Adapting Automated People Mover Capacity to Real-Time Demand Via Model-Based Predictive Control

Adapting Automated People Mover Capacity to Real-Time Demand via Model-Based Predictive Control

M.P. (Mark) van Doorne BSc. December 2015

Adapting Automated People Mover Capacity to Real-Time Demand via Model-Based Predictive Control

M.P. van Doorne BSc.

to obtain the degree of Master Master of Science in Transport, Infrastructure & Logistics at the Delft University of Technology.

to be defended publicly on Wednesday December 2nd, 2015 at 14:00 PM.

Graduation Committee TU Delft Prof. dr. ir. G. Lodewijks TU Delft (3ME) dr.ir. W.W.A. Beelaerts van Blokland TU Delft (3ME) dr.ir. G. Homem de Almeida Correia TU Delft (CiTG)

Supervision NACO Royal HaskoningDHV ir. T. van Vrijaldenhoven Airport Planning & Building Design ir. P.H. Ringersma Airport Planning & Building Design drs. ing. R. Lunstroo Airport Planning & Building Design

Executive Summary

This report contains the ﬁndings of a research on the control of an Automated People Mover (APM) system. APMs are combined passenger transit systems that are an important asset for large airports to support intra-terminal passenger movements and/or provide inter-terminal transit. While APMs were ﬁrst introduced at airports in the early ’70s, the physical and operational system characteristics have not changed much since.

The problem is that the control system of any APM attempts to replicate a predefined schedule by constantly measuring parts of the system (i.e. trains) and take actions to diminish any errors. These schedules are already defined in the design phase of a system and should cope with a peak demand throughout the operational life cycle. Such a design approach inevitably results in that there will be a gap between the actual demand and the available capacity. Especially in airports with large demand fluctuations, this will e.g. incur large over-capacities during down periods.

The objective of this research is to analyse the current ﬁxed schedule control of an APM system and design a more intelligent control logic that adapts the capacity available to the real-time demand. This research will not only address the technical aspect to obtain such an Adaptive Control System (ACS), but will also consider the change in the system state compared to a conventional system. Which system state is better or worse depends on the requirements set by the system owner (i.e. airport). These requirements will be a trade-oﬀ between aspects that concern the economic and sustainable impact of the system and the passenger experience. This is summarised in a research objective:

Determine the feasibility of adapting Automated People Mover capacity to real-time demand, by taking economic, sustainable, passenger comfort and implementation aspects into consideration.

To transform the objective into an actual design, a stepwise approach is taken in which ﬁrst a conceptual outline is made of the ACS, after which a detailed design is tested and evaluated. To determine the generality of the ACS, it is applied to two test case (proposed) APM systems at Amsterdam Schiphol International Airport (AMS) and Shenzhen Ba’oan International Airport (SZX). Thereby, a set of (Key) Performance Indicators (KPIs) is formulated to quantitatively support any conclusion on the performance of an ACS, by measuring:

Passenger experience: platform dwell time, platform & train Level of Service (LoS); System cost: capital cost & operational Cost;

External eﬀect: CO2 pollution.

Conceptual Design of an Adaptive Control System The technological development of Communication Based Train Control (CBTC) for APMs allows for a system that changes states according to a schedule. The goal of the ACS is to replace the currently used ﬁxed schedule with a ﬂexible schedule that adapts to real-time demand. It is therefore proposed that the available CBTC technology that currently controls train movements is combined with an hierarchically higher controller that can change the system capacity in terms of train frequency and train capacity.

iii A known example of a system that already has such a higher level controller is Personal Rapid Transit (PRT), which allocates small vehicles (pods) to a station with demand. However, the problem with the controller type used for this system, is that it reacts to demand initiated by a passenger with a push of a button. This is ﬁne for a system in which a high number of vehicles is available to keep the reaction time low, but will be problematic in a typical APM system with few trains that need more time to anticipate. The result is that passengers have to wait uncomfortably long and instead some form of proactive control is required to activate a train in time.

It is therefore chosen to design an Adaptive Control System (ACS) by means of Model-based Predictive Control (MPC) method. An MPC calculates a set of future actions that as a combination satisﬁes an overall objective based on passenger demand forecast models. The prime objective of the MPC is to minimise the diﬀerence between the system capacity for the next n time steps and the demand forecast for that same period n. This demand is determined by aircraft movements that induce a high and narrow arrival distribution and a lower but broader departure distribution.

While demand characteristics can roughly be calculated based on historical data and airport forecasts, it is preferred to place a sensor at an appropriate distance before the platform such that the minimum forecast period is met. The capacity can either be changed by running more/less trains or increase/decrease the amount of cars per train. The complete control structure is summarised in Figure2.

Figure 2: Adaptive Control System

The MPC can adjust train scheduling and train composition based on changes in demand. The primary action of the controller is to add a 1-car train to the schedule in response to the initial creation of demand, such that the first passenger has to wait a maximum acceptable waiting time (based on system owner’s requirement). If the demand for a scheduled train exceeds the available capacity, there are two approaches to increase capacity further; an additional train can be scheduled before or after the first train and effectively increase frequency, or the train can be extended by an extra car. If in either approach a maximum is reached (no more trains able to be scheduled or no more cars to be added), the other approach should be used to further

iv Graduation Thesis expand capacity. This results in two concept ACS alternatives which can be tested and evaluated together with a reference alternative:

Reference Case: a representation of a conventional ﬁxed schedule operation; ACS 1 - Frequency: an ACS that favours changing train frequency over train capacity; ACS 2 - Capacity: an ACS that favours changing train capacity over train frequency.

Table 1: Summary of Simulation Results AMS

KPI Passenger Experience Cost (daily) External

PI Wait(s) LoS Plat LoS Train Capital Operation kg CO2 Reference Case 90.611 A D $1,314 $560 9,424 ACS 1: Frequency 95.33 A D $2,012 $350 5,874 ACS 2: Capacity 94.18 A D $1,136 $346 5,806 1during daytime

Table 2: Summary of Simulation Results SZX

KPI Passenger Experience Cost (daily) External

PI Wait(s) LoS Plat LoS Train Capital Operation kg CO2 Reference Case 92.31 C D $3,504 $4,454 110,751 ACS 1: Frequency 80 D D $3,107 $1,678 41,714 ACS 2: Capacity 80.42 D D $2,888 $1,757 43,674

Detailed Design of an Adaptive Control System The conceptual design of the ACS is converted into detailed designs for proposed APM systems at Amsterdam Schiphol International Airport (AMS) and Shenzhen Bao’an International Airport (SZX). The simulation results, summarised in the tables1 and2, show a comparable result on the (Key) Performance Indicators (KPIs) for the two ACS alternatives in terms of dwell time, Level- of-Service (LoS), operational costs and CO2 pollution. Capital costs are however consistently lower in ACS alternative 2 (capacity), as fewer cars have to be acquired.

This diﬀerence is caused by the distinctive decision logic of ACS alternative 1 (frequency), which can add an extra earlier train to decrease the waiting time of passengers. A mismatch can occur between the demand forecast and capacity availability when this train is added but passengers miss it and an extra car in the following train is required to compensate. The expected lower waiting time that should result from the logic is thereby insigniﬁcant with comparable average dwell times for the ACS alternatives.

The dwell time results of the ACS alternatives also show a distinctive difference between the two test cases, in which the results are better for SZX than AMS. This is the result of the ACS decision logic that makes a primary decision to schedule a 1-car train when demand is created and then it waits as long as possible to optimally fill that train (180 seconds). Any further adjustments to the schedule or train composition are possible when the demand surpasses the capacity of the first scheduled 1-car train. This does however only sporadically occur in the low demand system at AMS and is more frequent in the control of the SZX system, hence resulting in better results of the ACS in the latter test case.

v The results of the ACS alternatives for the KPI passenger experience are on par with the reference case. The dwell time is similar for AMS and lower for SZX. However the ACS does come with a higher platform utilization, which results in a drop in level of service for a singular platform in the SZX test case (from LoS C to Los D, on a ranking from A=best to F=worst). The LoS D measured on this platform at SZX is acceptable according to IATA standards for short periods, but should generally be redesigned to a higher LoS to meet client requirements. The LoS D measured for the trains in all alternatives is also low, but due to the short transit period transit it is deemed as an appropriate minimum LoS by industry experts.

The ACS alternatives differ significantly from the reference case on the KPIs cost and sustainability. As was concluded before, the ACS alternative that favours frequency over capacity (ACS alternative 1) consistently requires more cars, which results in a larger capital and total cost in the AMS test case. In the SZX case, the costs in the ACS alternatives are consistently lower, though. As the ACS alternatives in both test cases significantly reduce the total run distance, directly proportional operational costs and CO2 pollution are reduced as well.

The eﬀect of an ACS is not only depended on the system scale but also on the design characteristics of an APM system. The relative reduction in run distance (and thus operational cost and CO2 pollution) of the ACS alternatives compared to the reference case is for instance larger for the APM system at SZX. This is due to the availability of parking spaces in the system and the single security status of APM passengers. AMS does not have parking spaces and has to make additional empty runs to vacate stations for scheduled train arrivals. There are also two passenger security states (Schengen and Non Schengen), which can result in an ineﬃciently large train composition. The AMS system on the other hand has much larger planned platforms, with as a result that the LoS remains the same for all alternatives.

Evaluating the feasibility of an ACS for APM systems As there was no information available on the preferences of the system owners (i.e. the airport authorities) during the research, it was not possible to perform a Multi Criteria Decision Analysis (MCDA) and single out a most desirable solution for any of the two test cases. However, based on the research it can still be concluded that the ACS is a feasible concept that can eﬀectively be implemented in APM systems and improve the performance thereof.

Both ACS alternatives show an overall better result compared to the reference case in terms of reducing costs (capital and operational) and increasing the sustainability of the system. While on the other hand the LoS decreases slightly for the SZX test case, the overall comfort that passengers experience is still adjudged to be acceptable. It should be noted that system demand and design characteristics have a considerable eﬀect on the relative result of the ACS implementation.

Favouring a change of train capacity before changing train frequency (ACS alternative 2) is the best approach for an ACS in both test cases. The alternative shows a consistently better result in respect to capital costs, while the expected increase in dwell time compared to the ﬁrst ACS alternative is only limited.

Recommendations This research is an initial step towards an intelligent control of Automated People Movers. Logically, additional research is required to further determine the opportunities of an ACS.

vi Graduation Thesis

It is important that NACO approaches companies within the APM industry and airport clients, to bring the idea of adapting APM capacity to demand in the attention. It should furthermore consider implementing any (future) intelligent control system in the physical design of new APM projects.

Scientiﬁc research is required on other APM systems and possibly on other combined passenger transit systems such as metros and (light) rail, to further test the generality of the ACS. Thereby, research is needed to test the robustness of an ACS to system failures and further test the eﬀect of changing the relatively sensitive headway variable. Furtermore, it is recommended to identify implementation obstacles in respect to CBTC and system design.

vii

Contents

1 Introduction 1 1.1 Research Context: Netherlands Airport Consultants...... 1 1.2 Research Context: Automated People Mover...... 2 1.3 Problem Description...... 3

2 Research Approach5 2.1 Research Relevance & Objective...... 5 2.2 Research Questions...... 6 2.3 Research Methodology: Engineering Design...... 7 2.3.1 Conceptual Design: Controller Types...... 7 2.3.2 Conceptual Design: Controller Structures...... 9 2.3.3 Detailed Design: APM System Test Case(s)...... 10 2.3.3.1 Process Analysis...... 10 2.3.3.2 Modelling & Simulation...... 11 2.3.3.3 Alternative Testing Framework...... 12 2.4 Research Scope & Boundaries...... 13 2.4.1 Automated People Movers...... 13 2.4.2 APM Airport Function...... 15 2.4.3 Test Case Airport Systems...... 17 2.4.4 0% System Failures...... 17 2.4.5 System Design Capacity...... 18 2.4.6 Basic System Schedule Design...... 18 2.5 Key Performance Indicators...... 18 2.5.1 Measuring Passenger Experience...... 18 2.5.2 Measuring Costs...... 19 2.5.3 Measuring External Inﬂuences...... 19

3 Conceptual Design of an Adaptive Control System for Automated People Movers in an Airport Environment 21 3.1 Available Control Systems...... 21 3.2 ACS Controller Structure...... 22 3.3 Central Model-Based Predictive Control Logic...... 25 3.3.1 Internal Workings of the Decision Logic...... 27 3.3.2 Demand Forecast Model...... 29

4 Detailed Test Case Design For Amsterdam Airport Schiphol 31 4.1 Airport Region...... 31 4.2 Airport Characteristics...... 31 4.3 APM System Characteristics...... 32 4.4 Alternative Description...... 34 4.4.1 Reference Case...... 34 4.4.2 ACS Alternative 1: Frequency...... 36 4.4.3 ACS Alternative 2: Capacity...... 37 4.5 Results...... 38 4.5.1 Passenger Experience...... 38 4.5.2 Cost...... 39 4.5.3 External Eﬀect...... 42

ix CONTENTS

5 Detailed Test Case Design For Shenzhen Bao’an International Airport 43 5.1 Airport Region...... 43 5.2 Airport Characteristics...... 44 5.3 APM System Characteristics...... 45 5.4 Alternative Description...... 47 5.4.1 Reference Case...... 47 5.4.2 ACS Alternative 1: Frequency...... 49 5.4.3 ACS Alternative 2: Capacity...... 50 5.5 Results...... 50 5.5.1 Passenger Experience...... 50 5.5.2 Cost...... 51 5.5.3 External Eﬀect...... 53

6 Conclusion & Recommendation 55 6.1 Recommendations for NACO...... 57 6.2 Recommendations for Further Scientiﬁc Research...... 58

Appendices 58

A Airport Pax Transit Systems 61 A.1 Personal/Group Rapid Transport...... 61 A.2 Metro...... 62 A.3 Other APTS Solutions...... 62

B APM System Benchmark 63 B.1 Hartsﬁeld-Jackson Atlanta International Airport...... 64 B.2 Beijing Capital International Airport...... 66 B.3 Birmingham International Airport...... 68 B.4 O’Hare International Airport...... 70 B.5 Dallas/Forth Worth International Airport...... 72 B.6 Detroit International Airport...... 74

C Arena Simulation Model 77 C.1 Arena Software Methodology...... 77 C.2 Model/Simulation Set Up, Validation & Veriﬁcation...... 77 C.3 Model Description...... 81 C.4 Amsterdam Airport Schiphol Model Set Up...... 88 C.5 Shenzhen Bao’an Airport Model Set Up...... 88

D Results 93

Bibliography 95

Glossary 95

x List of Figures 2 Adaptive Control System...... iv

1.1 Bombardier ...... 2 1.2 Mitsubishi...... 2 1.3 Siemens...... 2 1.4 Network layouts...... 3

2.1 Basic Control System principle (Negenborn and Hellendoorn, 2010)...... 8 2.2 Multi criteria decision support methodology (Vreeker et al., 2002)...... 13 2.3 Test Case APM vehicle...... 15 2.4 Available actions between APM and Aircraft movement...... 16

3.1 FBTC (Top) vs CBTC (Bottom)...... 22 3.2 Example Fixed Schedule for Singapore Changi (Lea+Elliot, 2014)...... 23 3.3 Hierarchy in control logic...... 24 3.4 ACS capacity change approaches...... 27 3.5 Increased Train Frequency Favoured Over Increased Train Capacity...... 28 3.6 Increased Train Capacity Favoured Over Increased Train Frequency...... 29

4.1 Arrival and Departure Pattern AMS...... 33 4.2 Example Block Schedule AMS...... 33 4.3 Proposed APM trajectory at AMS...... 33 4.4 Proposed APM Network Layout at AMS...... 34 4.5 Passenger Type distribution (Lea+Elliot, 2014)...... 35 4.6 APM Demand AMS (1 run)...... 37

5.1 Pearl River Delta...... 44 5.2 Arrival and Departure Pattern SZX...... 45 5.3 Proposed APM trajectory at SZX...... 46 5.4 Proposed APM Network Layout at SZX...... 46 5.5 APM demand SZX (1 run)...... 48 5.6 The location of the sensor at Terminal 3...... 50

A.1 WVU PRT (Wikipedia)...... 62 A.2 Heathrow PRT (Wikipedia)...... 62

C.1 Simple representation of a typical system analysed with SIMAN (Pegden, 1983). 77 C.2 Model Sections AMS...... 81 C.3 Module Schedule Generator...... 84 C.4 Module Passenger Generation...... 84 C.5 Module Platform Distributor...... 85 C.6 Module ACS Favour Frequency...... 85 C.7 Module ACS Favour Capacity...... 85 C.8 Module Platform Waiting Process...... 85 C.9 Module Outbound Station Process...... 86 C.10 Module Inbound Station Process...... 86 C.11 Module Parking Process...... 86 C.12 Module Train Continuation Process...... 86 C.13 Train Schedule Module...... 87 C.14 Module Scheduled Train Holding...... 87

xi LIST OF FIGURES

C.15 Tornado diagrams for AMS Sensitivity Analysis...... 90 C.16 Tornado diagrams for SZX Sensitivity Analysis...... 90

xii List of Tables 1 Summary of Simulation Results AMS...... v 2 Summary of Simulation Results SZX...... v

1.1 APM characteristics...... 2

2.1 Types of Systems (Ackoﬀ, 1999)...... 11 2.2 Airport Passenger Transit Systems...... 13 2.3 Diﬀerences Between APM and PRT systems (Furman et al., 2014)...... 14 2.4 Test Case Airports...... 17

4.1 Round trip Single Service AMS (v=50kph, acc=1m/s2, dec=1m/s2)...... 34 4.2 Headway and Train Composition Combinations AMS...... 36 4.3 Results Dwell Time (Day Time Results Between Brackets)...... 38 4.4 Results Level-of-Service...... 39 4.5 Results Cost...... 41 4.6 Results CO2 pollution...... 42

5.1 Round trip Single Service SZX (v=50kph, acc=1m/s2, dec=1m/s2)...... 46 5.2 Headway and Train Composition Combinations SZX...... 48 5.3 Results Dwell Time...... 51 5.4 Results Level-of-Service...... 52 5.5 Results Cost...... 54 5.6 Results CO2 pollution...... 54

6.1 Summary of Simulation Results AMS...... 56 6.2 Summary of Simulation Results SZX...... 56

B.1 APM test cases...... 63

C.1 Arena Modules...... 78 C.2 Results 18 run replications AMS...... 88 C.3 Results 10 Run replications SZX...... 89 C.4 Sensitivity Analaysis...... 91

D.1 Raw Output Data AMS Simulation...... 93 D.2 Raw Output Data SZX Simulation...... 93

xiii

Preface

The document in front of you is the final step towards the degree of Master of Science in Trans- port, Infrastructures & Logistics. For my graduation thesis, I was commissioned by the Airport Planning & Building Design department of NACO to improve the operations of Automated Peo- ple Movers at airports. After 8 months of thorough research, it is this document that contains the final report on my findings that entail two effective alternatives.

Thanks go out to my university committee; Gabriel Lodewijks, Wouter Beelaerts van Blokland and Goncalo Homem de Correia for their positive, but critical feedback during the project. I also want take them for recommending me to do my thesis on a subject that is somewhat out of my ’comfort zone’ (i.e. Automated People Movers), to challenge myself and widen my overall knowledge on airport systems.

I also want to thank my supervisors at NACO; Tim van Vrijaldenhoven, Piet Ringersma and Ronald Lunstroo. It has been a great experience to be part of this professional and exciting company. While agendas were full and meetings were short, there has always been someone to talk to, brainstorm with or to critically review my progress. The last 8 months gave me the opportunity to experience the normal work environment of an airport consultant, with as highlight the chance to participate in the site visit at Shenzhen Bao’an International Airport.

M.P. van Doorne BSc. November 17th 2015

1. Introduction

The air transportation sector has expanded rapidly in the last decades. After the deregulation of the industry in the U.S. in 1978 and the easing of many bilateral agreements in the following years, the sector has grown at an annual rate of about 5% globally (Reynolds-Feighan, 1998, De Neufville and Odoni, 2003, IATA, 2015b). This changing demand does not only mean that airlines have to expand their ﬂeets and operations, but airports need to adapt as well.

Airports are therefore expanding their terminal buildings and/or construct new terminals to accommodate future passenger demand. For such projects, they generally call in the help of an expert third party to consult, improve and develop. The Netherlands Airport Consultants (NACO) is such an expert party for airport (re)development and is therefore always in search of new technologies to design good solutions for their customers.

This thesis is on the improvement of airport passenger transit systems (APTS) and in particular Automated People Movers (APM), which are an integral part of many airport terminal designs. Knowledge within the company on ’simple’ systems such as travelators and buses is adequate but for projects that contain APMs, the help of partner companies such as Lea+Elliot1 is required. NACO is therefore interested to improve its in-house knowledge on APMs, as this will aid in creating better and more comprehensive designs for its clients.

This introduction contains a brief explanation of the company and APMs, so to familiarise the reader with the research context. The chapter is concluded with the problem description.

1.1 Research Context: Netherlands Airport Consultants

Netherlands Airport Consultants (NACO) was founded by Royal Dutch Airlines (KLM) director- president dr. Albert Plesman back in 1949 and the company became renowned for its design of e.g. Amsterdam Airport Schiphol. NACO was taken over by engineering ﬁrm DHV in 2003, which in 2012 merged with Royal Haskoning to Royal HaskoningDHV (de Voogt, 2014). The company has done some major airport projects that include the airports of Kuala Lumpur, Bei- jing and Hong. It has recently won a tender for another major airport together with architectural ﬁrms Foster & Partners and Fernando Romero for Mexico-City (Royal HaskoningDHV, 2015, Foster+Partners, 2015). With these and other projects, NACO has been involved in circa 550 airport developments in over 100 countries, making it one of the major parties in the airport design industry (NACO, nd).

The company has its head oﬃce in the Hague and the ± 150 people that work there account for the fast majority of the company’s employees. A couple of employees are stationed at smaller oﬃces outside the Netherlands to better serve the local markets in e.g. Mexico, Saudi Arabia, U.A.E. and South Africa. 1Lea+Elliot is responsible for the design of approximately 80% of systems in operation

1 Chapter 1 Introduction

1.2 Research Context: Automated People Mover

Automated People Mover (APM), also referred to as Automated Airport People mover, Auto- mated Transit System, Airport People Mover, Automatic People mover or simply People Mover, is a collective noun for systems that transport passengers without the interference and/or control of human beings (i.e. automated). APMs are part of a large group of airport passenger transit systems (APTS) that also includes e.g. buses, metros, personal rapid transit (PRT or Pods) and travelators (moving walkways).

The latter is probably the most common APTS and it functions as a backbone for many intra- terminal connections, allowing passengers to move quickly through lengthy hallways and piers. While technically both the travelator and PRT are fully automated and can thus be denoted as APMs, one generally refers to high capacity vehicles that run on a ﬁxed route and schedule.

Currently there are 48 of these systems in operation at airports, transporting passengers on landside, airside or both. The systems not only diﬀers from other guided transport on automation, but also on the short track distance (average <2km), narrow vehicles, high frequency and high standees to seating ratio.

APM Types Upon indexing the diﬀerent systems, it was found that the type of vehicles generally are quite similar based on capacities and vehicle sizes, but are very diﬀerent when it comes to the support, propulsion and guidance system. The used solutions are summarized in table 1.1.

Table 1.1: APM characteristics

Support Propulsion Guidance Rubber tires AC Electric motor Guide rail Steel wheel DC Electric motor Conventional rail Levitation Cable pulled Monorail Side wheels

The majority of systems make use of a rubber tired system. In most cases, this is either a Bombardier Innovia APM (versions C-100, CX-100, 100, 200 or 300), the Mitsubishi Crystal Mover or the Siemens Airval, of which examples are shown in Figures 1.1, 1.2 and 1.3. The Bombardier system and new Airval system make use of a central guide rail (Bombardier, 2015, Siemens, 2014) and the Mitsubishi and earlier Siemens systems using side-mounted wheels to ﬁnd their way (Mitsubishi HI, 2010, Siemens, nd). This type of system is a derivative of the VAL introduced back in 1983 by Matra, now Siemens (Lardennois, 1993).

Figure 1.1: Bombardier Figure 1.2: Mitsubishi Figure 1.3: Siemens

2 Graduation Thesis

Cable pulled APMs are also used at airports, mainly in combination with levitating vehicles. The cars are lifted 2 millimetres above the surface by means of a single 10 bhp electric motor and pulled along the track (Bares, nd). These systems are slightly more limited than the aforementioned rubber tired systems that can change tracks at will (assuming the track has change points).

Only a single application is found of a monorail system for inter-terminal transport at Tokyo Haneda, which in fact also functions as a transport system further towards the city. A marginally larger number is found for rail mounted APM system, with the 2 station (cable pulled) APM at Birmingham and the 3 station (cable pulled) APM at Toronto.

Network Layouts Different network layouts are used based on e.g. the vehicle characteristics, amount of stations and flexibility required. Elliott and Norton(1999) have determined the different layouts possible, which are shown in figure 1.4.

Figure 1.4: Network layouts

Systems can differ from a single vehicle that runs back and forth on a single rail up to multiple vehicles that operate at the same time on a pinched loop layout. Generally it is the latter that is used due to high capacity and vehicle flexibility (e.g. London Heathrow and Atlanta) in combination with the rubber tired APM vehicles. This layout is however not compatible with cable pulled APMs, which generally use a single lane with bypass or dual lane configuration (e.g. Detroit and Zurich).

Full details and characteristics of the diﬀerent APM systems currently in use at airports are given in AppendixB. An example case is included for all systems.

1.3 Problem Description

While the aforementioned APMs serve their goals in transporting passengers at an airport, their operations are still far from optimised. An important inefficiency is that all systems use a predefined capacity (ACRP, 2012b). In many cases, the capacity is fixed throughout the day based on the peak demand and in some cases this approach is marginally improved by running a set schedule.

3 Chapter 1 Introduction

Such an approach is relatively simple and generally fulfils the required task, but is de facto inefficient. Demand fluctuations throughout a day are a common phenomenon at airports (especially at transfer/hub airports), with substantial differences between peak and off-peak periods. When one would use an APM at an airport nowadays, he or she will notice that outside peak periods, the trains are as good as empty. On the other hand, riding the same train during peak periods can be a cramped experience and passengers might even have to wait for the next train. This problem should therefore be assessed in this research and an eventual solution should be designed.

4 2. Research Approach

This chapter contains a comprehensive explanation of the research approach. First, the scientific relevance of researching the problem stated in the former chapter is briefly discussed and a research objective is formulated that should be met to sufficiently solve the system problem (Section 2.1). The research objective is then transformed in a main research question and a set of supportive research questions that will function as a framework (Section 2.2). Thereafter, the scientific methods to analyse the system and design, test and validate alternatives is discussed in Section 2.3. The last section (2.4) of this chapter contains the scope & boundaries of the system.

2.1 Research Relevance & Objective

As was brieﬂy explained in the introduction, APM utilisation is increasing at large airports, but operations are still far from optimal. Although APMs have been around for almost 50 years, not so much has changed from a technical perspective. There have been some new concepts such as hovering vehicles, cable pulled systems and more complex track lay-outs, but vehicles have remained nearly identical and operations are unchanged. For example, the APM system at Capital Airport in Beijing was built only recently in 2008, but still makes use of the Bombardier Innovia CX-100 vehicle, which was introduced back in 1970 at Miami International Airport (Elliott and Norton, 1999).

It is not completely clear why airports have never required a more adaptive operation, but most likely this is due to the manufacturers that provide a homogeneous product. Thereby, APM system design is done by a only a couple of parties that compete in a market that is dominated by Lea+Elliot. When assessing the commercial solutions and scientiﬁc literature available, it is clear that the application of adaptive operations for APMs is not actively researched. No scientiﬁc papers could be found for any form of demand driven APM control and only a single manufacturer (Siemens) acknowledges that it can supply systems with some adaptive capability for train frequency only (?), but only in situations that the normal schedule cannot be operated (i.e. system failures).

A probable reason that research is still limited, is that technological enablers for eﬀective demand measurements and/or capacity adaptive control have not been around for a long time. However, technological developments on both aspects have been extensive in the last decade and as a result, some initial research on the opportunities of adaptive control is done for similar guided vehicle transit systems such as metro and common rail networks (Wang et al., 2010, Lin and Sheu, 2011, Sheu and Lin, 2012).

The recent introduction of for instance Communication Based Train Control (CBTC) has strongly increased the operational capabilities. Although CBTC is a generic concept that is used in a variety of guided vehicle transit systems, it can be ﬁtted with Automatic Train Protection (ATP), Automatic Train Operation (ATO) and Automatic Train Supervision (ATS) functions (Schifers and Hans, 2000). The overall goal of CBTC is to constantly update the system to a set timetable and adapt to it accordingly. Simply put, CBTC has the capability of running fully automated, but currently misses the right control logic that is able to make adaptive time table changes based

5 Chapter 2 Research Approach on system policies and demand measurements. In combination with new airport improvements such as Collaborative Decision Making (CDM) and advanced high resolution sensor systems that use WiFir, BluetoothTM, infra-red and/or CCTV technology, the technological means are however available to design such a control logic for an APM system in an airport environment (Eriksen, 2002, Malinovskiy et al., 2012, Kim et al., 2008, Woodman and Harle, 2008).

The objective of this research is to analyse the ’simple’ scheduled control system of an APM and design a new adaptive control systems, further referred to as ACS, that seeks an optimum system state to adapt capacity to real-time demand. The primary goal is to determine if an ACS will result in a better optimum system state than a conventional system.

What an optimum system state is, depends on what the system owner (i.e. airport) wants. This will most likely be driven by the economical aspect of the project, but due the global perception on sustainability and growing competition between airports, other important factors are energy use and passenger comfort.

A minor aspect to the design of an ACS is that it would be favourable if for the implementation of a solution, characteristics can be shared with current systems. It should however not be seen as a restricting factor if a design can only be obtained with a change of technology.

The whole research objective is summarised as:

Determine the feasibility of adapting Automated People Mover capacity to real-time demand, by taking economic, sustainable, passenger comfort and implementation aspects into consideration.

2.2 Research Questions

To achieve the stated research objective, a main research question is deﬁned:

What are the value drivers for the design of a control system to adapt Automated People Mover system capacity to real-time demand?

To answer this main research question, the content is broken up into smaller supportive questions that act as a framework for this thesis. Question 1 is important to allow any comparison of solutions, whereas question 2, 3 and 4 answer how the ACS should conceptually translate demand into an appropriate capacity (considering the current technology available). Question 5 results in a set of alternatives for the ACS, which in turn should be tested and compared with the current situation to answer question 6.

1. What measures the performance of an Automated People Mover system at an Airport?

2. What is the current approach to control an Automated People Mover system at an Airport?

3. What determines capacity and how can capacity be changed?

4. What determines demand and how can demand be measured?

5. How can capacity be adapted to demand?

6. Does an adaptive control systems alternative signiﬁcantly improve the performance of an Automated People Mover system compared to a conventional control system?

6 Graduation Thesis

2.3 Research Methodology: Engineering Design Paragraph Summary: The theory of Engineering Design will function as a framework for the research approach. In the conceptualisation of the ACS logic, a single or set of On-Off controller(s), Proportional controller(s), Proportional, Integral, Derivative (PID) controller(s) and/or Model-Based Predictive controllers (MPC) will be used in either a single-agent, distributed or hierarchical structure. The ACS will be tested with two test cases for which the process is first analysed with a flow chart and thereafter modelled and simulated in the Arena/SIMAN environment. A Cost-benefit Analysis is used as primary alternative comparison method and incorporates all monetary output parameters from the simulation. If any non-monetary results are measured, these should be prepared for a Saaty and/or Regime analysis. The latter two analyses should only be performed if reliable weights can be appointed to the different output parameters.

A robust research framework is needed to come from an initial problem to a comprehensive conclusion. As the focus of this research is to design an ACS, the Engineering Design theory is chosen as the framework for the overall project approach. Engineering Design is used to systematically transform a problem into a task and eventual product or theory in a stepwise process (Ertas and Jones, 1996, Pahl and Beitz, 2013).

It is important in any design research, to ﬁrst of all determine the objective and state a set of supportive research questions (Sections 2.1 and 2.2). While these are just the initial steps of the research, Verschuren and Doorewaard(2007) explain that the objective should be clearly formulated (what should be solved) and already contain a rough outline on what aspects of the system should be considered. Setting an outline in an early stage of the research could limit the solution space, but is a necessity to obtain a comprehensive result within the the time and scope available for a typical thesis research.

The next step is to analyse the available literature on and actual use of the design product (APMs, control theories, etcetera), after which the design is conceptualised. This conceptualisation is on a high level and is needed to ﬁnd the probable feasibility/implementability of the concept and set the minimum requirements that the design should have. This is followed by a more detailed design that can be tested and evaluated.

The last steps of the Engineering Design process are the production design and actual implementation of the system. These steps are however beyond the scope of this research in both terms of complexity and time.

This section contains a further explanation of the methods to perform the individual steps. This includes the control theory on which to conceptualise a design and the methods to deﬁne, test and evaluate a detailed design.

2.3.1 Conceptual Design: Controller Types

The basic idea of a control system is that some sort of control entity (e.g. a human or a computer) can perform an action u to adjust the state x of a system. A particular distinction can be made between open and closed loop control systems.

7 Chapter 2 Research Approach

In an open loop control system, a single action is executed by the controller after which the system state changes. The drawback of this logic is that the system cannot be adjusted for disturbances. A closed loop control system should be used instead, as it includes a feedback mechanism (Kuo, 1981). By means of measurements y the system state is checked after a controller action is made. It can then readjust the system based on the additional information if the system state has diverted from what was expected due to a disturbance d (ﬁgure 2.1). As the goal of the ACS is to control and adjust the system based on changing demand (i.e. disturbance), only closed feedback loop logics will be considered.

Figure 2.1: Basic Control System principle (Negenborn and Hellendoorn, 2010)

The action taken by a controller depends on the logic it uses. This logic can diﬀer from very simple calculations to complex model-based decisions. The possible controller types are therefore summarised in this section to give the reader a better understanding of the methods available. Which of these controllers should eventually be included in adaptive control system is depended on the system requirements and are determined at a later stage in the conceptualisation phase.

On-Oﬀ Controller The simplest control structure is an on-oﬀ or “bang-bang” controller that generates a

Boolean action u(t)min or u(t)max. The error e(t) that actuates the control is deﬁned as the

diﬀerence between the measurement y(t) and the set point (desired value) hsp (Bequette, 2003). The resulting algorithm is shown in equation 2.1, in which the factor δ represents a threshold value compared to set point.

if y(t) > hsp + δ, then u(t) = u(t)min

if y(t) 6 hsp − δ, then u(t) = u(t)max (2.1)

if hsp − δ < y(t) < hsp + δ, then u(t) = current value(u(t))

An on-oﬀ controller is far from optimal, especially for transport systems. For example, one would be reluctant to step in a train that can only fully engaged or disengage its propulsion or brakes.

Proportional Controller The proportional controller is a more advanced control logic that executes an action u(t) that is proportional to the error e(t). u(t) can take a value from a continuous range that is equal to the standard output required in a steady state system, also known as the bias

term b, combined with the error that is multiplied with the proportional gain kp.

u(t) = kp ∗ e(t) + b (2.2)

8 Graduation Thesis

While this control logic is more adaptable than the on-oﬀ controller, it still has some diﬃculties in calculating an optimal action.

Proportional-Integral-Derivative (PID) controller The PID controller is the most commonly used controller application in industrial systems. It incorporates the aforementioned proportional control and combines this with a integral component and a derivative component. As Araki(2002) explains, the proportional control only assesses current measurements to make decisions, whereas the PID control also considers earlier decisions and can to some point predict future decisions. This is possible as the integral part of the error will increase if the former action made by the controller is to small or large, thereby tuning the action over time. The diﬀerential part thereby corrects the following action based on the rate of change in the error, which will approach

0. The resulting formula is given in equation 2.3, where kc is equal to kp and τI and τD are the ratio of the integral gain factor kI and the derivative gain factor kd . kp kp

1 Z t de(t) u(k) = kc e(t) + e(t)dt + τD (2.3) τI 0 dt

Model-Based Predictive Controller (MPC) The shortcoming of the aforementioned control logics is that they solemnly rely on feedback measurements. This is viable in systems that are quickly adaptable or encounter only limited excessive disturbances, but are less optimal for systems that show more capricious behaviour. The model-based predictive control is therefore developed, which incorporates the fun- damental feedback logic to measure current state x(t) as is used in the former control solutions and combines this with an applicable model C with which it makes a prognosis of the future systems states x(t + n) and thereby determines an action u(t) (equation 2.4) (Morari and Lee, 1999). u(t) = C(y(t))) (2.4)

The value n is case speciﬁc and the appropriate action u(t) is determined by n objective function J that considers all future actions u(t + n). this objective function J is shown in equation 2.5 and consists of one or more sub-objectives which should meet an appropriate value (e.g. minimisation or maximisation). The constant α gives a weight to the respective sub-objective. N X J(y(t + n), u(t)) = α ∗ Jx(yx(t + n), ux(t)) (2.5) x=1

2.3.2 Conceptual Design: Controller Structures

A system can be controlled by a single controller or a combination of multiple controllers. The three controller structure types explained below are commonly used in theory and practice (Ne- genborn and Hellendoorn, 2010):

• Single-agent controller: In this control structure there is only one control agent, which controls all the actions and receives all the information from the system. In this structure the optimal performance can be reached most easily, as no negotiation between controllers

9 Chapter 2 Research Approach

is necessary. However, the computational requirements are high, as it has to process all measurements and actions simultaneously. Thereby, it is not robust to failures, which will shut down the whole system.

• Multi-agent single layer (distributed): In a multi-agent single layer controller there are multiple agents that each control their own part of the system. These agents only control the actions, and receive information from their own system. In this control structure it is possible that the systems can or cannot communicate with each other. It becomes more difficult to achieve optimal performance when multiple agents are involved, as each of the different systems might behave differently and if agents optimise there own part of the system, this could negatively affect the overall system performance.

• Multi-layer control structure (hierarchical): In this control structure there are multiple agents that operate at diﬀerent hierarchical layers. For instance, at a lower layer in the hierarchy the systems are controlled by controllers with a distributed control structure, these controllers try to optimize their own performance. The controller at a higher layer inﬂuences the controllers at the lower level, and this controller tries to optimize the performance of the whole system.

Similar to the controller types, the correct structure is depended on the system requirements and is determined at a later stage in the conceptualisation phase.

2.3.3 Detailed Design: APM System Test Case(s)

A (set of) test case(s) is required to validate the functioning of the ACS and measure the eﬀectiveness of real time adapting compared to conventional APM control. The goal is to ﬁrst conceptualise an ACS that is generic to any (realistic) APM operation and thereafter build a detailed design test case for an APM system at an airport. As time in this project is limited, the number of test cases is 2. The representative cases should together contain a diverse set of possible APM system characteristics.

2.3.3.1 Process Analysis

As the research is focused on a process, it is good to (graphically) comprehend the complexity and therefore several methods can be used such as; the ﬂow chart, IDEF0, UML, swim lane and value stream mapping (VSM). With changing demand and capacity as a main focus, it is important that the method allows for a decision logic and therefore the VSM method is not preferable. While the essence of Lean, of which VSM is a supportive tool, is something that should be incorporated in the overall research (i.e. standardisation, decreasing wastes, etc.), the processes assessed with VSM are generally repetitive (Womack and Jones, 2010).

UML and swim lane are possible approaches to analyse the system, but incorporate complex aspects to allow a system analysis with multiple stakeholders (Bergenti and Poggi, 2000). As there is only one stakeholder involved (i.e. only the airport), these features are however redundant and less complex methods will have the same result When choosing between the Flow Chart method and IDEF0, the former is preferred, as it is simpler but with similar eﬀectiveness. The advantage of IDEF0 is that it allows describing decision logics that are aﬀected by both internal and external information/support. As the system at hand concerns an automated system in an enclosed environment, all decisions are however made internally.

10 Graduation Thesis

2.3.3.2 Modelling & Simulation

Modelling & Simulation (M&S) is needed to test the system and obtain quantitative results to determine the eﬀectiveness of an ACS. It is therefore chosen to use discrete event modelling & simulation (DEM/S) in an Arena/SIMAN environment.

According to Ackoﬀ(1999), it is important to match the M&S decision method to the actual system decision method. He therefore proposes 4 types of systems in which decision can be made on system level and/or parts level (Table 2.1). The system level in an APM network is for instance the scheduling of the APM vehicles, whereas the parts can be the APM vehicles on the network, aircrafts arriving and/or passengers that make use of the system.

Table 2.1: Types of Systems (Ackoﬀ, 1999)

Parts Whole Example

Deterministic No Choice No Choice Clock Ecological Choice No Choice Nature Animate No Choice Choice Person Social Choice Choice Corporation

The ACS will represent an ’Ecological’ or ’Social’ system, as choices in the system are at least made by the parts such as, passengers walking at their own pace and aircraft arriving at a non- scheduled moment. It however depends on the type of control structure used in an alternative, if choices can be made at a system level.

The controller structure that is chosen from the former section can thus inﬂuence the system M&S method. If for instance a system is controlled with a ﬁxed schedule, no further choices can be made on a system level and the system will be considered ’Ecological’. If on the other hand an ACS is used, choices can be made on a system level and the system becomes ’Social’.

According to Verbraeck (Course lecture, 2014), it is best to use DEM/S for ’Ecological’ systems, whereas Agent Based Modelling & Simulation (ABM/S) would better suit a ’Social’ system. DEM/S represents a system by having entities endure a discrete sequence of events in time to alter their own state and/or that of the whole system (Banks et al., 1998).

In ABM/S the complexity of a true system can be much better represented as choices of an agent are not exclusively bounded to a predeﬁned range, but they can in fact show emergent behaviour (Macal and North, 2010). As it is however undesirable to build the alternatives in diﬀerent M&S environments, a choice is made to favour DEM/S over ABM/S for a couple of reasons:

1. The inﬂuence of individual agents that can show emergent behaviour (i.e. humans and aircraft) is limited; 2. Automated Agents in the APM network (i.e. controller and trains) are bound to rules to prevent emergent behaviour; 3. The researcher is more experienced with the DEM/S.

DEM/S is a popular method to represent systems and a wide array of tools and languages have therefore been developed over the years. Currently, the market leaders are SimioTM and Rockwell

11 Chapter 2 Research Approach

Arenar Simulation Software. The prime advantage of these two tools over the competition is their visual modelling dashboard and the ability to represent the system via easy animation. This will not only help in explaining the model to anyone else than the builder, but also help the latter in test and tune the model appropriately.

Arenar is chosen to model the system in this research due to availability of the program, experience of the researcher and easy implementation of ﬂow chart system descriptions. The software methodology and language is summarised in AppendixC.

2.3.3.3 Alternative Testing Framework

An alternative testing framework is needed to create a comprehensive conclusion on the eﬀec- tiveness of the ACS, based on the results of the simulation model.

The general approach to compare alternatives in infrastructural projects, is to perform a Cost Benefit Analysis (CBA) (Vickerman, 2007, Layard et al., 1994). The CBA method compares the cost and benefits of a project and thereby determines the most profitable (or least costly) solution. All factors that are not expressed in a monetary unit, are converted in the relative value (e.g. meter to AC).

As the costs are generally assumed too low, overspending is a large problem with infrastructural projects. To make the results more realistic, Salling and Banister(2009) propose an adjusted method known as the CBA-DK, which corrects the expected costs with statistical distributions of factor costs, based on other infrastructural projects.

Nonetheless, CBA or CBA-DK are limited and less reliable when more soft criteria (i.e. non- monetary) are included in the research. Vreeker et al.(2002) therefore propose a framework that combines CBA and Multi Criteria Decision Analysis (MCDA) to better comprehend the solution space. They state that especially airport developments and expansions require such an approach, as airports are projects where costs are not necessarily the prime criteria. Airports act as the entrance to a city or country and therefore soft criteria such as aesthetics, passenger comfort and/or sustainability are of importance too.

Non-monetary criteria are considered aside from a normal CBA and are included by using a proper MCDA method, notably Regime method, Saaty’s AHP method and the flag method (graphically explained in figure 2.2). Any MCDA method is based on (subjective) weights and can only be effectively used if these are reliable. Generally, the weights are set by a system owner and should be a trade-off of the importance of different output parameters.

It should be noted that if no reliable weights can be determined for a test case within the research time frame, it is of no use to conduct a MCDA. The non-monetary output parameters should in this case only be prepared for a MCDA, so that a comparison of the diﬀerent alternatives can be made at a latter stage.

12 Graduation Thesis

Figure 2.2: Multi criteria decision support methodology (Vreeker et al., 2002)

2.4 Research Scope & Boundaries Paragraph Summary: A conceptual design for an APM car is used as reference case in the research, for which the characteristics are determine with a benchmark study. No other APTS are considered in terms of physical characteristics, but the adaptive control logic of the PRT will be further researched. The test case APM system should only be accessible to passengers and honour the ’must-ride’ principle (i.e. only mode of transit between two locations), which makes it possible to create reliable demand patterns based on aircraft movements. The test case airports will be Amsterdam Schiphol International Airport (AMS) and Shenzhen Bao’an International Airport (SZX) that together contain a broad range of typical APM system characteristics. Furthermore, it is assumed that the automated parts of the system will endure no failures, the system should cope with peak demand for the day with the 30th annual peak hour and that a representation of a ﬁxed schedule will only distinguish a diﬀerent capacity for day and night.

To obtain a scientiﬁc relevant result, it is important that the research is made concise. A set of boundaries to the system is therefore set to limit the scope.

2.4.1 Automated People Movers

There is a large variety of APTS, which is sorted and summarised in table 2.2. An extensive explanation of APMs has already been introduced in chapter1 and the characteristics of all other APTS are found in AppendixA.

Table 2.2: Airport Passenger Transit Systems

Mode Availability Continuous Discrete Control Manual Walking Train, Metro, Bus, taxi Automated Travelator APM, Metro, PRT

13 Chapter 2 Research Approach

Table 2.3: Diﬀerences Between APM and PRT systems (Furman et al., 2014)

APM PRT Operates like an automated bus: fixed route, Operates like an automated taxi: no fixed vehicle may have multiple stops and starts route, vehicle travels non-stop from origin sta- from origin to destination, and stations may tion to destination station, and stations are be on or off the main line (but are typically located off the main line on the main line) Passengers gather in groups with strangers Passengers can travel alone or with chosen companions Passengers must wait for a vehicle on a fixed Passengers may schedule vehicles at their con- schedule venience

As it is not the goal of this research to make a comparison of APTS, only APM systems are considered in this research. The reason for this choices is summarised in this section.

Discrete System In essence, all systems could be assessed and tested (except for walking), as some sort of scheduling is required. Nonetheless, it is deemed unnecessary to consider a continuous system such as a travelator, as due to its continuous availability, an ACS is simple and many commercial solutions are already available. Adaptive travelators are widely used and can instantly start or stop operations when detecting passengers that enter the system (Mitsubishi Electric, 2015).

Instead, discrete systems impose a larger challenge as activating capacity requires considerable eﬀort (personnel availability, vehicle availability, vehicle location, etcetera).

Automated As is shown in table 2.2, there is a variety of discrete systems in use at airports, but for this research the system is bounded only to automated systems. The reasoning hereof is that whether the vehicle is controlled by human or computer, both have to follow strict rules and policies when executing actions. Any sort of control should therefore have the same result (e.g. a train starts moving or stops at a station). With the prime goal of this research in mind it is important to have as many factors kept the same, thus favouring automated systems that have a much higher repeatability.

The choice is made to consider the physical infrastructure and train characteristics of APM system for this research and use the currently used control approach as reference case. The reason for this choice is that the usage of APMs at airports is higher than either a metro or PRT.

It should be noted that an analysis of the control logic of PRT systems is beneﬁcial to this research. A comparison of PRTs and APMs (table 2.3) shows that whereas APMs use a ﬁxed operation, PRTs are well capable of operating adaptively.

APM Reference Case As it is not the goal of this thesis to determine the diﬀerences between diﬀerent APM system and/or favour one over the other, the APM characteristics are taken generic, albeit based on available systems.

14 Graduation Thesis

To allow for free movement through the network (i.e. forwards, backwards and between tracks), a rubber-tired system is the only available solution. The APMs that have such a feature are build by either Mitsubishi (Crystal Mover), Bombardier (Innovia APM 100/200/300) and Siemens (AirVal) and all share primary characteristics. A benchmark research is done (AppendixB) on these systems and the resulting mean characteristics are summarised below.

A single car is guided by a central rail and is roughly 12.00 metres in length, 2.80 metres in width and 3.40 metres in height. One car can transport some 60 passengers, based on the of 0.36 m2/pax, which is adequate for a short transit in peak periods (Lea + Elliot, 2009). Energy and environmental information is based on the most recent APM developed by Bombardier (Innovia

APM 300), which consumes 2,56 kWh/km, expels 1470 gr/km CO2 (Europe Average) and has a life cycle of 30 years (Bombardier, 2015). The acquisition value of an APM is $2.4 million, which corresponds with a single car cost for the Innovia APM CX-100 system (Kimley Horn, 2014).

The speed and acceleration that a vehicle can attain varies per system. It is hereby important to distinguish operational speed and maximum design speed, which can diﬀer substantially. The maximum design speed of an APM vehicle is mostly 80 kph (Bombardier, Siemens, Mitsubishi), but due to the distinctively short distances of an APM system, the operational speed is generally around 50 kph. For the acceleration and deceleration of the vehicles, the assumption is made m that both are 1 s2 . A render of the reference vehicle is shown in Figure 2.3.

Figure 2.3: Test Case APM vehicle

2.4.2 APM Airport Function

The demand that is generated for an APM system depends on the function that it has at an airport and the locations that are served. Based on a thorough survey of current APM systems (AppendixB) it is determined that APMs are used for 4 distinct functions:

1. Inter-terminal transit;

2. Intra-terminal transit;

3. Terminal-to-parking transit;

4. Terminal-to-public transport transit.

15 Chapter 2 Research Approach

(a) Departure

(b) Arrival

Figure 2.4: Available actions between APM and Aircraft movement

An important aspect of these functions is the must-ride principle that entails the necessity for people to make use of the APM system. A terminal to satellite connection is an example of such a must-ride as all passengers that ﬂy in and out from the satellite will make use of the APM (e.g. Atlanta & Zurich). It is therefore possible to set the APM demand equal to the satellite passenger throughput. If the APM is however used as an additional means of transportation and passengers can also walk to their destination, only part of the passengers will use it (e.g. Detroit).

In coherence with the must-ride principle, a distinction should also be made between APMs operating on landside, airside or both. The reason for this is that demand on airside is only dictated by passenger ﬂows, whereas landside systems can be used by anyone (e.g. meeters, greeters, employees or business). This means that the most apparent driver for airport APMs, the arrival and departure of an aircraft, is not necessarily the only reason for transport demand in all APM cases. Other driving forces could for instance be employment, auxiliary businesses, shopping and plane spotting. For this research, only systems are analysed that contain the must- ride principle and can only be accessed by passengers, so that demand patterns can eﬀectively be deducted from available aircraft movement data.

In ﬁgure 2.4a and 2.4b the relation between an aircraft movement and an APM movement of a single passenger is visualised. There is a variety of actions that could be made in the meantime, which dictate the time shift between a passenger being in the aircraft and being in the APM. Such a time shift can be represented by a stochastic distribution.

16 Graduation Thesis

2.4.3 Test Case Airport Systems

Amsterdam International Airport Schiphol (AMS) and Shenzhen Bao’an International airport (SZX) are chosen for the test cases. Due to the time limitation of this research, a mix is sought in two airports that together contain as many typical characteristics of airports. This will ensure a good understanding of the eﬀectiveness and generality of the ACS.

Neither airport has an APM in operation, but both have considered the implementation of one. NACO has been involved in both designs and adequate information is thus available on the ’would-be’ characteristics of the APM systems (size, length, stations, track lay-out, etcetera). The individual characteristics are shortly summarized in table 2.4. Especially the diﬀerence in

Table 2.4: Test Case Airports

Characteristic AMS SZX Annual Passengers (2013) 52.6 million 32.2 million Annual Passengers (2020) 54.9 million – Annual Passengers (2040) – 67.0 million Region Western Europe Far East Asia Design configuration Single Terminal Two terminals + Satellite Transfer rate 40% <2% Domestic/International* 33%/67% 98%/2% APM system length 0.8 km 2.7 km Stations 3 3 Track Lay-out Pinched Loop Pinched Loop *AMS differentiates Schengen/Non Schengen movement types (transfer/origin-destination and domestic/international) generate very different demand patterns, which are of interest to the research. It should be noted that the annual passenger values given in Table 2.4 are not representative for the research. Both APM systems are considered for their respective design year, which is 2022 for AMS and 2040 for SZX. Forecasts models for both passengers and aircraft movements are made by the respective airports and these are assumed to be adequate to generate a concise future demand.

2.4.4 0% System Failures

Failures such as car or network break downs in the system can heavily inﬂuence the operation, which can result in delayed or cancelled APMs. As Ledoux1 (personal communication, 2015) explains, the Automatic Train Supervision (ATS) is the governing control system that will ensure an optimal routing plan in such an event. It is therefore deemed unnecessary to pay further attention to system failures in this thesis, as the ACS will replace the normal system operation and it can thus be assumed that any back up plan executed by an ATS will still function in an adaptive situation. Thereby, APM systems show extreme reliability due to there enclosed environment and automated operations (ACRP, 2012a).

1technical advisor at Siemens S.A.S., Mobility Division

17 Chapter 2 Research Approach

2.4.5 System Design Capacity

Airports experience demand ﬂuctuations on a day, week, month and annual base and it is therefore not advisable to design system capacity for the absolute peak demand. If an airport would decide to do so, it will experience excessive overcapacity.

There are numerous methods to determine the maximum demand that the system should handle (Reichmuth et al., 2011). The methods are intrinsically based on client requirement, but the most common approach is to take the day that includes the 30th peak hour of the year (Wang and Pitﬁeld, 1999), which is in line with the 95% certainty of appropriate service provided, that is used for airport systems designs in general (ACRP, 2012a, Sloboda, 2009).

2.4.6 Basic System Schedule Design

All APM systems use a ﬁxed schedule with a high frequency service during the day and a low frequency service during down times (e.g. at night). While it is possible to prepare the system with a changing schedule for a day, the assumption is made for this research that only a day and a night operation is used. The day frequency is a usually a couple of minutes and is determined based on client requirements, whereas the night frequency is assumed to be 15 minutes to oﬀer minimal service.

2.5 Key Performance Indicators

It is important to have measurable criteria, also known as Key Performance Indicators (KPIs), to quantify and compare any alternative with a reference (current) APM operation. A distinction can be made of three KPIs:

• Passenger experience;

• System Cost;

• External eﬀect.

The KPIs give an overall indication of the respective system aspects and are combinations of several supporting Performance Indicators (PIs).

2.5.1 Measuring Passenger Experience

The overall passenger experience is measured by means of three PIs. The most basic PI is that passengers are transported in a reasonable time that is composed of a dwell time and a transit time. However, the assumption is made that trains run at 100% certainty, which automatically means that transit times will never differ. Therefore, only platform dwell time should be measured in this research to test the effective transit of passengers. The period of the acceptable waiting time is case specific and relies heavily on the customer requirements.

The two other Passenger Experience PIs concern the space that passengers have during their APM transit period on the platform while waiting and in the APM. By means of the IATA/Fruin standards it is possible to determine the level of service (LOS) in the respective areas with which a ranking can be made.

18 Graduation Thesis

2.5.2 Measuring Costs

A large infrastructural project like an APM system affects the airport owner financially and it is therefore important to measure and compare the financial implications of the alternatives. This financial impact is partly based on capital costs and partly on operational costs.

Capital Cost Performance Indicators The capital cost factors that are applicable to an APM system are:

• Tunnel system

• Guidance network

• Switch systems

• Control System

• Platforms

• Platform access

• Passenger Sensor systems

• APM Vehicles

Not all capital cost factors should be considered in this research, because the networks are considered to be the same in all alternatives and only the operations are aﬀected by the ACS. Those factors that will diﬀer and thus function as a PI, are the amount of passenger sensor systems and the required amount of APM vehicles. With the assumption that no failures occur, the amount of vehicles will not consider a surplus for maintenance and/or backup. To determine cost the costs, the PIs should be multiplied with the respective unit value.

Operational Cost Performance Indicators Human labour cost will not be considered as PI. The APM is an automated system and only a handful of employees reside in the control room to supervise the system, which will be the same for all alternatives.

Operational costs are instead dictated by the usage of the system, which is expressed in vehicle energy cost and vehicle maintenance cost. Both factors are PIs for this research and can be measured with the run distance or run time.

2.5.3 Measuring External Inﬂuences

The last aspect that should be measured is the external inﬂuence of the system, which is in this case the pollution of the system. This PI is calculated by multiplying the energy consumption with the average CO2 pollution of energy production that is characteristic for the airport region.

Pollution can be expressed in more emission types, such as NO2 and PM(x), but these emissions are just as CO2 directly proportional to the run distance. It is therefore chosen to only monitor the eﬀect in terms of CO2.

3. Conceptual Design of an Adaptive Control System for Automated People Movers in an Airport Environment

Chapter Summary The control of current APM operations attempts to replicate a predefined schedule by constantly measuring parts of the system (i.e. trains) and take actions to diminish any errors. A known example of a system that does adapt capacity to demand is Personal Rapid Transit, but its control logic is purely reactive, which is undesirable for higher capacity systems such as APMs. However, the introduction of Communication Based Train Control (CBTC) and its supporting systems (ATO, ATS, ATC) for APMs allows for a system that changes states according to a flexible schedule generated by a proactive Model- Base Predictive Controller (MPC). This MPC is part of a hierarchical control structure in which it only defines the (future) time and location of when and where a train is required with a certain amount of cars. All local decisions on i.e. transit progress is done by hierarchically lower controllers in the trains, the track changes and the supportive Automatic Train Supervision (ATS).

The MPC can adjust train scheduling and train composition based on changes in demand. The primary action of the controller is to add a 1-car train to the schedule in response to the initial creation of demand, such that the first passenger has to wait a maximum acceptable waiting time (client requirement). If the demand for a scheduled train exceeds the available capacity, there are two approaches to increase capacity further; an additional train can be scheduled before or after the first train and effectively increase frequency, or the train can be extended by an extra car. If in either approach a maximum is reached (no more trains able to be scheduled or no more cars to be added), the other approach should be used to further expand capacity. This results in two concept ACS designs which can be tested.

This chapter contains a comprehensive explanation of the Adaptive Control System (ACS). The current application of control systems for APMs is discussed ﬁrst in Section 3.1, followed by an applicable control structure for the ACS in Section 3.2. The conceptual decision logic of the ACS is thereafter explained in 3.3.

3.1 Available Control Systems

Control systems are used to guide, check and adapt the trains on the network. According to Lott et al.(2009), two types of control systems are used for guided APM systems, which are Fixed Block Train Control (FBTC) and Communication Based Train Control (CBTC). The latter is a relatively new innovation which is slowly being introduced in APM systems (Little and Ross, 2013). For example, CBTC has become standard for the Siemens Airval and optional for Bombardier systems with the CityFlo650 control system or the Mitsubishi Crystal Mover with SelTrac (Drolet and Jadhav, 2005, Siemens, 2014, Thales, 2015).

21 Chapter 3 Conceptual Design of an Adaptive Control System for Automated People Movers in an Airport Environment

Both CBTC and FBTC are part of a multi-layer Automated Train Control (ATC) structure, in which a central control dictates the movements of the vehicles on the network and the operation of switches. The primary advantage of the CBTC technology is that trains are directly connected to the central controller by means of wireless data transmission, instead of wayside sensors that collect information of the trains that pass by. With this innovation it is possible to reduce headways in the system, as is graphically explained in Figure 3.1.

Figure 3.1: FBTC (Top) vs CBTC (Bottom)

ATC that utilises CBTC can be equipped with Automatic Train Supervision (ATS) that automatically supervises and corrects operations (Morar, 2010). A schedule is fed to the ATS and based on the network data gathered by the controller on train locations, the error is calculated between the scheduled operation and actual operation. A reactive solution is thereafter calculated to speed up or slow down trains in the system appropriately (i.e. PID-control).

The motivation of this research is that ineﬃciencies are induced by the static and (daily) repetitive nature of a schedule currently used for an APM system. The capacity of the system is determined in the design phase of the system and is fully based on expected ridership. As day to day operations variate, this results in a design for a relatively constant capacity with only a few capacity changes throughout the day. An example is given in Figure 3.2 of such a rough schedule, which is proposed for the new APM at Singapore Changi International Airport (Lea+Elliot, 2014).

However, the technology of adapting capacity to demand is already used for Personal Rapid Transport, as was explained brieﬂy in 2.4.1. The operations of the Morgantown PRT, which is the earliest and largest PRT system, uses a demand driven operation in which a passenger requests a vehicle with the push of button (Baumgartner and Chu, 2013). The control system will then make a reactive (proportional) trade-of between the dwell time of that passenger and the total passenger to determine the urgency with which to activate a car (Raney and Young, 2005).

Thereby, some early studies for general train scheduling and control show the eﬀectiveness of MPC. Wang et al.(2010) conclude that it is possible to implement the MPC logic in current central train control systems and that hereby a much better alignment of the scheduled capacity and demand is feasible.

3.2 ACS Controller Structure

To select the correct controller type and structure, it is important to ﬁrst outline the measurements and actions that the system should make. The ACS should in essence control the system

22 Graduation Thesis

Figure 3.2: Example Fixed Schedule for Singapore Changi (Lea+Elliot, 2014) similar to the Morgantown PRT, in which demand is somehow measured at a platform and a vehicle is redirected to the station. The disadvantage of the logic used in Morgantown is that the system will only adapt in reaction to the demand, in line with the action span of a proportional controller. In the case of a PRT, this is a relatively minor problem as there is a large amount of cars, which increases the probability of one being nearby. Nonetheless, the chance still remains that the system cannot react in time using this method, which results in an uncomfortably long waiting period for a passenger. In an APM system this problem is even more likely due to the amount of trains being much lower and distances being larger, which makes the period to adapt to changing demand higher.

To eﬀectively adapt the system to the demand, it is therefore important to somehow forecast the moment that passengers wait at the platform and assign the appropriate capacity to the network in terms of train capacity and/or train frequency. The only controller type that can obtain such a result is the MPC, which is therefore chosen as prime method.

As the MPC will have to calculate the system actions for a certain forecast period, the processor requirements can become extensive. It is therefore undesirable to let all actions in the system be controlled by one central controller. Instead, a hierarchical controller structure is taken, in which the central MPC determines the required system capacity in terms of train capacity and train frequency and translates this in a new destination for a (set of) APM car(s). Cars can run independently by means of a closed loop control (i.e. PID) that continuously measures the car position with the ATS-CBTC (Siahvashi and Moaveni, 2010). The latter system should thereby operate any eventual track changes en route. The complete hierarchical structure is graphically summarized in Figure 3.3.

23 Chapter 3 Conceptual Design of an Adaptive Control System for Automated People Movers in an Airport Environment Hierarchy in control logic Figure 3.3:

24 Graduation Thesis

3.3 Central Model-Based Predictive Control Logic

While the car controllers and switch controllers are an essential part of the system, it is determined that no further eﬀort is required in detailing the precise control operations thereof. Due to hierarchical structure, all system changing decisions are made in the MPC and as the assumption is made that 0% failures occur, the functioning of the lower level parts is undisturbed.

The MPC diﬀerentiates from other controller types by considering the current action and a set of future actions based on a model-based forecast. As was brieﬂy introduced in 2.3.1, the MPC is in essence the same as any other controller by using system state measurements y(t) to adjust system state x(t) to a new system state x(t + 1). However, the action u(t) is not based solemnly on the single system measurement y(t) to obtain an optimal state at t + 1, but is part of a set of actions u(t), .., u(t + n) that will satisfy the system objective J over a time period n (?). The information y(t + n) from the forecast model will allow the MPC to anticipate on future disturbances (i.e. changes in demand). It is hereby important that a set of system constraints are considered that bound the action space.

The exact interrelation of the MPC aspects is further explained in this section and includes a detailed description of the algorithm used.

Measurements As was pointed out before, the MPC obtains real-time locations of trains T 1 via the ATS and gathers forecast demand information to measure the system state. An additional system measurement that is required, is the status of the car2. If a car is executing an action or is planned to do so, its status active status is denoted as 1, whereas an inactive car has a status attribute of 0.

The trains run independently of each other and their location and status should therefore be measured separately. The same holds for the diﬀerent entry platforms, that have their own demand patterns. The measurements are summarised in equation 3.1 and a part of the set y(t+n). It should be noted that an iterative process is required to calculate y(t + n) ∀ n ≥ 1, as the action taken at t will aﬀect the predicted measurements for t + n.

yc1(t + n) = Position of train T (network coordinate)

yc2(t + n) = Status of car C (active/inactive)

yl1(t + n) = last departure at location l (time)

yl2(t + n) = next departure(s) at location l (time) (3.1)

yp1(t + n) = Demand at platform p (persons)

yp2(t + n) = maximum waiting time at p (time)

y(t + n) = [yc1(t + n), yc2(t + n), yp1(t + n), yp2(t + n)]

Actions The MPC can adjust the departure time and destination of any car in the system based on

1a train is a single or set of cars in operation 2a car is a single vehicle and a set of them comprises a train

25 Chapter 3 Conceptual Design of an Adaptive Control System for Automated People Movers in an Airport Environment

the measurements y. It can thereby also adjust the train length in terms of cars. Due to the hierarchical structure of the control system, all local actions such as vehicle speed, acceleration, deceleration and on-line vehicle separation and trip progress is done by the car controller.

uc1(t + n) = Departure train C (time)

uc2(t + n) = Destination train C (network coordinate) (3.2)

uc3(t + n) = elongate train (cars/train)

While there are three actions that the MPC can execute, these are in fact part of two operational choices that combine 2 or 3 of the actions together. The ﬁrst composed action u1(t) is to initiate a single car train to execute a transit to location l at time t. The other composed action does the same, but initiates the movement of a multi-car train. The resulting actions are a function of all measurements and all or a selection of the partial actions combined (equation 3.3).

u1(kt + n) = single car train = f(y(t + n), uc1(t + n), uc2(t + n))

u2(t + n) = multi car train = f(y(t + n), uc1(t + n), uc2(t + n), uc3(t + n)) (3.3)

u(t + n) = [u1(t + n), u2(t + n)]

Constraints The actions that can be executed by the MPC are constrained. These constraints are summarised in Equation 3.4 and can either aﬀect availability for or limit the value range of the action.

Logically, the status yc2 of a car should be inactive to allow any action to be executed and a min next departure yl2 at location l should honour a minimal headway theadway. Thereby, the dwell max time on a platform yp2 should be equal to or lower than a maximum dwell time tdwell. Lastly, the action uc3 (cars/train) is constrained by a maximum length that is dictated by the system platform length.

yc2(t + n) = 0 min yl2(t + n) ≥ yl1(t + n) + theadway (3.4) max yp2(t + n) ≤ tdwell max uc3(t + n) ≤ clength

Optimal Performance Function The prime objective of the ACS has been stated in Section 2.1 and requires the controller to minimise the error of the system, i.e. adapt to the measured demand by taking a capacity-altering action.

What the optimal performance should be, depends on the client requirements. The system can either favour passenger experience by increasing the frequency and thereby reducing the average dwell time by calling up more trains, or it can favour a solution where frequency is kept low to

26 Graduation Thesis

(a) train capacity (b) train frequency

Figure 3.4: ACS capacity change approaches reduce e.g. operational costs and the amount of required cars. These two options to obtain an optimal performance are graphically shown in Figure 3.4.

The complete performance function should thus above all attempt to minimise the error J1 between capacity and demand. An additional part of the function should measure the error J2 between an optimally low passenger waiting time (0) and the actual waiting time. A third part max of the function should measure the error J3 of an full train composition (clength) and the actual train length.

The optimal performance function is summarised in equation 3.5. A choice between the two approaches can be made by changing the boolean and dissimilar weights α and β accordingly ([0,1] or [1,0]).

J1 = f(u(t + n), yp1(t + n), yc2(t + n))

J2 = f(u(t + n), yp2(t + n)) (3.5) J3 = f(u(t + n)) J = Min(J1 + α ∗ J2 + β ∗ J3)

3.3.1 Internal Workings of the Decision Logic

The former section sketches the algebraic outline of the control logic used by the MPC to obtain the optimal performance. To better comprehend the internal workings and the decision sequence of the MPC used for the ACS, a simpliﬁed explanation is given in this section for both optimisation approaches and graphical overviews are given in Figure 3.5 and Figure 3.6.

Increased Train Frequency Favoured Over Increased Train Capacity The central controller will use the measured forecast to determine demand for a period n. If demand takes a value of 1 or more at time n, the decision logic will check if a train is scheduled to arrive in time and with enough capacity. If no service is schedule, the controller will activate the nearest single car train and redirect it to the appropriate platform. The train will not move instantly, but will instead wait as long as possible to allow any further changes to be made if needed.

When at a later time step the demand at n has taken a value that is larger than the maximum capacity of the scheduled train, an additional train service is added. This train will arrive

27 Chapter 3 Conceptual Design of an Adaptive Control System for Automated People Movers in an Airport Environment before the train that is already scheduled, as long as this does not violate the minimum headway constraint set. If a train can not be scheduled before the ﬁrst train, the service will be a minimum headway time later. If it is not possible to add any more train services without passengers waiting longer than the maximum dwell time, scheduled trains are elongated with additional cars.

Figure 3.5: Increased Train Frequency Favoured Over Increased Train Capacity

The diﬃculty with adding extra train services is that schedules are generated independently for all platforms. It is therefore important to synchronise a new train service with eventual departures planned at the surrounding stations. For instance, a train that is optimally scheduled to depart at time t at station 2, can interfere with the arrival of an already scheduled train from station 1. This same scheduled departure time at station 2 might also interfere with the schedule of station 3. There are thus seven options to decide the departure time of a next train, which can be:

max 1. scheduled at the optimal time (t + theadway);

min 2. scheduled theadway before the arrival of a scheduled train from the last station;

min 3. scheduled theadway after the arrival of scheduled train from the last station;

4. synchronised with the arrival and use the scheduled train from the last station;

min 5. scheduled to arrive theadway before the departure of a scheduled train at the next station;

min 6. scheduled to arrive theadway after the departure of a scheduled train at the next station;

7. scheduled to synchronise with the departure of a scheduled train at the next station.

Increased Train Capacity Favoured Over Increased Train Frequency The second approach is to favour train capacity over train frequency, which logically has similar- ities with the ﬁrst approach. The main distinction in the decision logic is the priority it gives to capacity increasing measures. When the platform demand exceeds the capacity and an increase is required, the action taken is to add an additional car to train. Only if no more cars can be added or demand has risen after the planned departure time, will the controller add another train to the schedule. Again, it will ﬁrst increase the capacity of this train, before changing the frequency.

28 Graduation Thesis

Figure 3.6: Increased Train Capacity Favoured Over Increased Train Frequency

3.3.2 Demand Forecast Model

The eﬀectiveness and precision of forecasting future states is narrowly related to the length of the forecast period. If this period is long, this will increase the uncertainty in prediction and also increase the processing requirements of the controller. If the the forecast period is however too small, the system is incapable of adjusting fast enough. The forecast period of an APM system should therefore ideally be equal to the time it takes to cover the largest parking to station distance, with time to spare for passengers to get in.

An adjustment to this forecast period can be made in respect to the maximum dwell time of a passenger on the platform. If for instance one passenger instigates demand, it is ineﬃcient to adjust the system such that only this one passenger will be just in time to make it to the next train. Instead the train should be scheduled to depart the station after the maximum dwell time of this ﬁrst passenger and thus allow other passengers to board the vehicle as well. This is summarised in equation 3.6, where tforecast is determined by maximum transit period from any parking location xparking to a platform location xplatform, reduced with the maximum dwell max time tdwell.

n xparking − xplatform max tforecast = max − tdwell (3.6) n∈N vvehicle

With the forecast period known, it is needed to determine an appropriate method to model this period. Two methods are therefore proposed in this section.

Schedule based demand forecast The ﬁrst method is to use the same logic as explained for the driving forces in Section 2.4.2, in which stochastic data is used to create a suitable forecast model. By using the scheduled or actual aircraft movement information and combining this with a system speciﬁc stochastic distribution, it is possible to approximate the passengers movement.

The advantage of using a scheduled based demand forecast, is that the information of aircraft movements is well documented in tower logs and can deliver real time information. Furthermore,

29 Chapter 3 Conceptual Design of an Adaptive Control System for Automated People Movers in an Airport Environment

using schedule information will also give more information on the type of passengers and thus their expected behaviour.

There are however some downsides to such an approach, as the stochastic distribution will have to consider all eventual actions between the passenger being in the APM and in the aircraft (see Figure 2.4a and Figure 2.4b). When the amount of actions increases, the uncertainty in the distribution will increase as well. This will make the forecast unreliable and results in a less than optimal operation.

Sensor based demand forecast Another more eﬀective means of measurement would be to use a sensor system. This can be realised by placing the sensor system some distance before the platform entrance, so that the MPC can eﬀectively calculate the future demand and adapt the system. The distance required between the sensor location and the platform entrance should be of such a length that a passenger only has to wait the maximum dwell time period. As not all passengers walk the same speed, the forecast time of the passenger entering should be corrected with a walking speed distribution. This will mean that the larger the distance between the sensor location and the platform is, the larger the uncertainty becomes in the forecast.

As there is always a marginal share of the population that walks at signiﬁcantly fast speeds, it is possible to choose the distance such that a minimum percentage of the population instead max of everybody will have to wait longer than tdwell. Thereby, a simple method to reduce the uncertainty is to use continuous systems such as escalators for which the transit time is much more precise (tcont). Equation 3.7 summarises the calculation needed for the precise distance between the sensor and platform (assuming that the walking speed is normally distributed).

max xsensor = (tforecast − tdwell − tcont) ∗ ((Z ∗ σ) + µ) (3.7)

There is a wide variety of sensor system commercially available to measure passengers passing by a certain point. Technological solutions to count people can be as simple as a gate of button (such as used for PRT), up to more advanced systems that use infra-red, WiFir, BluetoothTM or CCTV technology.

For this research it is of utmost importance that a sensor should not obstruct the passenger ﬂow and should deliver high resolution passenger data to accurately create a forecast. As a representative sensor system in this research, the BlipTrackTM technology developed by Blip Systems is taken. Their system uses infrared, BluetoothTM and WiFir technology to create high resolution real-time and on-demand passenger ﬂow patterns (BLIP Systems, 2015).

30 4. Detailed Test Case Design For Amsterdam Airport Schiphol Chapter Summary:The test case of Amsterdam Airport Schiphol considers a proposed APM connection between the D-Pier, C-Pier and (future) A-Pier. Distinctive characteristics of the system are a separation of Schengen and Non-Schengen Passengers and that inactive trains are parked at the platform due to the absence of dedicated parking locations.

Both ACS alternatives eﬀectively reduce the total run distance of the cars in the system by only running trains when there is demand and thereby show good results for the Performance

Indicators operational costs and CO2 pollution. The passenger experience in terms of dwell time on the platform is similar for the ACS alternatives and the reference case.

Of the two ACS alternatives, the one in which train capacity changes are favoured over train frequency changes shows a signiﬁcantly better result on capital costs as less cars have to be acquired.

This chapter contains a comprehensive explanation of the test case for Amsterdam Schiphol International airport, further referred to as AMS (IATA code). The airport region and airport characteristics are given in Section 4.1 and 4.2, followed by an analysis of the proposed APM system that was researched for the airport back in 2009 (Section 4.3). A description is thereafter given of the reference case and the two ACS alternatives, which includes the data used and eventual assumptions that have been made in Section 4.4. The chapter is concluded with a summary of the simulation results in Section 4.5.

4.1 Airport Region

AMS is the primary airport in the Netherlands and is located in the metropolitan ’Randstad’ area that includes the 4 most populous cities of the country (Amsterdam, Rotterdam, The Hague and Utrecht). It is located in one of the busiest air spaces in the world, with the large airports London Heathrow, London Gatwick, Paris Charles-de Gaulle, Frankfurt, Düsseldorfand Brussels Zaventem within a range of 400km. Among these primary hubs, there are approximately 10-15 more airports located within the catchment area that serve frequent international flights as well. The Netherlands is part of the Schengen region, which means that most continental flights are not subject to custom checks and are effectively handled as a domestic.

4.2 Airport Characteristics

AMS is currently the 17th largest airport in the world and the 4th largest in Europe with 52.6 million passenger passing through annually (Air Council international, 2014). The airport has 6 runways and a single terminal with 7 piers (B,C,D,E,F,G,H) that connect a total of 98 contact stands (Schiphol Group, 2015a). The airport experiences a steady growth in air travel demand and has issued expansions plans for an additional pier (A) to be constructed in the coming years, followed later by the additional piers A’ and A” (Bijloo, 2012).

31 Chapter 4 Detailed Test Case Design For Amsterdam Airport Schiphol

The airport is not only a major access point to the Netherlands, but AMS also functions as a transfer hub that accounts for ± 40% of all passenger movements (Schiphol Group, 2015a). The majority of these transfer passengers use the KLM-Air France network, for which AMS is the primary hub. It thereby also functions as a primary hub for passenger airlines TUI Netherlands, Corendon Dutch Airlines and Transavia.com and as a secondary hub for SkyTeam alliance member Delta and low cost carriers Veuling and Easyjet. Martinair is the only (full) freighter operator that has its base at Schiphol, but a large amount of leading cargo airlines operate on a frequent base due to the strategic location of the airport

The daily movement pattern (Figure 4.1) fluctuates during the day, with the airport closed for traffic (outbound) between 11 p.m. and 6 a.m. The distinctive peak and down times are the result of the arrival and departure ‘blocks’ in which European and intercontinental flights connect. A typical bank starts off with a peak of inbound flights, followed by a peak of outbound flights, graphically explained in Figure 4.2 and clearly recognizable in Figure 4.1. The airport handles as much as 100 movements per hour during a peak, but on average it is near to half of that. The average seating capacity is ± 130 passengers per aircraft, which is the result of the high percentage of narrow body aircraft (±81%) used on continental connections to feed the intercontinental trunk routes (Schiphol Group, 2015b).

4.3 APM System Characteristics

Due to the single terminal concept used at the airport, distances are already quite long and with the new expansion some sort of assisted intra-terminal transport mode (e.g. people mover) would severely beneﬁt the customer experience (Kusumaningtyas and Lodewijks, 2013). Ben- themCrouwel/NACO and Lea+Elliot were therefore commissioned back in 2009 to do an extensive research on the feasibility of an APM. While no further plans for an APM have been made due to circumstances (i.e. economic crisis) since then, the report gives a detailed explanation of what the system characteristics would look like.

Based on the report it is determined that the total distance of the track is designed to be 800 metres long for the ﬁrst phase and will run on the roof of the existing buildings. The route from Pier D to Pier A is interrupted after 160 meters for a stop at Pier C (Figure 4.3). The challenge for the design of this APM is that it requires a separation of Schengen and non Schengen passengers. This means that trains have to consist of multiple exclusive cars and platforms need a separation wall. The Schengen passengers can only travel between Pier A and pier C and non Schengen passengers or OD passengers between pier A and D. The network layout is graphically summarised in Figure 4.4, including the switch and station locations. The round trip time of a single service is shown in Table 4.1.

The APM is located somewhat at the end of the of the departure process, which is generally not the most advisable location according to Ringersma, senior architect at NACO (personal communication, 2015). It is best to have stressful events as early as possible in the process to increase the comfort of a passenger. However, shopping and other pastime events are done at AMS in the central lounges located near the roots of pier C, D and E. Only minimal services are found inside the piers and thus the prime events happening after the APM journey are gate ﬁnding, waiting and boarding.

32 Graduation Thesis

Figure 4.1: Arrival and Departure Pattern AMS

Figure 4.2: Example Block Schedule AMS

Figure 4.3: Proposed APM trajectory at AMS

33 Chapter 4 Detailed Test Case Design For Amsterdam Airport Schiphol

Figure 4.4: Proposed APM Network Layout at AMS

Table 4.1: Round trip Single Service AMS (v=50kph, acc=1m/s2, dec=1m/s2)

Location Time (s) Cum. Time (s) Station D start 0 0 Station D (de)boarding 35 35 Transit D - C (E) 21 56 Station C (E)(de)boarding 35 91 Transit C (E) - A 54 145 Station A (de)boarding 35 180 Transit A - C (W) 54 234 Station C (W) (de)boarding 35 269 Transit C (W) - D 290

4.4 Alternative Description

The goal is to implement an adaptive control system (ACS) in the test case described for AMS. To understand the eﬀect of any alternative, the KPI results will be compared to a reference case that represents a conventional ﬁxed schedule operation.

As was explained in Chapter3, there are two approaches to implement an ACS. These ACS alternatives will further be referred to as:

• ACS alternative 1: Frequency, which favours a change in train frequency over a change in train capacity;

• ACS alternative 2: Capacity, which favours a change in train capacity over a change in train frequency.

4.4.1 Reference Case

The moment that a passenger makes use of the APM is based on the moment this passenger will arrive or depart the airport by aircraft, corrected with a time distribution. The time distribution for arriving passengers is chosen as the average distance a passenger has to walk from a gate to the APM platform at the satellite, combined with the time consumed by disembarking the aircraft, which is assumed to be a uniform process that takes 10 minutes (personal communication, P. Ringersma, 2015).

34 Graduation Thesis

Figure 4.5: Passenger Type distribution (Lea+Elliot, 2014)

Departing passengers are able to dwell in the lounges in the centre of the terminal and it is assumed that passengers will only head to the A pier for boarding. An appropriate ‘at-the- gate’ distribution is taken from an earlier assignment done by NACO for Frankfurt International Airport (FRA), which is a similar airport (geographical and operational) to Schiphol (Wijk, 2007). The additional travel time in the APM is connection speciﬁc and the additional walking distance to the correct pier is on average 80 meters. This can be transformed in time units with the walking speed distribution given by Young(1999)( N (1.347, 0.255) m/s).

The reference case represents a scheduled operation such as is normal for present APM systems. The schedule for the reference case can be determined by calculating the demand per direction for the design day over a speciﬁc time period (Ledoux and Le Picart, personal communication, 20151). The most common unit used in literature is ‘passengers per direction per hour’ (PPDPH) (ACRP, 2012b), but this does not suﬃce to calculate the actual capacity required in a peak period. Instead, the time period should be reduced to the minimum acceptable headway during peak operations, which is 180 seconds based on client requirements.

Figure 4.6 shows the demand patterns for the test case on the Schengen (C-A) and non-Schengen (D-A) connections. To obtain the correct demand for both connections, a selection of ﬂight 10 movements is attributed to the A-Pier based on the share in gates ( 108 ). The new A-Pier will have a Schengen/non-Schengen separation throughout and based on demand forecasts made by Lea+Elliot and Schiphol, it can be determined that roughly 33% of the passengers have a Schengen status (Figure 4.5).

A clear distinction in the demand patterns can be made between the eﬀect of arriving and departing aircraft, with the former resulting in narrow but high distributions and the latter resulting in broad and low distributions. To even out any extreme values, the actual design peak is determined on the average maximum value of 18 model runs (see AppendixC for further information on run repetitions). The resulting value is 73 for the Schengen connection and 157 for the non-Schengen connection and should therefore be covered by an appropriate capacity in the system. The vehicle combinations and headway periods that can be run at AMS are summarized in Table 4.2 and it is determined that the combination of running a train with 1 car for Schengen en 2 cars for non Schengen every 135 seconds ﬁts the required demand best. This will require a total of 9 cars that run in three-car trains.

1Ledoux, G. is technical advisor and Le Picart, G. is sales manager at Siemens S.A.S., Mobility Division

35 Chapter 4 Detailed Test Case Design For Amsterdam Airport Schiphol

Table 4.2: Headway and Train Composition Combinations AMS

Headway Trains 1 car 2 car 3 car 4 car 90 4 120 240 360 480 105 3 102 205 308 411 120 3 90 180 270 360 135 3 80 160 240 320 150 2 72 144 216 288 165 2 65 130 196 261 180 2 60 120 180 240

However, a correction should be made to replicate a more realistic scheduled operation. The 135 seconds frequency is a must if the airport wants to have an absolute 0% chance that a passenger has to wait more than 180 seconds (during day time). This requirement is in reality a less rigid boundary to the system design and should be met for a majority of passengers, which is generally assumed as 95% for airport systems (ACRP, 2012a, Sloboda, 2009). The standardised frequency could therefore be decreased (i.e. increase of headway) to positively aﬀect sensitive output parameters such as the required amount of vehicles and run distances, as long as the maximum waiting time is suﬃciently met (see AppendixC for sensitivity analysis).

Two simulation test runs are therefore conducted for a scheduled frequency of 135 seconds as is deemed appropriate to meet peak capacity and for a scheduled frequency of 180 seconds as is the maximum allowable waiting time. The comparative results conﬁrm that no passengers have to wait longer than 180 seconds when a frequency is taken of 135 seconds. When this frequency is changed to 180 seconds, there is only a very marginal share of passengers that will have to wait for an extended period (±20 out of ±10.500 passenger2). It can therefore be concluded that a scheduled operation of 1 car Schengen and 2 cars Non-Schengen running on a 180 seconds interval is suﬃcient.

The model will run this headway/train length combination for most of the day, with the exception of 6 hours between 00:00 a.m. and 06:00 a.m. Demand during this period is consistently low in all replications, with only 1 or 2 aircraft arriving and 0 departing. In this period, trains run every 15 minutes to serve any passengers that need transit during the down time. While this approach can be wasteful, it is a common approach to obtain a minimal service at night (Ledoux and Le Picart, personal communication, 2015).

4.4.2 ACS Alternative 1: Frequency

With the ﬁrst ACS alternative, the trains are called up based on sensor data that is located at the entrance of the platform areas (Equation 4.1).

max xsensor = (tforecast − tdwell − tcont) ∗ ((Z ∗ σ) + µ) = (35 − 180 − 0) ∗ ((1.96 ∗ 0.255) + 1.347) (4.1) = −286.25 → 0

2average of 18 replications

36 Graduation Thesis

(a) Schengen

(b) Non-Schengen & Origin-Destination

Figure 4.6: APM Demand AMS (1 run)

When demand exceeds the capacity of the ﬁrst train with 1 car (60 passengers), a second train with one car is requested and fed into service before or after the ﬁrst scheduled train. If no more trains can be added due to maximum waiting time requirements, scheduled trains are extended with additional cars. It should be noted that a maximum of 3 cars per train is possible in any combination to allow changes to be made at the terminus stations. Chapter3 contains a comprehensive explanation of the underlying logic and complications.

The dilemma for the ACS operation at AMS is that trains that move in opposite direction need to be synchronised. This is due to the fact that in an inactive system state, a maximum of 4 trains are located at both sides of the two terminus stations A and D. When per example a train movement is initiated at station A, it will be unable to enter station D unless a train at station D is set in motion as well. Since the parking locations for inactive trains are the sides of station A and D, this train originating from station D should therefore continue its journey and park at the vacant platform side of station A. It is this network characteristic that will logically induce a number of empty runs, which are undesirable but inevitable.

4.4.3 ACS Alternative 2: Capacity

The second ACS alternative is similar to the ﬁrst in any way but the logic that calls up new trains and cars. In this alternative trains will primarily be extended with additional vehicles, before increasing the frequency. Chapter3 contains a comprehensive explanation of the underlying logic and complications.

37 Chapter 4 Detailed Test Case Design For Amsterdam Airport Schiphol

4.5 Results

The simulation generates a number of output parameters to measure the (K)PIs. For most (K)PIs, output parameters have to be transformed into the correct unit and/or need to be multiplied with another output parameter. In this section the KPIs are discussed, with the raw output data from the Arena Model summarised in AppendixD.

As there is no reliable information available on the APM system owner preferences, it chosen not to execute a MUlti-Criteria Decision Analysis (MCDA), but only discuss the eﬀect of an ACS compared to the reference case per KPI instead.

4.5.1 Passenger Experience

The passenger experience KPI is supported by three PIs that are measured for the replication average peak operations. The dwell time results are summarised in Table 4.3 and it shows that both alternatives have a lower average dwell time than the reference case overall. However, the reference case is quite heavily inﬂuenced by the night schedule for the APM and therefore the average dwell time during the day is added in the table. During this day operation, the average dwell time is roughly similar for the reference case and the ACS alternatives. The average is near to 90 seconds, which is logical as the scheduled system runs a 180 second headway and the basic principle of the ACS is to wait as long as is allowable (i.e. 180 seconds).

Table 4.3: Results Dwell Time (Day Time Results Between Brackets)

Average Minimum Maximum % change Reference Case 100.80 (90.61) 0.00 900.00 – ACS Alternative 1: Frequency 95.33 0.00 237.00 -5.4% (+5.2%) ACS Alternative 2: Capacity 94.18 0.00 237.00 -6.4% (+3.5%) *excludes passengers that walk too slow and take next train (± 1%)

It should be noted that the maximum dwell time of 237 seconds for both ACS Alternatives are the result of a model discrepancy. The model construct as is used for the AMS test case is incapable of calling up an extra earlier train for demand at Station C, which results in passengers having to wait for the next planned train that arrives a maximum 57 seconds late (transit time Station D-C and boarding time). The aﬀected population is under 5%, with all other passenger having a maximum waiting time of 180 seconds. It is chosen not to change the model, as this will require an exceptional amount of work for only marginal improvements. Thereby, the conceptual theory of adding extra trains earlier to the system is proven for all other platforms and on a whole system level for the SZX test case described in Chapter5.

The PI platform space and PI train space are measured with the output parameters ‘maximum passenger waiting on the platform’ and ‘maximum car load factor’. These values are composed of the average maximum of the 18 replications and translated in the appropriate Level-of-Service as deﬁned by Fruin(1971) and IATA(2015a). The results are shown in Table 4.4 and it can be concluded that all three alternatives provide an excellent (A) experience for all platform areas. This is the result of the sizing of the platforms that is larger than required. The design requirements for the platform are prepared for the eventual expansion of the network to Piers A’ and A” in mind.

38 Graduation Thesis

Table 4.4: Results Level-of-Service

Reference Case Aspect Average Peak Area m2 m2/Pax % change LoS Platform Level-of-Service Platform A S 3 76 448 5.89 – A Platform A NS 7 166 448 2.70 – A Platform C 2 24 400 16.67 – A Platform D 5 48 400 8.33 – A Train Level-of-Service Train (per car) 4 60 22 0.36 – D

ACS Alternative 1: Frequency Aspect Average Peak Area m2 m2/Pax LoS Platform Level-of-Service Platform A S 2 62 448 7.22 +22.6% A Platform A NS 5 136 448 3.29 +21.9% A Platform C 2 21 400 19.05 +14.3% A Platform D 4 40 400 10.00 +20.0% A Train Level-of-Service Train (per car) 10 60 22 0.36 – D

ACS Alternative 2: Capacity Aspect Average Peak Area m2 m2/Pax LoS Platform Level-of-Service Platform A S 2 65 448 6.88 +16.8% A Platform A NS 4 136 448 3.29 +21.9% A Platform C 2 21 400 19.05 +14.3% A Platform D 4 40 400 10.00 +20.0% A Train Level-of-Service Train (per car) 9 60 22 0.36 – D

The trains on the other hand are ﬁlled to their maximum capacity in all alternatives, which results in a level D in all cases. The average load factor is higher for both ACS alternatives and indicates a more eﬀective use of the available capacity.

4.5.2 Cost

The system costs are composed of capital and operational costs, of which a selection is measured in the model. The capital costs are summarised in the upper part of Table 4.5 and are expressed in one time capital costs and daily depreciation costs. It is hereby assumed that the airport is able to ﬁnance the investments costs itself and is therefore not inﬂuenced by interest or discount rates. The operational costs are given in the lower part of the same table and a summation of the

39 Chapter 4 Detailed Test Case Design For Amsterdam Airport Schiphol

daily costs is given as well. It should be noted that the percentage changes given in the results only considers the measured costs and will be smaller when total project costs are considered.

Capital Cost The capital cost is composed of the cost of sensor systems and APM cars. The proposed BLIP system will cost an approximate $150,000 according to the company’s CEO Knudsen (personal communication, 2015) and the cost of a single car is $2.4 million (section 2.4.1). It is assumed that the sensor system does not have to change during the life cycle of the APM vehicle, which is 30 years (Bombardier, 2015). It is assumed that there is no market for second-hand APM vehicles or system specific sensors, so the residual value for both capital costs is assumed $0. The capital costs are almost exclusively depended on the required amount of cars, which are lower for the second ACS alternative (5), but significantly higher for the first ACS alternative (9). The reason that the maximum required amount of cars is consistently a higher for alternative 2 than alternative 3, is that in peak periods the system will reduce the headway to 90 seconds and thus requires 4 trains with 9 cars (3x[1 car Schengen+1 car Non Schengen]+1x[1 car Schengen + 2 cars Non Schengen]). In alternative 3 however, the system will first increase the amount cars/train without decreasing the headway lower than the threshold value for which three or four trains are required simultaneously ( headway < 145 seconds).

Operational Cost The operational cost is composed of the cost of energy and the cost of maintenance. Energy is used by the APM en route and consumes 2.56 kWh/km, as explained in section 2.4.1. The price of energy is fluctuating as a result of many factors such as changing oil prices and political stability, but has nonetheless shown a relative consistent value of ± AC0.09 or $0.10 in the Netherlands (EuroStat, 2015a). The cost per kilometre is thus 2.56 ∗ 0.10 = 0.256$/km. Kerr et al.(2014) state that the maintenance cost are approximately 70% of the power costs, which translates to a value of 0.256 ∗ 0.70 = 0.179$/km. The whole operational cost is thus directly proportional to the total distance travelled by the cars in the system and is lower for both ACS alternatives compared to the reference case. The significantly lower run distance is the result of 3 prime reasons: 1) trains do not run when there is no demand and 2) if there is demand they only serve the connection on which transport is required, after which they return to their idle parking location and 3) train combinations are a lot smaller, with on average just over 1 car per train for both ACS alternatives (instead of a fixed 3 cars per train in the reference case AppendixD). As can be expected, the amount of cars per train is (slightly) higher in the seconds ACS Alternative (capacity) compared to the first ACS alternative (frequency).

Only the ACS alternative that favours capacity over frequency (alternative 2) is beneﬁcial for the total cost reduction in terms of both capital investment and daily operation costs. Especially the reduction in vehicles required and the distance run on the system are strong drivers to reduce costs. The fact that the costs for the ACS alternative that favours frequency over capacity

40 Graduation Thesis

(alternative 1) is higher than the reference case, is completely the result of having to run 4 trains during peak period to meet the 90 seconds headway.

Table 4.5: Results Cost

Reference Case Cost Units Unit Cost Total Cost Depreciation Capital Cost BLIP sensor system 0 $150,000 $0 $0 APM Cars 6 $2,400,000 $14,400,000 $1,314 Total $14,400,000 $1,314 Operational Cost Energy 1,240 0.256 $316 NVT Maintenance 1,240 0.197 $244 NVT Total $560 Total Daily $1,874

ACS Alternative 1: Frequency Cost Units Unit Cost Total Cost Depreciation Capital Cost BLIP sensor system 3 $150,000 $450,000 $41 APM Cars 9 $2,400,000 $21,600,000 $1,971 Total $22,050,000 $2,012 Operational Cost Energy 773 0.256 $198 NVT Maintenance 773 0.197 $152 NVT Total $350 Total Daily $2,362 (+26.0%)

ACS Alternative 2: Capacity Cost Units Unit Cost Total Cost Depreciation Capital Cost BLIP sensor system 3 $150,000 $450,000 $41 APM Cars 5 $2,400,000 $12,000,000 $1,095 Total $12,450,000 $1,136 Operational Cost Energy 764 0.256 $195 NVT Maintenance 764 0.197 $151 NVT Total $346 Total Daily $1,482 (-20.9%)

41 Chapter 4 Detailed Test Case Design For Amsterdam Airport Schiphol

4.5.3 External Eﬀect

The KPI external eﬀect is solemnly measured with the PI CO2 pollution. As the APM propulsion is electric, the train itself does not expel any foul gasses. The energy is however sourced indirectly from a power plant which, if not renewable, impacts the environment.

The study performed by Bombardier(2015) on the pollution of their Innovia APM 300 product

states that a single car expels 1456 gr/km of CO2 when in operation at an European airport. Their study is based on the overall penetration of renewable and nuclear energy in Europe, which accounts for 46% of the total production. However, in the case of AMS, this percentage will be lower as the penetration of renewable energy in the Netherlands (4.2%) is in fact below the European average (11.8%), as well as the penetration of nuclear energy of only 4% compared to 31% (EuroStat, 2015b, M.J.J. Scheepers et al., 2007). Assuming that energy is sourced evenly from the available power suppliers, it can thus be said that the reference APM vehicle 31+11.8 expels 1, 456 ∗ 8.2 = 7, 600 gr/km at AMS. The resulting CO2 pollution per alternative is summarised in table 5.6, from which it is clear that the environmental impact of an ACS is signiﬁcant, with a reduction for both ACS alternatives of almost 40% compared to the reference case.

Table 4.6: Results CO2 pollution

Distance gr/km CO2 Total kg CO2 % change Reference Case 1,240 7,600 9,424 – ACS Alternative 1: Frequency 773 7,600 5,874 -37.7% ACS Alternative 2; Capacity 764 7,600 5,806 -38.4%

42 5. Detailed Test Case Design For Shenzhen Bao’an International Airport

Chapter Summary:The test case of Shenzhen Bao’an International Airport considers a proposed APM connection between terminal 3, a (future) mid-ﬁeld satellite and (future) terminal 4. Distinctive characteristics of the system are that passengers can only transit between a terminal and the satellite but never between the two terminals. Thereby, the system has multiple parking locations spread throughout the network to store inactive cars.

Both ACS alternatives eﬀectively reduce the total run distance of the cars in the system by only running trains when there is demand and thereby show good results for the Performance

Indicators passenger experience, operational costs and CO2 pollution.

Of the two ACS alternatives, the one in which train capacity changes are favoured over train frequency changes shows a signiﬁcantly better result on capital costs as less cars have to be acquired.

This chapter contains a comprehensive explanation of the test case for Shenzhen Bao’an In- ternational Airport, further referred to as SZX (IATA code). The airport region and airport characteristics are given in paragraph 5.1 and 5.2, followed by an analysis of the proposed APM system that is planned at the airport (5.3). The modelled system is thereafter described including the data used and eventual assumptions that have been made in Section 5.4. The chapter is concluded with a summary of the simulation results in 5.5.

5.1 Airport Region

SZX is an airport in the Shenzhen urban region and is built on the shoreline of the Pearl River Delta. Shenzhen is part of the populous Guangdong province and borders the Special Administrative Region of Hong Kong. The city has experienced an excessive growth in population since the Chinese government has designated it as Special Economic Zone (SEZ) in 1979 (Zhu, 1994). A SEZ has diﬀerent and more lenient rules on international trade and was created by the government to improve the global market position of the country (Wang, 2013). While the city was a only small ﬁshing village in the ‘70s, housing only a couple of thousand people, it has now grown to one of China’s largest cities with over 10,000,000 inhabitants (Shenzhen Government, nd). It is hereby the second largest city in the Pearl River Delta1 (PRD), only surpassed by Guangzhou. It is nonetheless closely followed in inhabitant numbers by other multimillion cities such as Shantou, Zhuhai (both also SEZ), Foshan, Dongguan and Zhongshan (based on World Bank Population Data). Since last year, the metropolitan PRD is the most populous region in the world and with an estimated 60% of the area still not urbanised , further growth is expected in which the total population number will surpass 100 million (Moore and Foster, 2011, Mead, 2015).

1Common denotation of the megalopolis that surrounds the eponymous delta (Enright et al., 2005)

43 Chapter 5 Detailed Test Case Design For Shenzhen Bao’an International Airport

Figure 5.1: Pearl River Delta

By far the largest airport is that of Hong Kong (HKG), which is located on the island of Chep Lap Kok. The airport handles almost 60 million passengers and 4 million tonnes of cargo annually, thereby ranking 11th and 1st in the world, respectively (Air Council international, 2014). Both Guangzhou and Shenzhen are in the top 50 of the busiest airports (16th and 47th), while Zhuhai and Macau complement air transportation in the region on a signiﬁcantly smaller scale. As transport secretary Anthony Cheung Bing-leung of the Hong Kong government explained to Hong Kong Standard(2015), the demand for air travel is growing extensively. This has resulted in many restructuring projects in the last years, including new terminals and additional runways at the airports. While new constructions are just ﬁnished at most airports, new plans are already executed to support the continuous growing demand, in which Shenzhen is no exception.

5.2 Airport Characteristics

SZX is a multi-terminal airport with two parallel runways. The airport has grown quickly in the last decade and has recently completed the construction of a second runway and the Terminal 3 building (Wong, 2013). Oﬃcially, the airport has three additional terminals (A, B & D), but these are in disuse for commercial aviation (Han, 2013). The airport is the hub for two passenger airlines; Shenzhen Airlines and Donghai Airlines and two cargo airlines; SF and UPS.

In the current situation of the airport, 32 million passengers pass through annually (ACI, 2014). The airport is focused on domestic OD2 traﬃc, resulting in an aircraft mix heavily shifted towards ICAO code C (e.g. Boeing 737, Airbus A32X). According to forecasts made by NACO, these narrow body aircraft with 100-200 seats, make up almost 90% of the movements at the

2Origin-Destination

44 Graduation Thesis

Figure 5.2: Arrival and Departure Pattern SZX airport. According to the Centre for Aviation(2013), SZX’s prime carriers all focus on point- to-point connections, resulting in minimal transfer passengers. It should be noted that larger aircraft (ICAO Code D, E and F) are generally also used to ﬂy domestic routes and international passengers account only for some percent.

The daily movement pattern (figure 5.2) shows an even spread of arrivals and departures, with the airport closed for traffic (outbound) between 11 p.m. and 6 a.m. A departure peak is seen in the morning and an arrival peak at night. The reason for these peaks is that the home carrier Shenzhen Airline has a significant amount of overnight parking of its aircraft that leave in the morning and come back in the evening. While the airport runways can handle some 60 movements on an hourly base, this number is currently not met due to busy air space that is shared with the surrounding airports. Law et al.(2007) explain that the capacity in the Pearl River Delta sky is mainly limited due to low cooperation of the different ATC (Hong Kong, Zhuhai and Guangzhou) and reforms in the system are required to efficiently handle traffic.

5.3 APM System Characteristics

This section will describe the modelled APM system and system parameters that are specific for the SZX test case. AppendixC contains the full model description including detailed flow charts of the decision logic executed in the Arena model, a model sensitivity analysis and the verification/validation of the model.

The planned APM system is part of a large master plan set for 2040. While the new terminal is just finished, growth forecasts show that the maximum capacity of the facility will already be met by 2020. Therefore, an additional master plan is prepared that entails the construction of a third parallel runway and two new terminal buildings: terminal 4 and a midfield satellite. The whole master plan should accommodate up to 70 million passengers annually by 2040 and endure peak traffic demands of over 100 aircraft movements per hour. These numbers are based on the forecasts made by NACO and CACC (China Airport Construction Group Corporation), in which it is assumed that the third runway will be in operation by 2022 and that air space restrictions are reduced/solved.

The APM network will be the connection between the current T3 terminal, the future satellite and T4 Terminal (Figure 5.3). It will run on a dedicated underground network which contains three stations at the terminals and satellite, 7 parking locations and a maintenance facility which is located at the outer perimeter of the airport. The total track length is approximately

45 Chapter 5 Detailed Test Case Design For Shenzhen Bao’an International Airport

Figure 5.3: Proposed APM trajectory at SZX

Figure 5.4: Proposed APM Network Layout at SZX

2700 meters, with T3 and the satellite 1800 meters and T4 and the satellite 900 meters apart. The airport has not yet set requirements for the headway and it is therefore assumed that the requirements used in the AMS test case are also appropriate for SZX. Some preparations for the system have already been made for a pinched loop network, with a tunnel running between the core of T3 up to just short of the future satellite location. Two side platforms are already in place at T3, which are accessible via an escalator system from the departure ﬂoor (level 4) down to the tunnel (level -1). However, the latest plans are to replace the two side platforms with a single track station at T3. The full network that is planned is graphically explained in Figure 5.4, including the switch, station and parking locations. The round trip time of a single service is shown in Table 5.1.

Table 5.1: Round trip Single Service SZX (v=50kph, acc=1m/s2, dec=1m/s2)

Location Time (s) Cum. Time (s) Station T3 start 0 0 Station T3 boarding 35 35 Station Satellite enter 142 177 Station Satellite boarding 35 212 Station T4 East enter 78 290 Station T4 East de-boarding 35 325 Station T4 West Enter 22 347 Station T4 West Boarding 35 382 Station Satellite Enter 77 459 Station Satellite boarding 35 494 Station T3 Enter 636

The APM system is designed such that passengers enter the APM directly after check-in (security is thus located at the Satellite). It is assumed that passengers will only transit between one of

46 Graduation Thesis the terminals and the satellite, but never between the terminals T3 and T4. The main reason is that a run from the satellite to a terminal has a different security status than a run from a terminal to the satellite. Thereby, airlines have their check-in at either T3 or T4, and use that same terminal or the satellite, but never the opposite terminal. Even more so, a good land side connection is offered between T3 and T4 via the metro line 11 (Airport Line) to serve those passengers that have chosen to transfer between two different airlines at the airport.

5.4 Alternative Description

The goal is to implement an adaptive control system (ACS) in the test case described for SZX. To understand the eﬀect of any alternative, the KPI results will be compared to a reference case that represents a conventional ﬁxed schedule operation.

As was explained in Chapter3, there are two approaches to implement an ACS. These ACS alternatives will further be referred to as:

• ACS alternative 1: Frequency, which favours a change in train frequency over a change in train capacity;

• ACS alternative 2: Capacity, which favours a change in train capacity over a change in train frequency.

5.4.1 Reference Case

The reference case represents a scheduled operation such as is normal for present APM systems. The schedule for the reference case is determined by plotting the demand curves per direction for the design day. The demand curves are composed of the individual moments that passengers enter the platform and will require transit to their destination

To accurately determine the time that a passenger uses the APM, the moment that this passenger will arrive or depart the airport by aircraft is shifted in time with an appropriate distribution. For arriving passengers the time distribution is chosen as the average distance a passenger has to walk from a gate to the APM platform at the satellite, combined with the time consumed by disembarking the aircraft, which is assumed to be a uniform process that takes 10 minutes (personal communication, P. Ringersma, 2015).

Although departing passengers are able to dwell for some time in the public areas in T3 and T4, it is assumed that passengers will enter the APM directly after check-in. There is no precise data known on the arrival pattern of passengers at SZX and instead a realistic assumption is designed in association with experts from NACO. Based on their experience at other Chinese airports, it is advised that passengers enter the system as early as 3 hours before the departure of the aircraft. The lower side of the distribution is 45 minutes before departure, which is normal for an airport where passengers have to transit between terminal buildings.

As the land side access by car or bus to a large Chinese airports is frequently congested and the average ﬂying experience of a Chinese citizen is low, passengers tend to arrive relatively early, which result in a skewed distribution with a most likely arrival time at 2 hours before departure. The additional walking distance to the platform is approximately 250 meters walking

47 Chapter 5 Detailed Test Case Design For Shenzhen Bao’an International Airport

and 5 stories down by elevator. This can be transformed in time units with a walking speed distribution (Young, 1999) and vertical transit speed of an elevator (OTIS, 2015), which are N (1.347, 0.255) m/s and 0.35 m/s, respectively.

The schedule for the reference case can be determined by calculating the demand per direction for the design day over a speciﬁc time period (Ledoux and Le Picart, personl communication, 20153). The most common unit used in literature is ‘passengers per direction per hour’ (ppdph) (ACRP, 2012b), but this does not suﬃce to calculate the actual capacity required in a peak period. Instead, the time period should be reduced the minimum acceptable headway during peak operations, which is 180 seconds based on client requirements.

Figure 5.5: APM demand SZX (1 run)

Figure 5.5 shows the demand pattern for the peak day in both north and south direction, based on the 2040 forecast prepared by the China Airport Construction Group Corporation of CAAC (CACC). A clear distinction can be made between the eﬀect of arriving and departing aircraft, with the former resulting in narrow but high distributions and the latter resulting in broad and low distributions. To even out any extreme values (such as the outlier just after 12:00 noon), the actual design peak is determined on the average maximum value of 10 model runs (see AppendixC for further information on run repetitions). The resulting value is 293, and should therefore be covered by an appropriate capacity in the system. The vehicle combinations and headway periods that can be run at SZX are summarized in Table 5.2 and it shows that there is a total of 6 combinations capable to meet demand. For the base alternative, the choice is made to use trains composed of 4 cars that run with 135 seconds of headway, as this requires the least amount of cars.

Table 5.2: Headway and Train Composition Combinations SZX

Headway Trains 1 car 2 cars 3 cars 4 cars 90 8 120 240 360 480 105 7 102 205 308 411 120 6 90 180 270 360 135 5 80 160 240 320 150 5 72 144 216 288 165 4 65 130 196 261 180 4 60 120 180 240

3Ledoux, G. is technical advisor and Le Picart, G. is sales manager at Siemens S.A.S., Mobility Division

48 Graduation Thesis

Two simulation test runs are therefore conducted for a scheduled frequency of 135 seconds as is deemed appropriate to meet peak capacity and for a scheduled frequency of 180 seconds as is the maximum allowable waiting time. The comparative results are found in AppendixD and it is conﬁrmed that no passengers have to wait longer than 180 seconds when a frequency is taken of 135 seconds. When this frequency is changed to 180 seconds, there is only a very marginal share of passengers that will have to wait for an extended period (±110 out of ±52.500 passenger4). It can therefore be concluded that a scheduled operation of 4 cars running on a 180 seconds interval is suﬃcient.

The model will run this headway/train length combination for most of the day; with the exception of 2 hours between 02:00 a.m. and 04:00 a.m. Demand during this period is consistently low in all replications, with only 1 or 2 aircraft arriving. In this period, trains run every 15 minutes to serve any passengers that need transit during the down time.

5.4.2 ACS Alternative 1: Frequency

The first ACS alternative incorporates the adaptive logic for an APM system. The trains are now called up based on sensor data that is collected some distance before the passenger enters the platform area. When demand exceeds the capacity of the first train with 1 car(60 passengers), a second train with one car is requested and fed into service before or after the first scheduled train. If no more trains can be added due to waiting time constrictions, scheduled trains are extended with additional cars, as long as these cars can be routed to the platform in time. Chapter3 contains a comprehensive explanation of the underlying logic and complications.

Unlike the AMS test case, the APM vehicles are unable to use the platforms as a parking location and are instead routed to dedicated parking locations along or near to the network. The system therefore requires an extended forecast period to call up vehicles when required. At SZX this period should be 370 seconds, which is equal to the transit time of the longest parking to platform connection (Parking 8 to Terminal 3), including a minimum boarding period of 35 seconds (Lea+Elliot, nd). This means that the sensor system should be located at a location which is passed by passengers 190 seconds before they enter the platform (equation 5.1).

max max tforecast = tdwell + twalking → twalking = tforecast − tdwell = 370s − 180s = 190s (5.1)

The 190 seconds is partly covered by walkways and partly by escalators. AT SZX, the vertical displacement in T3 is approximately 25 meters from the departure ﬂoor (level 4) down to the APM tunnel (level -1). A typical escalator moves at a rate of 0.5 m/s in transit direction,

4average of 10 replications

49 Chapter 5 Detailed Test Case Design For Shenzhen Bao’an International Airport

which translates in a vertical speed of approximately 0.35 m/s (assuming an elevator angle of 45 degrees). Therefore 71.4 seconds are covered by escalator, which means that another 118.6 seconds should be covered by a dedicated walkway. According to Young(1999), the average speed of a passenger in an airport environment is 1.347 m/s, with a standard deviation of 0.255 m/s for free-ﬂowing environments. The general rule of thumb for an APM transit service is that 95% of the users will not endure an exceedingly large waiting time, i.e. more than 180 seconds on the platform (ACRP, 2012a, Sloboda, 2009). This means that passengers walking at a pace of ≤1.84 m/s should arrive at the most ≥ 180 seconds prior to departure. The minimum walking distance should therefore be 220 meters. This distance is possible at SZX’s Terminal T3 and is visualised in Figure 5.6.

Figure 5.6: The location of the sensor at Terminal 3

The same distance should be used in the design of T4 and the satellite. It should be noted that in this specific test case, the distances are relatively large, but can be effectively reduced by either locating vehicle parking closer to the platforms, excluding specific parking to platform connections that are highly unlikely to occur or increasing the maximum platform dwell time.

5.4.3 ACS Alternative 2: Capacity

The second ACS alternative is similar to the first ACS alternative in any way but the logic that calls up new trains and cars. Different to the first ACS alternative, in this alternative trains will primarily be extended with additional vehicles, before increasing the frequency. Chapter3 contains a comprehensive explanation of the underlying logic and complications.

5.5 Results

5.5.1 Passenger Experience

The passenger experience KPI is supported by three PIs that are measured for the replication average peak operations. The dwell time is shown in Table 5.3 and it shows that both alternatives have a lower average dwell time than the reference case. The average is near to 90 seconds, which is logical as the scheduled system runs a 180 second headway and the basic principle of the ACS is to wait as long as is allowable (i.e. 180 seconds). However, the average dwell time for the ACS alternatives is some 10 seconds lower due to the eﬀect of the control logic. With the large

50 Graduation Thesis demand for the system, frequency changing choices are being made to lower average dwell times. This is diﬀerent to the AMS test case, in which the demand was lower and thus less additional choices were made (i.e. changing frequency or capacity after the initial train creation).

The PIs platform space and vehicles space are measured with the output parameters ‘maximum passenger waiting on the platform’ and ‘maximum car load factor’. These values are composed of the average maximum of the 10 replications and translated in the appropriate Level-of-Service as deﬁned by Fruin(1971) and IATA(2015a). The results are shown in table 5.4 for the separate alternatives and it can be concluded that all three deliver an excellent (A) experience for most areas.

Only on the west platform of the Satellite station and in the trains will passengers experience a lower LoS during peak periods. This is the result of the Sattelite-T3 connection serving the majority of arriving passengers. Due to the nature of the ACS to wait for extra passengers, The LoS drops down to D for both ACS alternatives, compared to a level C in the reference case. The trains are ﬁlled to their maximum capacity in all alternatives, which results in a loS D in all cases. While both ACS alternatives show a slightly lower LoS on the aforementioned system aspects during peak periods, the system is used more eﬀective. It is true that passengers have to stand in more crowded areas, but in none of the cases do these passengers have to endure exceptional discomfort. The result is that the load factor of the trains is higher.

Table 5.3: Results Dwell Time

Average Minimum Maximum % Change Reference Case 92.31 0.00 900.00 – ACS Alternative 1: Frequency 80.00 0.00 180.00 -12.5% ACS Alternative 2: Capacity 80.42 0.00 179.99 -12.5% *excludes passengers that walk too slow and take the next train (± 1%)

5.5.2 Cost

The system costs are composed of capital and operational costs, of which a selection is measured in the model. The capital costs are summarised in the upper part of Table 5.5 and are expressed in one time capital costs and daily depreciation costs. It is hereby assumed that the airport is able to ﬁnance the investments costs itself and is therefore not inﬂuenced by interest or discount rates. The operational costs are given in the lower part of the same table and a summation of the daily costs is given as well. It should be noted that the percentage changes given in the results only considers the measured costs and will be smaller when total project costs are considered.

51 Chapter 5 Detailed Test Case Design For Shenzhen Bao’an International Airport

Table 5.4: Results Level-of-Service

Reference Case Aspect Average Peak Area m2 m2/Pax % Change LoS Platform Level-of-Service Platform T3 22 124 477 3.85 – A Platform T4 5 37 248 6.70 – A Platform Sat (W) 23 311 248 0.79 – C Platform Sat (E) 5 68 248 3.65 – A Train Level-of-Service Train (per car) 6 60 22 0.36 – D

ACS Alternative 1: Frequency Aspect Average Peak Area m2 m2/Pax % Change LoS Platform Level-of-Service Platform T3 67 205 477 2.32 -39.7% A Platform T4 23 88 248 2.81 -58.1% A Platform Sat (W) 68 466 248 0.53 -32.3% D Platform Sat (E) 23 154 248 1.62 -55.6% A Train Level-of-Service Train (per car) 32 60 22 0.36 – D

ACS Alternative 2: Capacity Aspect Average Peak Area m2 m2/Pax % Change LoS Platform Level-of-Service Platform T3 67 208 477 2.30 -40.3% A Platform T4 23 78 248 3.17 -52.7% A Platform Sat (W) 67 448 248 0.55 -30.4% D Platform Sat (E) 23 159 248 1.56 -57.3% A Train Level-of-Service Train (per car) 30 60 22 0.36 – D

The capital costs are almost exclusively depended on the required amount of cars, which are signiﬁcantly lower for both ACS alternatives. The maximum required amount of cars is consistently a bit higher for ACS alternative 1 (Frequency) than ACS alternative 2 (Capac- ity), as in some cases increasing the frequency before increasing the train capacity results in an uneven spread of passengers over the two scheduled services. If per example 100 passengers require a train service within 180 seconds, the system will in both alternatives initially activate 2 vehicles. If however only ± 30 passengers actually make it to the train scheduled extra after 90 seconds, this means that the second train still has to run a 2-car train to transport the remaining 70 passengers.

Operational Cost The operational cost is composed of the cost of energy and the cost of maintenance. Energy is used by an APM car en route and it consumes 2.56 kWh/km, as explained in section 2.4.1. The price of energy is ﬂuctuating as a result of many factors such as changing oil prices

52 Graduation Thesis

and political stability, but has nonetheless shown a relative consistent value of ± $0.10 in China (Want China Times, 2010, Liu, 2013). The cost per kilometre is thus 2.56 ∗ 0.10 = 0.256$/km. Kerr et al.(2014) states that the maintenance cost are approximately 70% of the power costs, which translates to a value of 0.256 ∗ 0.70 = 0.179$/km. The whole operational cost is thus directly proportional to the total distance travelled by the cars in the system. This value is lower for both ACS alternatives, resulting in a large cost reduction compared to the reference case. 1) trains do not run when there is no demand and 2) if there is demand they only serve the connection on which transport is required, after which they return to their idle parking location and 3) train combinations are a lot smaller, with on average just over 1 car per train for both ACS alternatives (instead of a ﬁxed 4 cars per train in the reference case AppendixD). As can be expected, the amount of cars per train is (slightly) higher in the second ACS alternative (Capacity) compared to the ﬁrst (frequency).

The ACS is beneﬁcial to reduce costs in terms of both capital investment and daily operation. Especially the reduction in vehicles required and the distance run on the system are eﬀective.

5.5.3 External Eﬀect

The study performed by Bombardier(2015) on the pollution of their Innovia APM 300 product states that a single car expels 1456 gr/km of CO2 when in operation at an European airport. Their study is based on the overall penetration of renewable and nuclear energy in Europe, which accounts for 46% of the total production. However, in the case of SZX, energy is likely sourced from a polluting power plant, since renewable energy penetration in China only 6% Liu et al. (2011). According to Lu(2010) and the NEI(2015) another 22% is generated hydraulic and only 2% nuclear , which means that the environmental impact of running an APM in China will be signiﬁcantly higher than in Europe. Assuming that energy is sourced evenly from the available 46 power suppliers, it can thus be said that the reference APM vehicle expels 1456 ∗ 30 = 11, 262 gr/km at SZX. The resulting CO2 pollution per alternative is summarised in table 5.6, from which is clear that the environmental impact of an ACS is signiﬁcant, with a reduction for both ACS alternatives of > 60% compared to the reference case.

53 Chapter 5 Detailed Test Case Design For Shenzhen Bao’an International Airport

Table 5.5: Results Cost

Reference Case Cost Units Unit Cost Total Cost Depreciation Capital Cost BLIP sensor system 0 $150,000 $0 $0 APM Cars 16 $2,400,000 $38.400,000 $3,504 Total $48,000,000 $3,504 Operational Cost Energy 9,834 0.256 $2,517 NVT Maintenance 9,834 0.197 $1,937 NVT Total $4,454 Total Daily $7,958

ACS Alternative 1: Frequency Cost Units Unit Cost Total Cost Depreciation Capital Cost BLIP sensor system 3 $150,000 $450,000 $41 APM Cars 14 $2,400,000 $33,600,000 $3066 Total $34,050,000 $3,107 Operational Cost Energy 3.704 0.256 $948 NVT Maintenance 3,704 0.197 $730 NVT Total $1678 Total Daily $4,785 (-39.9%)

ACS Alternative 2: Capacity Cost Units Unit Cost Total Cost Depreciation Capital Cost BLIP sensor system 3 $150,000 $450,000 $41 APM Cars 13 $2,400,000 $31,200,000 $2.847 Total $31,650,000 $2,888 Operational Cost Energy 3,878 0.256 $993 NVT Maintenance 3,878 0.197 $763 NVT Total $1,757 Total Daily $4,645 (-41.6%)

Table 5.6: Results CO2 pollution

Distance gr/km CO2 Total kg CO2 % Change Reference Case 9,834 11,262 110,751 – ACS Alternative 1: Frequency 3,704 11,262 41,714 -62.3% ACS Alternative 2: Capacity 3,878 11,262 43,674 -60.6%

54 6. Conclusion & Recommendation

Automated People Movers (APMs) are an important asset for large airports to support intra- terminal passenger movements and/or provide inter-terminal transit. While APMs were first introduced at airports in the early ’70s, the physical and operational system characteristics have not changed much since. A large inefficiency is that trains run on fixed schedules, but with recent developments in technology such as Communication Based Train Control (CBTC) and high-resolution passenger flow sensors, it is possible to redesign the operational control and make it more intelligent.

The objective of this research is to utilise the new technologies and design a comprehensive control method to adapt the network capacity availability of an Automated People Mover in an airport environment to the real-time demand. The design is thereby not only tested on a technological level, but also includes economic, passenger comfort and sustainability aspects to determine the feasibility of such a control type.

Conceptual Design of an Adaptive Control System To design an Adaptive Control System (ACS), it is determined that Model-based Predictive Control (MPC) is the most favourable method. The MPC calculates a set of future actions based on passenger demand forecast models that as a combination satisﬁes an objective.

The prime objective of the MPC is to minimise the diﬀerence between the system capacity for the next n time steps and the demand forecast for that same period n. While demand characteristics can roughly be calculated based on historical data and airport forecasts, it is preferred to place a sensor at an appropriate distance before the platform such that the minimum forecast period is met. The capacity can either be changed by running more/less trains or increase/decrease the amount of cars per train. The sequence of such capacity changing actions depends on the system owner’s requirements and can favour changing train capacity over train frequency or vice versa.

Detailed Design of an Adaptive Control System The conceptual design of the ACS is converted into detailed designs for proposed APM systems at Amsterdam Schiphol International Airport (AMS) and Shenzhen Bao’an International Airport (SZX). These designs are tested and evaluated by means of system models for three alternatives: Reference Case: a representation of a conventional ﬁxed schedule operation; ACS 1 - Frequency: an ACS that favours changing train frequency over train capacity; ACS 2 - Capacity: an ACS that favours changing train capacity over train frequency.

The simulation results, summarised in Table 6.1 and 6.2, show a comparable result on the (Key) Performance Indicators (KPIs) for the two ACS alternatives in terms of dwell time, Level-of-

Service (LoS), operational costs and CO2 pollution. Capital costs are however consistently lower in the seconds ACS alternative (capacity) as fewer cars have to be acquired.

55 Chapter 6 Conclusion & Recommendation

Table 6.1: Summary of Simulation Results AMS

KPI Passenger Experience Cost (daily) External

Table 6.2: Summary of Simulation Results SZX

KPI Passenger Experience Cost (daily) External

PI Wait(s) LoS Plat LoS Train Capital Operation kg CO2 Reference Case 92.31 C D $3,504 $4,454 110,751 ACS 1: Frequency 80 D D $3,107 $1,678 41,714 ACS 2: Capacity 80.42 D D $2,888 $1,757 43,674 miss it and an extra car in the following train is required to compensate. The expected lower waiting time that should result from the logic is thereby insigniﬁcant with comparable average dwell times for both ACS alternatives.

The results of the ACS alternatives for the KPI passenger experience are on par with the reference case. The dwell time is similar for AMS and lower for SZX. However the ACS does come with a higher platform utilization, which results in a drop in level of service for a singular platform in the SZX test case (from LoS C to Los D, on a ranking from A=best to F=worst). The LoS D measured on this platform at SZX is acceptable according to IATA standards for short periods, but should generally be redesigned to a higher LoS to meet client requirements. The LoS D measured for the trains in all alternatives is also low, but due to the short transit period transit it is deemed as an appropriate minimum LoS by industry experts.

The ACS alternatives differ significantly from the reference case on the KPIs cost and sustainability. As was concluded before, the ACS alternative that favours frequency over capacity (ACS alternative 1) consistently requires more cars, which results in a larger capital and total cost of ACS alternative 1 in the AMS test case. In the SZX case, the costs in the ACS alternatives are consistently lower, though. As the ACS alternatives in both test cases significantly reduce the total run distance, directly proportional operational costs and CO2 pollution are reduced as well.

56 Graduation Thesis

Both ACS alternatives show an overall good result compared to the reference case in terms of reducing costs (capital and operational) and increasing the sustainability of the system. While on the other hand the LoS decreases slightly for the SZX test case, the overall comfort that passengers experience is still adjudged to be acceptable. It should be noted that system demand and design characteristics have a considerable eﬀect on the relative result of the ACS implementation.

6.1 Recommendations for NACO

This research has been conducted on the request of NACO to primarily increase the in-house knowledge of APM systems. While the research includes many aspects of the APM system functioning and the APM industry, it is recommended to have a periodic review or research to stay up-to-date with the evolving technology.

As for the outcome of this research, it is important that NACO approaches companies within the APM industry (specialists such as Lea+Elliot and manufacturers such as Bombardier and Siemens) and brings the idea of adapting capacity to demand to the attention. The effect of the concept is promising as is indicated in this chapter and it could be a distinctive product to offer clients. A validation is therefore first needed by manufacturers to ensure the technical feasibility of the design. Thereafter, it should be the function of NACO to research the market potential of a new APM operation through its client network.

NACO’s involvement in APM studies commonly considers the physical characteristics of the system and would therefore be recommended to adapt certain design considerations to the concept explained in this research. Design aspects such as a network with individual cars or trains, parking locations and the system peak capacity could be proposed such that future changes in the operational control could be implemented. A general recommendation on the design of an

57 Chapter 6 Conclusion & Recommendation

APM system is thereby that platforms are experience the highest peak period on the side where aircraft movements occur (e.g. the Satellite at SZX and Pier-A at AMS). This is caused by the narrow time period in which arriving passenger move from the arrival gate to the APM platform.

6.2 Recommendations for Further Scientiﬁc Research

The two chosen airports are a good representation of the general characteristics that an airport can have, but still fail to represent some possibly influential driving forces. It was outside the scope of this research to consider more test cases, but it is worth mentioning an airport that could be of particular interest for further research, which is Atlanta. The first shortcoming of both APM test cases is that they do not actually exist, which results in a theoretical reference case. It would be interesting to do an additional test case for the Atlanta APM system that is in operation. Thereby, there is a difference in the effect of the ACS for both airports which is the result of the size of the system. SZX already represents a large operation, but it would be interesting to see the effect of the ACS in the larger system at Atlanta, which has 2.5 times more stations and a daily ridership of 200.000 passengers.

A large omission in this thesis is neglecting system failures. Whereas this thesis shows that an ACS is feasible in a system that continuously operates a normal procedure, in reality it should also cope with eventual failures. Thereby, utilising an intelligent control system might increase system failure proneness on its own and it would therefore be very interesting to further research the eﬀects thereof.

Another interesting aspect of this thesis that might require further research, is the effect of changing the headway. As was concluded from the model simulation results, this variable can heavily affect the outcome of the system once the headway crosses a threshold that requires a change in amount of trains active. The first ACS alternative (frequency) for AMS for instance has a reduced effectiveness compared to the reference case, caused solemnly by the chosen minimum headway.

Other topics that could be considered for future research are:

• What is the eﬀect of parking locations in an APM network to optimally adapt capacity to changes in passenger demand?

• How should an intelligent control structure such an ACS be implemented in a ATC control system?

• What are the opportunities of an intelligent control structure such an ACS for large scale transit systems such as metros and (light) rail?

58 Appendices

A. Airport Pax Transit Systems

An APTS is deﬁned as any form of system used to transport passengers between various locations at an airport. Airports expand their terminal buildings and/or construct new terminals to accommodate future passenger demand and thereby passenger travel distances can grow to multiple kilometres.

To cover the distances, some kind of transport system is required. The most common solution used at virtually all airports is the moving walkway or travelator. This system functions as a backbone for many intra-terminal connections, allowing passengers to move quickly through lengthy hallways and piers. With speeds of up to 12 km/h, travel times can be reduced by 75% (Bryant, 2014).

While travelators are functional inside an airport terminal, some sort of transport vehicle is required to move people between terminals. These vehicles can range from simple bus systems that use the normal road network, up to fully automated train(like) systems that run on dedicated underground tracks. A reference literature survey is done for APTS, of which APM systems are explained in Chapter1 and the other systems are summarised in this section.

A.1 Personal/Group Rapid Transport

Personal and Group Rapid Transport, further referred to as PRT and GRT respectively, are small semi-autonomous vehicles that run on a dedicated track. The general application of this systems uses a multitude of cars that run on a demand base, by which passengers request a direct transport from origin to destination (Irving et al., 1978). The vehicles have a capacity that ranges from a 1 to 6 passengers for PRT (Gilbert and Perl, 2007) to a couple of dozen passenger(s) for GRT (Jeﬀery, 2010). Depending on the track layout and guidance system, vehicles are able to chose shortest routes in the network and bypass other vehicles on their journey. Currently, one PRT system is operated at London Heathrow, which runs from terminal 5 to the business parking lot on a three station track and serving 10,000 passengers daily.

While applications are limited and the conception is generally deemed something of the future, the idea of PRT/GRT is already more than half a century old. Fichter(1964) was one of the first who envisioned a PRT/GRT, by proposing a tram-like overhead transit system on which small individual vehicles could suffice in the transport needs of a city. The first true application was completed by Boeing at West Virginia University in Morgantown (WV) just a decade later (Crowley, 1974, Boeing, nd), but not many other systems have followed thereafter. A total of 5 systems is currently in operation of which two are those of Morgantown and London Heathrow (figures A.1 and A.2). All systems run on short networks and are generally chosen not so much for their effectiveness, but for their attractiveness. The report by Furman et al.(2014) on the current state of the industry gives a very bleak prospect, as due to the few applications, costs and uncertainties are still high and upcoming competition of autonomous vehicles are expected to be disruptive.

Nonetheless, 7 companies worldwide actively develop PRT/GRT systems, with another 9 companies showing their intention to enter the market. Thereby, several researches have shown that

61 Appendix A Airport Pax Transit Systems

Figure A.1: WVU PRT (Wikipedia) Figure A.2: Heathrow PRT (Wikipedia)

PRT/GRT can be a good substitute to other modes of transport due to advantages, such as; routing freedom, low capital costs, low infrastructure costs, sustainable footprint, small headways (i.e. high capacity), high level of service (i.e. passenger experience and visual intrusion (Buchanan et al., 2005, Kerr et al., 2014, Juster and Schonfeld, 2013).

A.2 Metro

Metro, also known as subway, rapid transit and underground, is a form of group transport system operating completely separate of other modes of transportation on an exclusive right-of- way (APTA, 2014). The metro is a popular mean for municipalities to provide good and reliable public transport. Applications are found in virtually all large cities (e.g. London, New York, Shanghai) and generally run on underground or overhead tracks and are therefore not disrupted by other traﬃc. In contrary to APM systems, metros are predominantly manned. While there is a variety of support, propulsion and guidance solutions, the majority of metro systems make use of a electric powered, steel wheel vehicle on a rail with power supplied trough a side rail (also known as ’third’ rail). Currently, there are some airports that use the metro as APTS. For instance the metro connects terminal 1 and terminal 2 at Shanghai Hongqiao airport and terminal 1,4 and 5 at London Heathrow. It should be noted that all metro APTS applications are strictly on landside and predominantly serve OD traﬃc that comes out of the city and wants to directly access the terminal of departure.

A.3 Other APTS Solutions

Other APTS solutions that are currently used at airports to transport passengers between the diﬀerent terminals and facilities are the bus, (executive/V.I.P.) taxi services and/or simple walkways (with or without travelators). All these systems are used on both landside and airside and are rather self explanatory.

62 B. APM System Benchmark

This appendix contains 6 benchmark studies of APM systems in operation. The explained airports all represent a speciﬁc APM system, and are part of a larger index including all 48 systems in operation at the moment. Table B.1 shows the airports that are used as benchmark and for which system this is.

Table B.1: APM test cases

Airport System Hartsﬁeld-Jackson Atlanta International Airport Mitsubishi Crystal mover Beijing Capital International Airport Bombardier Innovia APM 100/C(X)-100 Birmingham International Airport Doppelmayr CC Cable Liner O’Hare International Airport Siemens (Matra) VAL 256 Dallas/Forth Worth International Airport Bombardier Innovia APM 200 Detroit International Airport Poma-Leitner MiniMetro (Hovair)

63 Appendix B APM System Benchmark

B.1 Hartsﬁeld-Jackson Atlanta International Airport Airport Characteristics City: Atlanta, GA Country: United States of America Region: North America Passengers (2013: 94.4 million

PTS Characteristics - Network Name: ATL Skytrain Constructor: Mistubishi Heavy Industries Construction Year: 2009 Length at Airport (km): 2.1 Location: Landside Stations: 3 Conﬁguration: Pinched Loop Headway 120 Capacity (ppdph): 2,686 Ridership (ppd): Unknown Control System: Unknown PTS Characteristics - Vehicle Name: Crystal Mover Propulsion: Electric Motor (DC, 750V) Guidance Side Guided 4-wheel steering Suspension: Rubber Tyres Acceleration (ms-2): 0.97 Deceleration (ms-2): 0.97 Maximum Speed (kmh-1): 55 (L+E) / 80 (MIT) Size(m): 11.8 x 2.7 x 3.6 Weight (kg): 17,200 Vehicle Capacity: 93 (MIT) / 67 (L+E) / 52 (ATL) Vehicles per Train: 2 Fleet Size: 12

Mitsubishi Crystal Mover

Hartsﬁeld-Jackson(nd), Mitsubishi HI (2010) 64 Graduation Thesis

65 Appendix B APM System Benchmark

B.2 Beijing Capital International Airport Airport Characteristics City: Beijing Country: China PR Region: Asia Passengers: 83.7 million

PTS Characteristics - Network Name: Terminal 3 People Mover Constructor: Bombardier Transportation Construction Year: 2008 Length at Airport (km): 2 Location: Airside Stations: 3 Conﬁguration: Pinched Loop Headway 300 Capacity (ppdph): 4100 Ridership (ppd): Unknown Control System: CityFlo550

PTS Characteristics - Vehicle Name: CX-100 Propulsion: Electric Motor (AC, 600V) Guidance Centre Beam Suspension: Rubber Tyres Acceleration (ms-2): ∼1.00 Deceleration (ms-2): ∼1.00 Maximum Speed (kmh-1): 46 (L+E) / 80 (BD) Size(m): 12.8 x 2.8 x 3.4 Weight (kg): 14,900 Vehicle Capacity: 52 Vehicles per Train: 2 Fleet Size: 11

Bombardier Innovia APM 100 / C(X)-100

Sources: Bombardier (2015), ACRP(2012b)

66 Graduation Thesis

67 Appendix B APM System Benchmark

B.3 Birmingham International Airport Airport Characteristics City: Birmingham Country: United Kingdom Region: Europe Passengers (2013): 9.1 million

PTS Characteristics - Network Name: Air-Rail Link Constructor: Doppelmayr Cable Car Construction Year: 2003 Length at Airport (km): 0.6 Location: Airside Stations: 2 Conﬁguration: Dual Lane Headway 120 Capacity (ppdph): 1,620 Ridership (ppd): Unknown Control System: PLC based

PTS Characteristics - Vehicle Name: Cable Liner Propulsion: Cable (AC, 415V) Guidance Rail Suspension: Rail Acceleration (ms-2): ∼ 1.00 Deceleration (ms-2): ∼ 1.00 Maximum Speed (kmh-1): 36 (DCC) / 45 (L+E) Size(m): 6.0x2.8x3.4 Weight (kg): 4100 Vehicle Capacity: 27 (DCC) / 29 (L+E) Vehicles per Train: 2 Fleet Size: 4

DCC Cable Liner

Sources:DCC Dopelmayr(nd), ACRP(2012b)

68 Graduation Thesis

69 Appendix B APM System Benchmark

B.4 O’Hare International Airport Airport Characteristics City: Chicago, IL Country: United States Of America Region: North America Passengers (2013): 66.8 million

PTS Characteristics - Network Name: Airport Transit System Constructor: Siemens (Matra) Construction Year: 1993 Length at Airport (km): 4.3 Location: Landside Stations: 5 Conﬁguration: Pinched Loop Headway 206 Capacity (ppdph): 3,200 (SIE) / 2,400 (ACRP) Ridership (ppd): 35,000 Control System: Block Based PTS Characteristics - Vehicle Name: VAL256 Propulsion: Electric Motor (DC, 750V) Guidance Centre Beam Suspension: Rubber Tyres Acceleration (ms-2): 1.30 Deceleration (ms-2): 1.30 Maximum Speed (kmh-1): 73 Size(m): 13.8 x 2.6 x 3.5 Weight (kg): 20400 Vehicle Capacity: 114 (SIE) / 62 (L+E) Vehicles per Train: 6 Fleet Size: 15 Siemens VAL 256

Sources: ACRP(2012b), Siemens (2014)

70 Graduation Thesis

71 Appendix B APM System Benchmark

B.5 Dallas/Forth Worth International Airport Airport Characteristics City: Dallas/Fort Worth, TX Country: United States Of America Region: North America Passengers (2013): 60.5 million

PTS Characteristics - Network Name: DFW Skylink Constructor: Bombardier Transportation Construction Year: 2005 Length at Airport (km): 7.9 Location: Airside Stations: 10 Conﬁguration: Dual Lane Loop Headway (s): 120 Capacity (ppdph): 5,000 (ACRP) Ridership (ppd): Unkown Control System: CityFlo 650 PTS Characteristics - Vehicle Name: Innovia APM 200 Propulsion: Electric Motor (DC, 750V) Guidance: Centre Beam Suspension: Rubber Tyres Acceleration (ms-2): ∼ 1.00 Deceleration (ms-2): ∼ 1.00 Maximum Speed (kmh-1): 60 Size(m): 12.0x2.9x3.4 Weight (kg): 14,500 Vehicle Capacity: 69 (L+E) Vehicles per Train: 2 Fleet Size: 64 Bombardier Innovia APM 200

Sources: ACRP(2012b), Lea+Elliot(nd)

72 Graduation Thesis

73 Appendix B APM System Benchmark

B.6 Detroit International Airport Airport Characteristics City: Detroit Country: United States Of America Region: North America Passengers (2013): 32.4 million

PTS Characteristics - Network Name: Express Tram Constructor: Poma Leitner Construction Year: 2002 Length at Airport (km): 1.1 Location: Airside Stations: 3 Conﬁguration: Single Lane with Bypass Headway (s): 192 Capacity (ppdph): 4000 Ridership (ppd): Unknown Control System: PLC based PTS Characteristics - Vehicle Name: MiniMetro Propulsion: Cable Guidance: Guidance Rail Suspension: Air Suspended Acceleration (ms-2): 0.74 Deceleration (ms-2): 0.74 Maximum Speed (kmh-1): 48 Size(m): 12.8x2.9x3.6 Weight (kg): 15,400 Vehicle Capacity: 73 Vehicles per Train: 2 Fleet Size: 4 Poma-Leitner MiniMetro

Sources: ACRP(2012b), Bares(nd), Poma Leitner(nd)

74 Graduation Thesis

C. Arena Simulation Model

This appendix contains a thorough explanation on the inner workings of the Arena software and the models that have been build to support the research. The appendix ﬁrst introduces the functioning and basic principles of the software, followed by three models that haven been build herewith. The generic model and its mathematical functioning is explained in C.3, followed by speciﬁc model aspects for the two test cases Shenzhen and Amsterdam in section C.4 and C.5, respectively.

C.1 Arena Software Methodology

The system will be modelled with the Rockwell Arena Simulation software, further referred to as ’Arena’. This software allows the user to build a Discrete Event Model/Simulation (DEM/S) in which decisions in the system are based on individual entities. The software uses the SIMAN language which was developed in the early ’80s to SIMulate and ANalyse (SIMAN) manufacturing processes (Pegden, 1983).

SIMAN uses a set of modules which can alter the attributes of entities that are fed in and/or alter system variables based on entity actions. Thereby, the language has been developed to represent time consuming processes that halt and queue individual entities ( Figure C.1).

Figure C.1: Simple representation of a typical system analysed with SIMAN (Pegden, 1983)

Whereas the SIMAN language is dedicated to pure discrete event systems (i.e. decisions are made at predeﬁned time intervals), the Arena software incorporates continuous simulation capabilities. Herewith it becomes possible to accurately simulate vehicle/conveyor/entity movements and/or instantly react to continuous processes such as ﬁlling a tank (Kelton et al., 1998).

With Arena it is also possible to visualise the decisions made within the system by means of (simple) animation (Altiok and Melamed, 2010). This feature assists the user to verify the functioning of the model and can be used to explain the abstract decision logic of the system to third parties. All modelling modules used for this research are listed and explained in table C.1.

C.2 Model/Simulation Set Up, Validation & Veriﬁcation

The simulations run with the models build in Arena should represent a correct and realistic system. It is therefore needed to ﬁrst prepare the simulation in terms of run time, warm up time

77 Appendix C Arena Simulation Model

Table C.1: Arena Modules

Name Icon Decision Basic Process library The assign module is activated by an entity. In the simulation for this Assign report, the assign module is used to add/change an attribute of the passing entity or to change a variable in the system. The batch module is used to batch entities. A predefined amount of entities build up to a batch size, after which the entities are combined Batch and released from the module as one. These batched entities can either be temporary or permanent. The create module is the starting point for any Arena model, because it creates the entities that will later trigger all decisions in the system. Create The creation of entities can be done either by a (statistically distributed) inter arrival time or by using a predefined schedule. The decide module represents the simple IF structure. By testing a Decide condition in the system or the passing entity, it distributes the entity to n options. The dispose module is needed to demarcate the system and releases Dispose entities from the process. The separate module is somewhat the opposite of the batch module. When in the latter entities are batched on a temporarily basis, the batch Separate can be separated again with the separate module. The module is also creating one or more duplicates. Advanced process The allocate module requests the nearest transporter in the network to Allocate the entity and changes the state of that transporter to ’busy’ The free module changes the status of the transporter that is assigned Free to the passing entity to ’free’ The route module leads from one location in a system to another location Route with a predefined delay. Station The station module represents a location in the system. The transport module sends the allocated transporter to a predefined Transport station Advanced transfer

Delay This module delays the entity for a predeﬁned time.

The hold module queues entities until further notice. The entities can Hold be released based on a system state or a certain signal given by a signal module The ReadWrite module is needed to write out and extract data to and ReadWrite from a ﬁle (e.g. .txt or .xlsx). The signal module creates a release order for a hold module. To specify Signal which hold module should be triggered, a unique code can be assigned

78 Graduation Thesis and run replications and thereafter verify and validate the working of the system. The methods used to test these aspects is explained in this section, with the test results speciﬁed in the the respective model summary in sections C.5 and C.4.

Run Length The Run length is (logically) the time period that is simulated. The run length should be long enough to present all possible event at least 5 times (Al-Aomar et al., 2015). In the modelled environment the most important events are the arrival and departure of aircrafts, the entering and exiting of a vehicle, including the coupling of vehicles and synchronisation of the train scheduling at one platform with the train scheduling with the former and next station. All model runs show that these events occur frequently during a daily run, with most events happening in the first 12 hours. As the system is driven based on a (fixed) schedule on when aircraft should depart and arrive every day, it is however appropriate to take a run length of 24 hours to also represent the daily fluctuations.

Warm Up Period The simulation run starts of with an empty input set, which is an unrealistic situation in a continuously operating system. In the case of an APM system, the simulation run will start at 12:00 p.m. with no vehicles, aircraft and/or passengers. In reality it is however very well possible that an event has happened before the start time which should inﬂuence the PIs when the system is started. Therefore, a certain warm up period is required to let the simulation reach a state in which operations are fully functioning. Kelton and Law(2000) propose to take the simulation run length for the warm up period also to eliminate any bias in the system, which is 24 hours.

Run Replications Replications are needed to diminish the effects of variation on the model. Every individual replication runs with a different seed and delivers different outputs. As it is possible that some runs are far from an actual representation, it is best to run multiple replications to even out all excesses. To determine the amount of replications needed for a valid model, the PIs defined in Section 2.5 and summarised below, are calculated with the model.

• Passenger experience;

– Dwell time [seconds] – Platform Area LOS [pax waiting] – Vehicle Area LOS [load factor]

• System Cost;

– Sensor System [#] – APM vehicles [vehicles] – Energy cost [km] – Maintenance cost [$/km]

• External eﬀect.

– Energy use [km]

79 Appendix C Arena Simulation Model

Hoad et al.(2007) states that the half width of the confidence interval for any criteria should be smaller than a predefined percentage of the cumulative mean. Kortmann (personal communication, 2013) advised that the half width should be no more than 5% of the average, which coincides with the 95% certainty interval commonly used in APM development. If any of the values exceeds this limit, an extra replication is added and the model is run again. This process is repeated until all PIs are within the limits. As some PIs are influenced differently in the alternatives, a run replication test is done for all alternatives, and the highest amount of replications is taken for all models.

Sensitivity Analysis Sensitivity analysis is defined as the study of how uncertainty in the output of a model can be attributed to different sources of uncertainty in the model input (Saltelli et al., 2008). By means of the analysis it is possible to define sensitive input parameters which can (if possible) could be changed to effectively optimise the system. If any alternatives have been defined, it also possible to test the robustness of the solution.

The most common method used to test the sensitivity of the model is the one-at-a-time (OAT) method. By changing one input parameter at the the time it is possible to see what the relative influence on the PIs is of that parameter. This is a fairly straightforward approach that creates good insight on the effect of a particular factor. Saltelli and Annoni(2010) does state that using this approach is intrinsically wrong and proposes that sensitivity analysis could better be based on e.g. statistical theory, drawing from experimental design, regression analysis and sensitivity analysis proper. It is however not the focus of this research to fully comprehend the precise sensitivity of all input parameters, but instead get an initial understanding of an ACS for APM Systems. Therefore, a ’simple’ OAT sensitivity analysis is deemed satisfactory. There are no guidelines on the range of a sensitivity test, but (Eschenbach, 1992) proposes the tornado diagram as a visual representation of the effect of different input parameters on the different PIs. This representation is much clearer than having a large table and is one of the most visual representations, together with spider plots or Pareto charts (Jain and Singh, 2003).

To reduce the complexity of the sensitivity analysis, only those input parameters that are based on large uncertainties and/or pure assumptions are included. The largest uncertainty is the amount of movements, which is forecast for 25 years. The passenger seating capacity is another uncertainty which can change somewhat in the future years. A third large assumption made in this research is that 75% of the passengers will use Terminal 3 instead of Terminal4. The last input paramater which is tested, is the capacity of the vehicle, which can range quite a bit between manufacturers. There are of course many other assumptions made, but there relative inﬂuence is deemed low and/or near to actual value (e.g. deboarding distribution and operational vehicle speed).

Verification & Validation Verification and validation is needed before actual results and conclusions can be extracted from the simulation. As Boehm(1991) describes; verification answers the question if the product is build right, whereas validation answers the question if the right product is build.

Veriﬁcation is largely based on common sense and can best be tested by having a third party look at the model/code and determine if it makes the right decisions to transform the input to output. As the Arena tool is able to visually show the decisions made, the checks could be done

80 Graduation Thesis fairly well with all supervisors at NACO and the university. By means of a simple animation, all available events could be checked on the decision logic used and no failures have been counted in the ﬁnal models. Thereby, The output of all models generate correct distributions and value ranges for all PIs. It can therefore be safely assumed that the model build does not have any (inﬂuential) faults that hamper the correct functioning of the simulation.

If the model reflects the right system is two-sided. For the ACS alternatives, the model correctly reflect the designed decision logic. The actions made are tested in the simulation and are checked by supervisors at NACO and the university. However, it is debatable if the base models present an accurate simulation of standard operations. While peak demand calculations are relatively well defined, it is unknown from literature how systems could be scheduled to account for off- peak and night periods. An assumptions is therefore made which should be validated with manufacturers and users and eventual changes should be made based on the expert advise.

C.3 Model Description

The model is explained in this paragraph per model section. A passenger entity will come trough these sections in a predeﬁned sequence, which is visualised in Figure C.2. The position of the Adaptive Control Logic for AMS is highlighted in cyan and is discussed in detail in the main text (Chapter3).

Figure C.2: Model Sections AMS

Aircraft Schedule Generator The APM passenger demand is generated by aircraft movements. As both test cases contain a conceptual future APM system, it is required to create an applicable schedule based on forecasts. The aircraft schedule generator is therefore used to extrapolate an existing movement schedule and create an appropriate movement pattern to run the model with. All scheduled aircraft movements are put trough the model and some are duplicated and added to schedule to represent additional movements (Figure C.3). The amount of movements that should be duplicated is based on the expected percentage growth between the base and future schedule and aircraft are added to the schedule.

A tower log of November 2012 is taken as basic schedule for the AMS test case. This was then corrected for the actual peak month July and the diﬀerence in demand between the base year 2012 and the design year 2017 (forecast by NACO), which results in schedule growth of 131%. The same approach is taken for SZX, in which the peak day of 2014 (24 August) is extrapolated to the design year 2040 based on the diﬀerence in annual throughput (32 Million to 70 Million).

Passenger Generator The Passenger Generation section creates passenger entities based on aircraft movement data. This aircraft data is read from a .xlsx (MS Excel) file which has been prepared beforehand. This file contains the scheduled arrival times of an aircraft, its Schengen status, passenger capacity and Aircraft ICAO code. As can be seen in Figure C.4, an initial aircraft entity is created before any other aircraft entity, which is needed to reset the excel readout to the first line after every

81 Appendix C Arena Simulation Model simulation replication. After the aircraft is generated, it is allocated to a stand at a particular pier or terminal based on the share of positions available there. In this case this constitutes to 10 a chance of 108 ∗ 100% = 9.25% that an aircraft will park at Pier A. Hereafter, a distinction is made between narrow body and wide body aircraft, which have a different turnaround time distribution. With the turnaround time it is known when an aircraft will depart. However, if an aircraft is scheduled to depart during the night curfew (23:00 p.m – 6:00 a.m.), the flight is rescheduled to depart the following morning. The last step in this model section is to separate passenger entities from the aircraft entities. The amount of passengers destined and originating from and to an aircraft is calculated by multiplying the aircraft capacity with the average fill rate, which is 82% at AMS (Schiphol Group, 2015b). After assigning an attribute to distinguish arriving and departing passengers, the entity is routed to the next model section.

Passenger Distribution The Passenger Distribution module shown in Figure C.5 is used to translate an arrival or departure of an aircraft in an appropriate demand pattern for the APM System. The module considers for all passengers if they are transfer and if the transfer will be within the same terminal (i.e. no transit needed). For AMS, this value is 40%, of which a mere 8% (9/108) actuall remains within the A Pier. For SZX, the transfer values are very low, with only 2% transfer of which only 33% remains in the satellite.

The Passengers are thereafter distributed over the stations based on airport distributions and delayed for an distributed period until they pass the sensor location. Based on the Alternative chosen, passengers entities will either be further analysed in one of the two ACS modules or are directly forwarded to the Platform waiting module (reference case).

Adaptive Control System Alternative Favouring Frequency The first ACS module (Figure C.6) calculates the requirement if another train should be added to the schedule, by favouring Train Frequency changes over Train Capacity changes. After a first train is called up (blue), the second train can either be scheduled before (green) or after (red) the firs train.

Adaptive Control System Alternative Favouring Capacity The Second ACS module (Figure C.7) calculates the requirement if another train should be added to the schedule, by favouring Train Capacity changes over Train Frequency changes. After a ﬁrst train is called up (blue), the second train can either be scheduled before (green) or after (red) the ﬁrs train.

Platform Waiting Process Passengers are halted at the platform until a scheduled train is available. When the train arrives, passengers are let into the train until capacity is met. Any remaining passengers will stay on the platform and wait for the next train.

Station Process The Station Process comprises two Process modules: Outbound and Inbound (Figure C.9 and Figure C.10). Passengers are released from the Platform Waiting Process when a scheduled train arrives. They enter the outbound process and are batched together with the Train entity and this batch then continues to the next station. It is also possible that there are no passengers waiting and in this case the train is routed to a parking location.

82 Graduation Thesis

When a train enters a station, it will acknowledge its arrival and remains parked (held) at that station until further it is released for the next scheduled train or parking operation. The passengers are separated from the train and are either disposed because they arrived at the correct station, or are forwarded to the outbound process of that platform to continue their journey. The train entity is forwarded to the continuation module, in which the next action is determined.

Parking Process1 A set of trains can wait at a parking location. The trains are called up when a new schedule is created, but will wait until the latest departure time to allow for any changes in train length. When trains come back to the parking, they are put on hold until there services are required once again (Figure C.11.

Train Continuation Process1 The following action of a train that has arrived at a station, is determine with the Train continuation module (Figure C.12). If a new train is scheduled to synchronise with the arrival of the train entity, it will continue on the next leg of the system. If the train is however not required any further, the nearest available parking location is selected as next destination. The train entity is then routed to the outbound station process.

Train Scheduling The preferred scheduling of the next train is created in the ACS modules, but should be corrected to synchronise with the schedules of surrounding stations (Figure C.13. It is therefore checked if any train is scheduled to arrive close to the preferred departure time of the next train and or if the train will arrive at the next station near an already scheduled departure. If so, the train will either be planned before or after the arrival of the incoming train or planned to arrive before or after the departure of the train at the next station. Another option is to synchronise and utilise the same train for multiple network legs.

Scheduled Train Holding When an appropriate time is chosen, the train is put on hold until the departure (Figure C.14). Depending on the situation, an extra entity is created to request a train at a parking. Another entity is used to check the ﬁnal demand for the train and determine the appropriate amount of cars for the next train to meet demand.

1Only Applies to the SZX model

83 Appendix C Arena Simulation Model

Figure C.3: Module Schedule Generator

Figure C.4: Module Passenger Generation

84 Graduation Thesis

Figure C.5: Module Platform Distributor

Figure C.6: Module ACS Favour Frequency

Figure C.7: Module ACS Favour Capacity

Figure C.8: Module Platform Waiting Process

85 Appendix C Arena Simulation Model

Figure C.9: Module Outbound Station Process

Figure C.10: Module Inbound Station Process

Figure C.11: Module Parking Process

Figure C.12: Module Train Continuation Process

86 Graduation Thesis

Figure C.13: Train Schedule Module

Figure C.14: Module Scheduled Train Holding

87 Appendix C Arena Simulation Model

C.4 Amsterdam Airport Schiphol Model Set Up

Run replications The results of the 18 run replication chosen are shown in table C.3. Only the averages are measured for some indicators, as variation will not happen between the runs. As the failure rate is assumed to be 0, the run distance and run time are constant in the base run, as are the vehicles. The system also either requires 0 or 3 Sensor Systems.

The simulation requires a minimum of 18 replications for all PIs to be beneath the limit of 5%. This is found by iteratively increase the amount of replications, starting from 10.

Sensitivity Analysis The sensitivity analysis shows that the model is not excessively sensitive to most tested input parameters. Especially the dwell time shows no sensitivity to any input parameter but the headway, with the largest offset being less than 0.5%. The same characteristics are found for the total run distance, which seems to be affected mostly by the headway. The load factor of a train is logically influenced by a change in vehicle capacity and shows a directly proportional reaction for most input parameters. The analysis is visualised in Figure C.15 and summarised in C.4.

Table C.2: Results 18 run replications AMS

Indicator Replication Distributions Base Alternative Freq Alternative Cap Alternative 1 1 1 Avg. 2 w % Avg. 2 w % Avg. 2 w % Load Factor 8.86 0.04 0.45 16.69 0.14 0.84 16.69 0.14 0.84 Dwell time 100.8 0.00 0.00 95.33 0.00 0.00 94.18 0.00 0.00 Waiting A (S) 3.10 0.08 2.58 1.88 0.03 1.60 2.04 0.02 0.98 Waiting A (NS) 6.84 0.13 1.90 4.75 0.05 1.05 4.33 0.04 0.92 Waiting C 2.21 0.06 2.71 2.05 0.02 0.98 2.07 0.02 0.97 Waiting D 4.65 0.17 3.66 4.31 0.05 1.16 4.38 0.05 1.14 Sensor Systems 0.00 – 3.00 – – 3.00 – – Cars 6.00 – 9.00 – – 5.00 – – Distance 1,240.00 – 773.00 4.00 0.52 764.00 4.00 0.52 Train Length 3.00 – 2.00 0.00 0.00 2.03 0.00 0.00

C.5 Shenzhen Bao’an Airport Model Set Up

Run replications The results of the 10 run replication chosen initially are shown in table C.3. Only the averages are measured for some indicators, as variation will not happen between the runs. As the failure rate is assumed to be 0, the run distance and run time are constant in the base run, as are the vehicles. The system also either requires 0 or 3 Sensor Systems.

As is highlighted in the table, all PIs are well beneath the limit of 5%, thus it can be concluded that 10 replications is more than suﬃcient to use for the base model.

88 Graduation Thesis offset being less than 0.5%. The same characteristics are found for the total run distance, which seems to be affected mostly by the headway. The load factor of a train is logically influenced by a change in vehicle capacity and shows a directly proportional reaction for most input parameters. The analysis is visualised in Figure C.16 and summarised in C.4.

Table C.3: Results 10 Run replications SZX

Indicator Replication Distributions Base Alternative Freq Alternative Cap Alternative 1 1 1 Avg. 2 w % Avg. 2 w % Avg. 2 w % Load Factor 13.26 0.56 4.22 53.15 0.93 1.89 49.67 0.85 1.71 Dwell Time 92.31 0.23 0.25 80.04 0.44 0.55 79.80 0.65 0.81 Waiting T3 23.20 1.06 4.66 67.51 2.92 4.33 66.27 2.48 3.74 Waiting T4 7.70 0.37 3.76 22.91 0.94 4.10 22.49 0.89 3.96 Waiting Sat (W) 22.64 0.90 4.31 67.26 2.53 3.76 66.65 2.66 3.99 Waiting Sat (E) 7.64 0.34 2.98 22.42 0.85 3.79 22.25 0.89 4.00 Sensor Systems –– – 3.00 – – 3.00 – – Cars 16.00 – – 13.50 0.57 4.22 12.70 0.41 3.23 Distance 9,834.00 – – 3,704.00 52.00 1.40 3,878.90 38.50 0.99 Train Length 4.00 – – 1.11 0.03 2.70 1.21 0.01 0.83

89 Appendix C Arena Simulation Model Waiting Waiting (c) (c) Distance Distance (b) (b) Tornado diagrams for SZX Sensitivity Analysis Tornado diagrams for AMS Sensitivity Analysis Figure C.16: Figure C.15: Load Factor Load Factor (a) (a)

90 Graduation Thesis 92.18 (92.12 0% (92.42 0% ) (92.47 0% ) (101.2 0% ) ( 10% ) ) 9834 (9834 0% (9834 0% ) (9834 0% ) (8973 0% ) ( -9% ) ) 12.05 (14.59 -9% (14.35 ) 10% (13.26 ) 8% (14.53 0% ) ( 10% ) ) 92.62 (91.62 0% (91.77 -1% (92.24 ) -1% ) (83.19 0% ) ( -10% ) ) 98349834 ( 0%9834 ( 0%9834 ( ) 0%10930 ( ) ( 0% ) 11% ) ) 14.73 (11.75 11% (11.92 -11% ) (13.26 ) -10% (11.93 ) 0% ( -10% ) ) Total Run distance 92.31 92.31 92.31 92.31 92.31 9834 9834 9834 9834 9834 13.26 13.26 13.26 13.26 13.26 Sensitivity Analaysis Table C.4: 102.24 (103.58 0% (102.56 1% ) (102.96 0% ) (111.62 Vehicle Capacity 1% ) ( Aircraft Movements 9% ) Aircraft Capacity ) Share T3 1240 Headway 1240 (1240 0% (1240 0% ) (1132 0% ) ( Vehicle Capacity 0% ) ( Aircraft Movements -9% ) Aircraft ) Capacity Share8.05 T3 Headway 9.64 (9.73 -9% (8.56 ) 9% (9.71 10% ) ( Vehicle Capacity ) -3% ( Aircraft ) Movements 10% Aircraft Capacity ) Share T3 Headway Load Factor Load Factor 102.24 (104.76 0% (101.8 2% ) 101.52 ( ( 0% ) 93.65 -1% ( ) ) -8% ) 12401240 (1240 0% (1240 0% ( ) 1364 0% ( ) 0% ( ) 10% ) ) 9.847.85 (7.96 11% (9.13 -11% ) (8.05 ) -10% ( ) 3% ( -9% ) ) Total Run distance 102.24 102.24 102.24 102.24 102.24 Basic1240 1240 1240 -10%1240 1240 10%8.86 8.86 8.86 8.86 8.86 Basic -10% 10% Basic -10% 10%Basic Basic -10% -10% 10% 10% Basic -10% 10% Passenger Platfrom Dwell Time Passenger Platfrom Dwell Time Amsterdam Schiphol International Airport Shenzhen Bao’an International Airport Aircraft Movements Aircraft Capacity Share Schengen Headway Aircraft Movements Aircraft Capacity Share Schengen Headway Aircraft Movements Aircraft Capacity Share Schengen Headway Vehicle Capacity Vehicle Capacity Vehicle Capacity

D. Results

Table D.1: Raw Output Data AMS Simulation

Output Parameter Reference Case ACS - 1: Frequency ACS - 2: Capacity Avg Min Max Avg Min Max Avg Min Max Load Factor [%] 8.86 0.00 100.00 16.69 0.00 100.00 15.64 0.00 100.00 Dwell time [s]* 100.80 0.00 900.00 95.33 0.00 237.00 94.18 0.00 237.00 Pax A (S) 3.10 0.00 76.00 1.88 0.00 62.00 2.04 0.00 65.00 Pax A (NS) [#] 6.84 0.00 166.00 4.75 0.00 136.00 4.33 0.00 136.00 Pax C [#] 2.21 0.00 24.00 2.05 0.00 21.00 2.07 0.00 21.00 Pax D [#] 4.66 0.00 48.00 4.31 0.00 40.00 4.38 0.00 40.00 Sensors [#] 0.00 0.00 0.00 3.00 3.00 3.00 3.00 3.00 3.00 Cars [#] 6.00 6.00 6.00 9.00 9.00 9.00 5.00 5.00 5.00 Distance [km] 1,240.00 1,240.001,240.00773.00 737.00 795.00 764.00 727.00 785.00 Cars/Train [#] 3.00 3.00 3.00 2.00 2.00 3.00 2.03 2.00 3.00 *includes passengers that have to wait >180s (± 5%) in ACS alternative 1 and 2

Table D.2: Raw Output Data SZX Simulation

Output Parameter Reference Case ACS - 1: Frequency ACS - 2: Capacity Avg Min Max Avg Min Max Avg Min Max Load Factor [%] 13.26 0.00 100.00 53.17 0.00 100.00 49.75 0.00 100.00 Dwell Time [s] * 92.31 0.00 900.00 80.00 0.00 180.00 80.42 0.00 179.99 Pax T3 [#] 23.20 0.00 124.00 67.50 0.00 205.40 66.27 0.00 207.70 Pax T4 [#] 7.70 0.00 37.00 22.91 0.00 88.10 22.49 0.00 78.20 Pax Sat (W) [#] 22.64 0.00 311.00 67.26 0.00 465.90 66.65 0.00 447.70 Pax Sat (E) [#] 7.64 0.00 68.00 22.42 0.00 153.50 22.25 0.00 158.90 Sensors [#] 0.00 0.00 0.00 3.00 3.00 3.00 3.00 3.00 3.00 Cars [#] 16.00 16.00 16.00 13.50 11.00 16.00 12.70 10.00 14.00 Distance [km] 9,834.00 9,834.009,834.003,704.00 3,499.003,926.00 3,878.00 3,750.004,073.00 Cars/Train [#] 4.00 4.00 4.00 1.13 1.00 4.00 1.21 1.00 4.00 *excludes passengers who walk to slow and take the next train (± 1%) in ACS alternative 1 and 2

Glossary

AMS Amsterdam Schiphol International Airport APM Automated People Mover APTS Airport Passenger Transit System ATC Automatic Train Control ATO Automatic Train Operation ATS Automatic Train Supervision Car Single APM vehicle in a train CBTC Communication Based Train Control Contact stand Aircraft parking at a gate EUR European continental ﬂight FBTC Fixed Block Train Control headway time period between two trains IATA International Air Transport Association ICA Intercontinental ﬂight ICAO International Civil Aviation Organization MPC Controller Model-Based Predicitive Controller Narrow body aircraft Small Aircraft with single aisle PID Controller Proportional Integral Derivative Controller remote stand Aircraft parking away from the terminal SZX Shenzhen Bao’an International Airport Tower Log Log of aircraft movements Train Combination of cars Wide Body Aircraft Large aircraft with multiple aisles

Bibliography

Ackoﬀ, R. L. (1999). Ackoﬀ’s best: His classic writings on management, volume 1. Wiley.

ACRP (2012a). Guidebook for Measuring Performance of Automated People Mover Systems at Airports, volume 37A. Transportation Research Board. ACRP (2012b). Guidebook for Planning and Implementation of Automated People Mover Systems at Airports, volume 37. Transportation Research Board. Air Council international (2014). ACI Airport Data 2013.

Al-Aomar, R., Williams, E. J., and Ulgen,¨ O. M. (2015). Basic witness R modeling techniques. Process Simulation Using WITNESS R , pages 111–148. Altiok, T. and Melamed, B. (2010). Simulation modeling and analysis with Arena. Academic press.

APTA (2014). Glossary of transit terminology. American Public Transit Association.

Araki, M. (2002). Pid control. Control systems, robotics and automation, 2:1–23. Banks, J. et al. (1998). Handbook of simulation. Wiley Online Library.

Bares, F. P. (n.d.). Rope automatic people mover systems.

Baumgartner, J. and Chu, P. (2013). Personal rapid transit user interface. In Automated People Movers and Transit Systems 2013@ sHalf a Century of Automated Transit—Past, Present, and Future, pages 247–262. ASCE. Bequette, B. W. (2003). Process control: modeling, design, and simulation. Prentice Hall Professional.

Bergenti, F. and Poggi, A. (2000). Exploiting uml in the design of multi-agent systems. In Engineering Societies in the Agents World, pages 106–113. Springer. Bijloo, E. (2012). The Application of Passenger Transport Systems at Schiphol Airport to reduce Walking Distances. PhD thesis, TU Delft, Delft University of Technology. BLIP Systems (2015). BlipTrackIndoor Sensor. Retrieved from http://www.blipsystems.com/ wp-content/uploads/2015/10/BlipTrack-Indoor-Sensor.pdf. Boehm, B. W. (1991). Software risk management: principles and practices. Software, IEEE, 8(1):32–41.

Boeing (n.d.). Personal Rapid Transit System. Retrieved from http://www.boeing.com/history/ products/personal-rapid-transit-system.page. Bombardier (2015). Automated People Movers. Retrieved from http://www.bombardier.com/en/ transportation/products-services/transportation-systems/automated-people-movers.html. Bombardier (2015). Innovia apm 300:environmental product declaration. Bryant, C. (2014). Thyssenkrupp aims to accelerate travelators beyond the pedestrian. Financial Times.

Buchanan, M., Anderson, J., Tegn´er,G., Fabian, L., and Schweizer, J. (2005). Emerging personal rapid transit technologies–introduction, state-of-the-art, applications. In Proceedings of the AATS conference, Bologna, Italy, pages 7–8. Centre for Aviation (2013). Shenzhen Airport to expand by 58% – supporting nearby congested Guangzhou, Hong Kong Airports. Retrieved from http://centreforaviation.com/analysis/shenzhen_airport-to-expand-by_58_ -supporting-nearby-congested-guangzhou-hong-kong-airports-131132. Crowley, R. T. (1974). Personal rapid transit system in morgantown, west virginia. Technical report, SAE Technical Paper. DCC Dopelmayr (n.d.). Automated People Mover. Retrieved from http://www.dcc.at/references/ air-rail-link-birmingham-uk/. De Neufville, R. and Odoni, A. (2003). Airport Systems. Planning, Design and Management. de Voogt, S. (2014). Dit bedrijf begon met Schiphol, en ontwerpt nu vliegvelden over de hele wereld. Retrieved from http://www.nrcq.nl/2014/09/08/ dit-bedrijf-begon-met-schiphol-en-ontwerpt-nu-vliegvelden-over-de-hele-wereld.

97 BIBLIOGRAPHY

Drolet, M. and Jadhav, R. (2005). Cbtc: Bombardier cityﬂo solutions. In Rail Transit Conference, 2005.

Elliott, D. M. and Norton, J. (1999). An introduction to airport apm systems. Journal of advanced transportation, 33(1):35–50. Enright, M. J., Scott, E. E., and Chang, K.-m. (2005). Regional powerhouse: the greater Pearl River Delta and the rise of China. Wiley Chichester. Eriksen, P. (2002). Collaborative decision making information management in airports. In Digital Avionics Systems Conference, 2002. Proceedings. The 21st, volume 2, pages 7B4–1. IEEE. Ertas, A. and Jones, J. C. (1996). The engineering design process, volume 2. John Wiley & Sons. Eschenbach, T. G. (1992). Spiderplots versus tornado diagrams for sensitivity analysis. Interfaces, 22(6):40–46.

EuroStat (2015a). Energy price statistics. Retrieved from http://ec.europa.eu/eurostat/ statistics-explained/index.php/Energy_price_statistics. EuroStat (2015b). Renewable energy statistics. Retrieved from http://ec.europa.eu/eurostat/ statistics-explained/index.php/Renewable_energy_statistics. Fichter, D. (1964). Individualized automatic transit and the city.

Foster+Partners (2015). Foster and partners to design new international airport for Mexico-City. Retrieved from http://www.fosterandpartners.com/news/archive/2014/09/ httpwwwfosterandpartnerscomnewsarchive201409_foster_and-partners_to_design_new_ international_airport_for_mexico_city/. Fruin, J. J. (1971). Pedestrian planning and design. Technical report.

Furman, B., Fabian, L., Ellis, S., Muller, P., and Swenson, R. (2014). Automated transit networks (atn): A review of the state of the industry and prospects for the future. Gilbert, R. and Perl, A. (2007). Grid-connected vehicles as the core of future land-based transport systems. Energy policy, 35(5):3053–3060.

Han, X. (2013). New terminal opens thursday. Shenzhen Daily.

Hartsfield-Jackson (n.d.). Hartsfield-Jackson’s Rental Car Center (RCC). Retrieved from http://www. atlanta-airport.com/Airport/Construction/RentalCarCenter/SkyTrain.aspx. Hoad, K., Robinson, S., and Davies, R. (2007). Automating des output analysis: how many replications to run. In Proceedings of the 39th conference on Winter simulation: 40 years! The best is yet to come, pages 505–512. IEEE Press. Hong Kong Standard (2015). Beijing backs third runway. Retrieved from http://www.thestandard. com.hk/news_detail.asp?sid=44262600&art_id=156101&con_type=1&pp_cat=30, Hong Kong. IATA (2015a). Airport development reference manual, volume 10th edition. International Air Transport Association. IATA (2015b). Strong traffic growth continues in may.

Irving, J. H., Bernstein, H., Olson, C., and Buyan, J. (1978). Fundamentals of personal rapid transit.

Jain, S. K. and Singh, V. P. (2003). Water resources systems planning and management. Elsevier. Jeﬀery, D. (2010). Guidelines for implementers of group rapid transit systems.

Juster, R. and Schonfeld, P. (2013). Comparative analysis of personal rapid transit as an urban transportation mode. Transportation Research Record: Journal of the Transportation Research Board, 2350(1):128–135.

Kelton, W. D. and Law, A. M. (2000). Simulation modeling and analysis. McGraw Hill Boston.

Kelton, W. D., Sadowski, R. P., and Sadowski, D. A. (1998). Simulation with arena, wcb. Kerr, T., Lowson, M., and Smith, A. (2014). Heathrow airport’s personal rapid transit system proves to be a viable transport solution. In Proceedings of the ICE-Civil Engineering, volume 167, pages 66–73. Thomas Telford. Kim, B.-s., Lee, G.-G., Yoon, J.-Y., Kim, J.-J., and Kim, W.-Y. (2008). A method of counting pedestrians in crowded scenes. In Advanced Intelligent Computing Theories and Applications. With Aspects of Artiﬁcial Intelligence, pages 1117–1126. Springer.

98 Graduation Thesis

Kimley Horn (2014). Metromover System Expansion Study. Miami-Dade County Metropolitan Planning Organization. Kuo, B. C. (1981). Automatic control systems. Prentice Hall PTR. Kusumaningtyas, I. and Lodewijks, G. (2013). On the application of accelerating moving walkways to support passenger processes in amsterdam airport schiphol. Transportation Planning and Technology, 36(7):617–635.

Lardennois, R. (1993). Val automated guided transit characteristics and evolutions. Journal of advanced transportation, 27(1):103–120.

Law, C. K., Law, J. S., and Tse, D. S. M. (2007). An investigation of airspace congestion & possible air traﬃc management integration in the pearl river delta region of southern china. Layard, P. R. G., Layard, R., and Glaister, S. (1994). Cost-beneﬁt analysis. Cambridge University Press. Lea+Elliot (2014). T5 APM Planning Singapore Changi: Final report. Lea+Elliot.

Lea+Elliot (n.d.). DALLAS/FORT WORTH INTERNATIONAL AIRPORT SKYLINK. Retrieved from http://www.leaelliott.com/dfw-airport-skylink. Lin, W.-S. and Sheu, J.-W. (2011). Optimization of train regulation and energy usage of metro lines using an adaptive-optimal-control algorithm. Automation Science and Engineering, IEEE Transactions on, 8(4):855–864.

Little, D. D. and Ross, S. E. (2013). Apms and airport mobility—historic trends and future possibilities. In Automated People Movers and Transit Systems 2013@ sHalf a Century of Automated Transit—Past, Present, and Future, pages 1–13. ASCE. Liu, W., Lund, H., Mathiesen, B. V., and Zhang, X. (2011). Potential of renewable energy systems in china. Applied Energy, 88(2):518–525.

Liu, Z. (2013). Electric power and energy in China. John Wiley & Sons.

Lott, J. S., Gettman, D., and Tai, D. S. (2009). Simulation analysis of apm systems in dense urban environments—part 2: System operations. In Automated People Movers 2009@ sConnecting People, Connecting Places, Connecting Modes, pages 588–603. ASCE. Lu, D. (2010). The current status of chinese nuclear power industry and its future. North China Electric Power University, Beijing. E-Journal of Advanced Maintenance, 2(1).

Macal, C. M. and North, M. J. (2010). Tutorial on agent-based modelling and simulation. Journal of simulation, 4(3):151–162.

Malinovskiy, Y., Saunier, N., and Wang, Y. (2012). Analysis of pedestrian travel with static bluetooth sensors. Transportation Research Record: Journal of the Transportation Research Board, (2299):137– 149. Mead, N. (2015). China’s Pearl River Delta overtakes Tokyo as world’s largest megacity. Retrieved from http://www.theguardian.com/cities/2015/jan/28/ china-pearl-river-delta-overtake-tokyo-world-largest-megacity-urban-area, The Gau- rdian. Mitsubishi Electric (2015). Moving Walks. Mitsubishi HI (2010). “crystal mover” automated people mover for hartsﬁeld-jackson atlanta international airport. Mitsubishi Heavy Industries Technical Review, 47. M.J.J. Scheepers, M., Seebregts, A., Lako, p., Blom, F., and van Gemert, F. (2007). Fact ﬁnding nuclear energy. Moore, M. and Foster, P. (2011). China to create largest mega city in the world with 42 million people. Retrieved from http://www.telegraph.co.uk/news/worldnews/asia/china/8278315/ China-to-create-largest-mega-city-in-the-world-with-42-million-people.html, The Tele- graph. Shanghai/Beijing.

Morar, S. (2010). Evolution of communication based train control worldwide. In Railway Signalling and Control Systems (RSCS 2010), IET Professional Development Course on, pages 281–289. IET.

Morari, M. and Lee, J. H. (1999). Model predictive control: past, present and future. Computers & Chemical Engineering, 23(4):667–682.

NACO (n.d.). About NACO. Retrieved from http://www.naco.nl/english/company-info/ about-naco.html.

99 BIBLIOGRAPHY

Negenborn, R. and Hellendoorn, H. (2010). Intelligence in transportation infrastructures via model-based predictive control. pages 3–24. NEI (2015). Nuclear Energy in China. Nuclear Energy Institute.

OTIS (2015). About escalators.

Pahl, G. and Beitz, W. (2013). Engineering design: a systematic approach. Springer Science & Business Media. Pegden, C. D. (1983). Introduction to siman. In Proceedings of the 15th conference on Winter simulation- Volume 1, pages 231–241. IEEE Press. Poma Leitner (n.d.). Minimetro. Retrieved from http://epaper.holzweg.tv/title/MiniMetro+ brochure+/pdf/http://en.minimetro.com/content/download/13733/514403/version/7/file/ MiniMetro-en.pdf. Raney, S. and Young, S. E. (2005). Morgantown people mover–updated description. young, 785:296– 6959. Reichmuth, J., Berster, P., and Gelhausen, M. C. (2011). Airport capacity constraints: future avenues for growth of global traﬃc. CEAS Aeronautical Journal, 2(1-4):21–34.

Reynolds-Feighan, A. J. (1998). The impact of us airline deregulation on airport traﬃc patterns. Geo- graphical Analysis, 30(3):234–253.

Royal HaskoningDHV (2015). NACO wins a multi-million euro tender to design the airside infrastructure for Mexico City’s new airport. Re- trieved from http://www.royalhaskoningdhv.com/en-gb/news-room/news/ 20150116pr-naco-wins-tender-design-airside-infrastructure-mexico-citys-new-airport/ 4137. Salling, K. B. and Banister, D. (2009). Assessment of large transport infrastructure projects: the cba-dk model. Transportation Research Part A: Policy and Practice, 43(9):800–813.

Saltelli, A. and Annoni, P. (2010). How to avoid a perfunctory sensitivity analysis. Environmental Modelling & Software, 25(12):1508–1517. Saltelli, A., Ratto, M., Andres, T., Campolongo, F., Cariboni, J., Gatelli, D., Saisana, M., and Tarantola, S. (2008). Global sensitivity analysis: the primer. John Wiley & Sons.

Schifers, C. and Hans, G. (2000). Ieee standard for communications-based train control (cbtc) performance and functional requirements. In Vehicular Technology Conference Proceedings, VTC, pages 1581–1585. Schiphol Group (2015a). Facts & Figures 2014.

Schiphol Group (2015b). Traﬃc Review 2015. Retrieved from http://trafficreview2014. schipholmagazines.nl/passengers.html#transporttransferodtransitoloadfactora. Shenzhen Government (n.d.). General Info. Retrieved from http://english.sz.gov.cn/gi/. Sheu, J.-W. and Lin, W.-S. (2012). Adaptive optimal control for designing automatic train regulation for metro line. Control Systems Technology, IEEE Transactions on, 20(5):1319–1327.

Siahvashi, A. and Moaveni, B. (2010). Automatic train control based on the multi-agent control of cooperative systems. The Journal of Mathematics and Computer Science, 1(4):247–257.

Siemens (2014). Airval: Airport passenger fully automated rubber-tire transport solutions.

Siemens (n.d.). Automated People Mover. Retrieved from http://www.mobility.siemens. com/mobility/global/en/urban-mobility/rail-solutions/automated-people-mover/Pages/ automated-people-mover.aspx. Sloboda, B. W. (2009). Transportation statistics. J. Ross Publishing.

Thales (2015). SelTrac R CBTC: Communications-Based Train Control For Urban Rail. Thales.

Verschuren, P. J. M. and Doorewaard, H. (2007). Het ontwerpen van een onderzoek. Lemma. Vickerman, R. (2007). Cost-beneﬁt analysis and large-scale infrastructure projects: state of the art and challenges. Environment and Planning B Planning and Design, 34(4):598.

Vreeker, R., Nijkamp, P., and Ter Welle, C. (2002). A multicriteria decision support methodology for evaluating airport expansion plans. Transportation Research Part D: Transport and Environment, 7(1):27–47.

100 Graduation Thesis

Wang, J. (2013). The economic impact of special economic zones: Evidence from chinese municipalities. Journal of Development Economics, 101:133–147. Wang, P. and Pitﬁeld, D. (1999). The derivation and analysis of the passenger peak hour: an empirical application to brazil. Journal of Air Transport Management, 5(3):135–141.

Wang, P., Zhou, Y., Chen, J., Wang, Y., and Wu, P. (2010). Real-time train scheduling and control based on model predictive control. In Intelligent Systems (GCIS), 2010 Second WRI Global Congress on, volume 3, pages 47–50. IEEE. Want China Times (2010). The price of power: China’s confusing electricity rates. Retrieved from http: //www.wantchinatimes.com/news-subclass-cnt.aspx?id=20130515000080&cid=1102, Beijing. Wijk, M. v. (2007). Airports as cityports in the city-region: spatial-economic and institutional positions and institutional learning in randstad-schiphol (ams), frankfurt rhein-main (fra), tokyo haneda (hnd) and narita (nrt). Netherlands Geographical Studies, 353.

Womack, J. P. and Jones, D. T. (2010). Lean thinking: banish waste and create wealth in your corporation. Simon and Schuster. Wong, M. (2013). China opens futuristic airport terminal. Retrieved from http://travel.cnn.com/ most-futuristic-airport-terminal-china-763891. Woodman, O. and Harle, R. (2008). Pedestrian localisation for indoor environments. In Proceedings of the 10th international conference on Ubiquitous computing, pages 114–123. ACM. Young, S. (1999). Evaluation of pedestrian walking speeds in airport terminals. Transportation Research Record: Journal of the Transportation Research Board, (1674):20–26.

Zhu, J. (1994). Changing land policy and its impact on local growth: the experience of the shenzhen special economic zone, china, in the 1980s. Urban Studies, 31(10):1611–1623.

101 Adapting Automated People Mover Capacity to Real-Time Demand via Model Based Predictive Control