Vous êtes sur la page 1sur 14

Transportation Research Part C 35 (2013) 127140

Contents lists available at SciVerse ScienceDirect

Transportation Research Part C


journal homepage: www.elsevier.com/locate/trc

The impact of end-to-end communication delay on railway


trafc ow using cellular automata model
Jing Xun a,, Bin Ning a, Ke-ping Li a, Wei-bin Zhang b
a

State Key Laboratory of Rail Trafc Control and Safety, Beijing Jiaotong University, Beijing, China
California PATH Program, Institute of Transportation Studies, University of California, Berkeley, Richmond Field Station, Bldg. 452, 1357 S.46th Street,
Richmond, CA 94804-4648, USA
b

a r t i c l e

i n f o

Article history:
Received 16 March 2010
Received in revised form 18 June 2013
Accepted 21 June 2013

Keywords:
Communication-based train control system
End-to-end communication delay
Trafc ow
Cellular automata

a b s t r a c t
Increasing number of railroads have begun to operate with a moving block system that
relies on a data communication system to transmit real-time information such as train
position and velocity. The end-to-end communication is a critical component to ensure
the efciency and the safety of train operation. The delay in the end-to-end communication
affects every aspect of the specication and operation of the moving block system and
therefore deserves special attention. In this paper, an improved cellular automata (CA)
model is proposed as a tool to perform analysis of this delay issue through dening neighborhoods in the cellular automata railway trafc model. Based on this model, simulation of
trafc ow in a transit system is conducted and the results are presented to demonstrate
the dependability of the proposed model. The study shows that, if the end-to-end communication delay satises the requirement described in the Euroradio specication, the
throughput of railway line (in number of trains per hour) is nearly the same in the simulation scenarios when running the 3 schemes we derived. However the inuence on each
individual train is not negligible. As one of delay factors, d3 should be kept as small as possible to minimize the headway time.
2013 Elsevier Ltd. All rights reserved.

1. Introduction
In conventional railway signal systems, trains can only operate in xed blocks, which divide tracks into segments and
trains can only enter a block when one or more blocks ahead are clear depending on the operating speeds, block length and
the signal control scheme. In a xed block system, the headway is limited physically by blocks. Modern communicationbased train control (CBTC) systems use data communication over a variety of paths to gather movement data of the train
ahead and to safely implement the principle of the moving block based on real-time train speed, track geometry, direction
of movement, required safe braking distances, and stopping characteristics. In the moving block system (MBS), the distance
between trains no longer depends on the length of blocks and can be reduced to the braking distance with a safety distance.
For passenger railway systems, MBS allows more frequent and shorter trains to increase track capacity by allowing trains to
have smaller headways while maintaining the required safety margins and therefore reducing the waiting time. This can
substantially increase the efciency and the safety of the overall rail system. From these advantages, a signicant number
of MBS systems have been deployed worldwide in recent years, such as Londons Docklands Light Railway, New Yorks L Line
and Beijings Subway Line 2.
Corresponding author. Tel.: +86 010 51688004; fax: +86 010 51685196.
E-mail addresses: xunjing_0@hotmail.com (J. Xun), bning@center.bjtu.edu.cn (B. Ning), rtkpli@188.com (K.-p. Li), wbzhang@path.berkeley.edu
(W.-b. Zhang).
0968-090X/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.trc.2013.06.008

128

J. Xun et al. / Transportation Research Part C 35 (2013) 127140

Under the MBS, computers calculate a safe zone, dened as Movement Authority (MA), for each moving train. Each train
must not exceed the boundary of this zone and moreover, no other train is allowed to enter into this zone. The information
on the MA is sent to each train by the data communication system (DCS). The DCS is a wireless communication system and
will incur dynamic signal degradation due to factors such as thermal noise, contention, multi-path and signal blockage. A
kind of degradation is an increase of end-to-end communication delay.
The end-to-end communication delay (ECD) refers to the length of time passing between initiation of a communication
message at the point of origin and when it is received at the destination system. This message delay can limit the train
throughout. The objective for a train is to receive its next MA update before it gets within braking distance plus warning time
(distance) of the limit of its current authority. If the DCS delivers an MA update too late, the associated train may have to
reduce speed (Polivka et al., 2009; Zhu et al., 2012). The unnecessary braking results in a disturbance of the optimal speed
prole and a decrease of the line capacity, especially when trains run at high speeds and the communication delays are very
long. Consequently, studying the impact of the communication delay on the trafc ow is an important consideration for the
development of MBS.
The existing models for MBS lack the description of the ECD. Therefore in this paper, we propose a CA model which incorporates the ECD parameter. By considering the ECD, the incorporation offers a way to simulate the railway trafc. The simulation will help to understand the impact of the ECD on the trafc ow.
The paper is organized as follows: In Section 2 the ECD in train control is introduced. In Section 3 the CA models for railway trafc are reviewed. After that, the proposed model is described in Section 4. The denitions of cell state, neighborhoods,
boundary and rules are given. Section 5 presents the simulation results with the proposed model. Section 6 provides an analysis of the effect of the ECD upon system performance. In Section 7, the conclusions are given.
2. End-to-end communication delay in train control
With the advances in wireless communication development in the past years, radio communication has been adopted as a
promising way to transmit data between train and track-side infrastructure. During train movement, a train will report its
position to a center periodically. The center uses this information to generate the MA message and then transmit to the following train. Based on the MA message, the limit speed is computed at time step t + communicationdelay (see Fig. 1(a) and (b)
for an illustration). Two problems can be found as the ECD increases:
(1) Lost capacity in sections of railway line: At time step t + communicationdelay, the limit speed is calculated according to
the MA information at time step t. However, actually the rear end of the preceding train has been changed because the
preceding train has run communication delay(s)  train_speed meters forward. This distance, which increases with the
increase of ECD, cannot be utilized because the rear end is still the one at time step t to the following train.
(2) Unnecessary Disturbance to the speed prole of the following train: When a MA message is delayed by the ECD, the
following train, for safety reasons, needs to decide on when it brakes according to the previous MA message. If the ECD
is long enough and the two trains run closely enough, the following train i would be slowed down in order to avoid

Fig. 1. A scenario of the difference between train movement with considering ECD and without considering ECD.

J. Xun et al. / Transportation Research Part C 35 (2013) 127140

129

exceeding the boundary of its MA. For instance, its speed is probably reduced from 30 m/s to 28 m/s when the gap is
500 m. If the following train does not want to reduce speed, it has to keep further from the preceding train. This will
decrease the line capacity.
The two problems harm the safety and the efciency of train operation. Consequently, it is important to investigate the
impact of ECD. Studies have been conducted to analyze communication delays for CBTC systems. Zimmermann rst presented a simplied model of communication failure and recover behavior as well as safety-critical data exchange (Zimmermann and Hommel, 2003). He also evaluated the performance of the model and the evaluation results show the signicant
impact of packet delays and losses on the reliable operation of high-speed trains (Zimmermann and Hommel, 2005). Xu evaluated the inuence of communication collision and message length on the transfer delay by using a Petri Net model (Xu and
Tang, 2007). The Petri Net Model used in these studies can evaluate the reliability of train operation. These studies focus on
the impact of the communication delay on the safety and reliability of train operation. However, few studies of the impact on
the trafc ow have been done. Consequently, we proposed a Cellular Automata model to simulate the train movement
while considering the ECD. And the impact of the ECD on the trafc ow is analyzed.
3. Cellular Automata model for railway trafc
Cellular Automata (CA) is a tool for simulating trafc ow and modeling a trafc network. The notion of CA was initially
conceived by John Von Neumann in the late 1940s (Burks, 1970). Nagel and Schreckenberg developed a one-dimensional
probabilistic CA model, i.e. NaSch model, which is a model of trafc ow on a single-lane (Nagel and Schreckenberg,
1992). In NaSch model, the road is divided into L cells numbered by i = 1, 2, . . . , L, and time is discrete. Each site can be either
empty or occupied by a vehicle with integer speed Vi = 0, 1, . . . , Vmax, where Vmax is the maximum speed. Xi(t) is the position
where vehicle i is at time t, and gapi(t) = Xi+1(t)  Xi(t)  1 expresses the gap between vehicle i and i + 1 at time t. The underlying dynamics of NaSch model are governed by the updated rules applied at discrete time steps. All sites are simultaneously
updated according to four successive steps:
(1) Acceleration:

V i t 1 ! minV i t 1; V max ;
(2) Slowing down:

V i t 1 ! minV i t 1; gapi t;
(3) Randomization:

V i t 1 ! maxV i t  1; 0; (Decrease Vi(t) by 1 with randomization probability p if Vi(t) > 0);


(4) Movement:z

X i t 1 ! X i t V i t 1:
After Nagel and Schreckenberg developed the rst single-lane trafc CA model, scholars investigated the application of CA
in the transportation eld and several models based on CA were developed (Nassab et al., 2006; Li et al., 2006; Lan and Hsu,
2006).
From the NaSch model, the rst CA railway trafc model is obtained by Li et al. (2005) and Ning et al. (2005). In their
study, the NaSch model was extended to simulate selected complex phenomena of train-following. The simulation results
demonstrated that this extended NaSch model can be successfully used for simulating railway trafc ow. Later, Zhou simulated the trafc phenomenon of delay propagation using a CA trafc model for moving-like block systems (Zhou et al.,
2006). In moving-like block systems, the minimum length of the location unit (such as track circuit) is a critical factor. Its
impact on the trafc ow is investigated. Based on the NaSch model, a CA trafc model for xed-block systems was proposed
by Li et al. (2007). In this study, analysis was conducted on how the length of speed-limited sections, speed-limit value and
train time interval affect the trafc ow in the xed block system. A railway network CA model and station model was proposed by this author and his colleagues (Xun et al., 2007). They simulate the trafc ow with the variation of the train departure time interval at original station by using this model. Table 1 lists 14 parameters, some of them have been used in the
previous studies.
In the previous studies, the movement of train, including acceleration and deceleration, is achieved by the rules in their
models. The rules will enforce the gap between the two successive trains larger than the minimum instantaneous distance in
order to avoid collision (see Fig. 1(c)). At time step t + communication delay, the gap is based on the current information, i.e.,
the current position of train j can be known by the train i. The transmission of data is abstracted, i.e. once the state of the
system is changed, it will be updated in the computation immediately. This abstraction facilitates the simulation and, at
the same time, makes the model away from the real description of the data transmission between train and track-side.

130

J. Xun et al. / Transportation Research Part C 35 (2013) 127140

Table 1
The parameters in the previous CA models for railway trafc.
Type of model

Fixed block system

Moving-like block system

Moving block system

Maximum speed of train


Maximum acceleration of train
Maximum deceleration of train
Length of train
Mix of train type
Distance between stations
Multi-platform at station
Dwell time at station
Train departure time interval at original station
Limit speed on line
Safe distance
Length of block
Length of location unit
End-to-end communication delay

Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y

Y
Y
Y
Y
Y
Y
N
Y
Y
Y
Y

Y
N

Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y

The shortcoming of the description makes the model lose the capability of evaluating the impact of communication delay on
railway trafc ow.
The train movement described by the previous models is simplied. As a result, the train movement accuracy, to a certain
degree, is reduced (Li et al., 2005). In order to describe more detailed behavior e.g., the ECD, the set of fundamental rules has
to be expanded.
In next section, CA railway trafc model is expanded by redening the state, neighborhoods, rules and boundary. The
track with L meters long is represented by L cells. Cells state, which includes ECD as a parameter, can identify each train.
Each train could departures rst cell and moves forward based on the calculation of its speed and coordinate at next time
step. One trains speed and coordinate is calculated depending on the states of some nearby cells, the neighborhoods, as
determined by rules. In rules, ECD is considered. All cells use the same rules so that the system is homogeneous. Finally,
according to the boundary, trains move out of the system at Lth cell.
4. Development of the Proposed CA model for MBS
Before describing the proposed model, we need to dene related parameters. In this research, the computer generates an
MA for each train, which is dened as the distance between the front of a train and the rear of the preceding train or the
entrance of a station (we simplify the algorithm of the MA here).
For the proposed model, the state, neighborhoods, rules and boundary condition are four key parameters to be identied.
They are described below.
4.1. State
We dene the state of the i th cell at time step t by:

8
9
X i t >
>
>
>
>
>
>
>
>
>
>
< V i t >
=
bi t TYPi t ;
>
>
>
>
>
> C i t >
>
>
>
>
>
:
;
SLi t
where Xi(t) is the coordinate of the train at the i th cell at time step t. Vi(t) is the speed of the train at the ith cell at time step t.
It can be among null, 0, 1, . . . , Vmax, where Vmax is the maximum speed of the train. Here null means there is no train occupying this cell. TYPi(t) is the type of train that is numbered with the maximum speed for train operation, e.g., TYPi(t) = 20 if
the maximum speed for train operation is 20. In the simulation TYPi(t) equals Vmax if there is no special statement. Ci(t) is the
ECD when the data is transferred between the train and track-side at the ith cell at time step t. SLi(t) is the limit speed of the
line at the ith cell at time step t. Generally it is specied by the railway system operator.
4.2. Neighborhoods
In a CA model, the number of neighborhoods of a given cell is nite. The state of a cells neighborhoods and its own state
will jointly decide the state of this cell at the next time step. In the proposed model, there are two situations shown as
follows:

J. Xun et al. / Transportation Research Part C 35 (2013) 127140

131

 There is an MA region for a train. when a cell is occupied by the front end of a train, the cell state at the next time step is
related with all cells within the region of the MA.
 When a cell is not occupied by the front end of a train, this cells state at the next time step is decided by itself.
As a result, the neighborhoods of a cell i which is denoted as Ni(t), changes at different time steps. Fig. 2 illustrates the
neighborhoods of Cell A. Cell A is occupied by Train i at time step t. The MA region of Train A is the area between the front
of train i and the rear of the preceding Train j (excludes the safe distance Ls). According to the above denition, the neighborhoods of Cell A NA(t) include Cells B, C and D.

4.3. Rules
Based on Li et al. (2005), the motion of trains in the proposed model is described by the following rules:
(1) Acceleration: Vi(t + 1)
min (Vi(t) + a, Vmax)
(2) Deceleration: V i t 1
minV i t a; SLi t; TYP i t; V MA
i t
(3) Movement: Xi(t + 1) = Xi(t) + Vi(t + 1)s
where Vi(t) = ith train speed at time step t; Xi(t) = coordinate of the i th train at time step t; a = maximum acceleration;

s = time step interval; V MA


i t = a limit speed which is related to the MA message at time step t (detailed explain in next
paragraph).
Compared our model to the NaSch model, there are two differences. Firstly, the step 3 adopted in NaSch model is ignored,
i.e., the randomization probability p is p = 0. Secondly, in the step 2 trains should be slow down when its speed exceeds the
minimum among the limit speed SLi(t), the maximum speed for train operation TYPi(t) and the variable V MA
i t. The variable
V MA
i t denotes the limit speed related to the MA message at time step t. The MA message includes the state information of
cells in the MAs region. In their state information, there are coordinate Xj(t) and limit speed SLj(t) at these cells. According to
one of the equations of rectilinear motion Eq. (1), a relative limit speed Rj can be deduced at Xi(t).

m2 u2 2as;

where u = starting velocity (m/s), v = nal velocity (m/s), s = displacement (m), a = acceleration (m/s2).
When considering ECD, u equals SLj(t  Ci(t)), a equals b and s equals Xj(t  Ci(t))  Xi(t  Ci(t))  LT  Ls. Rj can be calculated by Eq. (2). Because the MA is sent to the train with an ECD, the MA information used by the onboard computer is not the
one at time step t but at time step t  Ci(t). The minimum in these several relative limit speed values is V MA
i t, i.e., Eq. (3).

Rj

q
SLj t  C i t2 2bX j t  C i t  X i t  C i t  LT  Ls ;

V MA
i t minfR1 ; . . . ; Rj ; . . .g;

CelljNi t  C i t;

2
3

where Xi(t) = coordinate of the i th train at time step t; Xj(t) = coordinate of one cell in the neighborhoods of the i th train at
time step t; b = maximum deceleration; LT = train length; Ls = safe distance; Ci(t) is the ECD for the i th train at time step t.
According to the rules, the update steps for the proposed model are illustrated with a simple example in Fig. 3. In Fig. 3,
steps 3 and 4 describe how trains move from one cell to another. When the cell is occupied by the front of a train, it will
update its own state based on its current state and neighborhoods states. After that, a new coordinate of the train will be
obtained. Then the front end of the train will be put into the cell corresponding to the coordinate. The other parts of the train
are put into the successive cells along the inverse of train running direction.
It depends on its schedule if a train were dispatched or not. When a train stays at station for a dwell time, it can leave the
station. So when the time for a train staying at station is less than the dwell time, the cell occupied by the train will not
update.

Fig. 2. Illustration of the neighborhoods.

132

J. Xun et al. / Transportation Research Part C 35 (2013) 127140

Fig. 3. An example of update steps.

4.4. Boundary
In our method, the boundary condition for the CA model is open. It is dened as:
(1) When the section from the rst cell to the Lsth cell is empty, a train with 0 m/s speed will be created if the amount of
time between the preceding trains departure time and current time is more than an interval I, which is specied
before simulation. Then this train moves forward according to the update rules.
(2) At Lth cell, where L is the total length of the railway line, the train simply moves out of the system.
Based on the above denitions, we use the proposed model below to simulate several scenarios of railway trafc ow and
discuss the effect of the ECD on trafc ow in the next section.
5. Simulation
The typical acceleration of a train approximates 1 m/s2. So we assign the iteration s to 1 s, and the length of a cell is
approximately 1 m. This means, e.g., that Vmax = 20 cell/update corresponds to Vmax = 20 m/s.
As we mentioned before, capacity loss and optimal speed prole disturbance are two reasons to drive us study ECD. By
using the proposed model we can duplicate these two scenarios and analysis what effect it will have on the following trains
speed prole and how much capacity is lost by considering ECD.
The comparison of train travel trajectories with considering ECD or not are shown in Fig. 4. The rst, second, and third
trains are denoted as Train 1, 2, and 3, respectively, in our simulation. The rst train travels without obstacles since there
is no train traveling in front of it. The distance between two stations is 2950 m. The reasons to choose 2950 m are:
(a) When the maximum speed is 50 m/s and the acceleration rate is 1 m/s2, the length of 450 m is needed so that trains
can speed up to the maximum speed. In addition, the braking distance is 450 m when trains run at 30 m/s and the
deceleration rate is 1 m/s2. Then the distance between stations must be longer than 900 m.
(b) The distance between stations is usually a few kilometers long in real world.
The other parameters used in the simulation are as follows:
(1)
(2)
(3)
(4)
(5)
(6)

Train acceleration a = 1 m/s2.


Train deceleration b = 1 m/s2.
Train length LT = 100 m.
Safe distance Ls = 60 m.
Maximum speed of train Vmax = 30 m/s.
Speed limit of line SLi = 30 m/s, i e (1, L).

J. Xun et al. / Transportation Research Part C 35 (2013) 127140

133

Fig. 4. The comparison of train travel trajectories with/without considering ECD.

(7) Interval of the trains departure time at station A I = 60 s (I is a variable when we calculate the minimum time
headway).
(8) Dwell time at the station B and C Tdw = 40 s.
(9) Length of the simulated model time is taken as T = 2000 s.
Except for the above parameters, there is still another parameter, ECD, which needs to be given during the simulation. The
ECD distribution is measured in an 802.11 wireless Mesh Network Test-bed (Najah et al., 2008). The measurement reveals
that the best ts for nearly 90% of its empirical distribution are two distributions: gamma and logistic. The probability density and distribution functions of the gamma distribution are

f x xk1

ex=h
hk Ck

and Fx

ck; x=h
Ck

; respectively:

The probability density and distribution functions of the logistic distribution are

f x

exl=s
s1 exl=s

and Fx

1
; respectively:
1 exl=s

Moreover, in the Euroradio specication (EEIG ERTMS User Group, 2003), which species the radio communication system requirements to the air gap interface between train and trackside equipment, the ECD for a message is at most 0.5 s with
95% condence, at most 1.2 s with 99% condence and at most 2.4 s with 99.99% condence in all cases. Following this specication, we can approximate the value of necessary parameters k  0.7 and h  0.2 for the gamma distribution and the value
of necessary parameters l  0.19 and s  0.1 for the logistic distribution. In the simulation and discussion section, we will
use the gamma distribution unless there are special reasons for using other distributions.
From Fig. 4, given the same train departure time interval, the travel trajectories of Train 2 and 3 have many unnecessary
accelerations and decelerations when considering ECD. Train 3s trajectory looks more far away from the ideal trajectory
(without considering ECD) than train 2. Thats an accumulative effect of ECD.
In order to present the accumulative effect of ECD, we use the proposed CA model to simulate a section of railway line,
which includes 3 stations. The stations A, B and C are located at 1 m, 4000 m and 8000 m respectively. Trains depart from
station A successively with interval I and stop at station B for a dwell time Tdw, then leave and run to station C. Finally they
move out of this system after staying at station C for Tdw. The other parameters used in the simulation are as same as we used
in the former simulation except for the maximum speed of train Vmax = 20 m/s.
Fig. 5 displays the simulation results around Station B. From Fig. 5(a), we nd that Train 1 enters the station without any
interference. Trains 2, 3, 4 and 5 experience interference by the preceding train because they are very close such that the
following train has to decrease its speed, even coming to a full stop. This interaction between the preceding train and the
following train should be avoided as much as possible to increase the operational efciency of the railway system. Moreover,
three out of ve trains have to stop outside of Station B because the station is occupied in Fig. 5(b). If the following trains
continuously arrive at Station B with the same interval, more trains will queue up outside of Station B. Fig. 5 presents a

134

J. Xun et al. / Transportation Research Part C 35 (2013) 127140

Fig. 5. Simulation results around Station B. (a) The velocitydistance diagram and (b) the timespace diagram.

situation under which the ow of arriving trains exceeds the capacity of Station B. The reason causing this situation is that
the departure time interval at the original station is smaller than the minimum headway of the system.
In studying the railway system, minimum headway is a key parameter for evaluation. The denition of headway is the
time interval of two successive trains passing the same location. The minimum headway occurs when the preceding train
leaves the station for a safe distance followed by another train entering the same station with the worst-case braking curve
prole. Different theoretical formulas for calculating the minimum headway with different maximum speeds are proposed
by Luo and Wu (2005) as follows:

Hmin

8 q
< 2Ls LT tan V max ;
a

: 2aLs LT V 2max
2aV max

p
2Ls LT =a
p ;
< 2Ls LT =a

V max P

t an V max
; V max
b

where tan, which is the system response time, equals zero because we did not consider the effect caused by tan in the proposed model.
The comparison between the simulation results and the theoretical results is shown in Fig. 6. The line with x-marks denotes the simulation values without considering the ECD, i.e., Ci(t) = 0, t e (0, T). The line with dots denotes the simulation
values using the proposed model with the gamma distribution of ECD. The solid line denotes the theoretical values calculated by Eq. (4).
Two parameters, which are used to measure the degree of dispersion, are dened by

r jX S  X T j;
g EX S  X T ;
where XS is the simulation value of the minimum time headway; XT is the theoretical value of the minimum time headway.
The results of r and g for two simulations (with and without the ECD) are presented in Table 2.

135

J. Xun et al. / Transportation Research Part C 35 (2013) 127140

Fig. 6. Comparisons between the simulation values and theoretical values.

From Table 2, we can see that the results of the simulation by the proposed model are closer to the theoretical values than
that obtained in the simulation without considering end-to-end delay. Specically, when Vmax is greater than 15 m/s, we see
that the simulation values of the minimum time headway with the ECD are closer to the theoretical values than the values
without considering ECD. This shows that the disturbance of minimum headway triggered by random end-to-end communication delay is more apparent when trains run at lower speed. There is a margin when Vmax is greater than 30 m/s, where
the simulation values are lower than the theoretical values. Furthermore, there is a minimum point of the time headway at
the maximum speed at approximately Vmax = 11 m/s or Vmax = 39.6 km/h. These results validate the dependability of the proposed model.

6. Analysis
According to the research by Federal Railroad Administration (Tse, 2008), the values of the communication delay in a Positive Train Control (PTC) system were allowed to vary from 5 to 20 seconds. So, next we will analyze the impact on line
capacity of the former section of railway line when the communication delays are increases from 5 to 30 s. The parameters
used in this analysis are as follows:
(1)
(2)
(3)
(4)
(5)
(6)
(7)

Simulated model time T = 4000 s.


Train length LT = 260 m.
Maximum speed of train Vmax = 20, 30, 40, 50 m/s, respectively.
Speed limit of line SLi = 50 m/s, i e (1, L).
Time interval of train departure at Station A equals the minimum headway corresponding to the maximum speed.
Dwell time Tdw = 60 at Stations B and C.
The ECD is a random number based on the gamma distribution and d3 = 5, 10, 15, 20, 25, 30 s, respectively.

The simulation results in Fig. 7 reveal that the increase of the maximum communication delay from 5 to 30 s can decrease
line capacity on a single-track line approximately to:





32.3% when trains run at 20 m/s;


31% when trains run at 30 m/s;
29.6% when trains run at 40 m/s;
28% when trains run at 50 m/s;

Table 2
Results of r and g.
Simulation

Maximum of r

I (With the ECD)


II (Without the ECD)

2.89
4.89

1.54
3.47

136

J. Xun et al. / Transportation Research Part C 35 (2013) 127140

Fig. 7. Decrease of line capacity when the communication delays vary from 5 to 30 s.

The model with considering ECD is more complex than that without considering ECD. On the one hand, this is helpful to
improve the train movement accuracy. On the other hand, the complex may result in increase of computational time. In
Fig. 8, the CPU time with two models is presented. The CPU time increases with the increases of the number of trains in simulation when the number of trains is from 20 to 30. Moreover, the increased CPU time while considering the ECD is presented
in Fig. 9. The maximum of increased CPI time is 4.5 s when the number of trains is 25. The average of increased CPU time is
3.3 s. All the simulations are conducted on a PC with an Intel Core Due T8100 2.10 GHz Processor and 3G RAM. The operating
system is the windows XP. The model is programmed in the MATLAB language.
In the rest of this section we will discuss the effect upon the trafc ow of using the gamma and logistic distributions of
ECD rst.
We have calculated the trafc ow in the section AC. The minimum value of trafc ow is 57.6 vehicles per hour for both,
no matter which satises the Euroradio specication or not when considering the ECD. This shows that little loss of system
performance will be caused by the ECD if the ECD satises the requirement described in the Euroradio specication.
However, the impact on individual trains cannot be neglected. Fig. 10 shows the total travel time of each train. In the gure, the solid line denotes the results without considering the ECD and the line with circles denotes the results using the
proposed model with the gamma distribution end-to-end communication delay while the line with dots denotes the results
using the proposed model with the logistic distribution end-to-end communication delay. Because of the ECD, except train 1
and 2, more travel time is needed for all trains. Furthermore, more and more delay accumulated on the following trains. The
propagation of the delay cannot be recovered and the most seriously affected train is the last train, train 25. From its spacetime diagram in Fig. 11, we nd that the delay is 17 s with the gamma distribution end-to-end communication delay and
23 s with the logistic distribution end-to-end communication delay respectively when it arrives at the destination. The delay
is a little over 2% of total travel time. These delays are caused by the decrease in train speed when train 25 is approaching
Station B and speed variance is due to the ECD.

Fig. 8. The CPU time for given number of trains.

J. Xun et al. / Transportation Research Part C 35 (2013) 127140

137

Fig. 9. Increased CPU time of the model with considering ECD compared to the model without considering ECD.

Fig. 10. Travel time of all trains.

Fig. 11. Spacetime diagram of train 25.

Although the ECD will not degrade overall system performance when it satises the Euroradio specication, its effect on
the punctuality of each individual train should be considered. Minimizing the ECD will keep each individual train movement
on schedule, furthermore ensuring the system performance.

138

J. Xun et al. / Transportation Research Part C 35 (2013) 127140

From the above discussion, we nd little loss of system performance will be caused by the ECD if it satises the requirement of the Euroradio specication. In the next section we will use the proposed model to investigate the maximum density
of the line and its dependency on the ECD and research into the ow variation with different end-to-end communication
delay requirements.
Here we assume that the ECD for a message is at most d1 seconds with 95% condence, at most d2 seconds with 99% condence and at most d3 seconds with 99.99% condence in all cases (0 < d1 < d2 < d3).
First of all, we would like to investigate the headway time between two consecutive trains and its dependency on the
ECD. The headway time is an important parameter to evaluate the theoretical line capacity. The theoretical line capacity indicates number of trains which can go through a particular point during a unit time under moving block operation. In (Zimmermann and Hommel, 2005), they investigate a deadline t and its dependency on the train head-to-head distance. The
minimum of train head-to-head distance is under a worst-case assumption. The worst-case assumption is that, after the last
integrity check has been completed, a part of the preceding trains carriages separate from the main train and stop where
they are or there is an accident.
Under this assumption, the movement of the following train can be divided into two stages under moving block operation
(Fig. 12):
(1) Keep running with its original speed until receive the next MA message: the maximum running distance is Sr = Vmax  (tic + 2td3 + tte + tft), where Vmax is the maximum speed of train.
(2) Braking to stop: the maximum running distance is Sb V 2max =2b, where b is the train deceleration rate.
Because in moving block operation the movement authority shall never exceed the safe rear end of the preceding train, so

Sh P Sr Sb Ls LT :
The necessary headway time tmin
thus becomes:
h

tmin

V max  t ic 2td3 t te t ft V 2max =2b Ls LT


Smin
h

V max
V max

2t d3 t ic t te tft V max =2b Ls LT =V max :

From Eq. (5), we nd that d3 should be kept as small as possible to minimize the headway time.
Another important factor of the system is the ow variation as a function of d1 and d2 when d3 is constant. According to
the Euroradio specication, we specify that d1 ranges between 0.5 and d2 seconds, d2 ranges to be between 1.2 and d3 seconds and d3 = 2.4 s. As such, we get a few sets of d1, d2 and d3. We select several typical values for each parameter and derive
3 schemes for the ECD requirement.
(I) The ECD for a message is at most 0.5 s with 95% condence, at most 1.2 s with 99% condence and at most 2.4 s with
99.99% condence in all cases.
(II) The ECD for message is at most 1.2 s with 95% condence, at most 2.3 s with 99% condence and at most 2.4 s with
99.99% condence in all cases.

Fig. 12. An illustration of minimum distance between two successive trains.

139

J. Xun et al. / Transportation Research Part C 35 (2013) 127140


Table 3
Minimum ow, total travel time and minimum time headway.
Scheme

Minimum ow (tph)

Total travel time (s)

Minimum time headway (s)

I
II
III

72
72
72

22000.6
22010.9
22006.9

37
38
37

(III) The ECD for message is at most 2.2 s with 95% condence, at most 2.3 s with 99% condence and at most 2.4 s with
99.99% condence in all cases.
We enlarge the length of the simulated model time taken as T = 4000 s and increase the number of trains to 80. Besides
these two parameters, the changed ones in this section are as follows:
(1)
(2)
(3)
(4)

The
The
The
The

maximum speed of train Vmax = 20 m/s.


time interval of train departure at Station A I = 37.
dwell time Tdw = 0 at Stations B and C.
ECD is a random number with the gamma distribution.

Other parameters are the same as described in the simulation section. Moreover, in order to minimize the effect of randomness, we conducted 10 rounds of simulation and averaged the results to obtain a mean value. The results are reported in
Table 3.
Table 3 gives us the minimum ow, total travel time and minimum time headway for each of the three schemes. From
these results, we nd that the minimum ow is the same for these schemes and there is a little variation of total travel time
and minimum time headway. In general, the trafc ow changes very little when d1 and d2 increase.
7. Conclusions
In this paper, a CA model has been proposed as a tool for simulating the MBS operation. In the proposed model, additional
parameters such as the type of the train, the speed limit of the track and the ECD are considered. We nd that if the ECD
satises the requirement described in the Euroradio specication, the overall trafc ow in the transit system would not
be affected while the inuence on each individual train cannot be neglected. Larger end-to-end communication delay will
impact the trains punctuality and increase the trains travel time. In addition, we assume that the ECD for a message is
at most d1 seconds with 95% condence, at most d2 seconds with 99% condence and at most d3 seconds with 99.99% condence in all cases and investigate the variation of trafc ow with these three parameters. The investigation results show
line capacity is insensitive to communication delay with the Euroradio specication. However the communication delays are
much longer than 2.4 seconds in the North American PTC system. Therefore, we analyzed the impact to line capacity when
the communication delays vary from 5 to 30 s. The analysis results show that the line capacity drops 32%, 31%, 29.6% and 28%
when trains run at 20, 30, 40 and 50 m/s, respectively. This study demonstrates that CA modeling is a dependable and effective tool for analyzing the impact of communication delays to train operations.
Acknowledgments
This work was performed by the PATH Program at the University of California in cooperation with China. The authors
thank Dr. Li Zhu and Li-jie Chen for their discussions. Last, the authors are grateful to the anonymous reviewers for their constructive comments. This research is partly supported by Beijing Jiaotong University (Contract No. 2013JBM121 & Contract
No. 2012JBZ014) and the National High Technology Research and Development Program of China (863 Program) No.
2011AA110502.
References
Burks, A., 1970. Essays on Cellular Automata. University of Illinois Press, Champaign, IL.
EEIG ERTMS User Group, 2003. Euroradio FFFIS, UIC, Brussels.
Lan, W.L., Hsu, C., 2006. Formation of Spatiotemporal Trafc Patterns with Cellular Automaton Simulation. 85th TRB Annual Meeting Compendium of
Papers, CD-ROM. Transportation Research Board, Washington DC.
Li, F., Gao, Z.Y., Li, K.P., 2007. Analysis of the property of train ow in the xed autoblock systems. Acta Physics Sinica 56 (6), 31583165.
Li, K.P., Gao, Z.Y., Ning, B., 2005. Cellular automaton model for railway trafc. Journal of Computational Physics 209 (1), 179192.
Li, X.G., Jia, B., Gao, Z.Y., Jiang, R., 2006. A realistic two-lane cellular automata trafc model considering aggressive lane-changing behavior of fast vehicle.
Physica A 367, 479486.
Luo, L.Y., Wu, W.Q., 2005. Analysis on the safety time interval of train with movable block system in urban rail transit. China Railway Science 26 (2), 119
123.
Nagel, K., Schreckenberg, M., 1992. A cellular automaton model for freeway trafc. Journal de Physique I, France 2, 22212229.

140

J. Xun et al. / Transportation Research Part C 35 (2013) 127140

Najah, A.L., Essa, E., Ali, E., Khaled, S., 2008. Measured Delay Distribution in a Wireless Mesh Network Test-bed. In: Sixth ACS/IEEE International Conference
on Computer Systems and Applications (AICCSA08), Doha, pp. 236240.
Nassab, K., Schreckenberg, M., Boulmakoul, A., Ouaskit, S., 2006. Effect of the lane reduction in the cellular automata models applied to the two-lane trafc.
Physica A 369 (2), 841852.
Ning, B., Li, K.P., Gao, Z.Y., 2005. Modeling xed-block railway signaling system using cellular automata model. International Journal of Modern Physics 16
(11), 17931801.
Polivka, A., Ede, B.M., Drapa, J., 2009. North American Joint Positive Train Control Project. Department of Transportation, FRA, US, DOT/FRA/ORD-09/04.
Tse, T., 2008. Safety Analysis of Communication Timeout and Latency in a Positive Train Control System. Research Results RR08-01. Department of
Transportation, FRA, U.S..
Xu, T.H., Tang, T., 2007. The modeling and Analysis of Data Communication System (DCS) in Communication Based Train Control (CBTC) with Colored Petri
Nets. In: Proceedings of the Eighth International Symposium on Autonomous Decentralized Systems (ISADS07), pp. 8392.
Xun, J., Ning, B., Li, K.P., 2007. Network-based train-following model and study of trains delay propagation, Beijing, China. Acta Physics Sinica 56 (9), 5158
5164.
Zimmermann, A., Hommel, G., 2003. A train control system case study in model-based real time system design. In: International Parallel and Distributed
Processing Symposium (IPDPS03), pp. 118b.
Zimmermann, A., Hommel, G., 2005. Towards modeling and evaluation of ETC real-time communication and operation. The Journal of Systems and Software
77, 4754.
Zhou, H.L., Gao, Z.Y., Li, K.P., 2006. Cellular automaton model for moving-like block system and study of trains delay propagation. Acta Physics Sinica 55 (4),
17061710.
Zhu, L., Yu, F.R., Ning, B., Tang, T., 2012. Cross-Layer Handoff Design in MIMO-Enabled WLANs for Communication-Based Train Control (CBTC) Systems. IEEE
J. Sel. Areas in Comm. (JSAC) 30 (4), 719728.

Vous aimerez peut-être aussi