213
Chapter 10
Cognitive Control for
Communications-Based
Train Control Systems
Hongwei Wang and F. Richard Yu
Contents
10.1 Introduction ..........................................................................................214
10.2 Overview of Cognitive Control .............................................................216
10.2.1 Cognitive Control Approach to CBTC Systems ......................217
10.3 Cognitive Control .................................................................................218
10.4 Formulation of Cognitive Control Approach to CBTC Systems ............ 221
10.4.1 Train Control Model ...............................................................221
10.4.2 Channel Model in MIMO-Enabled WLANs ..........................223
10.4.3 Q-Learning in the Cognitive Control Approach ......................225
10.4.3.1 System States and Actions ......................................225
10.4.3.2 Reward Function ................................................... 226
10.5 Simulation Results and Discussions ....................................................... 231
10.5.1 Parameters of Train Dynamics ................................................231
10.5.2 Parameters of the ATO ............................................................232
10.5.3 Parameters of the Wireless Channel ........................................233
10.5.4 Simulation Results and Discussions ......................................... 233
10.6 Conclusion ............................................................................................242
References .........................................................................................................243
214 ◾ Advances in Communications-Based Train Control Systems
10.1 Introduction
Urban rail transit systems have developed rapidly around the world in the recent
past. Due to the huge urban trac pressure, improving the eciency of urban rail
transit systems is in high demand. As a key subsystem of urban rail transit systems,
communications-based train control (CBTC) is an automated train control system
using train–ground communications to ensure the safe and ecient operation of
rail vehicles [1]. CBTC can improve the utilization of railway network infrastruc-
ture and enhance the level of service oered to customers [2].
As urban rail transit systems are built in a variety of environments (e.g., under-
ground tunnels, viaducts, etc.), there are dierent wireless network congurations
and propagation schemes. For tunnels, the free space is generally adopted as the
propagation medium. However, the leaky coaxial cable is also an option, such as
Tianjin Subway lines 1 and 2 built by Bombardier. For the viaduct scenarios, leaky
rectangular waveguide is a popular approach, as it can provide higher performance
and stronger anti-interference ability than the free space [3]. In addition, due to the
available commercial o-the-shelf (COTS) equipment, wireless local area networks
(WLANs) are often adopted as the main method of train–ground communications
for CBTC systems [4].
Building a train control system over wireless networks is a challenging task. Due
to unreliable wireless communications and train mobility, the train control perfor-
mance can be signicantly aected by wireless networks [5]. Because CBTC systems
are safety critical, trains usually run according to the front train’s state, including
velocity and position. When a wireless network brings large communication latency
caused by unreliable wireless communications or handos, the current train may
not be able to obtain the accurate state information of the front train, which would
severely aect train operation eciency, or even cause train emergency brake.
e performance issues in railway environments have attracted a lot of interest
recently. A fast hando algorithm suitable for passenger lines is proposed in [6] by
setting a neighboring cell list to facilitate hando operations. In [7], a novel hand-
o scheme based on on-vehicle antennas is introduced. e authors of [8] propose
a cross-layer hando design in for multiple-input and multiple-output (MIMO)-
enabled WLANs in CBTC systems. In [9], energy-ecient train control schemes
are studied in CBTC systems.
Although these above excellent works have been done to study CBTC systems
from both train–ground communication and train control perspectives, these two
important areas have traditionally been addressed separately in the CBTC literature.
However, as shown in the following, it is necessary to jointly consider these two
advanced technologies together to enhance the level of safety and services in CBTC
systems. e motivation behind our work is based on the following observations.
◾ Most existing CBTC systems are based on the COTS equipment, in which
traditional design criteria (e.g., network capacity) are used in the design and
Cognitive Control for CBTC Systems ◾ 215
conguration of these systems. Although traditional design criteria are suitable
for commercial networks (e.g., hot spots for Internet access), they may not
be suitable for CBTC systems due to their specic characteristics, such as
stringent requirements for communication availability and latency.
◾ Recent studies in cross-layer design show that maximizing network capacity
does not necessarily benet the application layer [10,11], which is train con-
trol in CBTC systems. From a CBTC perspective, the performance of train
control is more important than that of other layers.
◾ Most train control schemes in existing CBTC systems assume perfect train–
ground communication. However, random transmission delays and packet
drops are inevitable in train–ground communications, which will signi-
cantly aect train control performance in CBTC systems [12].
In this chapter, we propose to jointly study train–ground communication and train
control so as to improve CBTC performance. e distinct features of this work are
as follows:
◾ With recent advances in cognitive dynamic systems [13,14], we take a cog-
nitive control approach to CBTC systems considering both train–ground
communication and train control. Recently, cognitive dynamic systems have
emerged as a new engineering discipline, which builds on ideas in statistical
signal processing, stochastic control, and information theory. is discipline
has been successfully used in the design of dynamic systems (e.g., cognitive
radio and cognitive radar) with eciency, eectiveness, and robustness being
the hallmarks of performance [13].
◾ In our cognitive control approach, the notion of information gap [14] is
adopted to quantitatively describe the eects of train–ground communica-
tion on train control performance. Specically, as train–ground communi-
cation is used to exchange information between the train and the control
center, packet delay and drop lead to information gap, which is the dierence
between the actual state and the observed state of the train.
◾ Unlike the existing works that use network capacity as the design measure, in
this chapter, linear quadratic cost for the train control performance in CBTC
systems is considered in the performance measure. Reinforcement learning
(RL) is applied to obtain the optimal policy based on the performance mea-
sure, which includes linear quadratic cost and information gap.
◾ e wireless channel is modeled as nite-state Markov chains with multiple
state transition probability matrices, which can demonstrate the character-
istics of both large-scale fading and small-scale fading. e channel state
transition probability matrices are derived from real eld measurement
results.
◾ Simulation results show that the proposed cognitive control approach can
signicantly improve the train control performance in CBTC systems.
216 ◾ Advances in Communications-Based Train Control Systems
e rest of this chapter is organized as follows: Section 10.2 gives the introduction
of cognitive control. Section 10.3 describes the cognitive dynamic systems and the
RL. Section 10.4 discusses the models. Section 10.5 presents the optimal solu-
tions, simulation results, and some discussions. Finally, Section 10.6 concludes the
chapter.
10.2 Overview of Cognitive Control
Cognitive control was originally developed in neuroscience and psychology (e.g.,[15]).
Recently, it has emerged as a new engineering discipline [14]. e basic diagram of
a cognitive control system is shown in Figure 10.1.
Compared with other control methods, such as adaptive control [16] and
neuro-control [17], cognitive control has the following advantages: ere is no
memory block in the adaptive controller, which reduces the ability of learning.
e neuro-controller lacks intelligence, which is distributed throughout the cogni-
tive dynamic system and can make the system in an orderly fashion. e concept
of cognitive control has been successfully applied in cognitive radar systems and
cognitive radio systems [18]. According to Figure 10.1, the feedback information
Sensory
measurement
Information
processing
Determine
information gap
Get
relevant part
Select
optimal actions
System
Environment
Perceptual
part
Executive
part
Action
Figure10.1 Basic schematic structure of a cognitive control system.
Cognitive Control for CBTC Systems ◾ 217
plays the key role in a cognitive dynamic system, and cognitive control describes
a control system from the information ow perspective. e feedback informa-
tion obtained by the perceptual part is partitioned into relevant information and
redundant information. However, the required information is used to make correct
decisions, which is called sucient information in cognitive control. e informa-
tion gap is dened as the dierence between sucient information and relevant
information obtained from the measurements, and the basic procedure of cognitive
control is shown in Figure 10.2. e goal of cognitive control is to decrease the
information gap.
10.2.1 Cognitive Control Approach to CBTC Systems
Cognitive control can be applied in CBTC systems to jointly consider both train–
ground communication and train control. e information gap of CBTC systems
can be dened according to the train control procedure, in which the current train
is controlled according to the information of the front train. In CBTC systems,
the train is controlled by the command generated from automatic train operator
(ATO). At each control cycle, ATO determines the decision of the train operation
according to the automatic train protection (ATP) emergency braking prole and
the location of movement authority (MA), which means that the ATO model is
also necessary. As cognitive control improves the train eciency through select-
ing the optimized communication policy, the channel model of CBTC operation
environment is important, where the accuracy of channel model directly aects the
accuracy of control. erefore, adopting cognitive control in CBTC systems needs
the dynamic model, the ATO model, and the channel model.
Based on the principles of CBTC train control, whether the MA is transmitted
timely decides the performance of whole CBTC system. MA is the basis for ATO
and ATP decisions, which is generated from zone controller (ZC) according to the
state information of the front train. An MA is generally dened as a physical point
Feedback information
Action
Sensory
measurements
Perceptual part
Estimator
Perceptual memory
Executive part
Cognitive controller
Executive memory
Action
Environment
System
Figure10.2 Basic procedure of a cognitive control system.
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.