set p_comm (PedSpeed_Commuter / 0.4)
set n_stand (Stand_Percent / 100 * inflow)
Peter D. Kauffmann Appendix A: NetLogo Model Code 67

set n_walk (((100 - Stand_Percent) / 100) * ((100 - FastWalker_Percent) /


100) * inflow)
set n_comm (((100 - Stand_Percent) / 100) * (FastWalker_Percent / 100) *
inflow)
set standers_in 1
set walkers_in 1
set commuters_in 1
set p_decel (decel_probability / 100)
end

;; A.9 Draw Simulation Environment module


to draw-belt
ask patches
[
set pcolor gray
if ((pycor > -2) and (pycor < 3) and (pxcor < treads + 2) and (pxcor > -
3))
[ set pcolor black ]
if ((pycor > -1) and (pycor < 2))
[ set pcolor gray + 2 ]
if ((pycor = 0) and (pxcor > -1) and (pxcor < treads))
[ set pcolor blue]
if ((pycor = 1) and (pxcor > -1) and (pxcor < treads))
[ set pcolor sky]
if ((pycor = 0) and (pxcor > -1) and (pxcor < treads) and ((pxcor mod 2)
= 0))
[ set pcolor blue + 1]
if ((pycor = 1) and (pxcor > -1) and (pxcor < treads) and ((pxcor mod 2)
= 0))
[ set pcolor sky + 1]
]
end

;; A.10 Populate Belt module


to populate-belt
if (belt_rules != "Walk Only") [
set inc_create_stand random-poisson (n_stand / 3600 * (treads / (v_belt)))
create-standers inc_create_stand
[
set color black
ifelse (belt_rules = "None")
[ set lane 0]
[ set lane (random 2)]
setxy (random treads) lane
set heading 90
set speed_desired 0
set time_since_lane_change 0
set speed_desired_now speed_desired
set speed (speed_desired_now + v_belt)
set on_belt true
set time_since_lane_change Min_t_btwn_lanechange
ifelse o_ellipse = true
[ ifelse (p_ellipse >= (random-float 1))
[ set facial_ellipse 1 ]
[ set facial_ellipse 0 ] ]
[ set facial_ellipse 0 ]
] ]
Peter D. Kauffmann Appendix A: NetLogo Model Code 68

set inc_create_walk random-poisson (n_walk / 3600 * (treads / (v_belt +


v_walk)))
create-walkers inc_create_walk
[
set color (lime + 2)
ifelse (lanechange_allowed = false)
[ ifelse (ignore_lane_assign = false)
[ ifelse (belt_rules = "None") [ set lane 1 ] [ set lane 0 ] ]
[ set lane (random 2) ] ]
[ ifelse ((belt_rules = "None") and (ignore_lane_assign = false))
[ set lane 1 ]
[ set lane (random 2) ] ]
setxy (random treads) lane
set heading 90
set time_since_lane_change 0
set speed_desired v_walk
set speed_desired_now speed_desired
set speed (speed_desired_now + v_belt)
set tolerable_gap_back_other (mergespace_walker / 0.4)
set on_belt true
set time_since_lane_change Min_t_btwn_lanechange
ifelse o_ellipse = true
[ ifelse (p_ellipse >= (random-float 1))
[ set facial_ellipse 1 ]
[ set facial_ellipse 0 ] ]
[ set facial_ellipse 0 ]
]
set inc_create_comm random-poisson (n_comm / 3600 * (treads / (v_belt +
v_comm)))
create-commuters inc_create_comm
[
set color red
ifelse (lanechange_allowed = false)
[ ifelse (ignore_lane_assign = false)
[ set lane 1 ]
[ set lane (random 2) ] ]
[ ifelse ((belt_rules = "None") and (ignore_lane_assign = false))
[ set lane 1 ]
[ set lane (random 2) ] ]
setxy (random treads) lane
set heading 90
set time_since_lane_change 0
set speed_desired v_comm
set speed_desired_now speed_desired
set speed (speed_desired_now + v_belt)
set tolerable_gap_back_other (mergespace_commuter / 0.4)
set on_belt true
set time_since_lane_change Min_t_btwn_lanechange
ifelse o_ellipse = true
[ ifelse (p_ellipse >= (random-float 1))
[ set facial_ellipse 1 ]
[ set facial_ellipse 0 ] ]
[ set facial_ellipse 0 ]
]

ask turtles
[
Peter D. Kauffmann Appendix A: NetLogo Model Code 69

loop
[ ifelse any? other turtles-here
[ fd ((2 * (random 2)) - 1) ]
[ stop ]
]
]

set selected1 one-of turtles


output-type "Selected Ped 1 = " output-print selected1
ask selected1 [ set color orange ]
set selected2 one-of turtles
ask turtles [
loop [
ifelse (selected2 = selected1)
[ set selected2 one-of turtles ]
[ stop ]
] ]
output-type "Selected Ped 2 = " output-print selected2
ask selected2 [ set color yellow ]
set selected3 one-of turtles
ask turtles [
loop [
ifelse ((selected3 = selected2) or (selected3 = selected1))
[ set selected3 one-of turtles ]
[ stop ]
] ]
output-type "Selected Ped 3 = " output-print selected3
ask selected3 [ set color magenta ]

set peds_in (peds_in + inc_create_stand + inc_create_walk +


inc_create_comm)
set peds_in_belt peds_in
set standers_in (standers_in + inc_create_stand)
set walkers_in (walkers_in + inc_create_walk)
set commuters_in (commuters_in + inc_create_comm)
end

;; A.11 File Output module


to setup-file
let file user-new-file
;; We check to make sure we actually got a string just in case
;; the user hits the cancel button.
if not is-string? file
[ set File_Output_TimeSpace false stop ]
;; If the file already exists, we begin by deleting it, otherwise
;; new data would be appended to the old contents.
if file-exists? file
[ file-delete file ]
file-open file
;; record the initial turtle data
file-print "NetLogo Escalator Simulation Model Output"
file-print ""
file-print "INPUT PARAMETERS"
file-print (word "Belt Speed = " BeltSpeed)
file-print (word "Slow with Distance? = " Slow_with_Distance)
file-print (word "Stander Pedestrian Speed = " PedSpeed_Stander)
file-print (word "Walker Pedestrian Speed = " PedSpeed_Walker)
Peter D. Kauffmann Appendix A: NetLogo Model Code 70

file-print (word "Commuter Pedestrian Speed = " PedSpeed_Commuter)


file-print (word "Walker Climbing Speed = " ClimbSpeed_Walker)
file-print (word "Commuter Climbing Speed = " ClimbSpeed_Commuter)
file-print (word "Probability of Deceleration = " Decel_Probability)
file-print (word "Merge Space Back for Walkers = " MergeSpace_Walker)
file-print (word "Merge Space Back for Commuters = " MergeSpace_Commuter)
file-print (word "Lane Change Allowed? = " LaneChange_Allowed)
file-print (word "Minimum Time Between Lane Changes = "
Min_t_btwn_lanechange)
file-print (word "Percent Standers = " Stand_Percent)
file-print (word "Percent Commuters out of Non-Standers = "
FastWalker_Percent)
file-print (word "Difference in Queue Between Lanes for Arrivals to Ignore
Assignment = " Max_Queue_Diff)
file-print (word "Ignore Initial Lane Assignment? = " Ignore_Lane_Assign)
file-print (word "Arrivals per Hour = " Inflow)
file-print (word "Belt Length (m) = " Belt_Length)
file-print (word "Initially Populate Belt? = " Initially_Populate_Belt)
file-print (word "Belt Rules? = " Belt_Rules)
file-print (word "Percent Obeying Facial Ellipse = " Obey_Ellipse_Percent)
file-print (word "Stop After N Seconds = " StopAfter_Seconds)
file-print (word "Capacity Test? = " Capacity_Test)
file-print ""
file-print "SIMULATION RESULTS"
file-print "Turtle Number, Breed, Ticks (seconds), xcor (tread count),
Lane, Speed (treads/tick), Desired Speed"
file-print
"TIME,VALUE,ticks,peds_in,peds_in_belt,peds_out,Density,queue_total"
file-print ""
end

;; A.12 GO routine
to go
if ticks >= StopAfter_Seconds [stop]
if (queue_left >= 90) or (queue_right >= 90) ; queue overflow
[
ask patches [set pcolor orange]
stop
]
lane-change
move-belt
arrivals
move-queue
tick
departures
create-plots
end

;; A.13 Arrivals module


to arrivals
set inc_create_stand random-poisson (n_stand / 3600)
create-standers inc_create_stand
[
set color black
ifelse (belt_rules = "None")
[ ifelse (queue_right - queue_left) >= Max_Queue_Diff [ set lane 1 ] [
set lane 0 ] ]
Peter D. Kauffmann Appendix A: NetLogo Model Code 71

[ set lane (random 2)]


setxy 0 lane
set heading 90
set time_since_lane_change 0
set speed_desired 0
set speed_desired_now speed_desired
set speed p_stand
set on_belt false
ifelse o_ellipse = true
[ ifelse (p_ellipse >= (random-float 1))
[ set facial_ellipse 1 ]
[ set facial_ellipse 0 ] ]
[ set facial_ellipse 0 ]
loop
[ ifelse any? other turtles-here
[ fd -1 ]
[ stop ]
]
]
set inc_create_walk random-poisson (n_walk / 3600)
create-walkers inc_create_walk
[
set color (lime + 2)
ifelse (lanechange_allowed = false)
[ ifelse (ignore_lane_assign = false)
[ ifelse (belt_rules = "None")
[ ifelse (queue_left - queue_right) >= Max_Queue_Diff [ set lane 0 ]
[ set lane 1] ]
[ ifelse (queue_right - queue_left) >= Max_Queue_Diff [ set lane 1 ]
[ set lane 0] ] ]
[ set lane (random 2) ] ]
[ ifelse ((belt_rules = "None") and (ignore_lane_assign = false))
[ ifelse (queue_left - queue_right) >= Max_Queue_Diff [ set lane 0 ] [
set lane 1] ]
[ set lane (random 2) ] ]
setxy 0 lane
set heading 90
set time_since_lane_change 0
set speed_desired v_walk
set speed_desired_now speed_desired
set tolerable_gap_back_other (mergespace_walker / 0.4)
set speed p_walk
set on_belt false
ifelse o_ellipse = true
[ ifelse (p_ellipse >= (random-float 1))
[ set facial_ellipse 1 ]
[ set facial_ellipse 0 ] ]
[ set facial_ellipse 0 ]
loop
[ ifelse any? other turtles-here
[ fd -1 ]
[ stop ]
]
]
set inc_create_comm random-poisson (n_comm / 3600)
create-commuters inc_create_comm
[
Peter D. Kauffmann Appendix A: NetLogo Model Code 72

set color red


ifelse (lanechange_allowed = false)
[ ifelse (ignore_lane_assign = false)
[ ifelse (queue_left - queue_right) >= Max_Queue_Diff [ set lane 0 ] [
set lane 1] ]
[ set lane (random 2) ] ]
[ ifelse ((belt_rules = "None") and (ignore_lane_assign = false))
[ ifelse (queue_left - queue_right) >= Max_Queue_Diff [ set lane 0 ] [
set lane 1] ]
[ set lane (random 2) ] ]
setxy 0 lane
set heading 90
set time_since_lane_change 0
set speed_desired v_comm
set speed_desired_now speed_desired
set tolerable_gap_back_other (mergespace_commuter / 0.4)
set speed p_comm
set on_belt false
ifelse o_ellipse = true
[ ifelse (p_ellipse >= (random-float 1))
[ set facial_ellipse 1 ]
[ set facial_ellipse 0 ] ]
[ set facial_ellipse 0 ]
loop
[ ifelse any? other turtles-here
[ fd -1 ]
[ stop ]
]
]
set peds_in (peds_in + inc_create_stand + inc_create_walk +
inc_create_comm)
set standers_in (standers_in + inc_create_stand)
set walkers_in (walkers_in + inc_create_walk)
set commuters_in (commuters_in + inc_create_comm)
end

;; A.14 Move Queue module


to move-queue
;; move the queue towards the escalator
ask-concurrent turtles with [on_belt = false]
[
ifelse (min-one-of other turtles with [(xcor > [xcor] of myself) and
(lane = [lane] of myself)] [distance myself]) = nobody
[ set distance_forward 20 ]
[ set distance_forward (distance (min-one-of other turtles with [(xcor >
[xcor] of myself) and (lane = [lane] of myself)] [distance myself]))]
jump (min list speed distance_forward)
loop [
ifelse any? other turtles-here
[ fd -1 ]
[ stop ]
]
]

;; people on belt speed up and change on_belt state


ask turtles with [ (on_belt = false) and (xcor >= 0)]
[
Peter D. Kauffmann Appendix A: NetLogo Model Code 73

set speed (speed + v_belt)


set on_belt true
set peds_in_belt (peds_in_belt + 1)
set time_since_lane_change (max list 0 (Min_t_btwn_lanechange - 2))
]

;; if we're unconcerned about the queue, we prevent queue overflow


if Capacity_Test = true [
set queue_removed (queue_removed + count turtles with [ xcor <= -75 ])
ask-concurrent turtles with [ xcor <= -75 ] [die] ]

;; compute queue length


set queue_left (count turtles with [(xcor < 0) and (ycor = 1)])
set queue_right (count turtles with [(xcor < 0) and (ycor = 0)])
set queue (queue_left + queue_right)

set queue_total (queue + queue_removed)


set max_queue max list queue_total max_queue
end

;; A.15 Move Belt module


to move-belt
ask-concurrent turtles with [on_belt = true]
[
set rand_decel random-float 1
if slow_with_distance = true
[
if (xcor >= 50)
[ ifelse (xcor >= 75)
[ ifelse (xcor >= 100)
[ set speed_desired_now (max list (speed_desired - 3) 0) ] ; slow
more after 40m run
[ set speed_desired_now (max list (speed_desired - 2) 0) ] ] ; slow
more after 30m run
[ set speed_desired_now (max list (speed_desired - 1) 0) ] ; slow a
bit after 20 m run
]
]

; FOLLOWING BEHAVIOR
; Determine if there is another vehicle ahead (same lane, greater x-
coordinate)
ifelse (min-one-of other turtles with [(xcor > [xcor] of myself) and
(lane = [lane] of myself)] [distance myself]) = nobody
; if no vehicle, set to an aribtrarily large headway
[ set distance_forward 20 ]
; if there is a vehicle, set distance_forward equal to the distance
[ set distance_forward (distance (min-one-of other turtles with [(xcor >
[xcor] of myself) and (lane = [lane] of myself)] [distance myself]))]

set gap_forward (distance_forward - 1)

; if speed >= gap, slow down to gap, keep minimum speed at v_belt
ifelse ((speed + facial_ellipse) >= gap_forward)
[ ifelse (rand_decel <= p_decel) ; if speed is greater than gap, we must
slow down
Peter D. Kauffmann Appendix A: NetLogo Model Code 74

[ set speed (max list (gap_forward - 1) v_belt) ] ; some probability of


extra deceleration, floor is belt speed
[ set speed (max list (gap_forward - facial_ellipse) v_belt) ] ] ;
otherwise slow to gap, floor is belt speed
[ ifelse (speed < (speed_desired_now + v_belt)) ; if forward gap exists,
check speed vs. speed limit
[ ifelse (rand_decel <= p_decel) ; if not at maximum speed
[ set speed (max (list speed v_belt)) ] ; some probability of no
aceleration
[ set speed (max (list (min (list (speed + 1) (gap_forward -
facial_ellipse) (speed_desired_now + v_belt) )) v_belt)) ] ] ;; accelerate in
increments of v_belt
[ ifelse (rand_decel <= p_decel) ; if at or above maximum speed
[ set speed (max (list (min list (gap_forward - facial_ellipse)
(speed_desired_now + v_belt - 1)) v_belt)) ] ; some probability of
deceleration
[ set speed (max (list (min list (gap_forward - facial_ellipse)
(speed_desired_now + v_belt)) v_belt)) ] ] ] ; otherwise maintain speed

;; move people at appropriate relative speed


if belt_type_for_facial_ellipse != "Down Escalator"
[ if facial_ellipse = 1
[ if (any? turtles-at facial_ellipse 0)
[ set speed (speed - facial_ellipse) ] ] ]

;; move the users at appropriate relative speed


jump speed

;; double check for overlap


if any? other turtles-here
[ fd -1 ]
]
end

;; A.16 Lane Change module


to lane-change
if ( lanechange_allowed = true)
[
;; COMPUTE CONSTANTS
ask-concurrent turtles with [on_belt = true]
[
;; calculate gaps around ped
ifelse (min-one-of other turtles with [(xcor > [xcor] of myself) and
(lane = [lane] of myself)] [distance myself]) = nobody
[ set gap_forward 20 ]
[ set gap_forward ((distance (min-one-of other turtles with [(xcor >
[xcor] of myself) and (lane = [lane] of myself)] [distance myself])) - 1)]
ifelse (min-one-of other turtles with [(xcor < [xcor] of myself) and
(lane = [lane] of myself)] [distance myself]) = nobody
[ set gap_back 20 ]
[ set gap_back ((distance (min-one-of other turtles with [(xcor <
[xcor] of myself) and (lane = [lane] of myself)] [distance myself])) - 1)]

;; calculate speeds of surrounding peds


ifelse (min-one-of other turtles with [(xcor > [xcor] of myself) and
(lane = [lane] of myself)] [distance myself]) = nobody
[ set speed_desired_now_forward 20 ]
Peter D. Kauffmann Appendix A: NetLogo Model Code 75

[ set speed_desired_now_forward ([speed_desired_now] of (min-one-of


other turtles with [(xcor > [xcor] of myself) and (lane = [lane] of myself)]
[distance myself]))]
ifelse (min-one-of other turtles with [(xcor < [xcor] of myself) and
(lane = [lane] of myself)] [distance myself]) = nobody
[ set speed_desired_now_back 0 ]
[ set speed_desired_now_back ([speed_desired_now] of (min-one-of other
turtles with [(xcor < [xcor] of myself) and (lane = [lane] of myself)]
[distance myself]))]

;; calculate gaps in adjacent lane


ifelse (min-one-of other turtles with [(xcor > [xcor] of myself) and
(lane != [lane] of myself)] [distance myself]) = nobody
[ set gap_forward_other 20 ]
[ set gap_forward_other (([xcor] of (min-one-of other turtles with
[(xcor > [xcor] of myself) and (lane != [lane] of myself)] [distance
myself])) - xcor - 1)]
ifelse (min-one-of other turtles with [(xcor < [xcor] of myself) and
(lane != [lane] of myself)] [distance myself]) = nobody
[ set gap_back_other 20 ]
[ set gap_back_other (xcor - ([xcor] of (min-one-of other turtles with
[(xcor < [xcor] of myself) and (lane != [lane] of myself)] [distance
myself])) - 1)]

; calculate speeds in adjacent lane


ifelse (min-one-of other turtles with [(xcor > [xcor] of myself) and
(lane != [lane] of myself)] [distance myself]) = nobody
[ set speed_desired_now_forward_other 20 ]
[ set speed_desired_now_forward_other ([speed_desired_now] of (min-one-
of other turtles with [(xcor > [xcor] of myself) and (lane != [lane] of
myself)] [distance myself]))]
ifelse (min-one-of other turtles with [(xcor < [xcor] of myself) and
(lane != [lane] of myself)] [distance myself]) = nobody
[ set speed_desired_now_back_other 0 ]
[ set speed_desired_now_back_other ([speed_desired_now] of (min-one-of
other turtles with [(xcor < [xcor] of myself) and (lane != [lane] of myself)]
[distance myself]))]
]

;; LANE CHANGE RIGHT


ask-concurrent turtles with [(on_belt = true) and (ycor = 1) and
(time_since_lane_change >= Min_t_btwn_lanechange)] [
ifelse ((speed_desired_now < speed_desired_now_back) and (gap_back <=
(speed_desired_now_back - speed_desired_now)))
[
;; lane change while tailgated
if ((not any? turtles-at 0 -1)) ;; see if space exists
[
set lane 0
set facial_ellipse 0
]
]
[
;; standard lane change
if ((speed_desired_now >= gap_forward) and (gap_forward_other >
gap_forward) and (gap_back_other >= tolerable_gap_back_other)
Peter D. Kauffmann Appendix A: NetLogo Model Code 76

and ((speed - v_belt) <= gap_forward_other) and (gap_back_other >=


tolerable_gap_back_other)) [

if (not any? turtles-at 0 -1) ;; see if space exists


[ set lane 0 ] ]
]
]

;; LANE CHANGE LEFT ;; test if in right lane, time is over minimum;


also, standers not allowed to merge left
ask-concurrent turtles with [(on_belt = true) and (ycor = 0) and
(time_since_lane_change >= Min_t_btwn_lanechange) and (breed != standers)]
[
if ((speed_desired_now >= gap_forward) and (gap_forward_other >
gap_forward) and (gap_back_other >= tolerable_gap_back_other)
and ((speed - v_belt) <= gap_forward_other) and (gap_back_other >=
tolerable_gap_back_other))
[

if (not any? turtles-at 0 1)


[ set lane 1 ]
]
]

;; PERFORM LANE CHANGES SIMULTANEOUSLY


ask-concurrent turtles with [(on_belt = true) and (lane != ycor)]
[
set ycor lane
set time_since_lane_change 0
]

;; increase time since lane change


ask turtles with [on_belt = true]
[ set time_since_lane_change (time_since_lane_change + 1) ]

]
end

;; A.17 Departures module


to departures
set peds_out (peds_out + count turtles with [xcor >= treads])
set walkers_out (walkers_out + count walkers with [xcor >= treads])
set standers_out (standers_out + count standers with [xcor >= treads])
set commuters_out (commuters_out + count commuters with [xcor >= treads])
if ([xcor] of selected1) >= treads
[
set selected1 min-one-of turtles [xcor]
ask turtles [
loop [
ifelse ((selected2 = selected1) or (selected1 = selected3))
[ set selected1 one-of turtles ]
[ stop ]
] ]
output-type "Selected Ped 1 = " output-print selected1
ask selected1 [ set color orange ]
]
if ([xcor] of selected2) >= treads
Peter D. Kauffmann Appendix A: NetLogo Model Code 77

[
set selected2 min-one-of turtles [xcor]
ask turtles [
loop [
ifelse ((selected2 = selected1) or (selected2 = selected3))
[ set selected2 one-of turtles ]
[ stop ]
] ]
output-type "Selected Ped 2 = " output-print selected2
ask selected2 [ set color yellow ]
]
if ([xcor] of selected3) >= treads
[
set selected3 min-one-of turtles [xcor]
ask turtles [
loop [
ifelse ((selected3 = selected1) or (selected2 = selected3))
[ set selected3 one-of turtles ]
[ stop ]
] ]
output-type "Selected Ped 3 = " output-print selected3
ask selected3 [ set color magenta ]
]
ask-concurrent turtles [ if xcor >= treads [die] ]
set flow (peds_out / ticks * 3600)
if v_belt > 0 [
ifelse ticks >= (treads * 5 / v_belt)
[ set flow_max max list flow flow_max ]
[ set flow_max "stabilizing..." ] ]
set density ((peds_in_belt - peds_out) / Belt_Length)
end

;; A.18 Create Plots module


to create-plots
;; compute system average speeds
set avg_speed_stand ((sum [speed] of standers) / (standers_in -
standers_out) * .4)
set avg_speed_walk ((sum [speed] of walkers) / (walkers_in - walkers_out) *
.4)
set avg_speed_comm ((sum [speed] of commuters) / (commuters_in -
commuters_out) * .4)

;; plot speeds
set-current-plot "Speeds"
set-current-plot-pen "avg Stand"
plot avg_speed_stand
set-current-plot-pen "avg Walk"
plot avg_speed_walk
set-current-plot-pen "avg Comm"
plot avg_speed_comm

;; plot fundamental characteristics


set-current-plot "Fundamental Characteristics"
set-current-plot-pen "Walkers in System"
plot (walkers_in - walkers_out)
set-current-plot-pen "Standers in System"
plot (standers_in - standers_out)
Peter D. Kauffmann Appendix A: NetLogo Model Code 78

set-current-plot-pen "Commuters in System"


plot (commuters_in - commuters_out)
set-current-plot-pen "Density (ped/10 m)"
plot (density * 10)
set-current-plot-pen "Queue Length"
plot queue_total
set-current-plot-pen "Flow (1000 ped/hr)"
plot (flow / 1000)

;; plot time-space diagram


set-current-plot "Time-Space"
set-current-plot-pen "Ped1"
plot (([xcor] of selected1) * 0.4)
set-current-plot-pen "Ped2"
plot (([xcor] of selected2) * 0.4)
set-current-plot-pen "Ped3"
plot (([xcor] of selected3) * 0.4)

;; display time passed


set time ticks

;; output file parameters


if file_output_timespace = true [
file-print (word "TIME,VALUE," ticks "," peds_in "," peds_in_belt ","
peds_out "," Density "," queue_total)
foreach sort turtles [
ask ? [
file-print (word who "," breed "," ticks "," xcor "," Lane "," Speed ","
speed_desired_now)
] ] ]
end
Appendix B – Model Operation Instructions
The instructions contained within this section are taken directly from the support documentation
included with the moving belt system model. All documentation related to this thesis is currently
hosted online at http://dropbox.peterpages.net/thesis, with the final version of the model hosted
at http://dropbox.peterpages.net/thesis/Version_1.1.2.html. The information contained within
this section can be used to operate that model.

B.1 VERSION 1.1


This model was created between November 17-19, 2010 by Peter Kauffmann. It is based on
Version 1.0.3, released November 16, 2010. For internal record-keeping purposes, the specific
version is 1.1.2.

Improvements from Version 1.0.3 to 1.1.2 include:


• the ability to plot time-space diagrams for selected pedestrians
• the option to create a text file containing model parameters, system information, and
pedestrian characteristics at every tick for use in subsequent data analysis. This data can
be used to make more detailed plots as well. This option does not function when
operated within a web browser, only through the NetLogo software.

B.2 OPERATION
1) To configure the model, use the various inputs to set the user characteristics and
composition, operational parameters, and belt characteristics. Alternatively, select a
default scenario to automatically set these inputs to a realistic set of values. Descriptions
of the example scenarios are given below.
2) Press "Setup" to generate the simulation environment, including creating a belt at the
user-specified length and populating it with users.
3) Press "Go" to begin the simulation. To slow down the model, move the slider above the
simulation window to the left to decrease the number of model time steps ("ticks") that
occur per actual second.
4) For an even better view of the tick-by-tick operation of the model, press "Go" again to
stop the simulation and instead advance it one tick at a time by pressing the "Go Once"
button.

More information about the model's operations can be found further down the page.

B.3 EXPLANATION
After the parameters are set by the user in the upper section, the model is generated using the
"Setup" routine. A belt is generated at the user-specified length with alternating colors shown
every 0.4 meters, representing the code-specified tread depth on an escalator in the United States
according to ASME A17.1. The input parameters are also used to populate the belt so that the
flow is roughly accurate from the start.

After pressing "Go," the model generates escalator riders assuming Poisson distributed arrivals at
the user-specified arrival rate and composition. These riders are placed on the model using a
simplified linear queuing assumption at the base of the moving belt. These users move at the

79
Peter D. Kauffmann Appendix B: Model Operation Instructions 80

specified walking speed until reaching the belt, where they are assigned a "speed limit" equal to
the sum of the belt speed (v_belt = 0 for stairs) plus their on-belt speed, which can be specified
based on whether the belt is an escalator or moving walk as well as local parameters.

While on the belt, the user speed is determined by the available space in front of them in
conjunction with the maximum speed that the individual is able to travel as limited by the
previously calculated "speed limit.” Passing behavior is loosely based on TRANSIMS
automotive passing rules with allowances made for pedestrian-specific behaviors and
aggressiveness defined by the user. At the end of the belt, riders are tabulated and removed from
the simulation environment.

B.4 EXAMPLE SCENARIOS


Various example scenarios have been created to show the implementation of escalator
management strategies and different hypothesized aspects of pedestrian behavior. The code
name for each scenario is composed of several different components. For example:
T N V
D 2500 0.4
| | |
| | | |
| | |
| | | Other parameters noted here
| | |
| | Belt Speed (m/s)
| | |
| Initially simulated arrival rate (passengers/hour)
| | |
Belt Type (Down escalator/Up escalator (with ellipse observer %)
/Moving walk)
| | Ignore Initial Lane Assignment? (V=check/X)
| Belt Rules (None/Walk only/Stand only)
Lane Change Allowed? (True/False)

The calculated approximate capacity values were determined by turning on the "capacity test"
option. This prevents queue overflow from halting the model. Simulations were then run at
various arrival rates until the maximum steady-state condition was reached.

==== Scenario 0 - TNV D 2500 0.4 ====


Basic escalator operation: downhill operation, passing allowed, no belt rules, and randomized
walker/commuter lane assignment. Volume is kept low so passing behavior can be observed.
Slowing with distance is ignored.
Approximate Capacity: 6100 ped/hr, which closely matches the 5800 ped/hr value from the 80%
rule.
Lane changing will be turned off for subsequent scenarios in order to reduce variability in the
results.
==== Scenario 1 - FNX D 3500 0.4 ====
Original operation of this model before implementation of passing behavior: downhill, no
passing, no belt rules, strict lane assignment. Observe how the system's capacity is entirely
based on that of the individual lanes and the average speed of walkers and commuters by varying
the walking percentage and the climbing speeds.
-- Approximate Capacity: 7100 ped/hr. Notice how by removing passing behavior we make the
system more streamlined.
==== Scenario 2 - FNX U-95 3500 0.4 ====
Peter D. Kauffmann Appendix B: Model Operation Instructions 81

Same as Scenario 1 except on an uphill escalator with 95% of passengers observing the facial
ellipse (0.4m forward gap required for comfort). Notice how the riders space themselves out and
the resulting capacity drop.
-- Approximate Capacity: 3700 ped/hr, since the belt space is not being utilized as effectively.
Also, note the lowered density.
==== Scenario 3 - FNX U-95 3500 0.8 ====
Same as Scenario 2 except with higher belt speed of 0.8 m/s versus 0.4 m/s. This speed is used
in the Moscow subway because of their extremely deep stations. See how capacity is improved -
and actually doubled - by a faster belt even in a case where capacity has been lowered by the
required pedestrian spacing.
-- Approximate Capacity: 7500 ped/hr, assuming a simplified boarding delay.
==== Scenario 4 - FSX U-95 3500 0.4 ====
Same as Scenario 2 except with a rule implemented where users may only stand on the belt - no
walking/climbing is allowed. Observe the effect on capacity.
-- Approximate Capacity: 4400 ped/hr, which is slightly higher because of the removal of
imperfections in packing due to walking.
==== Scenario 5 - FWX U-95 3500 0.4 ====
Same as Scenario 2 (and Scenario 4) except with a rule implemented where all users must walk
up the escalator. Although this would be impractical to enforce in the real world, observe the
effect on packing efficiency and capacity.
-- Approximate Capacity: 3800 ped/hr - still slightly higher than Scenario 2 because of the
reduction of variability due to the presence of standers, but lower than packing in with standers
alone.
==== Scenario 6 - FNX U-0 3500 0.4 SLOW ====
Same as Scenario 1 except that it accounts for the endurance of the pedestrians. Based on speed
curves found in the literature, pedestrians will slow as they climb an escalator. Note that this
must therefore take place on an upwards escalator, but the percentage of pedestrians observing
the facial ellipse has been set to zero to remove the impact of this factor. Because of the discrete
nature of this CA model, speeds must be reduced in increments of 0.4 m/s (1 cell/second), and
the first of these drops takes place at 20m, which is ten treads from the exit of the escalator.
-- Approximate Capacity: 5100 ped/hr. This reduction is caused by a shockwave present at the
speed drop, which eventually propagates back causing a capacity drop relative to Scenario 1.

B.5 ERROR CODES


Several error codes exist to display problems with a model run.
• If a feature is selected that is not yet supported by the model, the background will turn
red.
• If there is a queue overflow beyond what is supported at the base of the belt (currently
100 people), the background will turn solid orange.
• If an unsupported button is pressed, the background will turn blue.
Peter D. Kauffmann Appendix B: Model Operation Instructions 82

B.6 FUTURE IMPROVEMENTS


If time permits, I plan to construct and implement a more realistic representation of the
bottleneck that users experience while boarding the escalator. My thoughts on this protocol
would be to use NetLogo's potential for continuous rather than discrete simulation to generate
users in a Poisson-distributed manner at a random location some distance away from the mouth
of the bottleneck and set them at a trajectory and speed that will take them towards the belt
entrance. More complicated pedestrian spacing rules would need to be developed to govern the
entrance to the bottleneck, but hopefully this would more accurately model the entry behavior
seen in Hoogendoorn's "Pedestrian Behavior at Bottlenecks" (2005).

The starting setup would look something like this, with pedestrians being generated on the
double lines and queuing into the belt:
===========
||
|| __
|| ====
|| __====
||
||
===========

More documentation on this improvement will be included at its release or in the proposed
research section, depending on my committee's decision.

B.7 RELATED MODELS


General inspiration was taken from the "Traffic 2 Lane" model in the NetLogo library, although
it must be said that there are several bugs with that model, most notably that vehicles will not
change lanes even though the model description says they should.
http://ccl.northwestern.edu/netlogo/models/Traffic2Lanes

The only code within my model that can be sourced directly to "Traffic 2 Lane" are the method
by which the POPULATE-BELT procedure adjusts pedestrians that are placed on an already
occupied cell and the way that the average speed of the various entities are calculated, although I
have had to modify the math involved in each of these processes to fit the configuration of my
model. Additionally, the method by which the user can select pedestrians for the time-space
diagram is influenced by the SELECTED-CAR routine of that model.

The general code structure by which the data is written to an external file was inspired by the
"File Output Example" tutorial model, available in NetLogo's internal Model Library.
Peter D. Kauffmann Appendix B: Model Operation Instructions 83

B.8 REVISION HISTORY


This model was created between November 17-19, 2010 by Peter Kauffmann. It is based on
Version 1.0.3, released November 16, 2010. For internal record-keeping purposes, the specific
version is 1.1.2.

Improvements from Version 1.0.3 to 1.1.2 include:


• the ability to plot time-space diagrams for selected pedestrians
• The option to create a text file containing model parameters, system information, and
pedestrian characteristics at every tick for use in subsequent data analysis. This data can
be used to make more detailed plots as well. This option does not function when
operated within a web browser, only through the NetLogo software.

Changes from Version 0.2.4 to 1.0.3 include:


• several default scenarios for analysis of rule-based and physical system constraints
(described below)
• inclusion of a "capacity test" option
• support for pedestrian passing behavior with user defined aggressiveness
• improved rules for when pedestrians ignore their initial lane assignment
The proposed bottleneck simulation has been suspended from this simulation and will be
attempted in a side project if time permits for future implementation in this model.

Modifications from Version 0.2.2 to 0.2.4 include:


• ability for users to ignore lane assignment if queue difference is too high
• accounting for the surprisingly important difference between "distance to forward
pedestrian" and "gap to forward pedestrian"
• more realistic following rules for close interaction
• acceleration behavior based on the literature instead of my faulty "infinite acceleration"
assumption
• generally, this model has a better implementation of following than v.0.2.2, which can be
seen by the fact that it is capable of passing a realistic number of passengers for a given
belt speed

Changes from Version 0.1 to 0.2.2 included:


• the inclusion of default scenarios (only one so far)
• the implementation of pedestrian following behavior. Before, pedestrians would just
walk right through one another, which is obviously inaccurate but took some time to
implement into the model
• support for rules that constrain passengers to walk-only or stand-only behavior
• the option to have climbers "tire" as they ascend stairs
• the ability to specify the minimum forward free space and the percentage of passengers
who obey this rule in order to model the "facial ellipse" that causes riders to increase the
forward space on a downwards traveling escalator
Peter D. Kauffmann Appendix B: Model Operation Instructions 84

Revisions from Version 0.0 to 0.1 included:


• the option to initially populate the belt with users
• improved output displays and plots.
• math errors (dividing by zero, mostly) involved in setting the belt speed to 0 to model
stairs have been addressed

B.9 CREDITS AND REFERENCES


Updates to this model should be available at http://dropbox.peterpages.net/thesis/

All work by Peter Kauffmann (c) 2010.