Hindawi

Discrete Dynamics in Nature and Society

Volume 2022, Article ID 2228197, 13 pages
https://doi.org/10.1155/2022/2228197

Research Article
Structure Optimization of Mixed-Speed Train Traffic for Cyclic
Timetable: Model and Algorithm Development

1
Jiamin Zhang and Jiarui Zhang2
1
College of Transportation, Shandong University of Science and Technology, Qingdao 266590, Shandong, China
2
Qingdao Locomotive Depot, Jinan Bureau of China Railway, Qingdao 266041, Shandong, China

Correspondence should be addressed to Jiamin Zhang; [email protected]

Received 13 September 2021; Accepted 27 November 2021; Published 4 January 2022

Academic Editor: Polinpapilinho Katina

Copyright © 2022 Jiamin Zhang and Jiarui Zhang. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
Trains can be optimally spread over the period of the cyclic timetable. By integrating sequencing issue with headway time together,
this paper studies the structure optimization of mixed-speed train traffic for a cyclic timetable. Firstly, by taking it as a job-shop
problem with sequence-dependent setup times on one machine, in the type of infinite capacity resource with headway (ICR + H),
the problem is transformed to alternative graph (AG) and then recast to the mixed-speed train traffic planning (MSTTP) model.
For the multiobjective in MSTTP, three indicators are optimized, i.e., heterogeneity, cycle time, and buffer time, which correspond
to diversity of train service toward passenger, capacity consumption of rail network, and stability of train operation, respectively.
Secondly, the random-key genetic algorithm (RKGA) is proposed to tackle the sequence and headway simultaneously. Finally,
RKGA is coded with visual studio C# and the proposed method is validated with a case study. The rail system considered is a line
section encompassing a territory of 180 km with 15 mixed-speed trains in each cycle of the timetable. Results indicate the
comprehensively balanced train plan for all stakeholders from random variations of train sequence and headway time. Both the
quantitative proportion of heterogeneity/homogeneity (e.g., 2.5) about the optimized distribution of the mixed train traffic and the
link between train headway time and the sequence for each traffic scenario are found. All the findings can be used to arrange the
mixed-speed train traffic more scientifically.

1. Introduction Capacity consumption may be different for each pro-

portional mix of trains and the order in which they run on
Usually, the railway industry provides the passenger services the route [3–6], as can be seen in Figure 2. It is possible to
in the form of the train timetable, in order to satisfy the achieve a high stability/punctuality on a railway network or
public travel demand and maintain the further development to provide a diverse service to the passengers. At the same
needs of the railway company. Any set of key performance time, it would also cost more capacity consumption. What is
indicators must be sufficient to adequately represent the demonstrated in Figure 2 are just two typical extreme cases.
requirements of all stakeholders. For timetabling, the key A cyclic timetable means that each line of passenger trains is
stakeholders include passengers, network operators, and operated in a cyclic pattern, e.g., with regularly sequenced
train operators. However, the requirements of these stake- repeating train lines [7]. In Figure 2, we may get the maximal
holders are prone to mutual conflict. The UIC capacity leaflet and minimal value of the cycle time, only benefiting the
[1] showed the capacity balance relationship among the interest of a certain part of the stakeholders, e.g., the railway
number of trains, the average speed, the heterogeneity, and operators. However, what we concern is to seek a com-
the stability (Figure 1). Landex [2] added an extra dimen- prehensively balanced solution as optimal as possible for all
sion, capacity consumption, which formed the “capacity of the involved stakeholders, by arranging the train traffic,
pyramid” as illustrated in Figure 1. which no doubt have great significance for train timetabling.
2 Discrete Dynamics in Nature and Society

number of trains capacity consumption

stability heterogeneity

average speed stability

number of trains average speed

heterogeneity
mixed train operations
metro train operation
(a) (b)

Figure 1: Railway capacity: capacity balance (a) and capacity pyramid (b) (sourcing: UIC [1]; Landex [2]).

departure station departure station

place place
Time for cycle
Time for cycle

time
time

high-speed trains
medium-speed trains
low-speed trains
(a) (b)

Figure 2: Mixed train traffic arrangement from perspective of passengers with very heterogeneous (a) and that of network operators with
slightly heterogeneous (b) (sourcing: Kaas [4]).

And, these concerns just motivate us to quantitatively problems, focusing on the planning objective and avail-
optimize the mixed-speed train traffic structure in an all- able data, regardless of the specific traffic carried or
round balanced way, so as to lay the foundation for a more network size. Zhang [9] summarized some strategies for
robust, stable, and efficient timetable with more reason- train operation plan under mixed traffic condition and
able transport capacity. According to whether they ex- analyzed them from both pros and cons aspects, which
plicitly model the track structure and whether the limited to a qualitative analysis of the mixed traffic
timetable is intended to be periodic or not, Harrod [8] structure rather than a quantitative optimization. Huang
addressed four fundamental timetable formulations [10] presented a time supplements allocation method that
suitable for optimization, e.g., mixed integer sequencing incorporates historical train operation data to optimize
linear programs, binary integer occupancy programs, buffer-time distribution in the sections and stations of a
hypergraph formulation, and periodic event scheduling published timetable.
Discrete Dynamics in Nature and Society 3

This paper studies the structure of the mixed-speed train problem space. Based on these characteristics, we propose
traffic for a cyclic timetable in a passenger railway line the modified random-key genetic algorithm (RKGA) to
section, integrating sequencing with headway optimization solve the mixed-speed train traffic planning (MSTTP)
for train lines. As was defined by UIC [10], in our research, a model, which can tackle the sequence and headway si-
railway line section is the part of a line in which the traffic multaneously. The objective of our study is to optimize the
mix and/or the number of trains or the infrastructure and structure of the mixed-speed train traffic for cyclic timetable,
signaling conditions do not change fundamentally. Infra- by integrating sequencing issue with headway time together.
structure elements can be modeled by using resources with The innovative contributions of this paper are as follows:
different properties in terms of capacity [11], one of which is
(i) The structure optimization problem of the mixed-
the kind of infinite capacity resource with headway
speed train traffic for cyclic timetable is innovatively
(ICR + H). In the type (ICR + H), the number of operations
transformed to an alternative graph (AG).
that can be simultaneously processed is not restricted.
However, the starting times of two consecutive operations (ii) The mixed-speed train traffic planning (MSTTP)
on the same resource are separated by a time interval. In this model is constructed innovatively, considering
study, we assume the railway line section as the one-machine three optimization indicators, i.e., maximization of
[12] type (ICR + H) and the trains as jobs to be processed. heterogeneity from the perspective of passenger
An alternative graph (AG) is a representation of a job- optimization, minimization of cycle time from the
shop scheduling model with additional operational con- perspective of capacity optimization, and maximi-
straints [13]. The alternative arcs of AG are used to model zation of the minimal buffer time from the per-
choice of orders between operations on the same machine spective of operational optimization, individually.
(open track segment) and are weighted with the setup time, This innovative model can ensure to achieve an all-
i.e., minimal or optimized headway time. By taking it as a round satisfied solution for all involved
job-shop problem with sequence-dependent setup times [14] stakeholders.
on one-machine type (ICR + H), we transform the problem (iii) The random-key genetic algorithm (RKGA) is
to an alternative graph (AG). Then, we recast it to the mixed- proposed to solve the mixed-speed train traffic
speed train traffic planning (MSTTP) model so as to opti- planning (MSTTP) model, which can tackle the
mize the temporal-spatial distribution of trains quantita- sequence and headway simultaneously. The RKGA
tively. To develop our model, we set up three objectives is an innovative construction and application for
considering the concerns of the stakeholders. They are the solving the kind of MSTTP model.
maximization of heterogeneity from the perspective of (iv) The RKGA was coded with visual studio C#, and the
passenger optimization, the minimization of cycle time from proposed method is validated to work out the op-
the perspective of capacity optimization, and the maximi- timal solutions for a case study. Both the quanti-
zation of the minimal buffer time from the perspective of tative proportion of heterogeneity/homogeneity
operational optimization, individually. For the level of in- (e.g., 2.5) about the optimized distribution of the
frastructure details considered in the models, we distinguish mixed train traffic and the link between train
among microscopic models (which consider detailed de- headway time and the sequence for each traffic
scription of infrastructure and train dynamics), macroscopic scenario are found.
models (which represent the railway network as a simplified
series of links connecting stations), and mesoscopic models The rest of the paper is organized as follows: Section 2
[15]. In this regard, our model MSTTP belongs to the kind of demonstrates the mathematical formulation (AG and
mesoscopic representation, using a level of abstraction be- MSTTP) of the problem. Section 3 reports the RKGA for
tween macroscopic and microscopic representation. We solving the problem, including one main algorithm and four
take the railway line section under study as a unidirectional subalgorithms. In Section 4, we apply the model and al-
open track segment without overtaking place. gorithm to a case study and investigate the results deeply.
The concept of genetic algorithm (GA) was first for- Finally, Section 5 concludes the paper.
malized by Holland [16] and then extended to optimization
by De Jong [17]. Bean [18] introduced the concept of 2. Mathematical Formulation
random keys, a method for representing solutions (chro-
mosomal encoding) that produces feasible offspring for 2.1. Problem Statement by Alternative Graph. We associate
many sequencing and optimization problems. For the with a sequence a modified alternative graph G � (Y, A) as
problem under study, any permutation of train sequence follows. The set Y of nodes is obtained by adding two fic-
would be a feasible solution. This would also be the typical titious nodes O and D to the set I of jobs (trains):
chromosome representation adopted in GA. However, when Y � I ∪ {O, D}; O is a job “beginning,” and D a job “end” in
applying crossover genetic operator, additional efforts the machine (unidirectional track section between two
should be taken to avoid duplication of trains in any feasible timetable points without overtaking). The set A of arcs
solution; i.e., a train is processed only once. The random-key includes three sets A � A1 ∪ A2 ∪ A3 . Let A1 � {(O, i)|i ∈ I};
genetic algorithm (RKGA) approach can help overcome this arc (O, i) is valued by yi , so that job i cannot start before the
shortcoming easily [19]. The unique feature of RKGA is that point in time yi . Let A2 � {(i, D)|i ∈ I}; arc (i, D) is valued
its search is over the random-key space instead of the by di , since job i has to spend an amount of time di in the
4 Discrete Dynamics in Nature and Society

system after its beginning of processing by the machine. Let a variety of train service for passengers within an acceptable
A3 � 􏼈(i, j)|job i proceeds to job j in the sequence􏼉; alterna- headway. According to [20], there are two measures for
tive arc (i, j) is valued by hij , which is a sequence-dependent representing the heterogeneity of trains, which is sum of
setup time, since it should keep an amount of headway time shortest headway reciprocals (SSHR) and sum of arrival
hij between two consecutive operations i and j; these arcs set headway reciprocals (SAHR), respectively. To be comple-
the sequence. The improved alternative graph is shown in mentary, we maximize both of them from the perspective of
Figure 3. It is formulated as a one-machine type (ICR + H) passenger optimization as follows:
scheduling problem in the unidirectional track section.
max F1 � αSSHR +(1 − α)SAHR, (1)

2.2. Notations. The notations used in the following dis- n n

1
cussion are presented in the following: SSHR � 􏽘 􏽘 , (2)
i�1 j�1 h−ij
F1 denotes the objective value of the heterogeneity from j≠i
the perspective of the customer.
n n
F2 denotes the objective value of cycle time in the cyclic 1
timetable from the perspective of capacity. SAHR � 􏽘 􏽘 A
.
i�1 j�1 hij (3)
F3 denotes the objective value of the stability from the j≠i
perspective of operation.
hij denotes the departure headway time between the
subsequent trains i and j.
2.3.2. Optimization from the Perspective of Capacity.
kij denotes the pairwise train scheduling decisions, Capacity consumption represents the utilization of an in-
which is a binary variable. When train i proceeds to frastructure’s physical attributes along a given section [22].
train j in the sequence, it equals 1, otherwise 0. UIC 406 method defines a methodology to measure the
h−ij denotes the smallest scheduled headway time be- capacity consumption based on compressing timetable
tween train i and j on the track section, and the final graphs. Zhang [23] updated the norm for capacity measure
train n is followed by train 1 in a cyclic manner. as the time needed to fulfill the task list in train service-
hAij denotes the headway time at arrival between train i demand intention set. Heydar et al. [24] and Petering et al.
and j on the track section, and the last train n proceeds [25] proposed a different definition of capacity; i.e., ca-
train 1 due to cyclicity. pacity is the minimal cycle length that can feasibly ac-
commodate a given number of trains over a given section of
Hij denotes the theoretical safety minimum headway track (double track, unidirectional) in each cycle. Zhang
time between the subsequent trains i and j. and Nie [26] proposed a minimal cycle time calculation
yi denotes the departure time of train i from the model on the basis of periodic event scheduling problem
starting point of the track section. for a given train line plan. In accordance with 2nd edition
di denotes the running time of train i on the track of UIC capacity leaflet [22], which defined the occupancy
section. time of a line section as the timescale between the first and
n denotes the number of trains to be scheduled. last train paths, we calculate the cycle length with formula
(4) to minimize the capacity consumption for the network
SSHR denotes sum of shortest headway reciprocals. operators.
SAHR denotes sum of arrival headway reciprocals. n n
α is a parameter of SSHR, and 0 < α < 1. min F2 � 􏽘 􏽘 hij kij .
i�1 j�1 (4)
M is a large positive number. j≠i

2.3. Model Description

2.3.1. Optimization from the Perspective of Passengers. 2.3.3. Optimization from the Perspective of Operation.
Regarding the permutation of mixed-speed train traffic, in This objective is to create a train plan that is as robust as
most cases we believe that heterogeneity could decrease the possible to achieve stability. That means it can both deal with
reliability of the train timetable comparing to its homoge- relatively small disturbances and achieve a high punctuality
neous counterpart [20]. However, we cannot deny that the in the real-time operations. Robustness reflects the degree to
customers’ preference should be the right paradigm in the which trains depart and arrive according to the plan.
market-oriented environment. Passenger travel behavior is The headway between subsequent trains is used by
far more complex and multifaceted. Assessing railway many scholars as a measure of robustness or a target for
performances by merely measuring punctuality or capacity robustness in various ways [27]. Thus, we maximize the
would provide an unfair picture of the level of service ex- minimal buffer time between train pairs with formula (5) to
perienced by passengers [21]. On the other hand, hetero- optimize the stability objective from the perspective of
geneity also implies diversity and balance, which can provide operation.
Discrete Dynamics in Nature and Society 5

y1 yn
y2 y3 yi

h1n

h12 h23 h3i hin

1 2 3 i n
h21
h32 hi3 hni

hn1

d2 d3 di dn
d1

Figure 3: Alternative graph (AG) statement for mixed-speed train structure.

F3 � max min 􏽮􏼐hij − Hij 􏼑kij 􏽯. For any considered subsequent train pairs, the rela-
1≤i≤n 1≤j≤n
(5) tionship of their departure times should follow the following
j≠i
formula:

Formula (5) is a max-min objective function. According yj ≥ yi + hij − Mkji , 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i. (10)
to Burggraeve [28], we introduce an auxiliary constraint as
For the pairwise train dispatching decisions, only one
follows:
alternative arc of the train pairs (i, j) can be selected in
F3 ≤ 􏼐hij − Hij 􏼑kij , 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i. (6) Figure 3, so formulas (11) and (12) should be held.

And, we change the objective function of formula (5) to 1, if yi < yj

the maximization of F3 as follows: kij � 􏼨 (11)
0, otherwise,
max F3 . (7)
kij + kji � 1, 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i. (12)

Regarding the security for trains running, any headway

2.3.4. Constraints. According to the notations mentioned time should meet the requirements of the theoretical safety
above, we put forward formulas (8) and (9) for calculation minimal headway time between the subsequent trains. Thus,
the value s of h−ij and hAij . formulas (13)–(15) should be obeyed.
h−ij � min􏽮hij , 􏼐􏼐yj + dj 􏼑 − yi + di 􏼁􏼑kij 􏽯, hij ≥ Hij , 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i, (13)
(8)
1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i,
h−ij ≥ Hij , 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i, (14)
hAij � 􏼐􏼐yj + dj 􏼑 − yi + di 􏼁􏼑kij , 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i.
(9) hAij ≥ Hij , 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i. (15)
6 Discrete Dynamics in Nature and Society

2.3.5. Complete MSTTP Model. As what are analyzed above,

we set up the complete mixed-speed train traffic planning
(MSTTP) model as follows.

Obj:
max F1 � αSSHR +(1 − α)SAHR,
n n
min F2 � 􏽘 􏽘 hij kij ,
i�1 j�1 (16)
j≠i

max F3 ,

s.t.
h−ij � min􏽮hij , 􏼐􏼐yj + dj 􏼑 − yi + di 􏼁􏼑kij 􏽯, 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i,

hAij � 􏼐􏼐yj + dj 􏼑 − yi + di 􏼁􏼑kij 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i,

F3 ≤ 􏼐hij − Hij 􏼑kij 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i,

yj ≥ yi + hij − Mkji , 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i,

⎧ 1, if yi < yj ,
⎨
kij � ⎩ (17)
0, otherwise,
kij + kji � 1, 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i,

hij ≥ Hij , 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i,

h−ij ≥ Hij , 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i,

hAij ≥ Hij , 1 ≤ i ≤ n, 1 ≤ j ≤ n, j≠i,

M is a large positive number Step 3. Generate P random-key chromosomes and train

α is a weight parameter and 0 < α < 1 sequences by iterating step 2 P times to build the initial
population.
Step 4. For each train sequence, generate the headway
3. Random-Key Genetic Algorithm time hij for the adjacent corresponding genes by in-
(RKGA) Development voking the subalgorithm for generation of train de-
parture headway time (GTDHT).
3.1. Main Algorithm of RKGA. The whole flowchart of the Step 5. For each sequence and headway time, check
main algorithm RKGA is demonstrated in Figure 4. whether the constraints are satisfied or not.
Step 0. Initialize, i.e., determine the parameters of the Step 6. For each sequence and headway time, calculate
total number of trains L, the population P, the the fitness value by invoking the subalgorithm for
crossover probability pc , the mutation probability pm , fitness estimation function of RKGA.
and the maximal iteration time N. Step 7. Sequence the fitness values descendingly.
Step 1. Number each train l, l � 1, 2, . . . , L, N � 1. Step 8. N � N + 1, check the value of N; if N does not
Step 2. Sequence the trains initially. reach the maximal iteration times, turn to step 9, else
stop calculation and output the solution with the largest
Step 2.1. Generate a random number between 0 and 1
fitness.
for each train number, and then take the random
number as the gene. Step 9. Select the top m best solutions according to the
Step 2.2. Combine the random numbers of L trains fitness value.
altogether to form the random-key chromosome. Step 10. For the selected m best solutions, use crossover
Step 2.3. Sequence the trains according to the random operator and mutation operator to generate the new
number genes (from smallest to largest) in the ran- solutions. The search is over the random-key space
dom-key chromosome. instead of the problem space.
Discrete Dynamics in Nature and Society 7

Initialization

Number each train

N=1

Initially sequence trains with random numbers

Build initial populations by generating P random-key chromosomes and train sequences

Generate headway time by sub-algorithm GTDHT

Check the contraints

Calculate fitness value with fitness estimation function

Sequence fitness value descendingly

N=N+1

Yes
N=maximun iteration times Stop

No

Select the top m best solutions

Crossover operation with crossover operator

Mutation operation with mutaton operator

Figure 4: Flowchart of RKGA.

Step 10.1. For the random-key chromosomes corre- Step 11. Repeat step 4～step 8.
sponding to the selected m best solutions, invoke
crossover operator so as to generate a new generation
of random-key chromosomes, and then sequence the 3.2. Fitness Estimation Function of RKGA. For the multi-
trains again according to the random number genes objective F1 , F2 , and F3 , it can be standardized according to
(from smallest to largest) in the random-key
⎪
⎧
⎪ Fi
chromosome. ⎪
⎪ for min Fi ,
Step 10.2. For the random-key chromosomes corre- ⎪
⎪ 1 + Fi
⎨
sponding to the selected m best solutions, invoke F′i � ⎪ (18)
⎪
⎪
mutation operator so as to generate a new generation ⎪
⎪ 1
⎪
⎩ for max Fi .
of random-key chromosomes, and then sequence the 1 + Fi
trains again according to the random number genes
(from smallest to largest) in the random-key Thus, the standardization for the objectives of MSTTP is
chromosome. expressed as
8 Discrete Dynamics in Nature and Society

⎪
⎧
⎪ 1 s s
⎪
⎪
⎪
F1′ � , hij � random􏼐Hij + F3 , Hhij 􏼑 + 􏼠 + 􏼡 ∗ 60. (26)
⎪ 1 + F1 v(i) v(j)
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨ F2
⎪ F2′ � , (19)
⎪
⎪ 1 + F2 3.4. Crossover Operator
⎪
⎪
⎪
⎪
⎪
⎪ Step 1. Assume pc max and pc min are the given upper and
⎪
⎪ 1
⎪
⎪ lower limits for crossover possibility. fl is the fitness
⎩ F3′ � 1 + F .
3 value for a certain chromosome individual. favg is the
average fitness value of the population; i.e.,
Thus, the vector of the standardized objectives for
favg � (􏽐Ll�1 fl /L) · fmax is the maximal fitness value of
MSTTP can be expressed as
the population. We adopt the self-adaption method to
min􏼈F1′, F2′, F3′􏼉. (20) calculate the crossover possibility Pc as

So far, we transform the multiobjective problem to its ⎪

⎧ 􏼐fl − favg 􏼑 pcmax − pcmin 􏼁
⎪
⎪
single-objective counterpart as formula (21) by introducing ⎪
⎪ pcmax − , fl > favg ,
⎨ 􏼐fmax − favg 􏼑
the linear weighting sum method. pc � ⎪
⎪
⎪
min F′ � aF1′ + bF2′ + cF3′, (21) ⎪
⎪
⎩p , fl ≤ favg .
cmax
where the weighting parameters 0 < a < 1, 0 < b < 1, 0 < c < 1, (27)
and a + b + c � 1.
Then, by using formula (22), we can get the fitness
function for the random-key genetic algorithm, which
means the larger the value of F is, the better is for the so- Step 2. Set i � 1.
lution to MSTTP model. Step 3. Judge whether we can impose the crossover
1 operation on chromosome individual i.
max F � . (22)
F′ Step 3.1. Select chromosome individual i, and calculate
However, any solution to MSTTP must satisfy its con- the crossover possibility pi using formula (27).
straints. Considering the possible violence toward the Step 3.2. Generate a random number (0, 1); if random
constraints, we modify the fitness function F by adding a (0, 1) < pi , then put chromosome individual i into the
penalty to it as arrays CROSS for crossover operation.

F � F + penalty, (23) Step 4. i � i + 1. If i ≤ m, turn to step 3, else turn to step 5.

Step 5. Reset i � 1. Assume the number of chromosome
where individuals in the arrays for crossover operation is mc .
0, if all constraints are satisfied, Step 6. Operate single-point crossover.
penalty � 􏼨
a large positive number, otherwise. Step 6.1. Select chromosome individuals i and i + 1
(24) (each contains L genes) from the arrays CROSS.
Step 6.2. An integer position k is selected randomly
between [1, L], and the feature information of chro-
3.3. Subalgorithm for Generation of Train Departure Headway mosome individuals i and i + 1 is exchanged between
Time (GTDHT). Headway time must guarantee that the flow the parents before this point.
of mixed-speed trains traveling at different rates along the Step 6.3. Put the new chromosome individuals into the
rail line section must be continuous. Let v(i) and v(j) denote new population after crossover operation.
the speed types (unit: km/h) of adjacent trains, Hhij the Step 7. i � i + 2. If i < mc , turn to step 6; else turn to step 8.
maximum reasonable headway time (unit: minute) between
adjacent trains (which can be estimated from railway op- Step 8. The crossover operation is over.
erational practice), s the length of the railway line section
(unit: km), and the expression random(δ, φ) generates a
3.5. Mutation Operator
random number in (δ, φ). The pseudocode for subalgorithm
for generation of train departure headway time (GTDHT) is Step 1. Assume pm max and pm min are the given upper
as follows. and lower limits for mutation possibility. fl is the
If v(i) ≥ v(j), fitness value for a certain chromosome individual. favg
is the average fitness value of the population; i.e., favg �
hij � random􏼐Hij + F3 , Hhij 􏼑. (25) (􏽐Ll�1 fl /L) · fmax is the maximal fitness value of the
population. We adopt the self-adaption method to
Else,
calculate the mutation possibility pm as
Discrete Dynamics in Nature and Society 9

⎪
⎧ 􏼐fl − favg 􏼑 pm max − pm min 􏼁 Table 1: Identifier and speed type of trains within one cycle of the
⎪
⎪ timetable.
⎪
⎪ pm max − , fl > favg ,
⎨ 􏼐fmax − favg 􏼑
pm � ⎪ Serial number Train identifier Speed type of train (km/h)
⎪
⎪
⎪
⎪ 1 L01 300
⎩p , f1 ≤ favg .
m max 2 L02 300
(28) 3 L03 300
4 L04 270
Step 2. Set i � 1. 5 L05 270
6 L06 270
Step 3. Judge whether we can impose the mutation 7 L07 250
operation on chromosome individual i. 8 L08 250
Step 3.1. Select chromosome individual i, and calculate 9 L09 250
the mutation possibility pi using formula (28). 10 L10 250
11 L11 250
Step 3.2. Generate a random number (0, 1) in (0, 1); if
12 L12 230
random (0, 1) < pi, then turn to step 4 because 13 L13 230
chromosome individual i can be executed by the 14 L14 200
mutation operation, else turn to step 5. 15 L15 200
Step 4. Operate mutation.
Step 4.1. For chromosome individual i, an integer Table 2: Theoretical safety minimal headway time of subsequent
position k is selected randomly between (1, L), and trains (unit: minute).
generate a new random number (0, 1) to replace its 300 270 250 230 200
counterpart corresponding to position k.
300 5 5 3 3 3
Step 4.2. Put the new chromosome individuals into the
270 9 5 4 5 3
new population after crossover operation. 250 15 10 4 4 3
Step 5. i � i + 1. If i < m, then turn to step 3, else turn to 230 18 14 12 4 3
step 6. 200 26 20 18 16 4
Note. The items in the first column represent the speed type of the pro-
Step 6. Mutation operation is over. ceeding train, the items in the first row represent the speed type of the
following train, and other numbers in the table represent the theoretical
4. Results and Discussion of a Case Study safety minimum headway time of adjacent trains.

4.1. Experiment Design. In the case study, the rail system

and the scenario s3 belongs to the kind of homogeneous
considered is a line section encompassing a territory of
traffic condition. Statistically, 42.86% of the 14 sequenced
180 km with 15 mixed-speed trains in each cycle of the
train pairs fall into the scenario s1, and the percentage of two
timetable. The details about the trains are shown in Table 1.
other scenarios are equal to each other, both of which are
According to Tian [29], we determine the theoretical safety
28.57%, as shown in Table 4. As can be seen, both the
minimum headway time of subsequent trains as Table 2.
And, we set the parameters for experimental trials as Table 3. heterogeneous traffic and the homogeneous traffic coexist in
the optimized train sequences. The homogeneous traffic
scenario s3 emerges in the top three high speeds of trains,
4.2. Results of Fitness and Multiobjective Value. The algo- e.g., 300 km/h, 270 km/h, and 250 km/h. Particularly, three
rithm described above is coded with visual studio C#, and it trains with the highest speed, i.e., 300 km/h, are arranged
can work out the optimized results within two minutes. The successively. Totally, the heterogeneity/homogeneity ratio is
results for fitness value in Figure 5 are obtained by executing 2.5.
the RKGA in 100, 150, and 200 generations individually.
After experimental trials in 200 generations, we get the
optimal heterogeneity value, i.e., 1.6121; the optimal cycle 4.4. Results of Train Headway Time. The optimized train
time, i.e., 179.7565 min; and the optimized minimal buffer departure headway time ranges from 4 min to 31 min, and its
time, i.e., 1 min, respectively. arrival counterpart disperses from 5 min to 20 min, as shown
in Figure 6. In Figure 6, the horizontal axis represents the
train sequence number, while the vertical axis represents the
4.3. Results of Train Sequence. In general, the speed rela- headway time. Among all of the departure/arrival headway
tionship between the sequential train pairs can be grouped times, i.e., the dep.time headway and the arr.time headway,
into three traffic scenarios such as s1, s2, and s3. In scenario the smallest departure headway time emerges between trains
s1, the speed of the previous train is greater than that of its L13 and L14, the speed of which is 230 km/h and 200 km/h,
following counterpart. In scenario s2, the speed of the prior respectively; the largest departure headway time emerges
train is less than that of its succeeding one, while in scenario between trains L15 and L07, the speed of which is 200 km/h
s3, the speed of two adjacent trains is identical to each other. and 250 km/h, respectively. Obviously, the variance of the
The former two scenarios produce the heterogeneous traffic, departure headway time is sharper than that of its arrival
10 Discrete Dynamics in Nature and Society

Table 3: Parameters for experimental trials.

Serial number Parameter name Parameter value Parameter meanings
1 P 100 Population scale
2 M 40 Number of selected chromosome individuals
3 pc min 0.4 Lower limit of crossover possibility
4 pc max 0.8 Upper limit of crossover possibility
5 pm min 0.001 Lower limit of mutation possibility
6 pm max 0.1 Upper limit of mutation possibility
7 α 0.6 Weight of SSHR
8 A 0.5 Weight of objective for passenger optimization
9 B 0.3 Weight of objective for capacity optimization
10 C 0.2 Weight of objective for operation optimization

1.7

1.695

1.69
fitness value

1.685

1.68

1.675

1.67
100 150 200
generation
Figure 5: The fitness value of generations.

Table 4: Optimized sequence of trains in one cycle of timetable.

L01 L02 L03 L04 L05 L06 L07 L08 L09 L10 L11 L12 L13 L14 L15
L01 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
L02 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
L03 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
L04 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
L05 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
L06 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
L07 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
L08 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0
L09 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0
L10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
L11 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
L12 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0
L13 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
L14 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
L15 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
Note. The first column represents the proceeding train in the sequence, the first row represents the following train in the sequence, the number 1 in the table
represents two trains are adjacent to each other, and the number 0 represents two trains are not adjacent.

counterpart. As far as the three scenarios defined before are 4.5. Discussion. In the current work, we built the MSTTP
concerned, in scenario s1, the departure headway time is model and RKGA for structure optimization of the mixed-
always less than its corresponding arrival headway time over speed train traffic for a cyclic timetable in a passenger railway
the trains’ running distance; in scenario s2, the situation is line section. They are of permutation nature. Our study
completely reverse; in scenario s3, they are equal. includes a high diversity of train speeds. With the known
Discrete Dynamics in Nature and Society 11

35

30

25

20

15

10

0
0 2 4 6 8 10 12 14 16
dep. time headway (min)
arr. time headway (min)
Figure 6: Spread of departure/arrival headway time.

average speed and given number of trains, we get the op- and arrival headway time. In turn, these findings would be
timized train plans from random variations of train se- instrumental in a further revision of the theoretical safety
quence and headway time, which can represent a more minimal headway time of subsequent trains as shown in
general significance from the mathematical point of view. Table 2, which is the prerequisite for train planning.
Moreover, it ensures the robustness of the timetable as
possible as it can by means of random permutation and 5. Conclusion
iteration.
Compared with the previous studies, we find the Trains can be optimally spread over the period of the cyclic
quantitative proportion of heterogeneity/homogeneity (e.g., timetable. Focused on the structure optimization of the
2.5) about the optimized distribution of the mixed train mixed-speed train traffic for cyclic timetable, this paper puts
traffic, instead of the merely qualitative assessment. It is a forward a new mixed-speed train traffic planning (MSTTP)
mathematical fact. The proportion is greater than 1, the model and the random-key genetic algorithm (RKGA). The
potential reason for which is that we allocate greater weight MSTTP model is recast from the alternative graph (AG) and
value for the objective of heterogeneity, i.e., parameter optimizes three indicators, i.e., heterogeneity, cycle time,
a � 0.5, from the perspective of passenger optimization. and buffer time, which correspond to diversity of train
Usually, rail operators tend to think in terms of moving service toward passenger, capacity consumption of rail
vehicles instead of people. However, putting customer first network, and stability of train operation, respectively.
and tailoring the system to their needs is critical to success. Compared with the conventional algorithms, the proposed
In tune with this viewpoint, we set the values for parameters model and algorithm have the following advantages:
a, b, and c as Table 3 in the phase of experiment design. To a
(i) Both sequence and headway time are taken into
more rational extent, we achieve a comprehensively bal-
account as the core components of the timetable
anced solution (Table 4, Figures 5 and 6) among all of the
structure simultaneously. Taken together, the key
timetabling stakeholders from the multiobjective optimi-
results of our study have been the optimal ar-
zation. Moreover, the higher the speed of the train is, the
rangement of the mixed-speed train traffic, by in-
higher the possibility is that the trains of the same speed are
tegrating the sequencing problem with the headway
sequentially bundled together to form the homogeneous
time optimizing together for a cyclic timetable, and
traffic in a mixed traffic condition.
demonstration of how they can be achieved.
We find the link between the train headway time and the
train sequence. For the mixed-speed train traffic, e.g., the (ii) The time complexity of the proposed algorithm is
traffic scenarios s1 and s2, the train headway time is not only low, and it can satisfy the requirement of real-time
a function of its speed, but also sequence dependent. In and fast computation.
scenario s1, the minimal headway time of the sequential (iii) Optimum headway time and train sequence can be
trains in the railway line section equals their corresponding recommended according to the concerns of the
departure headway time; in scenario s2, with other similar stakeholders, i.e., heterogeneity, cycle time, and
conditions, the minimal headway time of the sequential buffer time, which correspond to diversity of train
trains in the railway line section equals their corresponding service toward passenger, capacity consumption of
arrival headway time. However, for the homogeneous traffic rail network, and stability of train operation,
as scenario s3, there is no difference between the departure respectively.
12 Discrete Dynamics in Nature and Society

(iv) Furthermore, it can be meaningful for the reason- Research and Management Science, vol. 17, no. 2, pp. 85–96,
able allocation of buffer time, which is related to the 2012.
stability of the timetable. [9] J. Zhang, “Analysis on comprehensive balance of train op-
eration plan for high speed railway under mixed traffic
In addition, when taking the overtaking issues into condition,” in Proceedings of the SOL, pp. 397–342, Beijing,
account, further studies should explore the dynamic head- China, 2011.
way time according to the change of train sequence at the [10] P. Huang, C. Wen, Q. Peng, J. Lessan, L. Fu, and C. Jiang, “A
overtaking place. As enumerated in Landex and Kaas [5], we data-driven time supplements allocation model for train
can firstly divide the line section due to overtaking, then operations on high-speed railways,” International Journal of
dealing with each section as such a subproblem as we Rail Transportation, vol. 7, no. 2, pp. 140–157, 2019.
modeled in this article. In this way, the problem studied in [11] K. Pavle, Models for Predictive Railway Traffic Management,
Ph.D. Dissertation, Delft University of Technology, Delft,
this paper occurs as a subproblem in a decomposition
Netherland, 2014.
procedure for more complicated railway line sections with [12] J. Carlier, “The one-machine sequencing problem,” European
overtaking stations. Journal of Operational Research, vol. 11, no. 1, pp. 42–47, 1982.
[13] A. D’Ariano, Improving Real-Time Train Dispatching: Models,
Data Availability Algorithms and Applications, Ph.D. Dissertation, Delft Uni-
versity of Technology, Delft, Netherland, 2008.
The data that support the findings of this study are included [14] C. Artigues and D. Feillet, “A branch and bound method for
within the article. the job-shop problem with sequence-dependent setup times,”
Annals of Operations Research, vol. 159, no. 1, pp. 135–159,
2008.
Disclosure [15] W. J. Lars, Robustness Indicators and Capacity Models for
Railway Networks, Ph.D. Dissertation, Technical University of
The authors certify that the authors have participated suf- Denmark, Copenhagen, Denmark, 2015.
ficiently in the work to take public responsibility for the [16] J. H. Holland, Adaptation in Natural and Artificial Systems,
appropriateness of the experimental design and method and University of Michigan Press, Ann Arbor, Michigan, USA,
the collection, analysis, and interpretation of the data. 1975.
[17] K. A. De Jong, An Analysis of Behavior of a Class of Genetic
Adaptive Systems, Ph.D. Dissertation, University of Michigan,
Conflicts of Interest Michigan, USA, 1975.
The authors declare no conflicts of interest. [18] J. C. Bean, “Genetic algorithms and random keys for se-
quencing and optimization,” ORSA Journal on Computing,
vol. 6, no. 2, pp. 154–160, 1994.
Acknowledgments [19] O. Ghrayeb and P. Damodaran, “A hybrid random-key ge-
netic algorithm to minimize weighted number of late deliv-
The authors would like to thank the authors of the articles eries for a single machine,” International Journal of Advanced
mentioned in the reference list for their inspirations. Manufacturing Technology, vol. 66, no. 1-4, pp. 15–25, 2013.
[20] M. J. C. M. Vromans, R. Dekker, and L. G. Kroon, “Reliability
and heterogeneity of railway services,” European Journal of
References
Operational Research, vol. 172, no. 2, pp. 647–665, 2006.
[1] UIC, UIC Code 406), International Union of Railways(UIC), [21] J. Parbo, O. A. Nielsen, and C. G. Prato, “Passenger per-
Paris, France, 2004. spectives in railway timetabling: a literature review,” Trans-
[2] A. Landex, Methods to Estimate Railway Capacity and Pas- port Reviews, vol. 36, no. 4, pp. 500–526, 2016.
senger Delays, Ph.D. Dissertation, University of Denmark, [22] UIC, UIC Code 406), International Union of Railways(UIC),
Copenhagen, Denmark, 2008. Paris, France, 2013.
[3] R. L. Burdett and E. Kozan, “Techniques for absolute capacity [23] J. Zhang, “Analysis on line capacity usage for China high
determination in railways,” Transportation Research Part B: speed railway with optimization approach,” Transportation
Methodological, vol. 40, no. 8, pp. 616–632, 2006. Research Part A: Policy and Practice, vol. 77, pp. 336–349,
[4] A. H. Kaas, Methods of Calculation of Railway Capacity, Ph.D. 2015.
Dissertation, University of Denmark, Copenhagen, Denmark, [24] M. Heydar, M. E. H. Petering, and D. R. Bergmann, “Mixed
1998. integer programming for minimizing the period of a cyclic
[5] A. Landex, A. H. Kaas, B. Schittenhelm, and J. Schneider-Tilli, railway timetable for a single track with two train types,”
“Evaluation of railway capacity,” in Proceedings of the Annual Computers & Industrial Engineering, vol. 66, no. 1,
Transport Conference, vol. 13, no. 1, Aalborg, Denmark, 2006. pp. 171–185, 2013.
[6] M. Yaghini, M. Sarmadi, N. Nikoo, and M. Momeni, “Ca- [25] M. E. H. Petering, M. Heydar, and D. R. Bergmann, “Mixed-
pacity consumption analysis using heuristic solution method integer programming for railway capacity analysis and cyclic,
for under construction railway routes,” Networks and Spatial combined train timetabling and platforming,” Transportation
Economics, vol. 14, no. 3-4, pp. 317–333, 2014. Science, vol. 50, pp. 892–909, 2015.
[7] L. Kroon, D. Huisman, and G. Maróti, Railway Timetabling [26] X. Zhang and L. Nie, “Integrating capacity analysis with high-
from an Operations Research, Econometric Institute Research speed railway timetabling: a minimum cycle time calculation
Papers, Rootterdam, Netherlands, No. EI 2007-22, 2007. model with flexible overtaking constraints and intelligent
[8] S. S. Harrod, “A tutorial on fundamental model structures for enumeration,” Transportation Research Part C: Emerging
railway timetable optimization,” Surveys in Operations Technologies, vol. 68, pp. 509–531, 2016.
Discrete Dynamics in Nature and Society 13

[27] R. M. Lusby, J. Larsen, and S. Bull, “A survey on robustness in

railway planning,” European Journal of Operational Research,
vol. 266, pp. 1–15, 2017.
[28] S. Burggraeve, S. H. Bull, P. Vansteenwegen, and R. M. Lusby,
“Integrating robust timetabling in line plan optimization for
railway systems,” Transportation Research Part C: Emerging
Technologies, vol. 77, pp. 134–160, 2017.
[29] C. Tian, S. Zhang, S. Zhang, and X. Jiang, “Study on the train
headway on automatic block sections of high speed railway,”
Journal of the China Railway Society, vol. 10, pp. 1–6, 2015.