Transportation Research Part C 85 (2017) 548–572
Contents lists available at ScienceDirect
Transportation Research Part C
journal homepage: www.elsevier.com/locate/trc
Review
Research and development of automatic train operation for
railway transportation systems: A survey
Jiateng Yin, Tao Tang ⇑, Lixing Yang, Jing Xun, Yeran Huang, Ziyou Gao
State Key Laboratory of Traffic Control and Safety, Beijing Jiaotong University, Beijing 100044, China
a r t i c l e
i n f o
Article history:
Received 27 February 2017
Received in revised form 8 September 2017
Accepted 12 September 2017
Keywords:
Automatic train operation
Railway systems
Railway traffic control
Review paper
Speed profile optimization
Train speed control
a b s t r a c t
With the rapid development of communication, control and computer technologies in the
last several decades, automatic train operation (ATO), for which the driver no longer has to
cautiously operate the control handle, is emerging in many urban rail systems to replace
traditional manual driving in recent years. As technology advances in railway systems,
one theoretically challenging and practically significant problem is how to use the ATO system to make the current railway network more efficient with higher carrying capacity,
lower cost and improved quality of service by optimized railway traffic management
and train operation. In this review, we focus on this emerging technology of automatic
train operation (ATO) for its theoretical development and practical implementations.
Specifically, this study first presents the background of ATO technology in railways, which
involves the detailed description of its development and implementation in urban metro
systems, fundamental features and basic structure of a typical ATO system. Then, we present a comprehensive literature review in this area, in which the current studies are generally classified into three main aspects, i.e., train operation modeling techniques, train
trajectory optimization and train speed control methods. Finally, the emerging requirements for current ATO systems and the most promising research directions in this area
in the future are discussed explicitly, including (i) the practical implementation of ATO
in main line and high-speed railways, (ii) the cooperative train operation methods for
energy-saving issues and (iii) the integration of railway traffic control with advanced
ATO technology.
Ó 2017 Elsevier Ltd. All rights reserved.
Contents
1.
2.
3.
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The basics of ATO technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.
Railway traffic control and train operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.
ATO: State-of-the-art technology for train operation in urban rail systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.1.
Automatic train control systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.2.
From manual driving to automatic train operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.3.
Train speed control by ATO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Methods of automatic train operation: literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.
Classification of train operation models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
⇑ Corresponding author.
E-mail addresses: jiatyin@bjtu.edu.cn (J. Yin), ttang@bjtu.edu.cn (T. Tang).
https://doi.org/10.1016/j.trc.2017.09.009
0968-090X/Ó 2017 Elsevier Ltd. All rights reserved.
549
550
550
553
553
553
555
558
558
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
3.2.
4.
5.
Speed profile optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.1.
Speed profile optimization: mathematical formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.2.
Speed profile optimization: literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.
Train speed control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.1.
Proportional integral derivative controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.2.
Intelligent control methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.3.
Adaptive control methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Further discussions: opportunities and challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.1.
Implementation of ATO for main line railways and high-speed railways . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.
Cooperative train operation for energy savings. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.
Integration of railway traffic control and automatic train operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
549
559
559
560
563
563
564
565
566
566
567
568
569
569
569
1. Introduction
Railway transportation, involving the main line railway, urban rail transit (or metro, subway, underground system, etc) and
the new high-speed railway (HSR), is an efficient means of public transport mode by way of vehicles running on railway
tracks. In the past decades, rail transportation systems have changed dramatically in the technology level, total length, travel
speed, service qualities, etc. The current rail transport system is playing an essential role in driving sustainable economic
growth, providing access for passengers into and between the major economic centres and fulfilling a vital position in the
supply chain (TSAG, 2010). In particular, due to the environmental and traffic congestion problems in large cities, urban rail
transport is being stressed as an ideally green and convenient transport mode in large cities, such as Beijing, New York,
Tokyo, London, etc (Wang et al., 2015). According to the published data by The International Association of Public Transport
(UITP), there are more than 148 cities around the world that have a metro system until late 2014, adding up to nearly 540
metro lines, 9,000 stations and 11,000 km of line infrastructure (UITP, 2014). As one of the busiest metro system in the world,
Beijing Subway has operated 19 lines with a total length of 574 km, as shown in Fig. 1(a). According to the published data by
Beijing Subway, it carried over 10:28 million passengers to their destinations during a single day in April 21, 2017 (see in
Fig. 1(b)), which greatly releases the public transport pressures for the alleviation of urban traffic congestions.
How to control the running of trains in order to achieve safe and efficient operation for a railway system is a long-lasting
issue dating back to the birth of rail transportation. For traditional railway systems, this is usually realized through (i) a timetable and rolling stock plan that are formulated by an extensive planning process (Cacchiani et al., 2014), which is often
made long time ahead of the real-time operations, and (ii) real-time train operation by drivers with the help of fixed signal
devices that can deliver train movement authorities, routes, etc, to the on-board drivers (Clark, 2012). Nevertheless, this
manual labor based train operation framework has a lot of drawbacks with the rising concerns on increasing transport
demand with limited railway infrastructures. In particular, manual driving is generally based on training and experience,
which is short of rigorous computation and systematic consideration, and thus, it is difficult to guarantee safety, service
quality (e.g., carrying capacity, punctuality, station stopping accuracy) and operational costs (e.g., energy consumption,
infrastructure occupation). And this issue is especially severe in urban rail systems, where the passenger demand is extremely high and train departure headway is very short.
With the development of communication, control and computer technologies in the last several decades, automatic train
operation (ATO) is considered as an emerging technology to replace traditional manual driving in many urban rail systems
(Dong et al., 2010; Miyatake and Ko, 2010; Yasunobu and Miyamoto, 1985). Typically, ATO aims to improve the efficiency of
railway traffic operations by automatically making real-time decisions of the optimized train accelerating, coasting and braking commands. With the increasingly serious environmental problems and energy issues, ATO is also widely recognized to be
a very promising approach by optimized train control decisions, to reduce the energy consumption and carbon emissions
while delivering an improved quality of services (TSAG, 2010). Currently, this important technology has been widely applied
to many new established urban rail transit lines, for example the Paris Métro, London Underground, Beijing Subway and
Tokyo metro, and has shown its great advantages in reducing the manual labor, enhancing the transport capacity of the
infrastructure and improving the service quality for passengers (higher punctuality and more precise train station stopping,
etc.). In Beijing Subway, all the lines (except for Line 5, Line 13 and Batong Line) are operated by ATO systems under normal
circumstances. In particular, since Line 5 in Beijing Subway is still operated by drivers, the train arrival punctuality is only
about 99.5% in the first quarter of 2017. Meanwhile, with ATO systems, the train departure headway of Line 1 in Beijing Subway can be as short as nearly 2 min and the punctuality is kept as high as 99.99%, which illustrates the improvement by
implementing ATO systems in urban rail systems.
Meanwhile, the real-world implementations of ATO technologies are still limited to the urban rail lines. To apply ATO system in main line railway or HSRs, a lot of theoretical studies and field tests are still being conducted to verify its feasibility
and effectiveness. At the same time, a large number of mathematical optimization models and ATO methods have also been
550
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
(a) Beijing Subway network
(b) Passenger amount of Beijing Subway in
April 21, 2017
Fig. 1. Beijing Subway and its passenger volume in a weekday of 2017.
studied for both urban rail systems, mainline rails and high-speed railways, in order to realize safe, environmentally friendly
and efficient train operation. In this paper we aim to present an overview of these recent methodologies in train operation
problems, and we especially intend to fill the gap between the technical elements of ATO which are documented in technical
reports, and its potential applications in being more environmentally friendly, efficient and convenient for the future development of rail transportation systems.
This paper is structured as follows. Section 2 presents some background information on the ATO technology in railway
systems, involving (i) the relationship between railway traffic control and train operation, and (ii) the development, definition and implementations of ATO in urban transit systems. Section 3 reviews the works that have been conducted over the
past years in this area. In particular, the topics of this section are classified by three main issues in academic research areas of
ATO, involving the research in modeling techniques of train operation, train speed trajectory optimization and train control
methods for tracking the trajectories. Further discussions that synthesize the main features of current works and future
research directions are presented in Section 4. This paper is concluded in Section 5.
2. The basics of ATO technologies
This section first describes some background information of ATO, and then focuses on the general definition, development
and basic functions of ATO in urban rail systems.
2.1. Railway traffic control and train operations
In general, a railway transportation system is formulated on the basis of an extensive planning stage (Cacchiani et al.,
2014), which consists of the planning of a timetable, the rolling stock plans, crew duties, etc. All of these plans are carried
out a long time before the real-time operations and aim to optimize their corresponding objectives. For example, the timetable specifies the conflict-free trips that are to be carried out by the trains in the system, as well as the detailed routes of the
trains through the railway stations (Cordeau et al., 1998). The objectives of railway timetable optimization involve the train
travel time (e.g., Zhou and Zhong, 2005, 2007), total energy consumption (e.g., Huang et al., 2016), transfer waiting time (e.g.,
Wong et al., 2008), or the trade-off among these indicators (Yin et al., 2017; Huang et al., 2017). After the timetable is given,
the railway managers need to assign the available resources, including the rolling stock and crew duties to the trips in this
timetable (Cacchiani et al., 2012).
In daily operations, the timetable and rolling stock plans are always disrupted by various kinds of perturbations (e.g.,
equipment failure, extreme weather) that may cause the infeasibility of the original plan. Therefore, one basic task in daily
railway operation is to reschedule or adjust the train timetable and rolling stock plan with real-time train information and
perturbation estimations, which is called ‘‘railway traffic control” (or rail traffic management). The railway traffic control is
usually considered in a macroscopic view with the objective of minimizing the negative effects of unexpected disturbances
by dispatchers in a railway traffic control center. For example, Fig. 2(a) shows the Beijing Subway Operating Control Center
(OCC),1 which is in charge of all the rail traffic control tasks for the lines in Beijing Subway. With the given scheduled (or
rescheduled) plan, the other significant issue is to execute this plan for all the in-service trains, i.e., moving these trains to their
planned destinations with given time constraints. The goal is to conduct a safe, scheduled and efficient travel for each single
train, which can be viewed as a microscopic level, called ‘‘train operation”. As shown in Fig. 2(b), the train operation in a metro
1
http://news.163.com/08/1227/14/4U66P9H0000120GU.html.
551
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
On-board
train control
equipment
Train-ground
communication
Track-side signal
equipment
Operating control center
(b) Illustration of train operation in railway
(a) Operating control center (OCC) of Beijing Subway
Fig. 2. Railway traffic control and train operations. http://www.tranbbs.com/news/cnnews/news_36442.shtml.
Rail traffic control
Traffic
management
Dispatchers
Outer control loop
Timetable
Train
positions
Rolling
stock plan
Timings
Crew plan
Disturbances,
etc
Rail transport
plan
Real-time
information
Train timetable
rescheduling
Rolling stock
rescheduling
Rail traffic network
Train operation
Train model
dynamics
Timetable
Given train
speed profile
Others
Line infrastructure
data
Local data
Driver
Train on-board
computer
Train driving
command
Inner control loop
Train on-board
sensors
Real-time train
positioning
Fig. 3. Relation between railway traffic control and train operation.
system is realized by the train dispatching orders from OCC through train-ground communication, and finally accomplished by
on-board train control equipment. These two fundamental categories, i.e., the railway traffic control and train operation, as indicated by Luthi (2009) and Rao (2015), can be described explicitly by using the concept of two control loops: an outer control and
an inner control loop (shown in Fig. 3).
Outer control loop: Railway traffic control
In practice, the train timetables or rolling stock plans are inevitable disrupted by some disturbances in practical operations,
causing some delays to the planned train schedule (Yang et al., 2014b; Li et al., 2017). The outer control loop, i.e., railway
traffic control, aims to supervise the status of traffic and infrastructures, detect deviations and conflicts, and develop a
552
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
conflict-free train rescheduling plan (timetable, rolling stock, crew duties, etc. (Cacchiani et al., 2014)) in order to make
support decisions for railway dispatchers to optimize capacity, train arriving punctuality and avoid conflicts with other
trains (Corman and Meng, 2015). As demonstrated in Fig. 3, the input data in this control loop consist of the original railway
transport plan (i.e., a timetable, rolling stock plans, crew duties) and online feedback data that refer to (i) the information of
potential disturbances or disruptions (Corman and Quaglietta, 2015; Meng and Zhou, 2014; Nielsen et al., 2012) and (ii) the
positions, speed (Corman et al., 2011) and timings (Meng and Zhou, 2011) of all the trains in this system (Tornquist, 2007).
Generally, the rail traffic control is accomplished through the railway OCC (see Fig. 2(a)), in which the train dispatchers
receive the online input data and takes real-time control actions to change the railway traffic to a desirable state. Here,
the control actions refer to the rescheduled plan that is typically related to the choices of train arrival and departure times
at each station, allowable maximum speed at each segment, and rerouting of trains. We can refer to two important reviews
by (Cacchiani et al., 2014) and (Corman and Meng, 2015) for more details on this point.
Inner control loop: train operation
With the scheduled (or rescheduled) plan, the train operation (i.e., the inner control loop) focuses on the safe and efficient
train movements on each block on a microscopic level by the determination of all the basic train control commands (i.e.,
accelerating, cruising, coasting and braking). In particular, the inner control loop can be regarded as an ‘‘executor” for the
outer control loop. In other words, it connects the fixed signaling infrastructure with the moving trains (Zimmermann and
Hommel, 2005) to execute the generated scheduled (or rescheduled) plan for each single train. For example, in a traditional railway system, the train operations are usually undertaken by drivers in individual trains. Each driver in a train
directly interacts with wayside signal devices, receives track occupation information and movement authorities, and then
switches the control handle of train to make the train move or stop. As can be seen in Fig. 3, the input data for each train in
the inner control loop are the scheduled (or rescheduled) plans and real-time train information (e.g., train speed, distance
to the next stop, time, train dynamics, line infrastructure information). In this control loop, the control actions refer to the
specific control commands taken on a train for its movement or stop. For a typical electric train, the control commands
correspond to the train traction and braking forces that are outputted by electric motors and make the train accelerate,
cruise, coast or brake. The primary goal of train operation is to ensure the safe movements of trains on a given network of
tracks. Some other evaluating indicators involve punctuality, riding comfort (e.g., Chang and Sim, 1997) and energy consumption (e.g., Howlett and Pudney, 1995; Liu and Golovitcher, 2003), which will be specified in the following section.
It is worth to mention that these two control loops are closely related to each other, and both of them are significantly complimentary for the safe and efficient operation of a railway system. The railway traffic control can centralize and supervise all
the train’s status in a rail line or rail network, and generate the real-time rescheduled train movement plans for these trains.
Based on the plan, the train operation is to make speed control commands under restrictions of moving authority and limited
speed for all the moving trains to execute this plan. Therefore, the typical evaluation indexes of railway traffic management
refer to delays, involving the average or maximum delays of trains (Corman et al., 2011; Pellegrini et al., 2012), travel time
of passengers (Tornquist, 2007; Gao et al., 2016), minimal time to recover from disturbances (Meng and Zhou, 2011), and transport capacity, etc. These indicators are generally defined on the network level rather than individual trains. On the other hand,
train operation focuses on providing the specific microscopic operations for each in-service train to guarantee the safety of railway traffic systems, and moreover to follow the pre-specified (or rescheduled) operational plans (Li et al., 2015). If the trains
are operated with poor quality (e.g., bad punctuality), they will in turn affect the performance of the entire railway traffic.
In the past, conventional trains were normally controlled and operated manually by drivers, who operate trains based on
track-side interlocking and blocking in conjunction with various train signalling and supervisory devices (Dong et al., 2010;
Matsumoto, 2005). This kind of train control system is mainly based on the human experience and skills of engineers. For
example, in order to ensure that the moving trains are separated, the authority to occupy a block is conveyed to train drivers
by using wayside signals that are manually controlled by human operators. The train drivers need to keep the forward signals in sight and then make appropriate train control commands. In essence, the wayside logic and dispatchers only know
which circuit the train is running on, and prevent other trains from entering the same track circuits. Moreover, since this
process lacks strict supervision and rigorous computation, it is usually vulnerable with respect to external factors (e.g., psychological condition of driver, extreme weather that may affect the sights of drivers) leading to insecurity and inefficiency.
In order to make full use of railway infrastructure to reduce the train following interval and increase transport capacity, a
communication based train control (CBTC) system was developed and firstly implemented in urban metro systems to improve
the capacity, safety and flexibility of rail transportation. Different from traditional train control systems, the CBTC system is a
continuous, automatic train control system that utilizes (i) high-resolution train positioning devices (independent of track circuits); (ii) continuous, high-capacity, bidirectional train-to-wayside data communications; (iii) and wayside processors that
are capable of implementing Automatic Train Control (ATC) that involves Automatic Train Protection (ATP), automatic train
operation (ATO) and Automatic Train Supervision (ATS) functions 2 (Greenway and Sheldon, 1974; IEEE Standard, 2004;
Moreno et al., 2015). Here, ATO is responsible for all the traction and braking controls of moving trains to guarantee punctuality
(according to the schedule), comfort and energy-efficiency during normal operations (Dong et al., 2010). Hence, ATO is a key to,
2
ATS is also the abbreviation of Automatic Train Stop in North America. To avoid potential confliction, the term ‘‘ATS” represents Automatic Train
Supervision throughout this study.
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
553
and has a direct impact on the development of train operation technology in railways. Currently, many researchers are focusing
on the train control methods to improve the performance of ATO (seeAlbrecht et al., 2016a,b; Chang and Sim, 1997; Faieghi et al.,
2014; Gao et al., 2013, etc), and meanwhile, the real-world implementations of ATO in urban rail lines, main line railways and
HSRs become more and more promising from industrial community (seeBienfait et al., 2012; IEC 62290-1, 2004; UITP, 2011;
TSAG, 2010; ERTMS, 2010; UIC, 2000a, etc). In the following content, we first present the state-of-the-art ATO technologies
for implementations in urban rail transit systems. Then, we conduct a comprehensive literature review for the existing train
operation methods, and we further discuss the foremost research directions in the field of ATO technology.
2.2. ATO: State-of-the-art technology for train operation in urban rail systems
2.2.1. Automatic train control systems
From the earliest days of railways, the safe movement of a train is accomplished by manual labor with the help of wayside
visual signals. In this case, the train is totally controlled by a driver in a cabin, who receives track state information from
wayside equipment (involving on-track balises, track circuits, etc (Midya and Thottappillil, 2008)) in real-time, and continuously takes corresponding train control commands in accordance with the train dispatching orders from a railway traffic
control center. Nevertheless, with the development of railway systems, this train control strategy has two main issues.
On the one hand, using visual signals brings the possibility of the driver overlooking, misreading or just ignoring the signal
giving him authority to proceed (Clark, 2012). If the driver fails to see or obey the signals, it could cause serious disasters. On
the other hand, manual driving is mainly based on the experiences and professional judgements of drivers, which is in particular lack of rigorous computation and optimization. Therefore, in many cases, for instance a heavy congested metro line
with short headway, manual train operations cannot achieve good performances.
As a mainline railway network, the Great Western Railway (GWR) first announced to develop an ‘‘automatic train control
system (ATC)”, which uses electromechanical equipments to automatically warn the driver by a horn in the cabin when
approaching a wayside signal. In addition, if the driver did not acknowledge the horn within 2-3 s, the train will automatically brakes to a complete stop. Different from this so-called ‘‘ATC” system that is more likely to be an automatic train warning system, the modern ATC system has become an integrated signalling system that combines railway train control,
supervision and management to help drivers (or completely substituting divers) control the train movements automatically
to guarantee the safe and efficient movement of railway trains (Greenway and Sheldon, 1974). In particular, for automated
urban metro systems, ATC contains three subsystems, i.e., automatic train protection (ATP), automatic train operation (ATO)
and automatic train supervision (ATS) (Tang and Huang, 2003), as illustrated in Fig. 4.
ATS system is responsible for monitoring the train movement to ensure that the trains conform to an intended schedule
and traffic pattern. In particular, ATS system connects train dispatching and train operation in urban rail systems that
helps to avoid or reduce damage resulting from system abnormalities and equipment malfunctions by performing the
following tasks: supervision of train status, automatic routing selection, automatic schedule creation, automatic operations logging, statistics and report generation and automatic system status monitoring.
ATP system is a fail-safe (vital) system, which is responsible for the safe movement of individual trains. ATP imposes
speed limits on the moving trains, not only to maintain a safe operating distance between the trains, but also to comply
with safety and speed requirements. Once the train exceeds the speed limit, ATP will automatically execute braking (or
emergency braking) order to stop the train to keep safety.
ATO system performs the on-board functions instead of a train driver to ensure a smooth acceleration of the train to the
running speed, speed regulation and stopping the train at the destination platform precisely. In normal situations, ATO is
responsible for all the train traction and braking control commands, and thus, it is a key to the operational efficiency and
profitability of train operation systems (Dong et al., 2010).
These three subsystems of ATC work together to ensure the safe and efficient running of trains. In particular, ATS gives
train routing and schedule adherence instructions to ATO according to the train state and schedule. ATO will gather the relevant information, such as train speed, programmed stop and dwelling time, and then decides the brake or accelerating rates
of the train. Meanwhile, ATP keeps monitoring the real-time train running status, and corrects the train operation commands
or triggers emergency brake if necessary. In other words, ATP guarantees the safe train separation and overspeed protection,
ATS keeps the trains in the predetermined schedule, and ATO is concerned on the train operational strategies that are directly
related to the train operation efficiency.
2.2.2. From manual driving to automatic train operation
Different from manual driving where the train is totally controlled by a driver (see in Fig. 5a), ATO is a system that uses
integrated logic between the wayside system where the train speeds are regulated without intervention from the operator
(Technical report, 2013). From the 1950s, the automation of train operation was already experimented and implemented on
the 42nd Street Shuttle from Grand Central to Time Square in New York City Transit.3 Meanwhile, the ATO system developed
3
https://en.wikipedia.org/wiki/Signaling_of_the_New_York_City_Subway.
554
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
Fig. 4. Illustration of ATC structure in metro systems.
(a) Train operation with manual labor
(b) Train operation with ATO
Fig. 5. Two modes of train operation in railways.
by Victoria line of London underground can automatically control the speed of travelling trains with the supervision of on-board
staff.4 This innovation was subsequently proved to be very effective in daily operations. It not only reduces manual labor costs,
but also improves operational efficiency. Compared with traditional manual driving, the ATO system in Victoria line enables to
dispatch a total number of 33 in-service trains every hours instead of 27. Also, this technology increases the line capacity overall
by 21%, which is equivalent to transporting a total of extra 10,000 passengers per hour. After that, ATO has been successful
developed in many new metro lines throughout the world. In particular, the San Francisco BART (Bay Area Rapid Transit)
was one of the first transit systems in the United States that developed the ATC system, which realizes the automation of train
operation under normal circumstances, as well as automatic routing and dispatching of trains with supervision of BART’s OCC.5
Essentially, the basic idea of ATO is to use computer programming and control techniques, helping drivers (or completely
substituting divers) to automatically control the train movements with the supervision of ATP and ATS (see in Fig. 5b). With
the support of relevant infrastructure (i.e., signalling system implemented at the wayside and on-board ATO computer) in
urban rail systems, the ATO function features can generally cover the following aspects.
The most important function of ATO system is to realize automatic train driving between two stations. In the train operation process, ATO receives real-time information, including speed limits, train speed and position information from ATP,
and movement authority, travel direction and destination information from ATS. Typically, the train operation is currently
realized through two ways. (1) Manual driving supported by a driver advisory system (DAS), which can provide a recommended speed for on-board drivers in order to help drivers with better driving strategies. (2) Semi-automated or fully
automated mode by ATO system in urban rail lines with higher automation level, which can partially (or completely)
replace manual driving. In these urban rail lines, ATO system first generates a recommended train speed profile before
4
5
https://en.wikipedia.org/wiki/Victoria_line.
https://en.wikipedia.org/wiki/Bay_Area_Rapid_Transit.
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
555
train departure, by which the train can arrive at the next station on time. After the train departs from the station, the
speed controller of ATO will automatically adjust the control commands for train acceleration, coasting, cruising or braking through a feedback control loop.
The other important function of ATO is automatic train station stopping. When the train enters the station stopping area
that is detected by the connect between wayside equipments (e.g., a parking beacon, balise), ATO will switch to the train
station stopping mode. In this mode, ATO adjusts the train braking rates dynamically based on the train speed and distance to the stopping point in order to stop the train at the station precisely.
Through the system connection with wayside equipments, e.g., balises, platform screen doors (PSDs), which enable the
information exchange between the train and the platform in metro stations, the ATO system for passenger transportation
is in charge of the opening and closing of train doors automatically when the train dwells or departs from each station.
Besides, the additional functions of ATO involve train automatic reverse at terminals, positive train identification, etc.
Remark 2.1. It is worth noting that the train operation with DAS is essentially realized by train drivers manually, which does
not belong to ATO systems. Nevertheless, the studies on DAS mainly focus on the optimization of speed profile with real-time
computational requirements that actually share the same features for ATO systems in urban transit lines. For example, the
commercial Energymiser Driver Advice System that was invented by SCG (Scheduling and Control Group at the University of
South Australia) and TTG Transportation Technology (TTG) (Albrecht et al., 2016a,b) can measure the real-time location and
speed of the train using global positioning system, and calculate the optimal train driving profiles instantly for the recommendation of drivers. Therefore, this study will also review these important studies on train trajectory optimization for DAS
in Section 3.
In addition, to clarify the automated level of train operation systems, the international standard IEC 62290-1 2014 defines
five grades of automation (i.e., GoA0 to GoA4) arising from apportioning responsibility for given basic functions of train operation between operation staff and the system (IEC 62290-1, 2004). According to this standard, GoA0 and GoA1 are essentially
non-automated train operation levels that need the drivers in the cabin to operate the trains manually. In GoA2, acceleration
and braking are automated while the driver in the cabin is responsible for the safe departure of the train and door control.
GoA3 is driverless train operation in which there is no driver in the front cabin of the train. Instead, there is only a member of
operation staff for the safe departure of train. Most of existing ATO systems generally achieve GoA2 or GoA3 (UITP, 2011).
The highest level of train operation automation is the unattended train operation (i.e., UTO or GoA4), in which there is completely no driver or operation staff, and the trains are operated fully automatically. In this sense, we can see that the main
feature of an ATO system that differs from manual driving is automatically delivering accelerating, cruising, coasting and
braking commands to keep the train running on each segment (as shown in Fig. 6a) (Dong et al., 2010; Zhou et al., 2017).
In particular, ATO also aims to keep the train always in the ‘‘optimal” running state, improve the riding comfort of onboard passengers, guarantee the punctuality and reduce energy consumption as much as possible. And we specify the commonly used indicators to evaluate the ATO performance, described as follows.
Punctuality: Since the railways are operated rigorously based on a pre-specified schedule, the most important aim of ATO
is to guarantee that the train arrives at each station punctually according to the running time determined by the train
timetable (see in Fig. 6b).
Riding comfort: Ensuring the comfort of passengers directly relates to the quality of service in urban rail systems. As indicated in many railway standards, (e.g., BSI EN 12299, 12299:2009), the comfort of passengers in a railway train is influenced by the vibrations and motions of the vehicle. For example, the average change of acceleration and deceleration (or
jerk) can be used to evaluate the riding comfort of a train operation process.
Energy efficiency: Energy-efficient train operation is critical as the rising of energy prices and environmental concerns. In
particular, the train operations account for about 80% in the whole energy consumption of metro systems (Yang et al.,
2016). This is one of the core considerations in developing an ATO system, since optimized train driving strategies could
reduce as much as 20% energy consumption according to the research byAlbrecht et al. (2016a,b).
Station stopping accuracy:Since many new established metro stations are installed with PSDs, station stopping accuracy
has become another important character for the ATO system design.
In the following part, we will particularly stress how these performance indicators are considered by researchers to
design more effective ATO methods in railway systems.
2.2.3. Train speed control by ATO
As we described above, the key function of an ATO system is ‘‘train speed control” (i.e., train driving), which is defined to
be the most important function for train operation in railways. On one hand, its main aim is to guarantee that the train
arrives at every station punctually according to the schedule. On the other hand, ATO enables the speed control of trains
as they travel along the track so that the jerk rates are within passenger comfort limits and the train speed is below the speed
limit imposed by the safety system (IEC 62290-1, 2004). On the basis of punctuality and comfort constraints, as indicated by
Chang and Sim (1997),Howlett and Pudney (1995), and Liu and Golovitcher (2003), different train speed control strategies,
556
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
si
s tm
st
st
t
(a) A speed profile for train operations between
two stations
t
v tm
vt
vt
t
t
t
tm
(b) State transition on a discrete train operation model
Fig. 6. Typical train operation process between two stations.
Fig. 7. Train speed regulation by ATO between two stations.
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
557
Fig. 8. Diagram of train speed control in an ATO system.
(e.g., accelerating distance, coasting and braking points) will directly lead to different values of energy-consumption, punctuality and riding comfort. Also, it is proved the existence of a unique optimal train operation strategy that minimizes the
energy-consumption for the train movement at each segment (Albrecht et al., 2016b). For example, the recent works by
González-Gil et al. (2014) show the reduction of energy-consumption by 5%–15% through implementing the optimized train
speed control with ATO. Note that the less energy-consumption will directly reduce the carbon emissions of railway systems,
making the railway transportation more environmentally friendly. Therefore, the speed controls of trains along the tracks
have direct impacts on the service quality, operational efficiency of railway systems and environment simultaneously.
Here, we use Fig. 7 to demonstrate an illustrative railway segment between stations i and i þ 1, and the train, which is
operated by an ATO system, is prepared to depart from station i. Before the train departs, the on-board computer will first
receive the confirmation messages, involving status of train doors and PSDs, status of ATO and ATP equipment, and travel
information of the next segment (distance to the next station, line gradients, speed limits, arrival time, etc). Then, the onboard computer can automatically generate a recommended speed profile under the speed limits, as shown in Fig. 7a. Usually, there are more than one speed profile that satisfies the journey time, and generating an optimal profile with low energy
consumption and improved riding comfort is a critical problem in ATO systems(Wang et al., 2013). This is a hot research
topic since 1990s, and many speed profile optimization methods are proposed based on Pontryagin maximum principle
(Howlett and Pudney, 1995; Albrecht et al., 2016a,b), dynamic programming (Ko et al., 2004), tabu search (Liu et al.,
2015), etc.
After confirming that all the departure procedures are correct, the driver presses the button in the cabin, which automatically starts the traction motor for train departure. In train driving process, as shown in Fig. 8, ATO system receives the realtime feedback data (i.e., train position x and speed v) and line information (e.g., speed limits, gradients) by on-board speed
sensors (e.g., radars, acceleration sensors) and wayside positioning devices (e.g., balises). The speed controller of ATO is
essentially based on control algorithms that are embedded in the on-board computer. The speed controller or driver com^ in position x to determine the control command u, so that
pares the feedback information v with the recommended speed v
the train can precisely track the recommended speed profile as much as possible. For example, if the train speed is much
lower than the recommended speed, the train should accelerate to increase the speed. In contrast, the train will decrease
the speed by braking or coasting. The implemented speed tracking methods involve PID (proportional integral derivative)
control, predictive fuzzy control (Yasunobu and Miyamoto, 1985; Yasunobu et al., 1983), etc. Afterwards, the control command u drives the train motors to generate traction force F þ or braking force B to keep the train speed tracking errors. Likewise, the train states are measured, transmitted and updated in real-time in this control loop until the train arrives at the
next station. The station stopping is controlled in a similar way for ATO system, which first generates a braking curve
and then employs the speed controller to track this curve (see in Fig. 7b). The train stop control methods generally involve
fuzzy control (Yasunobu et al., 1983), iterative learning control (Hou et al., 2011), etc.
Although the real-world implementations of ATO systems in many urban metro systems have already demonstrated the
effectiveness of ATO technologies, currently, there are basically two intractable issues that lead to a large amount of studies
for the further performance improvement of ATO systems. On one hand, the optimization of recommended speed profile is
inherently a complex problem with multiple objectives and constraints (Chang and Sim, 1997; Liu and Golovitcher, 2003).
For meeting the practical operation environments, a series of factors, including gradients, speed limits, track curvature, traction efficiency and regenerative energy, should be further considered in specifying the speed profiles (Li and Lo, 2014), which
makes the speed optimization problem more difficult. One the other hand, the control methods for train speed tracking are
subject to a variety of influencing factors, such as input nonlinearities, actuator failures and parameter uncertainties in the
train dynamic model, variable gradients and aerodynamic resistances(Song and Song, 2011; Gao et al., 2013; Faieghi et al.,
2014). For high-speed trains, the train speed tracking is even more difficult since the dynamic models of a high-speed train
(HST) is much more complex than metro trains for its extremely high-speed and complex external environment. Hence,
558
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
many researchers have proposed different kinds of control methods to stress the speed tracking problem for HSTs with complex or unknown train dynamic models (Gao et al., 2015; Li et al., 2014a,b). In the following content of Section 3, we focus on
the review of recent research works in three topics: (i) train operation models, (ii) speed profile optimization and (iii) control
methods for speed tracking.
3. Methods of automatic train operation: literature review
3.1. Classification of train operation models
Since the topic of this review is subject to the operation of trains, we first present a review of typically implemented train
operation models, which specify the dynamic train control process. In general, the existing train operation models for ATO
can be divided into two classes, i.e., single-point train control models and multi-point train control models.
The single-point train control model is the most commonly used model in solving train operation problems (e.g., Albrecht
et al., 2016b; Liu and Golovitcher, 2003; Su et al., 2013, 2015; Wang et al., 2013). For this kind of train control models, a train
that consists of multiple vehicles (e.g., locomotives and carriages) is simplified as a single-point-mass object and thus, its
longitudinal motion can be approximately characterized by a Newton equation. If we consider a train with continuous control rate (i.e., the train can output any continuous value within limits), the typical formulation of train movement on a segment can be given as
Mv_ ðtÞ ¼ FðtÞ BðtÞ wðv Þ gðxÞ;
ð1Þ
_
xðtÞ
¼ v;
ð2Þ
where M is the mass of the train, x; v , and t respectively represent the train position, speed and time, FðtÞ and BðtÞ denote the
train traction force and train braking force, wðv Þ ¼ Mðc0 þ c1 v þ c2 v 2 Þ is the Davis formula that represents the aerodynamic
drag and rolling mechanical resistances Dong et al. (2010), and gðxÞ represents the gradient resistances and curve resistances
with respect to train position x. In this model, the multiple vehicles that make up a train are represented by a single point
with identical position and speed. We note that the above single-point train control model is only a basic model, which can
be transformed into different formulations in different situations. For example, a high-speed train control model needs to
consider the additional traction and braking characteristics with respect to the train running speed. In this situation, the
train control model can be modified into Mv_ ðtÞ ¼ aav FðtÞ abv BðtÞ wðv Þ gðxÞ, in which aav and abv denote the relative accelerating and braking coefficients, respectively. In some other studies (Gao et al., 2013), the resistance parameters, including
c0 ; c1 ; c2 ; gðÞ, are treated as uncertain values, since they are difficult to determine in real-world applications.
This single-point train model can achieve good result in developing ATO methods in urban rail transit systems, where the
running resistances are much smaller than the traction force and braking force (Su et al., 2013). However, this simplified
model is usually infeasible for heavy-haul trains, in which each train is very long and consists of many vehicles and locomotives. It is unpractical to use a single point to represent all positions and speeds of the involved vehicles. On the other hand,
since the couplers that connect adjacent vehicles are not perfectly rigid, the interactive forces (or in-train forces) among the
connected vehicles become an important factor in order to avoid coupler failure for locomotives. Hence, the multi-point train
models proposed inAstolfi and Menini (2002);Yang and Sun (2001); Zhuan and Xia (2008);Faieghi et al. (2014) consider the
interactive impacts of in-train forces and the different position and speed for each vehicles (see in Fig. 9). Here, we consider a
typical train that consists of n vehicles with n 1 couplers connecting the adjacent vehicles, and in a multi-point train modeling, the train is formulated as a nonlinear multi-input multi-output (MIMO) model, i.e.,
8
a
2
>
< m1 €x1 ¼ F 1 B1 kDx1;2 m1 ðc0 þ c1 x_ 1 þ c2 x_ 1 Þ þ R1 ðxÞ
mi €xi ¼ F i Bi kðDxi;iþ1 Dxi1;i Þ mi ðc0 þ c1 x_ i þ c2 x_ 2i Þ þ Rai ðxÞ; i ¼ 2; . . . ; n 1
>
: €
mn xn ¼ F n Bn þ kDxn1;n mn ðc0 þ c1 x_ n þ c2 x_ 2n Þ þ Ran ðxÞ
ð3Þ
where xi ; mi ; F i and Bi denote position, weight, traction force and braking force of the i-th vehicle, respectively, and Dxi;iþ1
represents the relative spring displacement between the neighboring vehicle i and i þ 1. mi ðc0 þ ci x_ i þ c2 x_ 2i Þ denotes the aerodynamic drag and rolling mechanical resistances, and Rai ðxÞ represents the additional resistances with respect to train position, involving gradient resistance, curve resistance, etc. In this model, the behavior of couplers between two adjacent
vehicles is approximately described by a linear spring with stiffness coefficient k, and each vehicle’s speed and position
are modelled specifically. Hence, this multi-point train model can be more effective in characterizing the dynamic behavior
of trains.
In addition, this multi-point train model was validated against experimental data collected on a heavy-haul train with
200 wagons by Chou et al. (2007), and consequently applied to regulating heavy-haul train speed (Chou and Xia, 2007;
Zhuan and Xia, 2008). The applications based on this model in HSRs can be found in Yang and Sun (2001);Faieghi et al.
(2014); Song et al. (2011) and Li et al. (2014a), etc.
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
559
Fig. 9. Illustration for the multi-point train model.
3.2. Speed profile optimization
In this section, we introduce the speed profile optimization for ATO systems. Firstly, we describe the general mathematical formulation for speed profile optimization. Then, a comprehensive review of the solution methodologies is given.
3.2.1. Speed profile optimization: mathematical formulation
As two complementary parts of ATO system, the recommended speed profile and train speed controller cooperatively
achieve the operation requirement of automation and efficiency. The recommended speed profile (or train trajectory) optimization is usually formulated as an optimal control problem based on the train operation model in Eq. (1). The control variables are the accelerating force F and braking force B. The state variables are the train position x and speed v. The objective
function could be the minimization of energy-consumption with a given trip time, riding comfort, switching frequency, or a
trade-off among these indicators. We consider that a train travels from station i to station i þ 1 within a given allowable time
T that is determined by the scheduled or rescheduled timetable. si and siþ1 denote the positions of station i and station i þ 1,
respectively. The energy consumption for the train from station i to i þ 1 can be expressed as
Z
T
E¼
0
½FðtÞ v ðtÞdt:
In addition, the riding comfort of passengers can be considered as a function of the change of the control variable F, i.e., the
total jerk, given by
Z
R¼
0
T
1 dFðtÞ
dt:
M dt The above equation indicates that, reducing the switching frequency and the rate of change u may improve passenger comfort (Wang et al., 2013). Then, the objectives need to be minimized with subject to the following constraints.
Punctuality constraint
This constraint guarantees that the running time by recommended speed profile is coincident with that by the timetable,
expressed as
xð0Þ ¼ si ; v ð0Þ ¼ 0;
xðTÞ ¼ siþ1 ; v ðTÞ ¼ 0:
ð4Þ
ð5Þ
Speed limit constraint
In order to guarantee the safety of train operation, the train speed is prohibited to exceed the speed limit at any time
t 2 ½0; T, i.e.,
v ðtÞ < v i ; 8t 2 ½0; T:
ð6Þ
Here we note that the speed limit v i could also be varying values in each segment in practical railways. And we can refer
to Wang and Goverde (2016) for more details in generating train trajectory with varying speed limit and signaling
constraints.
Control variable constraint
Due to train motor power constraints, the control variables, i.e., traction force FðtÞ and braking force BðtÞ are bounded by
the maximum traction force F max and the maximum braking force Bmax , respectively, expressed by
FðtÞ 6 F max ;
8t 2 ½0; T;
ð7Þ
BðtÞ 6 Bmax ;
8t 2 ½0; T:
ð8Þ
560
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
Fig. 10. Maximum traction force in dependency of train speed (Wang et al., 2013).
In practice, the maximum traction and braking forces F max and Bmax are usually not constant, but a function of the train
velocity. This is mainly caused by the limited train motor power P (Gao et al., 2013), which decreases the maximum traction force when the train speed is very high. An illustrative curve that represents the relation between the maximum traction force and train speed is presented in Fig. 10.
Therefore, it is shown that the speed profile optimization of ATO is essentially a multi-objective optimization problem:
min yðB; FÞ ¼ ðE; RÞ;
s:t: B; F 2 A # Rn ;
Constraints ð4Þ—ð8Þ
ð9Þ
where B; F 2 A # Rn is the decision variable, i.e., the train accelerating and braking orders, A denotes its feasible set. There
may be many speed profiles that satisfy the above constraints between two successive stations. The speed profile optimization problem is to determine a Pareto optimal speed profile from these feasible solutions.
3.2.2. Speed profile optimization: literature
A large amount of literature is available on speed profile optimization models. A detailed classification of the various
mathematical models is shown in Fig. 11 to get a comprehensive overview on this topic.
As we mentioned above, the performance of train operation is directly related to two aspects, i.e., off-line train trajectory
optimization and real-time train speed control, and these two aspects are closely connected to each other. In this sense, the
Fig. 11. Mathematical model classification for train speed profile optimization.
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
561
Table 1
Publications on train trajectory optimization problem.
Publication
Model
Algorithm
Main contribution
Howlett (2000)
1
Pontryagin principle
Prove the optimal speed profile for a contiguous control
Kuhn-Tucker conditions
Energy-efficient driving strategies with discrete control
Khmelnitsky (2000)
Liu and Golovitcher (2003)
Continuous
train model
2
Dieselelectric
locomotives
Continuous
Continuous
Pontryagin principle
Pontryagin principle
Albrecht et al. (2016a,b)
Continuous
Pontryagin principle
Su et al. (2013)
Metro train
Pontryagin principle
Aradi et al. (2013, 2014)
Continuous
Predictive optimization method
Ko et al. (2004)
Dynamic programming (DP)
DP
Miyatake and Matsuda
(2009), Miyatake and Ko
(2010)
Maialen (2012)
Discrete train
model
Discrete train
model
Continuous
Consider the variable gradients and speed limits
Consider the continuous traction force with constant
efficiency
A more general case with steep grades, and the existence of a
unique solution is proved
Minimize the energy consumption for a single train on the
entire routes
A optimization model that aims to reduce energy
consumption while keep computational time
Application of DP to directly search for the optimal speed
profile
Show the relationship between optimization performance
and computational time by using DP
A optimization model that can obtain the energy-efficient
speed profile with energy storage devices
Continuous
A direct numerical approach and DP
Domínguez et al. (2012)
Rodrigo et al. (2013)
Continuous
Continuous
Numerical algorithm
Lagrange multiplier algorithm
Wang et al. (2013)
Continuous
Wang and Goverde (2016)
Continuous
Chang and Sim (1997)
Continuous
Pseudospectral method and a mixed
integer linear programming
formulation
Multiple-phase optimal control
model and pseudospectral method
Genetic algorithm (GA)
Wong and Ho (2004)
Sicre et al. (2012)
GA and heuristics
A simulation model with GA
Lu et al. (2013)
Continuous
Discrete train
model
Discrete
metro train
Continuous
Kim and Chien (2011)
Continuous
Simulated annealing algorithm (SA)
Liu et al. (2015)
Continuous
Tabu search (TS)
Calderaro et al. (2014)
Ke et al. (2009, 2010)
Sequential quadratic programming
(SQP)
Ant colony optimization (ACO) and
MAX-MIN ant system
GA, ACO and DP
Minimization of consumed energy with guaranteed riding
comfort
consider recovered energy of on-board storage devices
Discretization of the original optimal control model in an Rn
optimization problem that is solved by Lagrange multipliers
Propose an efficient MILP reformulation approach to
optimize energy consumption and riding comfort
Train trajectory optimization with signalling constraints
under delay and no-delay situations
Consider trip time, passenger loads and track voltages to
jointly optimize riding comfort, punctuality and energy
consumption
Compare the performance of different search methods
A simulation model that considers manual driving behaviors
for better punctuality and energy saving
Consider the energy-efficient speed profiles both in a fixedblock mode and a moving-block mode
Consider various trip time and compare the performance of
different evolutionary algorithms
Consider track alignment, speed limit, and schedule
adherence
Consider inaccurate speed tracking to achieve energy
consumption reduction
research in train trajectory optimization can be first classified into two categories by considering the practical speed control
process of a train, as shown in Fig. 11. In other words, some existing studies assume that the train can track the optimized
trajectory accurately, while the others consider that the train speed control process is subject to external or uncertain influencing factors and the train cannot track the optimized speed curve accurately. Specifically, when a train is controlled by a
driver with a recommended speed profile, they consider that the trajectory tracking by the driver may be inaccurate with
reaction time and different driving behavior. For example, Sicre et al. (2012), Liu et al. (2015) and Albrecht et al. (2013) considered that the trains are operated by drivers manually with DAS, and the drivers cannot guarantee the completely accurate
tracking. In this case, the speed profile optimization is subject to some additional constraints. For example, the train speed
control handle has only several discrete train accelerating and braking levels that correspond to different accelerating and
braking rates. In such cases, the frequent changes of train accelerating and braking levels should be considered for drivers
to easily handle the train. In addition, Carvajal-Carreño et al. (2014) considered the influence of train-mass uncertainty to
the optimized train trajectory in ATO system, and a fuzzy NAGA-II algorithm was developed to solve the problem. A case
study of Metro de Madrid was adopted to illustrate the effectiveness of the proposed approach in reducing energyconsumption for train operation.
On the other hand, the majority of existing research formulated train trajectory optimization models based on completely
accurate tracking cases, which is indeed the typical situation for ATO systems with advanced speed control methods. These
studies assumed that the train actual speed curve is exactly the same with the target speed profile (see Howlett, 2000; Tang
et al., 2015; Ke et al., 2009, etc). Therefore, those indicators, i.e. punctuality, riding comfort, energy consumption, switching
562
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
frequency, can be considered in the optimization, where the punctuality is always defined as a constraint. Also, specific line
conditions (e.g., variable speed limits, gradient), signal block modes (Ke et al., 2009, 2010) and discrete/continuous control
settings (Howlett, 2000), are taken into account in the optimization models. For instance, Ke et al. (2009, 2010) considered
speed curve optimization of a train on two different signal block modes, i.e., a fixed block scenario and a moving block
scenario.
On the basis of different mathematical formulations, a variety of methods were proposed to solve the problem of the
speed profile optimization, and these methods are mainly divided into three categories, i.e., analytical algorithm, numerical
algorithm and evolutionary algorithm. Some recent literatures are listed in Table 1 and the detailed descriptions are presented below.
(i) Analytic methods One of the main solution methodologies for speed profile optimization is the analytical algorithm,
which is typically based on the optimal control theory and solved by Pontryagin maximum principle ((Howlett and Pudney,
1995; Khmelnitsky, 2000; Liu and Golovitcher, 2003)). This kind of solution method can obtain the theoretically optimal
solution, but it requires rigid properties of the formulated mathematical models. Therefore, the existing analytical algorithms commonly consider only two objectives, i.e., train energy consumption and punctuality, under simplified environment in the modeling process. For instance, to achieve the minimal fuel consumption within a given trip time, Howlett
and Pudney (1995) and Howlett (2000) formulated a discrete train control model on a relatively flat slope. Based on Pontryagin maximum principle, the optimal solution was confirmed to be an ‘‘maximum acceleration-cruising-coasting-maxi
mum braking ” control sequence with unique switching points. Considering a practical situation with variable gradients
and arbitrary speed limits, Khmelnitsky (2000) designed a practically efficient algorithm based on the maximum principle
to find the optimal speed profile that minimizes the energy consumption. Liu and Golovitcher (2003) proposed a continuous
traction force model with constant efficiency to find the optimal switching points on a multilayer state-variable plain. More
recently, Albrecht et al. (2016a,b) discussed the energy-efficient train operation profile based on a more general model on an
undulating track with steep grades. They first used Pontryagin maximum principle to denote necessary conditions on an
optimal strategy, and proved that an optimal strategy always exists. Then, an intrinsic local energy minimization principle
for determination of optimal switching points is derived, which defines a unique optimal train driving strategy. In addition,
Su et al. (2013) proposed to optimize the overall energy consumption that considers the speed profiles of a single train on
every segment. This is inherently a bi-level programming approach. In the first level, an efficient algorithm with fast computing speed is developed based on Pontryagin maximum principle. In the second level, the distribution of total trip time is
optimized for the energy-efficient speed profile on the entire route.
(ii) Numerical algorithms Numerical algorithms, involving dynamic programming (DP) (Ko et al., 2004), sequential
quadratic programming (Miyatake and Matsuda, 2009), Lagrange multiplier-based method (Rodrigo et al., 2013), have relatively less requirements for the objective function, and can make a trade-off between optimization performance and computational time. For example, to provide the on-board drivers with energy-efficient speed profiles in real-time, Aradi et al.
(2013, 2014) developed a predictive optimization method, for which the optimization criterion is to reduce energy consumption while keeping the computational time. Since the methods by Pontryagin maximum principle often meet difficulties
accounting for the complex train dynamic models, variable speed limits and gradients, etc., Ko et al. (2004) reformulated
the train running process into a multi-stage decision process, and applied dynamic programming (DP) to searching for
the optimal control strategy directly. The optimal speed profile can be obtained within practically acceptable computational
time even when the method is applied to actual complicated running conditions. Calderaro et al. (2014) showed the trade-off
relationship between optimization performance and computational time with a dynamic programming method. They proposed a two-step approach to obtain a more efficient solution with a given time. The fist step aims to find a set of pseudooptimal speed with minimum energy consumption, and the second step evaluates these solutions by a simulation tool. To
optimize the energy consumption and riding comfort with fixed trip time, Wang et al. (2013) proposed two numerical
approaches to generate the speed profiles with good performance and short computational time. First, a pseudospectral
method is used to solve this nonlinear optimization problem. To shorten the computational time, the speed profile optimization problem was reformulated as a mixed integer linear programming problem by approximating with piecewise affine
functions and solved by the commercial software. Considering the complex operational constraints and signaling constraints,
Wang and Goverde (2016) formulated the train trajectory optimization problem into a multiple-phase optimal control
model that was solved by a pseudospectral method. Particularly, this approach is able to calculate the optimal train trajectories under both delay and no-delay situations for the minimization of train delay time and energy consumption.
(iii) Evolutionary algorithms Compared with the former two kinds of methods, evolutionary algorithms, e.g., genetic
algorithm (GA) (Bocharnikov et al., 2010; Chang and Sim, 1997; Wong and Ho, 2004), ant colony optimization (ACO) (Ke
et al., 2009), tabu search (TS) (Liu et al., 2015), and simulated annealing algorithm (SA) (Kim and Chien, 2011), have least
requirements for models in train speed profile optimization. Nevertheless, most of these algorithms cannot guarantee the
optimality and convergence of the solutions. We present some representative literature that use evolutionary algorithms
to solve the speed profile optimization problem in railway systems. Chang and Sim (1997) applied the GA to determine
the coast-accelerate-brake point by joint evaluation of energy consumption, punctuality and riding comfort. Wong and
Ho (2004) showed that, the GA can achieve a lower average number of iterations and a fitter solution with multiple switching points compared with the classical Nelder and Mead method. Ke et al. (2009) formulated a combinatorial optimization
model with constraints of fixed-block signaling system to minimize the computational time and energy consumption. A
Max-min ant system (MMAS) of ACO was developed to search for the optimal train speed profile, demonstrated to be more
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
563
time-efficient than GA. This ACO-based algorithm was further extended into a metro line with the consideration of movingblock system (Ke et al., 2010). The new MMAS algorithm could realize online optimization, and the results showed a reduction of 19.6% in energy consumption with the former ACO algorithm. To compare the performance of different evolutionary
algorithms, Lu et al. (2013) applied three algorithms, i.e., GA, DP and ACO, respectively, to optimize the energy-efficient train
speed profile. The results indicated that, each algorithm has advantage in some specific aspects (i.e., deviation, performance,
computational time), and multiple algorithms should be employed to obtain more efficient solutions. In addition, Kim and
Chien (2011) developed an SA algorithm for the optimization of speed profiles for energy-efficient train operation with the
consideration of tracking alignment, speed limit and schedule adherence, and this method was applied to a segment of the
New Haven line of the Metro-North Commuter Railway between Woodlawn and New York. Further sensitive analysis was
conducted with different trip times, train weight, maximum train speed, etc. With the consideration of inaccurate speed
tracking of ATO controller (or driver), Liu et al. (2015) developed two improved TS algorithms to calculate the energyefficient train speed profile. The two algorithms could save energy for 8.93% and 2.54% by comparing with the practical train
operation data in Beijing subway, and more importantly, the computational times of these two algorithms are only about
1–2 s, which is short enough to be applied in real-world applications.
Remark 3.1. It is worth noting that these above three kinds of methods actually have different characteristics with respect
to the magnitudes of gap between performance and optimality. For analytic methods, the theoretically optimal solution can
be proved and obtained under each specific mathematical model with different line parameters (e.g., gradients, speed limits),
although the mathematical model always needs to be simplified to make the model solvable. The numerical algorithms, e.g.,
dynamic programming, mixed integer linear programming, can obtain a near-optimal solution with evaluative benchmark.
Theoretically, this kind of solution approach can obtain the optimal solution in case of enough computational time. The third
kind of methods, i.e., evolutionary algorithms, cannot always guarantee the optimality of the solutions and they cannot
provide theoretical benchmarks for evaluating the solution algorithms. Therefore, most of these evolutionary algorithms use
practical case studies or real-world train trajectory as benchmarks for evaluating the performance of these solution
approaches.
3.3. Train speed control
After generating the optimal recommended speed profile, the next procedure of train operation is to develop an efficient
method to control the movements of the train with respect to different train models (metro trains, high-speed trains, etc)
and running conditions (e.g., tunnels, curves, sharp gradients), in order that the train can track the speed profile precisely
and run safely and smoothly. In a typical railway system, the train speed control is usually accomplished by one of the following two approaches.
In most main line railways, HSRs, and metro lines under GoA1 or GoA0, the trains are predominantly controlled by drivers, and thus a driver advisory system (DAS) that is embedded into the train control system can provide drivers with
additional driving advice to keep the train at the recommended speed (see Albrecht et al., 2013; Howlett and Pudney,
1995; Panou et al., 2013; Rao, 2015; Quaglietta et al., 2016).
In urban metro lines with high automation levels (i.e., GoA2, GoA3, or GoA4), the train speed control function is realized
by an on-board computer with pre-designed speed control algorithms. In practice, the on-board computer receives realtime train states (i.e., speed, time, velocity) and automatically output the optimal control commands (accelerating, coasting or braking) to the train traction and braking motors by comparing with the recommended speed.
3.3.1. Proportional integral derivative controller
The most widely used train speed control method of ATO is the PID controller (e.g., Beijing Subway Yizhuang line, Changping line, etc), which continuously calculates the error value between the measured train speed v and recommended speed
v^ , and adjusts the control command to minimize the speed tracking error over time. Although PID-based controller can
achieve a relatively good tracking performance in a wide variety of industrial implementations, the engineers particularly
focus on two issues in designing PID-based controllers in practical applications for ATO systems.
On the one hand, how to determine the best PID coefficients is a challenging task, and most existing methods are based on
manual experiences and professional judgements through many times of repeatedly field tests. In addition, the parameters
of train models are always affected by some external factors in daily operations, such as weather condition, normal deterioration and mechanical wear. The parameter variations will inevitably reduce the performance of PID controller if the PID
coefficients are fixed.
On the other hand, PID-based controller always has poor riding comfort due to the frequent switches of PID control
commands, which could also increase the energy-consumption for train operations. Therefore, engineers in practice have
to formulate some additional constraints for PID controller in order to jointly improve the performance of multiple objectives
(i.e., tracking accuracy, punctuality, comfort, energy-efficiency). For example, a maximum accelerating rule is designed in
Beijing Subway Yizhuang line to avoid that the train accelerates too sharp during train accelerating phase. And the
accelerating commands are allowed only when the speed error is larger than a pre-defined threshold. Even though, these
564
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
Table 2
Review of studies on train speed control algorithms for automatic train operations.
Publication
Model
Algorithm
Main benefits
Yasunobu and
Miyamoto
(1985)
Yasunobu et al.
(1983)
Sekine et al.
(1995)
Ke et al. (2011)
Metro
trains
Predictive fuzzy control
Improve riding comfort, station stopping accuracy and energy-savings
Metro
trains
Metro
trains
Mass rapid
transit
MP-HSTa
Locomotive
trains
Metro
trains
Metro
trains
MP-HST
Fuzzy control with
linguistical control rules
Fuzzy neural network
control
Fuzzy PID control
Guarantee a comfortable and accurate train stop control
Fuzzy logic control
Expert knowledge
Improve tracking accuracy and riding comfort
Obtain control strategies from expert drivers and engineers to improve driving
performance
Develop two ITO algorithms that improve riding comfort, punctuality and
energy-savings
Use data mining technique to improve riding comfort, punctuality and energysavings
High accuracy speed tracking method with coupler forces and resistances
Dong et al. (2013)
Meulen (2008)
Yin et al. (2014)
Yin et al. (2016b)
Yang and Sun
(2001)
Chou and Xia
(2007)
Zhuan and Xia
(2008)
Faieghi et al.
(2014)
Song et al. (2011)
a
b
c
d
Heavy-haul
trains
Heavy-haul
trains
MP-HST
MP-HST
Song et al. (2014)
MP-HST
Gao et al. (2013)
HST
Kaviarasan et al.
(2016)
Yang et al. (2014a)
MP-HST
HST
Hou et al. (2011,
2013)
Li et al. (2014a,b)
Main line
trains
MP-HST
Expert system and
reinforcement learning
Expert knowledge and
Ensemble CARTb
A mixed H2 =H1 controller
A closed-loop cruise
controller
Measurement feedback
output regulation
Robust adaptive cruise
control
Robust adaptive control with
optimal distribution
Backstepping adaptive
control
Robust adaptive control by
RBF-NNc
Reliable dissipative control
Predictive control method
with an ANFISd
Iterative learning control
(ILC)
Robust cruise control based
on sampled-data
Reduce fuzzy rules of train operations on the basis of the research in Yasunobu
and Miyamoto (1985)
Reduce energy consumption
Improve speed tracking accuracy, in-train force management and energy usage
An accessible approach for its simplicity, cost effectiveness and implementation
convenience
Achieve asymptotic tracking and disturbance rejection for MP-HSTs
Good robustness with external disturbances, resistances, and unknown system
parameters
Achieve high-accuracy tracking control under time-varying traction/braking
faults with complete unknown system paramaters
Stress the actuator saturation nonlinearity with unknown system parameters in
HST models
Establish a sufficient condition that guarantees the stability with probabilistic
time-varying delays
Improve tracking accuracy by modeling dynamics of HSTs using an ANFIS
Propose a model-free ILC based train trajectory tracking algorithm that
guarantee the asymptotic convergence and tracking effectiveness
Guarantee speed tracking accuracy with uncertain disturbances and stability of
relative spring displacement
MP-HST: Multi point high-speed train model.
CART: Classification and regression trees.
RBF-NN: Radial basis function neural network.
ANFIS: Adaptive neurofuzzy inference system.
formulated rules are all based on experiences, which may be infeasible with different types of railway trains and their
parameter variations.
In recent years, a lot of researchers have developed different automatic train speed control methods to solve the above
problems. These studies are characterized by different types of train control models and objectives, and different kinds of
algorithms, depending on their specific applications. In Table 2, we present a schematic overview of these studies. In general,
these studies can be classified into two categories. One kind of these methods take advantage of empirical knowledge and
professional experiences to summarize or represent these experiences. These methods are principally based on fuzzy logic,
expert system or data mining algorithms to improve the multiple objectives for train operations, termed as intelligent speed
control methods in this review. Meanwhile, the second type of methods takes advantage of the train model information to
design an efficient adaptive or robust speed controller to guarantee the trajectory tracking accuracy, named by adaptive control methods in the following discussion.
3.3.2. Intelligent control methods
As we have mentioned above, the train driving process needs to consider multiple objectives, i.e., energy consumption,
punctuality and riding comfort. Since a single PID controller is difficult to meet these goals synchronously, many researchers
began to use some intelligent control methods (e.g., fuzzy control, expert systems) to transform the driving knowledge and
experiences into a series of domain rules in order to improve the riding comfort of passengers and reduce energy consumption. For example, Yasunobu and Miyamoto (1985) and Yasunobu et al. (1983) applied the predictive fuzzy control method
for automatic train speed control in ATO system. First, they defined the fuzzy sets of performance indices, involving safety,
riding comfort, tracking accuracy and energy consumption. Then, a predictive fuzzy control method was proposed, which
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
565
predicts the result of each candidate control command and selects the best control rule based on the fuzzy sets of performance indices. Moreover, this fuzzy ATO system was put into operation in the Sendai Subway, Japan in 1987, and the results
indicated that this fuzzy ATO system can operate trains automatically without drivers, and it can be as effective as manual
driving. Ke et al. (2011) presented a fuzzy-PID gain method to track the recommended speed profile, and the speed profile in
their research was generated by an MAX-MIN ant system. In their research, the control rules were determined by a fuzzy
inference system with railway gradient, speed error and its changing rate. In order to achieve high tracking accuracy with
unknown parameters of train models, Dong et al. (2013) designed two fuzzy approaches, i.e., a direct fuzzy logic controller
for single-point train model and an implication fuzzy logic controller for multi-point train model, and both of these two
methods are model-free.
Besides, there are also some other studies that employ knowledge representation methods (expert system, data mining,
etc) to simulate experienced driving strategies to realize automatic train speed control. As indicated by Yasunobu et al.
(2000) and McClanachan and Cole (2011), manual driving strategies can be better than automatic speed control methods
because experienced humans can easily control a railway train (i.e., a nonlinear dynamic system with uncertain parameters).
Moreover, human driving is more flexible than ATO in complex situations, for instance the train in an inaccurate station stopping scenario. Therefore, using human expertise and judgements by means of knowledge representation is an important
method of designing speed control methodologies. According to the research by Meulen (2008), expert drivers and engineers
were used to develop and refine the control strategies through the use of train operation simulations and experiments, and
the formulated control method was shown to be effective in operating long distributed power trains. More recently, Yin et al.
(2014, 2016a,b) proposed to use some machine learning algorithms to learn the driving experiences from historical raw data
by means of knowledge representation (Chandrasegaran et al., 2013). On the basis of some domain empirical rules,Yin et al.
(2016b) first applied a regression algorithm, i.e., CART (Classification and regression tree) and ensemble learning methods
(i.e., Bagging and LSBoost) in order to represent the valuable expert knowledge from historical train driving data. In addition,
two intelligent train operation (ITO) algorithms that are respectively based on an expert system and reinforcement learning
were further proposed (Yin et al., 2014, 2016a). These two ITO algorithms, combined the advantages of both manual driving
and automatic speed control methods, were shown to be effective in improving the performances for metro train operations.
3.3.3. Adaptive control methods
It is very interesting to note that the above speed control methods that are based on knowledge experiences mainly focus
on the metro train operations, in which the train dynamic models are relatively simple and the train running speed is usually
lower than 60 km/h. The trains are usually treated as a single-point train model in the above studies. Nevertheless, in the
main line railway or HSRs, realizing train speed control is much more challenging due to the operation complexity and
high-speed characteristics of train dynamics (e.g., the aerodynamic drag forces, interactive impacts among the vehicles,
and the nonlinearity of actuator saturation). Therefore, the speed control of main line trains or HSTs is particularly regarded
as one of the main issues in improving the automation grade in railway systems. In this part, we focus on some state-of-theart studies, which generally use adaptive control methods in order to handle the complexity and uncertainty of train operation models.
Considering the interactive forces among the connected vehicles of a train, the multi-point high-speed train operation
model (MP-HST) that treats an HST as a cascade of masses connected by flexible couplers was first proposed by Yang and
Sun (1999). This model is refined on a single-point train model, and it can better match the reality to describe the motion
of trains. In addition, they also designed a mixed H2 =H1 cruise controller based on this model, which was synthesized by
linear matrix inequalities to satisfy a mixed objective of speed tracking accuracy and gust attenuation (Yang and Sun,
2001). Chou and Xia (2007) extended this model to the heavy-haul trains and developed a closed-loop cruise controller with
electronically controlled pneumatic brake systems. The objective is to improve velocity tracking accuracy, in-train force
management and energy usage. Focusing on the speed control of multi-point heavy-haul trains, Zhuan and Xia (2008) proposed an approach of output regulation with measurement feedback, which regulated all vehicles’ speeds to a recommended
speed profile.
Recently, some researchers have focused on the adaptive speed control for HSTs with unknown parameters and external
disturbances, which is more realistic in real-world applications. For example, by considering the factors of input nonlinearities, actuator failures and in-train forces, Song et al. (2011) developed a neuroadaptive fault-tolerant control, which is capable of automatically generating the intermediate control parameters and producing the control command based on input
and response data. In addition, Song et al. (2014) further considered the effects of time-varying failures in both train traction
and braking phases, and proposed an adaptive backstepping control method, which was fully parameter-dependent and was
shown to be effective in achieving good speed tracking performance. Besides, actuator saturation that may constrain the output of train serving motors is an important issue in ATO systems. Gao et al. (2013) addressed this problem by designing an
on-line approximation-based robust adaptive controller, in which an RBF-NN was utilized as the approximator. This
approach is capable of online-estimating the unknown system parameters and keeping the stability of the closed-loop system with output saturations. Through a different approach, Yang et al. (2014a) derived an adaptive neurofuzzy inference system (ANFIS) to model the running process of a complex HST with uncertain and nonlinear dynamics. This data-driven model
was trained and validated by historical train operation data in Beijing-Shanghai HSR. On the basis of this model, a generalized predictive control algorithm was proposed to improve the speed tracking accuracy for HSTs. Furthermore, iterative
learning control (ILC), which requires less system model knowledge and is an almost model-free method, was first
566
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
introduced by Hou et al. (2011) and Sun et al. (2013) into the field of ATO systems. They proposed two ILC based speed control methods, one for train automatic stop control (Hou et al., 2011) and the other for train automatic driving (Sun et al.,
2013). These two methods, similarly, can learn to improve control performances from previous executions. In other words,
the two ILC based methods can gradually reduce the speed tracking errors via repeated operations of the train.
Different from the above studies that are mainly confined to the continuous-time control design for continuous-time
models, Li et al. (2014a,b) proposed the idea that, a sample-data control method, for which the control commands are constant during the sampling period and change only at the sampling instant (given positions of sensors), can be more practical
and important than the continuous-time control approach. In their research, the problem of speed control for HSTs with
sampled-data was transformed to the problem of stability analysis of time-varying system by converting the sampling
period into a bounded time-varying delays system. Based on Lyapunov stability theory, they proved that the robust
sampled-data cruise control method can (i) keep the train tracking accuracy with recommended speed profile, (ii) guarantee
the stability of the relative spring displacement between two neighboring vehicles, and (iii) be robust with a prescribed H1
disturbance attenuation level with respect to the wind gust, which could thus guarantee the safety and comfort of the
operations of HSTs.
4. Further discussions: opportunities and challenges
As the gap between the increasing railway transport demand and limited infrastructure is becoming larger and larger in
recent years, we anticipate that research for ATO systems will continue. In light of this review, the future of ATO systems
could be improved on the following aspects in industrial implementations in the years to come:
Real-time improvements: The recommended train trajectories of current ATO systems are commonly optimized and
embedded in advance in the train on-board computers. In practice, the real-time train operation process is always subject
to many external uncertain factors, e.g., the train departure (arrival) delay, stochastic train mass, etc (Karvonen et al.,
2011). Therefore, an urgent question is to develop train trajectory optimization methods that can produce optimal (or
near-optimal) train speed profile in a very short time when the train is dwelling at station (usually 20 to 40 s in metro
systems).
Fully automated (or unattended) train operation (termed as UTO): The use of UTO is a growing trend for the future development of urban metro systems, which achieves GoA4 according to the IEC-62290 standard. In particular, Cohen et al.
claimed that, GoA4 systems can either be UTO without onboard staff or attended GoA4, where there is an attendant
on board who is not essential to the operation of the train. On the one hand, UTO lines have about 70% fewer staff than
conventional lines with staffed stations, and enable a more regular service with high train frequency and passenger carrying capacity (Cohen et al., 2015). On the other hand, how to reduce the risks of UTO lines and improve the reliability of
equipment in UTO systems remain to be new challenges in the future (Powell et al., 2016; Wang et al., 2016).
Big data analysis for train operations: With the increased automated levels of train operation in railways, a lot of new
sensors have been installed in railway systems that enable us to access to a comprehensive historical archive of realtime train operation data from these sensors. The advanced technology has provided a unique opportunity to demonstrate how big data can be used to offer new insights for developing and improving automatic train operations. The possible research directions involve data visualization for better understanding and evaluating the performance of train
operation, data mining techniques to supervise the ATO system and find potential faults to improve the reliability of train
operations, etc.
In addition, we note that the ATO systems that are related to high-speed railways, environmental considerations, and the
integration with railway traffic control have been frequently stressed in the most recent research papers and technical
reports, which are likely to shape further directions of ATO and its related technologies.
4.1. Implementation of ATO for main line railways and high-speed railways
Currently, ATO system that replaces drivers to operate the trains automatically has already become a mature technology
and been widely implemented in the field of urban transport systems. Nearly a quarter of the world’s metro systems exploit
at least one line in UTO (unattended train operation) mode, for instance Paris Métro Line 14 in Europe, Beijing Subway Yanfang Line and Dubai Metro in Asia, and Sydney Metro in Australia (to be operated in 2019) (Fraszczyk and Mulley, 2017).
Through the utilization of advanced optimisation and control methods (e.g., fuzzy control methods in Sendai Metro, fuzzy
PID-based controller), ATO can help drivers to deliver improved train operation performance in saving energy, enhancing
service quality and releasing manual labor. However, the train driving tasks in main line or high-speed railways are still
achieved by manual driving or manual driving with DAS. In such cases, the physiological conditions of drivers would have
direct impacts on the safety and efficiency of train operations. Although some advanced speed control methods for main line
trains and HSTs are found in existing literature, and the successful experiences of ATO implementation in urban metro systems are employed as references, the real-world application of ATO in main line or high-speed trains still faces some tough
challenges.
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
567
Compared with urban rail transit networks, in which each line is a relatively enclosed system that is independent from
each other and each train moves like a ‘‘shuttle” on the fixed tracks, the main line railway or HSR networks are relatively
open systems with heterogeneous trains. The trains usually have different circulation plans, and each train may travel
among different railway lines according to its circulation plan. In Europe, a main line railway may even cross more than
one region or nation. Therefore, the ATO framework design for these open systems would be definitely much more complex than that of metro systems in order to be flexible with various external circumstances.
The track layout is larger and more complex, which brings complicated operational constraints (e.g., signalling constraints, interlocking of routes) in mainline railways and HSRs. In this sense, it is particularly required to rigorously consider these complicated yet practical constraints in order to generate practicable train speed trajectories in the
development of ATO systems for main line and high-speed railways.
As indicated by Bienfait et al. (2012), the new generation of ATO for main line railways is needed to respect the fundamental principles of scalability and interpretability of the train control systems (e.g., ETCS in European, Chinese Train
Control System in China (Ning et al., 2004), ATC in Japan). Different from urban metro lines, in which the main functions
of ATO are clearly defined, there is currently no uniform standard to clarify what ATO functions are needed to achieve the
operational goals for main line railways.
Several recent projects have focused on the implementation of ATO systems into suburban railways, which already takes
the first step in this direction. For example, the Thameslink, a railway line that connects the suburban and center of London,
is to be updated by an ATO system (UITP, 2015). This ATO system, developed by Siemens, is expected to realize the message
exchange between the on-board ATO with the traffic control center (TCC) via ETCS, and exactly control the optimized speed
of the train. Furthermore, shorter headways through time-optimized travel, energy-efficient train operations, exact train station stopping, automatic door control are the main functions in this ATO system. Meanwhile, NGTC (Next Generation Train
Control) project funded by the European Commission is paving the way for standardized train control systems for mainline
railways, which could provide the complete ATP, ATO and ATS functions to reduce the total operational costs and improve
the efficiency of mainline railways. Besides, the other preliminary works involve the GreenRail program leaded by Alstom,
which is testing the real-time train optimal driving on actual Belgian Railway lines (Bienfait et al., 2012) and the ERTMS project that aims to combine ATO with ETCS in a segment of main line railway in Kolín, German (UIC, 2012), etc.
4.2. Cooperative train operation for energy savings
In recent years, the new indicator called carbon performance is implemented in some railway systems in Europe and Asia
(TSAG, 2010). The carbon performance, identified by the greenhouse gas emissions, is directly related to the energy usage of
a railway system. More specifically, traction energy is the major emissions source (Yang et al., 2016). Since conventional
speed profile optimization in ATO focuses on the speed profile at each interstation separately, the cooperative train operation
that integrates energy-efficient train operation with timetable design for ATO is paid more and more attention in recent
years. This new approach aims to find a global optimal solution that minimizes the total energy consumption for the entire
route instead of a single segment (e.g., Albrecht, 2014; Su et al., 2013; Goverde et al., 2016; Yan et al., 2016; Zhao et al., 2015),
and thus, it could be more energy-efficiency through system-level optimization.
Su et al. (2013) developed a bilevel programming model, in which its first level determines the optimal train timetable
(i.e., a set of planned trip time in the segments), and the second level aims to find the energy-efficient train speed profiles
with respect to the given trip time. In addition, they designed a fast algorithm based on Pontryagin maximum principle to
obtain the speed profile that minimizes the traction energy consumption. Then, in order to find the global optimal solution,
an efficient algorithm was developed, which first calculates the minimum trip time of a train in each segment, and then distributes the reserve time into the segments that have the largest ratio between variation of trip time and energy consumption. The numerical experiments on a real-world case demonstrated the energy reduction in the entire route for 14:5%, and
the computational time is short enough to be applied into practical applications. Goverde et al. (2016) proposed a three-level
framework for designing a conflict-free and energy-efficient timetable, which takes account into the train speed profile optimization with respect to the scheduled running times, microscopic infrastructure and rolling stock characteristics. A case
study of a central part of the railway network in the Netherlands was adopted. The numerical results show that, this
approach saves 8.7% energy consumption with respect to the minimum running times, and the energy saving percentage
is 35.3% using the energy-optimal speed profile. Also, Wang and Goverde (2016) developed a multiple-phase optimal control
model that considers both the operational constraints but also the signalling constraints. In addition, the optimal train trajectory can be generated by a pseudospectral method under both delay and no-delay situations for minimizing the train
energy consumption and delays with the original timetable. An integrated train control and scheduling approach was considered by Ye and Liu (2016). With pre-given train departure sequences, the problem was formulated as a multiphase optimal control problem incorporated with practical train running conditions (e.g., limitations of train speed, traction and
braking forces). In particular, this approach enables to obtain the theoretically optimal train control strategies and train
schedules (i.e., arrival and departure times at stations) that minimize the traction energy consumption of multiple trains
on railway lines.
More recently, regenerative braking is being applied in ATO systems in urban rail trains for carbon reduction (Domínguez
et al., 2012; González-Gil et al., 2013). In conventional metro systems, the kinetic energy of the running train is converted to
568
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
heat by friction and braking force. Regenerative braking enables that we can use the electric motor as an electric generator
when the train is braking, and then feed a part of regenerated electricity energy back to contact lines, or store the regenerative braking energy into an on-board energy storage device. Using the example given by Yang et al. (2016), the regenerative
energy accounts for 33.6% in the total traction energy consumption in Beijing subway. Thus, how to consider the regenerative braking energy into ATO is a very promising approach to save energy and release carbon emission in railway systems.
On one hand, the regenerative braking energy can be saved by implementing on-board or wayside energy storage devices,
or reversible substations when the train is braking (González-Gil et al., 2013; Miyatake and Matsuda, 2009). Then, the train
can use the storage energy to accelerate during the next start. For example, considering a train with on-board energy storage
device, Domínguez et al. (2012) proposed a network model that calculates the total energy recovered by energy storage, and
design a practical method to generate the optimal ATO speed profiles that minimize the energy consumption.
Due to the high expense and limited capacity of energy storage devices, most trains in real-world applications are currently not equipped with these devices. The alternative approach is to use the regenerative braking energy that is generated
by a braking train for the acceleration of another train in the same substation. This approach has no need for the energy storage devices, and thus it is regarded as a preferential method to utilize the regenerative energy (González-Gil et al., 2013).
However, if the feedback regenerative braking energy cannot be utilized immediately, it will be wasted by heat resistors
(Yang et al., 2016). Therefore, a hot research topic is the cooperative train operations with collaboratively optimized train
speed profiles, so that the utilization of regenerative braking energy can be maximized (see Su et al., 2015; Li and Lo,
2014; Tang et al., 2015).
Su et al. (2015) proposed a cooperative train operation model based on the scenario that an accelerating train can reuse
the regenerative energy from a braking train on the opposite track. Two algorithms, i.e., a numerical algorithm to calculate
the optimal train speed profile and a bisection method to determine the train departure time, were given to minimize the
total energy consumption.
In addition, Li and Lo (2014) developed an integrated model to jointly optimize the train timetable and the corresponding
speed profiles of trains in every segment. In this model, the decision variables are the recommended speed profiles, arrival
times and departure times at every track segments, and the objective is the deviation between the energy consumption and
utilized regenerative energy. The results indicated that, by considering regenerative energy, the integrated optimization
model can reduce energy consumption around 20% compared with the two-step approach optimizing the timetable and
speed profile separately. Combined with train operation strategies, Yang et al. (2015) formulated an integer programming
model to minimize the trains’ energy consumption in an integrated running cycle. By using an allocation algorithm and a
genetic algorithm, the model was solved based on the real-life operation data in Beijing subway. Numerical examples
demonstrated the energy-efficiency of this new model compared with the models that only consider each two adjacent
trains.
Furthermore, for better understanding of the energy-efficient train operation and timetabling problem, we can refer to
Scheepmaker et al. (2017), which presents a comprehensive review of existing modeling and solution approaches for this
integrated optimization problem in railways.
4.3. Integration of railway traffic control and automatic train operation
Most existing studies focus on optimisation of railway traffic control and train operations separately, but not collaboratively (Rao, 2015). For example, automated train dispatching systems are implemented in many railway systems. These systems are essentially management information systems, which only focus on setting arrival/departure routes for the trains
(Corman and Meng, 2015). Meanwhile, the train control systems are in charge of specific train control actions separately,
as we described above. A very promising approach is the centralization of railway traffic control and train operations into
a single system or ‘‘guiding” mind. The key idea is that, each train reports its status in real-time to the centralized center,
which then guides the train to operate automatically with conflict-free plan and optimized speed profile. As indicated by
Telecommunication Standardization Advisory Group (TSAG) (TSAG, 2010), the next generation railway transportation system will bring the operation and rescheduling of trains together to increase the rail’s delivery potential, reliability and customer satisfaction, and guarantee operation safety and energy-efficiency through better management.
In the industrial field, European Standard IEC 62290 published in 2014 has presented the framework for an Urban Guided
Transport Management and Command/Control Systems (UGTMS), involving the functional, system and interface requirements (IEC 62290-1, 2004). UGTMS is defined to be an integrated train command, control and management system that
can achieve the comprehensive functions of both railway traffic control and train operations on urban, guided passenger
transport lines and networks in the future.
Even though much literature formulating various models and algorithms can be found in these two fields, until now, there
is few integrated optimisation models to deal with rail traffic control and train operations collaboratively (Rao, 2015). Rao
et al. (2016) proposed a new integrated approach for traffic management and train automated operations in order to determine a set of conflict-free train trajectories as well as the corresponding traction and braking forces of trains. Then, the best
train trajectory by overall evaluation of different optimisation objectives can thus be obtained for the decision-making of
train operators. Nevertheless, the considered problem was a relatively simple case that only involves two trains on a given
segment of rail blocks. More recently, an integrated framework for the automatic real-time management of railway traffic
was reported by Quaglietta et al. (2016) as a main output of the European project ON-TIME. The framework is composed
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
569
by several interacting modules, involving the traffic state monitoring, perturbation management module, automatic routing
setting module and driver advisory system module. Three case studies that are based on the East Coast Main Line in the UK,
the Utrecht-Eindhoven-Tilburg-Nijmegen corridor in the Netherlands and the Iron Ore line at the border between Sweden
and Norway are adopt, which all demonstrate the effectiveness and potential benefits by integration of railway traffic control
with train operations. In the view of authors, the following research issues are identified as potential challenges for the integrated optimisation of rail traffic control and train operations:
One of the hot research directions in this field is to formulate integrated mathematical models that synchronously consider real-time timetable, rolling stock and crew rescheduling with train speed trajectory regeneration to realize a global
optimal (or at least feasible) solution. Due to the computational intensity rising from these integrated models, one possible strategy can be to design a close-loop framework that enables to solve each aspect of this problem from the outercontrol loop to the inner control loop, and then get feedback to iteratively update these solutions to obtain a feasible solution with good global performance (Cacchiani et al., 2014; Xu et al., 2015).
Up to now, many recent studies are observed in the field of railway traffic control or train operations that concentrate on
different disturbance or disruption scenarios in single, double railways or bi-directional metro lines. Nevertheless, due to
the inherent uncertainty of these disturbance or disruptions, how to identify the realistic train delay distributions, clarify
different disturbance or disruption scenarios and then carry out appropriate emergency plans accordingly is actually a
very complicated issue. In this sense, collecting and analyzing the realistic train operation and passenger smart card data
have practical significance for the qualitatively evaluation and guidance of rail traffic control and train operations.
In practice, railway traffic control and train operation require high level of real-time property for the solution approach
(Yin et al., 2016c). For example, in case of a disruption in high-speed railways, a rescheduled timetable must be obtained
in several minutes for reducing the impacts caused the disruption as much as possible (Zhan et al., 2015). Furthermore, as
metro systems are operated with high train frequency in peak-hours, a train dispatching plan should be determined by
the dispatchers within one minute to avoid delay propagation. Meanwhile, real-time optimization methods are also
required to adjust train trajectories in such cases for the ATO system. To this end, the main difficultly is still to develop
real-time and exact solution approaches for solving these practical yet commonly large-scale mathematical models for
the integrated optimization of rail traffic control and automatic train operations.
In particular, as unattended train operation (or fully-automated train operation) systems are being widely implemented
in urban transit lines throughout the world (Powell et al., 2016; Fraszczyk and Mulley, 2017), we believe that the UTO or
fully-automated train operation systems will hopefully pave the way for the integration of rail traffic control and automatic
train operation by the centralization of real-time train information in a unified traffic control center.
5. Conclusion
This paper reviewed the recent research related to the use of ATO technologies in rail transportation systems. In particular, we analyzed approaches for the problem of train operation, that is typically realized by an optimized speed profile and a
train speed controller. The current literature on these two aspects was reviewed, respectively. In light of this review, we
pointed out a few directions of future research: implementation of ATO in main line railways or HSRs, cooperative train operations for energy savings, and the integration of railway traffic control and train operations.
In the coming years, it is very promising that the integration of ATO and railway traffic management will become available for the future rail transportation systems. How to make use of these two parts collaboratively to optimise the railway
network’s capacity and increased customer satisfaction, reliability and efficiency is an urgent problem to be solved. Although
there is currently no common model or approach that seeks to control the whole railway network, an emerging trend is to
simultaneously generate train control actions (e.g., recommended speed profile) and rescheduled plans with integrated optimisation models. Furthermore, since the long-term evolution for railway (LTE-R) technology is emerging to enable crossborder train-ground and train-train communications, the integrated railway traffic control can also benefit from the future
application of this new technology, helping to improve the railway transportation systems.
Acknowledgement
This research was supported by the National Natural Science Foundation of China (Nos. 61403020, U1434209, 71422002,
71621001), the Beijing Laboratory of Urban Rail Transit, the Beijing Key Laboratory of Urban Rail Transit Automation and
Control, and the Major Program of Beijing Municipal Science & Technology Commission under Grant Z161100001016006.
References
Albrecht, T. et al, 2013. Applications of real-time speed control in rail-bound public transportation systems. IET Intell. Transp. Syst. 7 (3), 305–314.
Albrecht, T., 2014. Energy-efficient train operation. In: Hansen, I.A., Pachl, J. (Eds.), Railway Timetabling & Operations, Hamburg, pp. 91–116.
Albrecht, A., Howlett, H., Pudney, P., Vu, Xuan, Zhou, P., 2016a. The key principles of optimal train control – Part 1: Formulation of the model, strategies of
optimal type, evolutionary lines, location of optimal switching points. Transp. Res. Part B 94, 482–508.
570
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
Albrecht, A., Howlett, H., Pudney, P., Vu, Xuan, Zhou, P., 2016b. The key principles of optimal train control – Part 2: Existing of an optimal strategy, the local
energy minimization principle, uniqueness, computational techniques. Transp. Res. Part B 94, 509–538.
Aradi, S., et al., 2013. A predictive optimization method for energy-optimal speed profile generation for trains. In: IEEE International Symposium in
Computational Intelligence and Informatics (CINTI), pp. 135–139.
Aradi, S., et al., 2014. Design of predictive optimization method for energy-efficient operation of trains. In: IEEE European Control Conference (ECC 2014), pp.
2490–2495.
Astolfi, A., Menini, L., 2002. Input/output decoupling problems for high speed trains. In: Proceedings of the American Control Conference (ACC), pp. 549–
554.
Bienfait, B., Zeotardt, P., Barnard, B., 2012. Automatic train operation: the mandatory improvement for ETCS application. In: ASPECT. London: IRSE 2012.
Bocharnikov, Y.V., Tobias, A.M., Roberts, C., 2010. Reduction of train and net energy consumption using genetic algorithms for trajectory optimisation. In:
IET Conference on Railway Traction Systems (RTS 2010), pp. 32–36.
BSI, 2009. BS EN 12299:2009: Railway applications. Ride comfort for passengers. Measurement and evaluation.
Cacchiani, V. et al, 2012. Railway rolling stock planning: robustness against large disruptions. Transp. Sci. 46 (2), 232–271.
Cacchiani, V. et al, 2014. An overview of recovery models and algorithms for real-time railway rescheduling. Transp. Res. Part B 63, 15–37.
Calderaro, V., et al., 2014. An algorithm to optimize speed profiles of the metro vehicles for minimizing energy consumption. In: IEEE International
Symposium on Power Electronics, Electrical Drives, Automation and Motion (SPEEDAM), pp. 813–819.
Carvajal-Carreño, W., Cucala, A.P., Fernàndez-Cardador, A., 2014. Optimal design of energy-efficient ATO CBTC driving for metro lines based on NSGA-II with
fuzzy parameters. Eng. Appl. Artif. Intell. 38, 164–177.
Chandrasegaran, S.K. et al, 2013. The evolution, challenges, and future of knowledge representation in product design systems. Comput. Aided Des. 45, 204–
228.
Chang, C.S., Sim, S.S., 1997. Optimising train movements through coast control using genetic algorithms. IEE Proc. Electric Power Appl. 144 (1), 65–73.
Chou, M., Xia, X., 2007. Optimal cruise control of heavy-haul trains equipped with electronically controlled pneumatic brake systems. Control Eng. Pract. 15,
511–519.
Chou, M., Xia, X., Kayser, C., 2007. Modelling and model validation of heavy-haul trains equipped with electronically controlled pneumatic brake systems.
Control Eng. Pract. 15, 501–509.
Clark, S., 2012. A history of railway signalling (from the bobby to the balise). In: IET Professional Development Course on Railway Signalling and Control
Systems (RSCS), pp. 6–25.
Cohen, J.M., Barron, A.S., Anderson, R., Graham, D.J., 2015. Impacts of unattended train operations on productivity and efficiency in metropolitan railways.
In: Transportation Research Board of the National Academies, Washington, DC, pp. 75–83.
Cordeau, J., Toth, P., Vigo, D., 1998. A survey of optimization models for train routing and scheduling. Transp. Sci. 32 (4), 380–404.
Corman, F., Meng, L., 2015. A review of online dynamic models and algorithms for railway traffic management. IEEE Trans. Intell. Transp. Syst. 16 (3), 1274–
1284.
Corman, F., Quaglietta, E., 2015. Closing the loop in real-time railway control: framework design and impacts on operations. Transp. Res. Part C 54, 15–39.
Corman, F., D’Ariano, A., Pranzo, M., Hansen, I.A., 2011. Effectiveness of dynamic recordering and rerouting of trains in a complicated and densely occupied
station area. Transp. Plann. Technol. 34 (4), 341–362.
Domínguez, M. et al, 2012. Energy savings in metropolitan railway substations through regenerative energy recovery and optimal design of ATO speed
profiles. IEEE Trans. Automat. Sci. Eng. 9 (3), 496–504.
Dong, H., Ning, B., Cai, B., Hou, Z., 2010. Automatic train control system development and simulation for high-speed railways. IEEE Circuits Syst. Mag., 6–18
Dong, H., Gao, S., Ning, B., Li, L., 2013. Extended fuzzy logic controller for high speed train. Neural Comput. Appl. 22 (2), 321–328.
ERTMS, 2010. Technical Document: Baseline 3 first consolidation.
Faieghi, M., Jalali, A., Mashhadi, S.K., 2014. Robust adaptive cruise control of high speed trains. ISA Trans. 53, 533–541.
Fraszczyk, A., Mulley, C., 2017. Public perception of and attitude to driverless train: a case study of Sydney, Australia. Urban Rail Transit, 1–12.
Gao, S. et al, 2013. Approximation-based robust adaptive automatic train control: an approach for actuator saturation. IEEE Trans. Intell. Transp. Syst. 14 (4),
1733–1742.
Gao, S., Dong, H., Ning, B., Sun, X., 2015. Neural adaptive control for uncertain MIMO systems with constrained input via intercepted adaptation and single
learning parameter approach. Nonlinear Dyn. 82, 1109–1116.
Gao, Y., Kroon, L., Schmidt, M., Yang, L., 2016. Rescheduling a metro line in an over-crowded situation after disruptions. Transp. Res. Part B 93, 425–449.
González-Gil, A., Palacin, R., Batty, P., 2013. Sustainable urban rail systems: Strategies and technologies for optimal management of regenerative braking
energy. Energy Convers. Manage. 75, 374–388.
González-Gil, A., Palacin, R., Batty, P., Powell, J.P., 2014. A system approach to reduce urban rail energy consumption. Energy Convers. Manage. 80, 509–524.
Goverde, R.M.P., Besinovic, N., Binder, A., Cacchiani, V., Quaglietta, E., Roberti, R., Toth, P., 2016. A three-framework for performance-based railway
timetabling. Transp. Res. Part C 67, 62–83.
Greenway, J., Sheldon, R., 1974. Automatic train control and communications for Washington Metro. Commun. Soc. 12 (6), 14–21.
Hou, Z., Wang, Y., Yin, C., Tang, T., 2011. Terminal iterative learning control based station stop control of a train. Int. J. Control 84 (7), 1263–1274.
Howlett, P.G., 2000. The optimal control of a train. Ann. Oper. Res. 98, 1–4.
Howlett, P.G., Pudney, P.J., 1995. Energy-efficient train control. Advances in Industrial Control. Springer, London.
Huang, Y., Yang, L., Tang, T., Cao, F., Gao, Z., 2016. Saving energy and improving service quality: bicriteria train scheduling in urban rail transit systems. IEEE
Trans. Intell. Transp. Syst. 17 (12), 3364–3379.
Huang, Y., Yang, L., Tang, T., Gao, Z., Cao, F., 2017. Joint train scheduling optimization with service quality and energy efficiency in urban rail transit
networks. Energy 138, 1124–1147.
International Electrotechnical Commission, IEC 62290-1, 2014. Railway applications: urban guided transport management and command/control systems.
IEEE Standard 1474.1, 2004, Communications-based train control (CBTC) performance and functional requirements.
Karvonen, H. et al, 2011. Hidden roles of the train driver: a challenge for metro automation. Interact. Comput. 23, 289–298.
Kaviarasan, B., Sakthivel, R., Shi, Y., 2016. Reliable dissipative control of high-speed train with probabilistic time-varying delays. Int. J. Syst. Sci. 47 (16),
3940–3951.
Ke, B.-R. et al, 2010. Optimisation of train energy-efficient operation for mass rapid transit systems. IET Intell. Transp. Syst. 6 (1), 58–66.
Ke, B.-R., Chen, M.-C., Lin, C.-L., 2009. Block-layout design using MAX-MIN ant system for saving energy on mass rapid transit systems. IEEE Trans. Intell.
Transp. Syst. 10 (2), 226–235.
Ke, B.-R., Lin, C.-L., Lai, C.-W., 2011. Optimization of train-speed trajectory and control for mass rapid transit systems. Control Eng. Pract. 19, 675–687.
Khmelnitsky, E., 2000. On an optimal control problem of train operation. IEEE Trans. Automat. Control 45 (7), 1257–1266.
Kim, K., Chien, S., 2011. Optimal train operation for minimum energy consumption considering track alignment, speed limit, and schedule adherence. J.
Transp. Eng. 137 (9), 665–674.
Ko, H., Koseki, T., Miyatake, M., 2004. Application of dynamic programming to optimization of running profile of a train. Comput. Railways IX, 103–112.
Li, X., Lo, H., 2014. An energy-efficient scheduling and speed control approach for metro rail operations. Transp. Res. Part B 64, 73–89.
Li, S., Yang, L., Li, K., Gao, Z., 2014a. Robust sampled-data cruise control scheduling of high-speed train. Transp. Res. Part C 46, 274–283.
Li, S., Yang, L., Gao, Z., Li, K., 2014b. Stabilization strategies of a general nonlinear car-following model with varying reaction-time delay of the drivers. ISA
Trans. 53, 1739–1745.
Li, S., Yang, L., Gao, Z., 2015. Coordinated cruise control for high-speed train movements based on a multi-agent model. Transp. Res. Part C 56, 281–292.
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
571
Li, S., Dessouky, M.M., Yang, L., Gao, Z., 2017. Joint optimal train regulation and passenger flow control strategy for high-frequency metro lines. Transp. Res.
Part B 99, 113–137.
Liu, R., Golovitcher, I., 2003. Energy-efficient operation of rail vehicles. Transp. Res. Part A 37 (10), 917–932.
Liu, S., et al., 2015. Energy-efficient operation of single train based on the control strategy of ATO. In: IEEE International Conference on Intelligent
Transportation System (ITSC 2015), pp. 2580–2586.
Lu, S., Hillmansen, S., Ho, T.K., Roberts, C., 2013. Single-train trajectory optimization. IEEE Trans. Intell. Transp. Syst. 14 (2), 743–750.
Luthi, M., 2009. Improving the efficiency of heavily used railway networks through integrated real-time rescheduling (Doctoral dissertation). ETH ZURICH.
Maialen, L., 2012. Optimization Techniques Applied to Railway Systems (Master Thesis), Universidad del Pais Vasco.
Matsumoto, M., 2005. The revolution of train control system in JAPAN. In: Proceedings in Autonomous Decentralized Systems, ISADS IEEE, pp.599–606.
McClanachan, M., Cole, C., 2011. Current train control optimization methods with a view for application in heavy haul railways. Proc. Inst. Mech. Eng. Part F:
J. Rail Rapid Transit 225, 1–12.
Meng, L., Zhou, X., 2011. Robust single-track train dispacthing model under a dynamic and stochastic environment: a scenario-based rolling horizon
solution approach. Transp. Res. Part B 45, 1080–1102.
Meng, L., Zhou, X., 2014. Simultaneous train rerouting and rescheduling on an N-track network: a model reformulation with network-based cumulative flow
variables. Transp. Res. Part B 67, 208–234.
Meulen, R.D., 2008. Selected heavy haul insights: some South African perspectives. In: Conference on Rolling Stock of XX1 Century, pp. 117–122.
Midya, S., Thottappillil, R., 2008. An overview of electromagnetic compatibility challenges in European Rail Traffic Management System. Transp. Res. Part C
16 (5), 515–534.
Miyatake, M., Ko, H., 2010. Optimization of train speed profile for minimum energy consumption. IEEJ Trans. Electr. Electron. Eng. 5 (3), 263–269.
Miyatake, M., Matsuda, K., 2009. Energy saving speed and charge/discharge control of a railway vehicle with on-board energy storage by means of an
optimization model. IEEJ Trans. Electr. Electron. Eng. 4 (9), 771–778.
Moreno, J., Riera, J.M., Haro, L., Rodriguez, C., 2015. A survey on future railway radio communications services: challenges and opportunities. IEEE Commun.
Mag. 53 (10), 62–68.
Nielsen, L.K., Kroon, L., Maroti, G., 2012. A rolling horizon approach for disruption management of railway rolling stock. Eur. J. Oper. Res. 220, 496–509.
Ning, B., Tang, T., Qiu, C., Wang, G., Wang, Q., 2004. CTCS-Chinese Train Control System. WIT press.
Panou, K., Tzieropoulos, P., Emery, D., 2013. Railway driver advice systems: evaluation of methods, tools and systems. J. Rail Transp. Plann. Manage. 3, 150–
162.
Pellegrini, P., Marliere, G., Rodriguez, J., 2012. Real time railway traffic management modeling track-circuits. In: ATOMOS 2012, 12th Workshop on
Algorithmic Approaches for Transportation Modelling, Optimization, and Systems, pp. 23–34.
Powell, J.P., Fraszczyk, A., Cheong, C.N., Yeung, H.K., 2016. Potential benefits and obstacles of implementing driverless train operation on the Tyne and Wear
Metro: a simulation exercise. Urban Rail Transit 2 (3–4), 114–127.
Quaglietta, E., Pellegrini, P., Goverde, R.M.P., Albrecht, T., Jaekel, B., Marli??re, G., Rodriguez, J., Dollevoet, T., Ambrogio, B., Carcasole, D., Giaroli, M.,
Nicholson, G., 2016. The ON-TIME real-time railway traffic management framework: a proof-of-concept using a scalable standardised data
communication architecture. Transp. Res. Part C 63, 23–50.
Rao, X., 2015. Holistic rail network operation by integration of train automation and traffic management (Doctoral dissertation), Diss., Eidgenossische
Technische Hochschule ETH Zurich, Nr. 22706.
Rao, X., Montigel, M., Weidmann, U., 2016. A new rail optimisation model by integration of traffic management and train automation. Transp. Res. Part C 71,
382–405.
Rodrigo, E. et al, 2013. Optimizing electric rail energy consumption using the lagrange multiplier technique. J. Transp. Eng. 139 (3), 321–329.
Scheepmaker, G.M., Goverde, R.M.P., Kroon, L.G., 2017. Review of energy-efficient train control and timetabling. Eur. J. Oper. Res. 257, 355–376.
Sekine, S., Imasaki, N., Endo, T., 1995. Application of fuzzy neural network control to automatic train operation and tuning of its control rules. In: Fuzzy
Systems. International Joint Conference of the Fourth IEEE International Conference on Fuzzy Systems and The Second International Fuzzy Engineering
Symposium 4, pp. 1741–1746.
Sicre, C., Cucala, A.P., Fernández, A., Lukaszewicz, P., 2012. Modeling and optimizing energy-efficient manual driving on high-speed lines. IEEJ Trans. Electr.
Electron. Eng. 7 (6), 633–640.
Song, Q., Song, Y.D., 2011. Data-based fault-tolerant control of high-speed trains with traction/braking notch nonlinearities and actuator failures. IEEE Trans.
Neural Netw. 22 (12), 2250–2261.
Song, Q., Song, Y.D., Tang, T., Ning, B., 2011. Computationally inexpensive tracking control of high-speed trains with traction/braking saturation. IEEE Trans.
Intell. Transp. Syst. 12 (4), 1116–1125.
Song, Y., Song, Q., Cai, W., 2014. Fault-tolerant adaptive control of high-speed trains under traction/braking failures: a virtual parameter-based approach.
IEEE Trans. Intell. Transp. Syst. 15 (2), 737–748.
Su, S., Li, X., Tang, T., Gao, Z., 2013. A subway train timetable optimization approach based on energy-efficient operation strategy. IEEE Trans. Intell. Transp.
Syst. 14 (2), 883–893.
Su, S., Tang, T., Robert, C., 2015. A cooperative train control model for energy saving. IEEE Trans. Intell. Transp. Syst. 16 (2), 622–631.
Sun, H., Hou, Z., Li, D., 2013. Coordinated iterative learning control schemes for train trajectory tracking with overspeed protection. IEEE Trans. Automat. Sci.
Eng. 10 (2), 2013.
Tang, T., Huang, L., 2003. A survey of control algorithm for automatic train operation. J. China Railway Soc. 25 (2), 98–102 (In Chinese).
Tang, H., Tyler Dick, P.E., Feng, X., 2015. A coordinated train control algorithm to improve regenerative energy receptivity in metro transit systems. Transp.
Res. Rec.: J. Transp. Res. Board, 15–1318.
Technical report: Automatic train control system update, 2013. Washington Metropolitan Area Transit Authority Board Action/Information Summary.
<https://www.wmata.com/about_metro/board_of_directors/board_docs/061313_SSATC20130613.pdf>.
Tornquist, J., 2007. Railway traffic disturbance management – an experimental analysis of disturbance complexity, management objectives and limitations
in planning horizon. Transp. Res. Part A 41 (3), 249–266.
Transportation Safety Advancement Group (TSAG), 2010. Shaping the 30-year Rail Technical Strategy. Available at: <http://www.i-n-w.org/
talking_collaboration/further_reading/Consultation%20RTS%20TSAG.pdf>.
UIC, 2000a. Euroradio FFFIS, EEIG ERTMS User Group. UIC, Brussels.
UIC, 2012. ATO and ETCS can work together, V Kampík. UIC, Stockholm.
UITP, 2011. Technical report: Metro automation facts, figures and trends – a global bid for automation: UITP observatory of automated metros confirms
sustained growth rates for the coming years. International Association of Public Transport (UITP).
UITP, 2014. Technical report: Metros: Keeping pace with 21st century cities. Available at: <http://www.uitp.org/metros-keeping-pace-21st-century-cities>.
UITP, 2015. Technical report: Thameslink: Faster and more comfortable service right across London, Siemens. International Association of Public Transport
(UITP). Available at: <http://www.siemens.com/press/pool/de/events/2015/mobility/2015-06-uitp/background-thameslink-e.pdf>.
Wang, P., Goverde, R.M.P., 2016. Multiple-phase train trajectory optimization with signalling and operational constraints. Transp. Res. Part C 69, 23–50.
Wang, Y., De Schutter, B., van den Boom, T.J.J., Tang, T., Ning, B., 2013. Optimal trajectory planning for trains – a pseudospectral method and a mixed integer
linear programming approach. Transp. Res. Part C 29, 97–114.
Wang, Y., Tang, T., Ning, B., van den Boom, T.J.J., Schutter, B., 2015. Passenger-demands-oriented train scheduling for an urban rail transit network. Transp.
Res. Part C 60, 1–23.
Wang, Y., Zhang, M., Ma, J., Zhou, X., 2016. Survey on driverless train operation for urban rail transit systems. Urban Rail Transit 2 (3–4), 106–113.
Wong, K.K., Ho, T.K., 2004. Coast control for mass rapid transit railways with searching methods. IEE Proc.-Electr. Power Appl. 151 (3), 365–376.
572
J. Yin et al. / Transportation Research Part C 85 (2017) 548–572
Wong, R., Yuen, T., Fung, K., Leung, J., 2008. Optimizing timetable synchronization for rail mass transit. Transp. Sci. 42 (1), 57–69.
Xu, X., Li, K., Yang, L., 2015. Scheduling heterogeneous train traffic on double tracks with efficient dispatching rules. Transp. Res. Part B 78, 364–384.
Yan, X., Cai, B., Ning, B., Shang-Guan, W., 2016. Online distributed cooperative model predictive control of energy-saving trajectory planning for multiple
high-speed train movements. Transp. Res. Part C 69, 60–78.
Yang, C.-D., Sun, Y.-P., 1999. Robust cruise control of high speed train with hardening/softening nonlinear coupler. American Control Conference (ACC), vol.
3, pp. 2200–2204.
Yang, C.-D., Sun, Y.-P., 2001. Mixed H2/H1 cruise controller design for high speed train. Int. J. Control 74 (9), 905–920.
Yang, H., Fu, Y.-T., Zhang, K.-P., Li, Z.-Q., 2014a. Speed tracking control using an ANFIS model for high-speed electric multiple unit. Control Eng. Pract. 23, 57–
65.
Yang, L., Zhou, X., Gao, Z., 2014b. Credibility-based rescheduling model in a double-track railway network: a fuzzy reliable optimization approach. Omega
48, 75–93.
Yang, X., Chen, A., Li, X., Ning, B., Tang, T., 2015. An energy-efficient scheduling approach to improve the utilization of regenerative energy for metro
systems. Transp. Res. Part C 57, 13–29.
Yang, X., Li, X., Ning, B., Tang, T., 2016. A survey on energy-efficient train operation for urban rail transit. IEEE Trans. Intell. Transp. Syst. 17 (1), 2–13.
Yasunobu, S., Miyamoto, S., 1985. Automatic train operation system by predictive fuzzy control. Ind. Appl. Fuzzy Control 1 (18), 1–18.
Yasunobu, S., Miyamoto, S., Ihara, H., 1983. Fuzzy control for automatic train operation system. In: Proc. 4th. IFAC/IFIP/IFORS Int. Congress on Control in
Transportation Systems, Baden-Baden, pp. 33–39.
Yasunobu S., Saitou S., Suryana Y., 2000. Intelligent vehicle control in narrow area based on human control strategy. In: Proc World Multi Conf Systemics
Cybernetics and Informatics, vol. 7, pp. 309–314.
Ye, H., Liu, R., 2016. A multiphase optimal control method for multi-train control and scheduling on railway lines. Transp. Res. Part B 93, 377–393.
Yin, J., Chen, D., Li, L., 2014. Intelligent train operation algorithms for subway by expert system and reinforcement learning. IEEE Trans. Intell. Transp. Syst.
15 (6), 2561–2571.
Yin, J., Chen, D., Yang, L., Tang, T., Ran, B., 2016a. Efficient real-time train operation algorithms with uncertain passenger demands. IEEE Trans. Intell. Transp.
Syst. 17 (9), : 2600–2612.
Yin, J., Chen, D., Li, Y., 2016b. Smart train operation algorithm based on expert knowledge and ensemble CART for the electric locomotive. Knowl.-Based Syst.
92, 78–91.
Yin, J., Tang, T., Yang, L., Gao, Z., Ran, B., 2016c. Energy-efficient metro train rescheduling with uncertain time-variant passenger demands: an approximate
dynamic programming approach. Transp. Res. Part B 91, 178–210.
Yin, J., Yang, L., Tang, T., Gao, Z., Ran, B., 2017. Dynamic passenger demand scheduling with energy-efficiency and waiting time minimization: mixed-integer
linear programming approaches. Transp. Res. Part B 97, 182–213, 201.
Zhan, S., Kroon, L.G., Veelenturf, L.P., Wagenaar, J.C., 2015. Real-time high-speed train rescheduling in case of a complete blockage. Transp. Res. Part B 78,
182–201.
Zhao, N., Roberts, C., Hillmansen, S., Nicholson, G., 2015. A multiple train trajectory optimization to minimize energy consumption and delay. IEEE Trans.
Intell. Transp. Syst. 16 (5), 2363–2372.
Zhou, X., Zhong, M., 2005. Bicriteria train scheduling for high-speed passenger railroad planning applications. Eur. J. Oper. Res. 167, 752–771.
Zhou, X., Zhong, M., 2007. Single-track train timetabling with guaranteed optimality: Branch-and-bound algorithms with enhanced lower bounds. Transp.
Res. Part B 41, 320–341.
Zhou, L., Tong, L., Chen, J., Tang, J., Zhou, X., 2017. Joint optimization of high-speed train timetables and speed profiles: a unified modelling approach using
space-time-speed grid networks. Transp. Res. Part B 97, 157–181.
Zhuan, X., Xia, X., 2008. Speed regulation with measured output feedback in the control of heavy haul trains. Automatica 44 (1), 242–247.
Zimmermann, A., Hommel, G., 2005. Towards modeling and evaluation of ETCS real-time communications and operation. J. Syst. Softw. 77, 47–54.