INTRODUCTION
In the rail system, the control of train operations is the core part to ensure
operational security and improvement of operational efficiency. In the automata
fixedblocking signal system, train trackingup is a complicated and nonlinear
process (Jia et al., 2007).
Cellular Automata (CA) theory initiated in 1940s is a key tool for the studies
on transportation and other complicated nonlinear systems (Wolfram,
2002). In CA model, the cellular making up the model are dispersed in time
and sp ace and of limited status. They are evolving in time and space according
to certain local rules. Therefore, the CA model is especially suitable for simulation
of the dynamic evolvement process of time and space conditions of complicated
nonlinear system. Thanks to the advantages of CA model, so far, it has made
fast progress in theories and in practices home and abroad (Chen
and Zhang, 2011; Mazzarello and Ottaviani, 2007;
Zhou et al., 2009; Fu et
al., 2008; Zhou et al., 2004; Ayanzadeh
et al., 2009; Ziabari et al., 2008;
Mahmoud, 2011; Azghadi et al.,
2007). The most famous CA model is NaSch model proposed by Nagel
and Schreckenberg (1992). The model is featured with simple algorithm and
capability to simulate the macro and micro traffic phenomena. It has been widely
applied in the urban road traffics and rail traffic flows. Li
et al. (2005a, b) have analyzed train tracing
model and rail traffic flows and proposed a CA model proper for rail traffic
system. In traffic flow and freight transport, many national and foreign scholars
have done a lot of relevant research to the railway transport (Alhassan
and BenEdigbe, 2011; Abdelouahab et al., 2006;
Mohan and Ravichandran, 2007; Msogoya
and Maerere, 2006; D’Acierno et al., 2011;
Mohammaditabar and Teimoury, 2008; Lin
and Liang, 2011) which also provided a good foundation for this research.
The article is based on the fouraspect fixedblock signal system, constructs the CA model for Daqin Railway. By simulating the tracing process of trains from Xiazhuang to Chawu station, different marshaling models are analyzed to sum up the relationship between train ratio, average speed and marshaling intensity. Matching schemes for different conditions are provided and actual operational plans are compared. Through comparative studies, the schemes are optimized for different carrying capacity, speed and intensity conditions.
CA MODEL FOR DAQIN RAILWAY
Signal system for Daqin Railway adopts the fouraspect fixedblock mode.

Fig. 1: 
The diagram of the signal in the fouraspect fixed block system 
In the signal system, each semaphore has four display statuses: red light,
yellow light, greenyellow light and green light, as indicated in Fig.
1. When the semaphore is green light, it suggests at least three block sections
are available and the train shall move on at the required speed. If not limited
by the line conditions, it is allowed to run at the maximum speed allowable,
otherwise, it shall run at the limited speed. When the semaphore is greenyellow
light, it suggests two block sections available ahead. In this case, the passenger
train shall slow down to the required speed before moving to the greenyellow
light, while the freight train does not need to slow down. It may pass the greenyellow
light at the regular speed. When the semaphore is yellow, the passenger trains
and freight trains are required to reduce to the required speed. If the semaphore
is red light, both passenger and freight trains are required to stop before
reaching the red light. During the operations, the colors of protection semaphores
in each block section will automatically change according to the position of
trains on the line.
The up direction can be divided into L cellular, each of the same size, as i = 0,1,2,..., L. Each cellular is empty or occupied by a train.
With the application of cellular automata model, the rail traffic of fouraspect block system can be simulated. When defining the speedrestriction function, the article proposes a speedrestriction function closer to the reality:
• 
Redlight speedrestriction function: 
• 
Yellowlight speedrestriction function: 
• 
Greenyellow light speedrestriction function: 
Where:
s 
= 
Distance between trains and protection semaphores ahead 
v_{l, 1} 
= 
Speedrestriction value for yellowlight signals 
v_{l, 1} 
= 
Speedrestriction value for greenyellow light signals 
b_{1} 
= 
Train’s regular braking deceleration 
b_{2} 
= 
Train’s urgent braking deceleration 
a 
= 
Train’s acceleration 
v_{max} 
= 
Train’s maximum speed value 
Table 1: 
Rules for updating of train’s speed 

The train’s position updated as x_{n} = x_{n}+v_{n} 
The above three is the maximum speed allowable for the train. The above three
speedrestriction functions can ensure that the train’s speed will be reduced
to the required at different semaphores.
As for the fouraspect fixed block, the iteration rules are divided into two steps:
• 
Updating of the train’s speed and position: In
each time step, the updating of train’s speed is determined by the
color of the semaphore ahead and the updating rules are indicated as follows
in Table 1. 
• 
Semaphore’s color updating: Mainly by the distribution
of trains on the railway line to update the colors of semaphore on each
block section. If the block section is occupied by the trains, the protection
semaphore will be red, while the semaphores for the followup block sections
are yellow, greenyellow and green. In the simulation system, the semaphore’s
color will be indicated with a number, 4 for green, 3 for yellow, 2 for
greenyellow and 1 for red 
The above is about the train in operations wayside. If it is near a station, when the station is occupied by a train closely ahead, the train (n) will have to keep a safe distance from the train closely ahead. If the station ahead is available, the train will move into the station directly at the decelerated speed b and when it stays in the station for a time of T_{d}, it will take leave. T_{d} is the time at station. Therefore, the iterative rules for trains at station are:
• 
Train at a certain position before reaching the station, as
below: 
When the train is within the station, the rules are as follows:
When the train stays in the station longer than T_{d}:
• 
Step1 Acceleration: v_{n} = v_{n}+a 
• 
Step 2 movement: x_{n} = x_{n}+v_{n} 
When the train stays in the station for no longer than T_{d}; v_{n} = 0. d_{s }suggests the distance between No. n train and the station ahead and d_{b} is the braking distance.
The model applies open boundary condition. Initially, there is no train running on the line and all semaphores are green. When t = 1, a train moves into the system. Then, a train starts off when green is on or at fixed interval. If the end of a train moves out of the up direction Chawu Station, it is believed to move out of the system and will be deleted.
Before the simulation results are analyzed, it is necessary to explain the
two concepts about train intensity. According to Wang (1990),
two definitions are given:
First intensity concept: Statistically, the train intensity means within a certain period of time and in a certain scope (such as full trip or on a certain line), averagely for each kilometer each day, the number of passenger and freight trains allocated. It is an average number by dividing the total of kilometers all the allocated trains are running each day by the total length of the line (operational kilometer).
Train intensity:
where, q is the number of passenger and freight trains passing by within a
certain period of time t, v is the average speed, L is the total length of the
line. There are two ways for computation of v: arithmetic mean and harmonic
mean.
Arithmetic mean:
where, v_{i} is the speed of No. i train. The disadvantage of this method is: The intensity calculated against the background of congested traffics will be excessively low.
Harmonic mean:
Intensity calculated by this method has a much higher scope than the intensity obtained by arithmetic mean method. Since in this article, only the trains running within the specified scope are considered and the train’s startstop time is not included, the speed is all about the train’s operational speed, a harmonic mean method is applied to calculate the train’s mean speed.
Second intensity concept: Train intensity means the number of passenger and freight trains that might pass by one day and one night within the controlled space. Generally, when the train’s operational speed is high, the interval is short and the passthrough competence is large, the train intensity is large. The following formula is given:
Train intensity:
where, T is the observation time. In this article, it is one day and one night, that is, 24 h. N is the number of trains passing by the observation points within the monitored time.
In the rail traffic system, the train’s intensity is always depicted with the marshaling interval. The shorter the marshaling interval is, the higher the train’s intensity is.
ANALYSES OF SIMULATION RESULTS
Without any influences on the simulation results, for the simulated space,
the most complicated part in the middle of Daqin Railway is chosen: the up direction
from Xiazhuang to Chawu, a total length of 14.895 km which means 169,320 cellular.
Table 2: 
Coordinates of semaphores from Xiazhuang to Chawu 

Xiazhuang is an intermediate station and Chawu is a district station. Eleven
semaphores are provided for the up direction, the coordinates of each semaphore
indicated in Table 2.
According to technical specifications for Heavyhaul on Daqin Railway, it is learnt that the maximum speed limit is 80, 55 km h^{1} for the yellow light, 75 km h^{1} for the greenyellow light. The 20,000ton train is 2,672 m long and 10,000ton train is 1,400 m.
In the simulation system, one cellular is 0.088 m and the system updating time step is 1s. The acceleration is 1 cell s^{2}, the regular deceleration speed is 2 cell s^{2}, the braking distance is 1400 m which means 15,909 cells. As it is supposed that all the trains are having the stop operations at Chawu station, the actual 20,000ton train is passing directly by. To reduce the influences on simulation results, four lines are set for up direction, two more than the actually adopted (two lines). Suppose the train, due to technical considerations, will stop and stay in the station for 25 min, that is, 1,500 steps. The system evolvement time is 12,000 steps. The initial conditions are: no trains running in the space, all semaphores being green, the first train starting off right at the time t = 1, the initial speed being 40 km h^{1} which means 126 cells sec^{1}. Therefore, to minimize the influences of initial conditions, consider the simulation started within one shift time (3 h). For statistical purpose, the data in first 1,200 time steps are abandoned and only the data for the followup 10,800 time steps are adopted.
Departure at green light: For departure at green light, the train delays are obvious. Therefore, it is relatively convenient to analyze the relationship between train proportion, speed and intensity against such a background.
Figure 2 is a timespace diagram about a 20,000ton train running within the controlled space. It shows obvious train delays and continuous spreading of the delays. It indicates that the simulation system can well simulate the train’s operational features.
Figure 3 is about the relationship between trains of 10,000
and 20,000 tons planned to operate and the average interval when the departureatgreenlight
mode is adopted.

Fig. 2: 
Timespace diagram of 20, 000ton train 

Fig. 3: 
Relationship between train proportion and average interval
(departure at green light) 
In the Fig. 3, the smooth curve is obtained after fitting
the higherorder function and the fold is based on simulations. From the Fig.
3, it is learnt that the interval is the longest when all trains are of
20,000 tons. With more 10,000ton trains are in operations, the mean interval
is reduced which is consistent with the actual operations. The length of 20,000ton
train doubles that of the 10,000ton train, therefore in the operations, it
will occupy two block sections while the 10,000ton train has just one block
space. For departure at green light, actually one more block section is needed
for the 20,000ton train than for 10,000ton train. Of course, the interval
is also longer.
In Fig. 4, the highest tons of cargo are carried when all
trains are 20,000ton train. However, according to Fig. 3,
the interval is the longest when all trains are of 20,000 tons and the interval
is the shortest when all trains are of 10,000 tons.

Fig. 4: 
Relationship between train proportions and tons of cargos
carried (departure at green light) 

Fig. 5: 
Relationship between train proportions and average speed of
trains (departure at green light) 
Although, the maximum number of trains is dispatched when all trains are 10,000
tons, the total tons of cargo carried are the smallest.
In Fig. 5, with the increase in the proportion of 10,000ton trains, the average speed of trains is slowed down. It is because the proportion of 10,000ton train is rising and the number of trains into the space within the unit time is also increasing. Therefore, the arrival and departure lines are unavailable for the trains planning to stop. It results in delays from time to time and slowingdown of the trains. Thus the conclusion is when the trains need to stop at stations, increased number of arrival and departure lines will reduce the chances of delays.
From Fig. 6a and b, it is learnt that when
the second train intensity concept is applied, with the increase in 10,000ton
trains, the train intensity is rising. When the first train intensity concept
is applied, with the increase in 10,000ton trains, the train intensity is reducing.
It is because the first train intensity has taken into consideration the average
speed and it is the demonstration of the line’s utilization rate. The higher
the train intensity is, the utilization rate is higher and the organization
will be optimized. According to the figure in left, when trains of 20,000tons
account for 0.4~0.5, the train intensity is reducing dramatically. Therefore
when conditions allow, trains of 20,000tons are to be dispatched as much as
possible, to ensure the high train intensity and full utilization of the line.
To make full use of the line, when cargo source and loading operations are under
restrictions, it is proper to select this ratio for marshaling which is actually
consistent with the actuality of Daqin Railway (38:46, first being the number
of train couples of 20,000 tons).
Departure mode at fixed interval: Considering the departure mode adopted for Daqin Railway at fixed intervals, the matching relationship between train proportion, speed and intensity is simply discussed against the background of fixed intervals. According to the current transportation mode for Daqin Railway, the traced interval for 10,000ton train is 10 min and the traced interval for composite combination of trains is 12 min. As the 20,000ton train is relatively long, the traced interval is 16 min, while the average traced interval for various trains is 14 min. According to above discussion, a matching scheme for the relationship between train proportion, speed and intensity at different train interval is proposed. Four groups are specified according to the intervals, as indicated in Fig. 3. Scheme for 8 min is simply for comparison, as in reality it is impossible.
When the interval is 14 min, the shortest time, as obtained from the simulation
test, for train running from Xiazhuang to Chawu is 701 s, with an average speed
of 76.4936 km h^{1}. The actual operational time is: 26 min for a combination
of trains, 24 min for other trains, with an average speed of 34.3731 and 37.2375
km h^{1}, respectively. There is a major difference between the simulation
value and actual value. When the departure interval is 14 min, in the simulation
system, all trains show no delay. They are accelerating to the highest speed
and then decelerate when close to the station. During the operations, all semaphores
are green lights, with only the maximum limit provided. However, the actuality
is much more complicated than the simulated. There are two slopes (12‰)
from Xiazhuang to Chawu, in addition to curves of small radius and slopes of
smaller rates of inclination.

Fig. 6 (ab): 
(a) Relationship between train proportion and first train
intensity (departure at green light) and (b) Relationship between train
proportion and second train intensity (departure at green light) 
Table 3: 
Matching scheme and transport competence 

Therefore in actual operations, trains cannot always run at the maximum speed.
In the simulation system, curves and speed restrictions are not considered which
is partly responsible for the speed given from the simulation much higher than
reality.
In the 16 schemes provided in Table 3, scheme 2 is most close
to the operational scheme specified in Daqin Railway 400millionton organizational
scheme (Feng, 2011) and the number of trains in up direction
is matched with the 90 trains as indicated in the actuality.
The train intensity after considering the train speed shows no apparent changes in the train speed as specified in scheme for intervals over 10 min. Therefore, the train intensity with consideration of the speed is not high. However, the general trend that the train intensity will reduce with the reduction in the proportion of 20,000ton trains is consistent with the above discussion. As for the scheme specifying a traced interval of more than 12 min, considering the relatively small number of trains dispatched, the train intensity even having taken into consideration the speed is relatively low.
On the prerequisite that the annual cargo carrying capacity is 400millionton, scheme 4 and 8 are not compliant. With consideration of the 14min interval, only scheme 1~3 are compliant. From the data above, when more 20,000ton trains are used, the annual cargo carrying capacity is higher. However, in view of the cargo source organizing, empty train organizing and available equipment, all being 20,000ton trains are impossible.
Of the four schemes for 8 mininterval, the train’s speed is much smaller than those specified in other schemes. Reduction in the interval will allow more trains to be dispatched and therefore more cargo to be carried. However, reduction in the train’s sped is a restraining factor.
As for which scheme is to be adopted, except for the above considerations, the economical and technological assessment is also necessary, including the scheme for train set and expenses. A decision is to be made with full consideration of the economic, operational costs, technical demands and task indexes.
CONCLUSION
As the line conditions, such as slope, radius of slopes and speed restrictions, deceleration and acceleration, are not considered and only one section of Daqin Railway is simulated, the data obtained from this simulation system shows difference from the actuality. However, the simulation system can well simulate the trains in operation and the relationship indicated between train proportion, speed and train intensity is consistent with reality. Through simulations, an optimized matching scheme for train speed, intensity and proportion is provided. Therefore, CA model is significant for simulation of Daqin Railway, so as to obtain more precise data and to improve the system.
Simulation results suggest that more 20,000ton trains are adopted, the scheme is more effective. However in reality, it is almost impossible to all applying the 20,000ton trains. Firstly, it is out of the consideration for cargo source and secondly the scheme for empty train on the returning. With the optimization of cargo source and organizational mode, as well as the improvement of scheme for empty train on the returning and the enhancement of railway technologies, the trend is that a higher proportion of 20,000ton trains will be used for Daqin Railway transport.
ACKNOWLEDGMENTS
Project supported by the Basic Scientific Research Funds allocated by the Ministry of Education for universities (2010QZZD021) and the Ministry of Railway Science and Technology Research Development Program (2009F021, 2010X014), China.