You are on page 1of 15

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/347519139

Integrated Overnight Train Scheduling and Maintenance Planning for High-


Speed Railway Lines

Article  in  Transportation Research Record Journal of the Transportation Research Board · December 2020
DOI: 10.1177/0361198120968830

CITATIONS READS

0 162

1 author:

Shuguang Zhan
Southwest Jiaotong University
20 PUBLICATIONS   206 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Passenger-oriented train rescheduling and passenger oriented train timetabling View project

All content following this page was uploaded by Shuguang Zhan on 23 December 2020.

The user has requested enhancement of the downloaded file.


Journal Title
XX(X):1–14
Integrated Over-night Train Scheduling c The Author(s) 2020
Reprints and permission:
and Maintenance Planning for sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/ToBeAssigned

High-speed Railway Lines www.sagepub.com/

SAGE

Dian Wang1 , Shuguang Zhan1,2,3 , Qiyuan Peng1,2,3 and Wentao Zhou1

Abstract
Over-night high-speed trains are very popular and convenient for passengers in countries with a large territory like
China. However, the over-night high-speed train operation is inevitably conflict with the evening regular maintenance.
We focus on both over-night high-speed train scheduling and maintenance planning to eliminate the conflict. Because
part of day-time high-speed trains that run early in the morning or late in the evening also interact with over-night
high-speed trains and maintenance, we also allow to slightly reschedule them to improve both the quality of over-night
train timetable and the maintenance plan. Our integrated optimization problem is formulated as a mixed integer linear
programming model, which can be solved efficiently by commercial solver CPLEX. Finally, we validate our model on a
large real-world case constructed based on the Beijing-Guangzhou high-speed railway line in China.

Keywords
High-speed railway, over-night train scheduling, railway maintenance planning, mixed integer linear programming

Introduction is finished, leading to conflicts between over-night trains


and the maintenance. To eliminate these conflicts, the
High-speed railway network expands fast in the world,
maintenance time of some segments is compressed to less
especially in China during the past 10 years. According to
than 4 hours, and the maintenance might be postponed, e.g.,
the short-term plan of Chinese high-speed railway, the total
the maintenance of the segment tracks between stations m2
length of high-speed railway will reach 30,000 km in the year
and m3 in Figure 1 continues after 06:00. Therefore, over-
of 2020, and most cities with a population above 500,000 will
night trains can use the spare time between the maintenance
be served by high-speed railway. Although high-speed trains
of two consecutive railway segments, which is shown by G in
run very fast with a speed of 250∼350km/h in China, it may
Figure 1. Note that the end time of maintenance is restricted
take a long time to travel from one city to another due to the
by the trains operated in the early morning, e.g., day-time
large size of the country. There are more than 1,000 origin
regular trains r3 and r4 in Figure 1.
and destination pairs in China, among which 328 pairs are
longer than 2,000 km. For such a long OD pair, it is quite 22:00 00:00 06:00 08:00
m1 m1
preferable to operate over-night high-speed trains (trains that i1 i2
r1 o1 o3 Maintenance
o2 o4 r3
depart during 19:00-22:00 in the evening and arrive during m2 m2
o1 o3 o2 o4 i1 r4
06:00-08:00 in the next morning), so that passengers do not G r3
i4
spend their daytime on these trains. Besides, the over-night m3 m3
i3 o1 o3 r4
r2 o2 o4
train can provide higher level of service because it is a sleeper i4
m4 m4
train and passengers can have a good sleep and start their regular train over-night train inspection train
work in the next morning. Until now, 26 over-night high-
speed trains are operated on the Chinese high-speed railway Figure 1. A sample train timetable and maintenance plan
network. These trains are quite popular, and all the tickets
of the over-night trains operating on the Beijing-Guangzhou Because the maintenance time of some segment tracks
high-speed railway line in China are sold out several days is compressed (less than 4 hours), the railway maintenance
before the planned departure time from the ticket booking staffs cannot finish all the required maintenance tasks during
website (https://www.12306.cn/index/).
Though the over-night high-speed trains are convenient
1 School of Transportation and Logistics, Southwest Jiaotong University,
and popular for passengers, they also bring out some
problems. One of the most important issues is that the over- Chengdu, Sichuan, China
2 National United Engineering Laboratory of Integrated and Intelligent
night trains are conflict with the regular evening maintenance Transportation, Southwest Jiaotong University, Chengdu, Sichuan, China
and the day-time regular trains operated in the late evening 3 National Engineering Laboratory of Integrated Transportation Big

and early morning. To ensure the safety of train operations, Data Application Technology, Southwest Jiaotong University, Chengdu,
maintenance is taken every evening normally during 00:00 to Sichuan, China
06:00 in the Chinese high-speed railway and lasts for 4 hours Corresponding author:
in principle. During the maintenance of a railway segment, Shuguang Zhan
trains cannot run on the segment until the maintenance Email: shuguangzhan@my.swjtu.edu.cn

Prepared using sagej.cls [Version: 2017/01/17 v1.20]


2 Journal Title XX(X)

the evening, and the train operation safety cannot be well trains and allow to reasonably revise them, so that a better
ensured. To this end, over-night trains can only operate on train timetable can be obtained. Third, after the maintenance
a certain number of days instead of every day. Currently, of each segment track, a comprehensive inspection train
the Chinese high-speed railway operators only operate over- (e.g., trains i1 ∼i4 in Figure 1) must be operated before the
night high-speed trains on Friday, Saturday, Sunday and regular and over-night trains in China to ensure the safety
Monday every week with a compressed maintenance time, of train operations. Thus, we consider the comprehensive
and maintenance can take a longer time during other three inspection trains as well. Fourth, because the maintenance
days to finish all the required maintenance tasks in the whole of each segment requires a continuous period of time, we
week. Although the whole maintenance tasks in a week are do not allow to separate the maintenance time of a segment.
finished, over-night trains cannot operate from Tuesday to As it is preferred to do the maintenance of consecutive
Thursday. In addition, the spare time during two consecutive segments simultaneously to facilitate the maintenance tasks,
segments by compressing maintenance time is limited, and the difference of the maintenance start time of consecutive
only a limited number of over-night trains can operate by segments is considered and minimized in this study. Overall,
using this spare time on the days when over-night trains are we will investigate when the considered (day-time regular,
operated. Thus, the capacity of over-night trains cannot meet over-night, and inspection) trains will arrive at and depart
the growing passengers’ requirement. from the stations along their planned path, which station
Theoretically two main strategies may be available to track is used to stop each of the considered train, and when
solve the conflict between over-night high-speed train to start the maintenance on each segment, with the aim of
operation and maintenance tasks. The first one (called optimizing particular objectives.
waiting strategy) is that when a segment is under The contributions of this paper are threefold. First, we
maintenance, trains stop at stations before the maintenance investigate the ITSMP problem with considerations of the
location to wait for the end of the maintenance. The above practical requirements, so that the train timetable
second one (called switching strategy) is to use the parallel and maintenance plan obtained in this paper are practically
conventional railway lines (normally of a lower design acceptable. Second, we formulate the investigated problem
speed) to organize high-speed trains during the maintenance. as a novel mixed integer linear programming (MILP)
That is, if a segment of a high-speed railway line is under model, in which all of the above practical requirements are
maintenance, over-night high-speed trains can switch to addressed. The proposed model can be efficiently solved by
run on the conventional line. And then they can either commercial solvers, and tight lower bounds can be found to
keep running on the conventional line to their destination assess the quality of a (practical) feasible solution. Third,
or switch back to the high-speed railway line after the a large realistic instance, constructed based on one of the
maintenance is finished. We refer to 1 for details about these longest and busiest real-world high-speed railway lines in
two strategies. However, because the standard and the control China, is used to test the capability of our model. The
system of the conventional railway lines are quite different benefit of applying new electric multiple units (EMUs) is
from those of the high-speed railway lines, high-speed trains also investigated, which is motivated by our project from
are not allowed to switch to conventional lines in practice the CRRC Tangshan Corporation Limited to explore the
on Chinese railway until now. Therefore, in this research we probability of operating over-night trains every evening by
focus on the waiting strategy. Under this strategy, one of the using new high-speed EMUs.
main decisions is how to stop high-speed trains during the The remainder of this paper is organized as follows.
maintenance. Section Literature review gives the related literature review.
The integrated over-night train scheduling and mainte- We briefly describe our problem in Section Problem
nance planning (ITSMP) problem is currently addressed by description and formulate it as a MILP model in Section
the Chinese railway operators manually based on experi- Model formulation. Section Case study presents the
ence with the help of a visualize tool, and it has attracted computational results, and we finally conclude this research
attention among the academia in recent years. In this paper, and discuss some future research directions in Section
we investigate the ITSMP problem on Chinese high-speed Conclusions.
railway lines by using the operations research methods, with
the aims of obtaining good quality solutions and releasing
Literature review
the solution burden of the railway operators. The novelty
of our work lies on the following considerations to reflect Our problem needs to schedule the over-night trains and
the practical requirement as much as possible. First, we reschedule the day-time regular trains if necessary while
consider all of the stations along a railway line, because the the regular maintenance is implemented in the evening.
capacity of each station affects the practical feasibility of the In the macroscopic perspective, the train scheduling and
obtained solution. Considering that there are usually a few rescheduling problems mainly lie on determining the arrival
tracks in some of the stations along a high-speed railway and departure times of trains at their visited stations. Until
line, we suppose that over-night trains could occupy a station now, the train scheduling and rescheduling problems have
track which is usually utilized by trains in the opposite received extensive attentions. Interested readers can refer
direction (coincide with the practice). Second, because the to 2–4 for train scheduling and to 5,6 for train rescheduling.
day-time regular trains that run late in the evening before Among the previous train scheduling/rescheduling papers,
the maintenance and early in the next morning just after the those focusing on scheduling additional trains on an existing
maintenance (e.g., trains r1 ∼r4 in Figure 1) interact with the timetable are most relevant to our study. Trains in the
over-night high-speed trains, we also consider these regular existing timetable can be either unchanged 7 or changed 8–11 .

Prepared using sagej.cls


Wang, Zhan, Peng and Zhou 3

In these previous studies, few of them consider the additional however, the capacity of every station should be checked. In
accelerating and decelerating times 10,11 and rerouting of addition, considering that synchronous maintenance window
trains at visited stations 7,11 . However, to obtain a high- is more convenient for maintenance, it is better to start the
quality over-night train schedule, day-time regular trains are maintenance of several consecutive sections at the same time.
allowed to be reasonably revised in our problem. Besides, In other words, the difference of the maintenance start time
the additional accelerating and decelerating times as well as of consecutive segments should be minimized, which is out
the route of trains at stations need to be exactly considered to of consideration in 1 . Finally, 1 assumed that all the regular
improve the practical feasibility of the obtained timetable. day-time high-speed trains are fixed when inserting the over-
We also need to schedule the maintenances which cause night high-speed trains. But the interaction between a regular
unavailability of segment tracks in some periods. Railway day-time high-speed train and an over-night high-speed train
infrastructure maintenance is a complex task, and it is indeed exists. Thus, it is possible to revise the schedule of
important for maintenance budget and train operations. some day-time high-speed trains a little bit (as done in our
Though some previous papers have solved the maintenance work) to improve the quality of the over-night high-speed
planning problem, most of them only focus on optimizing train timetable and the maintenance plan.
the maintenance plan, and trains are usually scheduled at a
later stage 12 . To the best of our knowledge, few contributions Problem description
have been made to deal with how to organize trains due to the
unavailability of tracks caused by maintenances. Generally, We focus on double-track high-speed railway lines. Each
the previous research focusing on both maintenance planning segment track is used by trains in one direction, and a station
and train scheduling can be classified into three categories: track can be used by trains in both directions. Our problem
fixed timetable and variable maintenance plan 13–15 , fixed mainly contains three parts. One is train scheduling, another
maintenance plan and variable timetable 16–20 , and variable is regular maintenance planning on high-speed railways, and
maintenance plan and variable timetable 21–26 . Different from the third is how to integrate the former two parts. Hence, we
the railways in Australia and Europe, the train scheduling explain our problem from three aspects in detail.
and maintenance planning are implemented by the same
railway operator in China, which makes it possible and Over-night high-speed train scheduling
necessary to simultaneously schedule maintenance plan and As mentioned previously, we need to operate over-night
train schedule to obtain better results. Therefore, our problem high-speed trains to meet the passengers’ demand due
belongs to the third category, i.e., variable maintenance plan to large territory in China. Over-night high-speed trains
and variable timetable. normally operate on a high-speed railway line with a length
In the most relevant studies that focus on variable of about 2,000 km. Trains depart from a large city in the
maintenance plan and variable timetable, 21 optimized the evening between 19:00 and 22:00, and finally arrive at
train schedule and maintenance of a single-track railway another large city in the next morning between 06:00 and
network. The maintenance task can be interrupted and 08:00, with a limited number of stops in some intermediate
arranged in several time slots. Therefore, the influence of stations.
maintenance on train operations can be reduced. 22 focused To schedule over-night high-speed trains, there are several
on the integrated problem for a railway network with both key issues to be considered. First, we need to determine
single-track and double-track lines. They allowed to cancel which high-speed railway line is suitable to operate this
some trains in the planned timetable. Similarly, 23 and 24 kind of trains considering both the appropriate departure and
allowed to split the maintenance task and cancel some trains arrival times and the passengers’ demand. Second, we need
in the planned timetable simultaneously. The considered to determine how many trains should be operated on each
situations in the above studies are different from that in line every day with limited line capacity. Third, the stopping
Chinese high-speed railway, and their methods are not able pattern (to board and alight passengers) of each train should
to be directly used to solve our investigated problem. 25 and 26 be carefully designed so that it is convenient for passengers
optimized train schedule and infrastructure maintenance plan to travel by these trains. Normally, over-night trains do not
from a microscopic perspective. Because there are usually have planned stops to board and alight passengers during
too many variables and constraints in a microscopic model, 01:00 and 05:00. However, they might have to stop to wait
it is quite difficult to be applied to a large real-world case. In for the maintenance. Fourth, the departure and arrival time
addition, the model in 25 can be large when a fine granularity of each train at each station as well as the station track to be
of seconds is used as the time interval. However, a fine occupied should be decided without any conflict with others.
granularity is necessary for high-speed railways. When it Fifth, because over-night trains partly run together with day-
comes to adding extra over-night high-speed trains into time regular trains and comprehensive inspection trains, the
the existing train timetable while considering maintenance, conflict among these trains should be strictly eliminated.
only 1 studied this problem for Chinese high-speed railway Considering that the number of over-night trains, their
lines. They introduced three strategies to organize over- paths and their planned stopping patterns are predetermined
night high-speed trains. Over-night high-speed trains were before making the detailed train timetable and maintenance
allowed to switch to parallel conventional railway lines when plan in China, they are regarded as input in our study. The
the high-speed railway segment was under maintenance. timetable of day-time regular trains is also given and we try
However, this is not currently allowed on Chinese railway. to add over-night trains to the planned timetable. To improve
Besides, they only consider the station capacity of some the quality of over-night train timetable and maintenance
stations. To ensure the feasibility of train operations, plan, we allow to slightly revise the timetable of day-time

Prepared using sagej.cls


4 Journal Title XX(X)

regular trains. Here, only the regular trains operated in the want to pass this track have to stop at previous stations to
late evening and early morning are considered to be revised wait until the end of the maintenance. Therefore, we need to
as they might be affected by over-night trains, and the decide the order of trains passing each segment track and the
others are unchangeable to ensure feasibility of the obtained start time of its maintenance. In addition, due to the limited
solutions within the whole day. station capacity, we also need to decide which train should
stop at which station track to wait for the maintenance. Our
Maintenance planning on high-speed railway goal is to appropriately arrange the required maintenance
tasks and minimize the influence of maintenance on train
Maintenance activities on a high-speed railway line mainly
operations.
refer to track maintenance, power-supply maintenance,
signaling-system maintenance and other fixed equipment
maintenance 1 . To ensure the safety of train operations, Model formulation
maintenance should be done every evening, and 4 hours is
Notation
required to fully inspect and repair the involved facilities
on the Chinese high-speed railway. When a segment track We formulate the ITSMP problem as a mixed integer linear
is under maintenance, no train is allowed to enter it. Thus, programming (MILP) model in this section. In what follows,
the maintenance is usually taken between 00:00 and 06:00 the over-night train, the regular train and the inspection
because day-time regular trains do not operate during this train represent the over-night high-speed train, the day-time
time period on the Chinese high-speed railway. Different regular high-speed train and the comprehensive inspection
from the low-speed railway, as the speed of a high-speed train train, respectively. The notations used in the model are
is very high, when one segment track is under maintenance, explained in Tables 1 and 2.
the other parallel track in the same segment is not allowed
to be used for train operation due to safety reasons. Objective function
Instead, maintenance is taken on both tracks in a segment Based on problem definition, we need to simultaneously
simultaneously for a double-track high-speed railway. minimize the total travel time of over-night trains, the total
In this situation, we have two types of maintenance win- deviation of regular trains from the planned timetable, and
dows: synchronous maintenance window and asynchronous the total absolute difference between maintenance start times
maintenance window. The former is that all tracks in all of consecutive segments. We only focus on downstream
segments of a high-speed railway line start and end the main- segment tracks as the maintenances are simultaneously taken
tenance at the same time. This type of maintenance window on both tracks in a segment. The objective function is as
is convenient and safe for maintenance tasks. However, it not follows.
only significantly affects the operation of over-night trains X X X
as they need to wait for a long time, but also increases min β1 · (yisi − xioi ) + β2 · (δim + εim )
passengers’ travel time. The latter is to divide the whole i∈T1 i∈T2 m∈Ai
railway line into several maintenance sections, each of which X
+ β3 · αmn
contains at least one segment between two stations. Differ-
(m,n)∈E|n6=|S|
ent maintenance sections can start maintenance at different
(1)
times, see in Figure 1. This type of maintenance window can
reduce the waiting time of over-night trains, but might affect The first part of the objective function is the total travel
the maintenance efficiency and safety. The maintenance of time of over-night trains, where (yisi − xioi ) represents
the whole railway line will last for a longer time as well. the travel time of train i ∈ T1 . The second part is the
Currently, the second type of maintenance window is applied total absolute arrival and departure deviation of the regular
on the Chinese high-speed railway because it is quite flexible. trains to the planned timetable, and the third part is the
Therefore, our research focuses on the second type. The total absolute difference between the maintenance start
maintenance window should be well organized to facilitate times of consecutive segments. The weights β1 ∼β3 can be
both the maintenance tasks and over-night train operations. It determined by the decision makers (e.g., railway operators)
is preferred to simultaneously start the maintenance of con- based on their preference, where β1 + β2 + β3 = 1.
secutive segments in the maintenance aspect. Overall, in the
maintenance planning problem, we mainly need to determine
how many maintenance sections a high-speed railway line
Constraints
should be divided into, and the start time of maintenance Train dwell and running constraints. To maintain practical
for each segment, so as to minimize the influence of the feasibility of train timetable, specified train dwell and
maintenance on over-night train operations and to facilitate running requirements at visited stations and segment tracks
maintenance tasks as much as possible. should be strictly respected, which are as follows.

yim − xim ≥ bim ∀i ∈ T, ∀m ∈ Ai (2)


Integrated over-night train scheduling and
maintenance planning min
rimn + qi1 · wim + qi2 · win ≤ xin − yim
(3)
Because the time for maintenance and that for over-night ∀i ∈ T, ∀(m, n) ∈ Bi
train operations are overlapping, we should appropriately
max
arrange these two tasks. As mentioned previously, if one xin − yim ≤ rimn + qi1 · wim + qi2 · win
(4)
segment track is under maintenance, over-night trains that ∀i ∈ T, ∀(m, n) ∈ Bi

Prepared using sagej.cls


Wang, Zhan, Peng and Zhou 5

Table 1. Definition of Sets, Indices and Parameters Table 2. Definition of Decision Variables
Notations Description Notations Description
Sets xim Nonnegative integer variable, represents the actual arrival
S Set of stations. Stations are indexed sequentially by the time of train i ∈ T at station m ∈ Ai .
downstream direction, S = {1, 2, · · · , |S|}, where |S| yim Nonnegative integer variable, represents the actual depar-
is the total number of stations. ture time of train i ∈ T at station m ∈ Ai .
T Set of trains, T = T1 ∪ T2 ∪ T3 , T = wim Binary variable, 1 if train i ∈ T stops at station m ∈ Ai ;
{1, 2, . . . , |T |}. 0 otherwise.
T1 , T2 , T3 Set of over-night trains, regular trains and inspection uijmn Binary variable, for every two trains i and j (ϕij = 0,
trains, respectively. Each type of trains is further i < j) and every segment track (m, n) ∈ Bi ∩ Bj , 1 if
composed of downstream trains and upstream trains. the actual departure time of train i to enter the segment
E Set of segment tracks which consists of downstream track is earlier than that of train j; 0 otherwise.
segment tracks and upstream segment tracks. pijm Binary variable, for every two trains i and j (i 6= j) and
Ai Set of stations visited by train i ∈ T based on the every station m ∈ Ai ∩ Aj , 1 if the actual departure time
planned path of the train, Ai ⊂ S. of train i at the station is earlier than the actual arrival time
Bi Set of segment tracks visited by train i ∈ T based on of train j at the station; 0 otherwise.
the planned path of the train, Bi ⊂ E. zmn Nonnegative integer variable, represents the maintenance
Km Set of station tracks at station m ∈ S. It contains the start time of segment track (m, n) ∈ E.
downstream track set Km down and the upstream track eimn Binary variable, for each train i ∈ T and each segment
up down and K up are further composed of a
set Km . Km m track (m, n) ∈ Bi , 1 if the actual arrival time of train i
main track and several platform tracks. at station n is not later than the maintenance start time of
rev
Kim Set of station tracks at station m ∈ Ai which are the segment track; 0 otherwise.
normally used by trains in a direction opposite to that vimk Binary variable, 1 if train i occupies station track k ∈ Km
of train i ∈ T . at station m ∈ Ai ; 0 otherwise.
Kmmain Set of main tracks at station m ∈ S, Km main ⊂ K αmn Nonnegative continuous variable, represents the absolute
m
Indices difference between the maintenance start time of down-
m0 , m, n, n0 Index of stations, m0 ∈ S, m ∈ S, n ∈ S, n0 ∈ S, stream segment track (m, n) ∈ E and its successor down-
m0 6= m 6= n 6= n0 . stream segment track (n, n0 ).
(m, n) Index of segment tracks, (m, n) ∈ E. If m < n, it is a δim Nonnegative continuous variable, represents the absolute
downstream segment track, otherwise, an upstream one. deviation of the actual arrival time of regular train i ∈ T2
i, j Index of trains, i ∈ T , j ∈ T , i 6= j. at station m ∈ Ai to the planned arrival time.
k Index of station tracks. εim Nonnegative continuous variable, represents the absolute
Parameters deviation of the actual departure time of regular train i ∈
oi , s i Origin and destination station of train i ∈ T . T2 at station m ∈ Ai to the planned departure time.
ϕij Binary constant, 1 if trains i ∈ T and j ∈ T have
different running directions; 0 otherwise.
bim Minimum dwell time of train i ∈ T at station m ∈ Ai . wim ≤ yim − xim ≤ M · wim ∀i ∈ T, ∀m ∈ Ai (7)
If train i does not stop at the station, it equals to 0.
aear lat
im , aim Earliest and latest arrival time of train i ∈ T at station δim ≥ xim − asch
im , δim ≥ asch
im − xim
m ∈ Ai , respectively. (8)
dear lat Earliest and latest departure time of train i ∈ T at ∀i ∈ T2 , ∀m ∈ Ai
im , dim
station m ∈ Ai , respectively.
asch sch Scheduled arrival and departure time of train i ∈ T2 at
εim ≥ yim − dsch sch
im , εim ≥ dim − yim
im , dim (9)
station m ∈ Ai , respectively. ∀i ∈ T2 , ∀m ∈ Ai
max , r min
rimn Maximum and minimum running time of train i ∈ T
imn
on segment track (m, n) ∈ Bi , respectively. Constraint (2) means that the actual dwell time of each
tmn Maintenance duration time of segment track (m, n) ∈ train at each visited station should be no less than the
E. minimum dwell time. Constraints (3)-(4) represent that the
ear , f lat
fmn Earliest and latest maintenance start time of segment
mn actual running time of each train on each passed segment
track (m, n) ∈ E.
hm Minimum departure headway between two trains to
track should be within a reasonable range which depends on
1
enter the same segment track from station m ∈ S in the train type mainly distinguished by its speed grade. Here,
the same direction. the additional accelerating and decelerating times are exactly
hm
2 Minimum arrival headway between two trains to arrive considered once a train stops at a station. Constraints (5)-
at station m ∈ S from the same segment track in the (6) ensure that the actual arrival and departure time of each
same direction.
hm Minimum departure-arrival headway between two
train at each visited station is within a reasonable arrival and
3
trains to leave and arrive at station m ∈ S. departure time slot respectively, to maintain the train service
hm
4 Minimum arrival-departure headway between two level for passengers. Here, the time slot of an over-night train
trains to arrive at and leave station m ∈ S. i ∈ T1 at an intermediate station m ∈ Ai \ {oi , si } can be
qi1 , qi2 Additional accelerating and decelerating time of train determined by that at the origin/destination stations and the
i ∈ T if it stops at a visited station.
∆1 Minimum headway between the leaving time of a train
minimum/maximum segment running time. We note that, for
from a segment track and its maintenance start time. each passenger train i ∈ T1 ∪ T2 and each station m ∈ Ai ,
∆2 Minimum headway between the maintenance end time our model can consider the passengers’ demand arising from
of a segment track and the entering time of a train to it. different time periods, by reasonably setting the time slot
M Large positive constant for big-M constraints. [dear lat
im , dim ] according to passengers’ demand. Constraint (7)
reflects the relationship between decision variables xim , yim
and wim . Specifically, if xim = yim , train i does not stop at
aear lat
im ≤ xim ≤ aim ∀i ∈ T, ∀m ∈ Ai (5) station m ∈ Ai and we have wim = 0; otherwise, the train
must stop at the station and wim = 1 according to constraint
dear lat
im ≤ yim ≤ dim ∀i ∈ T, ∀m ∈ Ai (6) (2). Constraints (8)-(9) calculate the absolute deviation of

Prepared using sagej.cls


6 Journal Title XX(X)

each regular train’s actual arrival and departure time at each yim + hm
3 ≤ xjm + M · (1 − pijm )
(14)
visited station to its planned time, respectively. ∀i, j ∈ T |ϕij = 0, ∀m ∈ Ai ∩ Aj
X
Train headway constraints. Series of headway between two yim + hm
3 ≤ xjm + M · (2 − pijm − vimk )
trains should be fulfilled to prevent potential conflicts at rev
k∈Kim (15)
stations and on segment tracks. We take the station illustrated ∀i, j ∈ T |ϕij = 1, ∀m ∈ Ai ∩ Aj
in Figure 2 as an example to explain different types of X
headways. The layout of station m is exhibited by Figure xjm + hm
4 ≤ yim + M · (pijm + 1 − vimk )
2(a), where double crossovers are set behind and in front rev
k∈Kim (16)
of the station, so that a train can arrive at each track of ∀i, j ∈ T |ϕij = 1, ∀m ∈ Ai ∩ Aj
the station. For example, downstream train i can occupy
upstream main track 2 or platform track 4 if both of the Constraints (10)-(11) reflect the departure headway
two downstream tracks (i.e., main track 1 and platform track between two trains to enter the same segment track in the
3) are not available when it arrives at the station. Note that same direction. Based on the variable definition, yim +
usually tracks either in a segment or in a station are used hm1 ≤ yjm should be hold if the departure time of train i
separately, i.e., some tracks are normally used by upstream to enter segment track (m, n) is earlier than that of train
trains, called upstream tracks, such as tracks 2 and 4 in Figure j (i.e., uijmn = 1) according to constraint (10); otherwise,
2(a), and the others are usually used by downstream trains, we have yjm + hm 1 ≤ yim from constraint (11). Similarly,
called downstream tracks, e.g., tracks 1 and 3 in Figure 2(a). the arrival headway between two trains to arrive at a station
The train headways are illustrated by Figure 2(b), where from the same segment track in the same direction is assured
situations (I)-(VI) are the headways between trains running by constraints (12)-(13). Note that overtaking between trains
in the same direction, and the others are those between trains in a segment track is prevented by these constraints. If two
running in different directions. trains occupy the same station track in the same direction,
First, for trains i and i0 in the same direction, the departure the departure-arrival headway between them is ensured by
headway hm constraint (14). Here, whether two trains occupy the same
1 to enter segment track (m, n) ∈ Bi ∩ Bj from
station m and the arrival headway hm station track or not is determined by constraint (19), which
2 to arrive at station
m from the same segment track (m0 , m) ∈ Bi ∩ Bj should will be explained latter. The departure-arrival headway and
always be respected, to ensure safety of trains in segment the arrival-departure headway, between two trains with
tracks and at stations. Besides, as expressed by situations different directions, are formulated by constraints (15)-(16),
(I)-(II) in Figure 2(b), the departure-arrival headway hm respectively. Specifically, either headway hm 3 or headway
3
between the departure of the former train and the arrival of hm4 should be satisfied only when train i occupies a station
the latter train should be fulfilled if these two trains occupy track
P normally used by trains in the opposite direction, i.e.,
rev vimk = 1. Therefore, constraint (15) transforms
the same station track at station m. Note that the arrival of k∈Kim
train i(i0 ) and the departure of train i0 (i) can be executed to yim + hm 3 ≤ xjn + M · (1 − pijm ) and constraint (16)
simultaneously on the Chinese high-speed railway when they changes to xjm + hm 4 ≤ yim + M · pijm . Obviously, if the
occupy different station tracks. In this case, the departure- departure time of train i at station m is earlier than the
arrival headway hm 3 and the arrival-departure headway h4
m arrival time of train j at the station, i.e., pijm = 1 expressed
are out of consideration, see in situations (III)-(VI). Second, by situation (VII), the departure-arrival headway is ensured
we consider trains i and j in both directions. Obviously, if by constraint (15), from which we have yim + hm 3 ≤ xin .
the downstream train occupies a downstream station track Otherwise, if pijm = 0 (see the situation (IX)), constraint
and the upstream train occupies an upstream station track, (16) ensures the arrival-departure headway.
it is not needed to introduce headway constraints between Station capacity constraints. A train needs to occupy a
these trains. However, once train i arrives at station track 2 station track at each visited station, and the station capacity
or 4, Conflict 1 (Conflict 2) in Figure 2(a) occurs between constraints should be fulfilled to ensure the practicability of
the departure (arrival) of train i and the arrival (departure) the resulting train timetable. As the headway between trains
of train j at the right (left) double crossover. Therefore, at every station can be maintained by the train headway
it is necessary to consider headways hm m
3 and h4 between constraints explained above, the station capacity constraint
these trains, see the situations (VII)-(IX). The train headway is to ensure that each station track cannot be simultaneously
constraints are as follows. occupied by more than one train at any time. The station
yim + hm capacity constraints are as follows.
1 ≤ yjm + M · (1 − uijmn )
(10)
∀i, j ∈ T |ϕij = 0, i < j, ∀(m, n) ∈ Bi ∩ Bj
X
vimk = 1 ∀i ∈ T, ∀m ∈ Ai (17)
k∈Km
yjm + hm
1 ≤ yim + M · uijmn
(11) X
∀i, j ∈ T |ϕij = 0, i < j, ∀(m, n) ∈ Bi ∩ Bj wim ≤ 1 − vimk ∀i ∈ T, ∀m ∈ Ai (18)
main
k∈Km
xin + hm
2 ≤ xjn + M · (1 − uijmn )
(12) vimk + vjmk ≤ 1 + pijm + pjim
∀i, j ∈ T |ϕij = 0, i < j, ∀(m, n) ∈ Bi ∩ Bj (19)
∀i, j ∈ T |i < j, ∀m ∈ Ai ∩ Aj , ∀k ∈ Km
xjn + hm
2 ≤ xin + M · uijmn Constraint (17) denotes that a train must occupy one and
(13)
∀i, j ∈ T |ϕij = 0, i < j, ∀(m, n) ∈ Bi ∩ Bj only one station track at each visited station. Constraint (18)

Prepared using sagej.cls


Wang, Zhan, Peng and Zhou 7

Platform
3 Conflict 1
i i’ Downstream
1
Upstream direction
2
direction j
Conflict 2 4
Platform
Station m
(a) Layout of station m

m’ m’ m’
i i’ i’ i i i’
m
hm2
m
h2
m
h2 h1
m m m m
m
h3 h3
m m
h1 h1
n n n
(I) (II) (III)
m’ m’ m’
i’ i i’ i i i’
m m m
m
h2 h1 h2m h1 hm2 h1
m m m

n n n
(IV) (V) (VI)
m’ m’ m’
i i
i
hm3 h4m
m m m
m
h3 hm4

j j j
n n n
(VII) (VIII) (IX)
(b) Headway between two consecutive trains

Figure 2. Illustration of the train headway constraints

ensures that a train that 1 − ejmn ≤ uijmn


P stops at a station cannot occupy (23)
the main track, i.e., k∈K main vimk = 0 if wim = 1. The ∀i ∈ T3 , ∀j ∈ T1 ∪ T2 , ∀(m, n) ∈ Bi
m
station capacity requirement is ensured by constraint (19).
uijmn ≤ eimn
Specifically, if trains i and j want to occupy the same station (24)
track at station m ∈ Ai ∩ Aj , the departure time of train i at ∀i ∈ T1 ∪ T2 , ∀j ∈ T3 , ∀(m, n) ∈ Bi
the station should be earlier than the arrival time of train j ear
fmn lat
≤ zmn ≤ fmn ∀(m, n) ∈ E (25)
at the station (i.e., pijm = 1), or the departure time of train j
at the station should be earlier than the arrival time of train i zmn = znm ∀(m, n) ∈ E|m < n (26)
at the station (i.e., pjim = 1). Here, the requirement that the
αmn ≥ zmn − znn0
station track has been cleared for a while (no less than hm 3 ) (27)
before it can be used again is strictly ensured by constraints ∀(m, n), (n, n ) ∈ E|m < n, n < n0
0

(14)-(15). Otherwise, if pijm = pjim = 0 (see the situations αmn ≥ znn0 − zmn
(III)-(VI) and (IX) in Figure 2(b)), at most one of these (28)
two trains can occupy station track k and vimk + vjmk ≤ 1 ∀(m, n), (n, n ) ∈ E|m < n, n < n0
0

strictly holds. The situation that a segment track is blocked and not
available for any train when it is under maintenance is
Segment maintenance constraints. When a segment track formulated by constraints (20)-(21). Specifically, if train
is under maintenance, it is not available for trains until the i ∈ T needs to pass segment track (m, n) ∈ Bi , the train
maintenance is finished. Besides, the start and end time of should have passed and leaved the segment track at least ∆1
the maintenance should be reasonably arranged to reduce seconds before the maintenance start time of the segment
the negative influence on train operations. The segment track, i.e., xin + ∆1 ≤ zmn if eimn = 1, or the train is
maintenance constraints are as follows. not allowed to enter the segment track until at least ∆2
seconds after the maintenance is finished, i.e., zmn + tmn +
xin + ∆1 ≤ zmn + M · (1 − eimn )
(20) ∆2 ≤ yim if eimn = 0. Constraint (22) indicates that the
∀i ∈ T, ∀(m, n) ∈ Bi inspection train should be operated on a segment track
after its maintenance. Constraints (23)-(24) guarantee that
zmn + tmn + ∆2 ≤ yim + M · eimn the inspection train is earlier than other types of trains
(21)
∀i ∈ T, ∀(m, n) ∈ Bi to enter a segment track after maintenance. Constraint
(25) makes sure that the maintenance start time of each
eimn = 0 ∀i ∈ T3 , ∀(m, n) ∈ Bi (22) segment track is within a given time range, and constraint

Prepared using sagej.cls


8 Journal Title XX(X)

(26) requires that both tracks in the same segment are and 5 minutes, respectively. hm m
1 -h4 (∀m ∈ S) are set to 3,
maintained simultaneously. Constraints (27)-(28) calculate 3, 2 and 2 minutes, respectively. The maintenance of each
the absolute difference between the maintenance start times segment track should start within 23:30-04:00(+1d), where
of consecutive segments. “+1d” means the next day. The objective weights β1 -β3 are
all set to 1/3. The timetable adopted by the test line uses
one second as a time interval, and the integer time values
Case study
represent seconds in our tests.
In this section, we describe our computational experiments
based on the Beijing-Guangzhou high-speed railway line
Practical train timetable and maintenance plan
in China. We first describe the test instance and parameter
setting. Then, the practical solution is explained, followed by The considered timetable is the actual one used in practice
the computational results obtained by our model. We finally from April 2019 on the day when over-night trains need to be
investigate the effect of new EMUs on train scheduling and scheduled. The railway operators usually spend a relatively
maintenance planning. long time to obtain a feasible practical train timetable and
The model is coded by MATLAB R2016b and directly maintenance plan (without guaranteed quality). The practical
solved by CPLEX 12.8, because CPLEX can find high solution is shown in Figure 4, where the over-night trains
quality solutions of the test instance according to preliminary are expressed by red lines, and the regular trains and the
computational results. The CPLEX parameters are set to the inspection trains are illustrated by blue lines and magenta
default values except for the maximum computational time lines, respectively. Meanwhile, the gray rectangles represent
which is limited to 4 hours. The Dynamic Search, which the maintenance of segments. In total 14 over-night trains are
is a branch-and-cut based built-in algorithm in CPLEX, is scheduled, where downstream train 305 and upstream train
invoked to solve our MILP model. We refer the readers 312 (expressed by red dotted lines) are cross-line over-night
to 27 for more details about CPLEX. The computational trains, and the range of their time to enter and leave the test
experiments are executed on a PC with Inter Core i5-7500 line is restricted by the maintenance plan of the connected
3.4 GHz CPU and 32 GB RAM. The readers can download line. We assume that they can be modified within a tough
the code and the input/output data of our model from range around the current one shown in Figure 4. Specifically,
https://github.com/dw-nueliit301/ITSMP-code. train 305 should depart from station 27 within 04:20-04:50,
and train 312 should arrive at station 27 within 22:30-23:00.
Background and parameter setting From Figure 4 we can see that, the practical solution is
rather intuitive to cluster the over-night trains into a narrow
The Beijing-Guangzhou high-speed railway line is 2,118 km, range, so that they can use the limited spare time effectively
and it is one of the longest and busiest high-speed railway without waiting the maintenance for a long time. However,
lines in China. The location of the Beijing-Guangzhou clustering over-night trains does not fully meet passengers’
high-speed railway line in the Chinese high-speed railway demand, especially for the late departures. It may also create
network is shown in Figure 3. Currently, in total 38 stations a capacity issue for passengers travelling shorter distances,
are set for passengers’ boarding and alighting along the e.g., the passengers travelling from station 38 to station 34
line, which divide the line into 37 segments. In Figure need to wait for about 45 minutes if they miss the short-
3, the numbers in brackets near each station represent the distance train 236 at station 38. When it comes to the
down
number of downstream station tracks |Km |, the number performance indicators of the practical solution, the total
up
of upstream station track |Km | and the minimum dwell travel time of over-night trains is 515,932 sec. Each over-
time bm , respectively. If train i needs to stop at station night train stops at a certain station where it is not necessary
m for passengers’ boarding and alighting, then bim = bm . to stop for passengers’ boarding and alighting, increasing
Otherwise, bim = 0. When a station connects more than one the total travel time of over-night trains. In terms of the
line, multiple yards might be set at the station (e.g., Beijing practical maintenance plan, the 37 segments are divided
West Station and Guangzhou South Station), each of which into 21 maintenance sections. The total absolute difference
is dedicated for the trains scheduled on each connected line. between the maintenance start time of consecutive segments
Thus, the number of station tracks shown in this Figure is reaches 43,800 sec. Finally, the maintenance time of 22
that in the yard where the associated train operations are segments are compressed. Therefore, the maintenance staffs
executed. Besides, an over-night train needs to stop at some cannot finish the maintenance tasks of these 22 segments on
of the intermediate stations (the stations with a red cycle in the days when over-night trains are operated.
Figure 3) for passengers’ boarding and alighting, and it might
also stop at other stations to wait for the maintenance.
The values of parameters are set according to the practice. Computational results
To reduce the influence of planning over-night trains on The over-night trains are preferred by the passengers to
regular trains, we assume that the arrival and departure time depart from their origin station within the range of 19:00-
of train i ∈ T2 can be modified by at most 20 minutes 22:00. However, based on our preliminary computational
before and after the planned one. The minimum running results on the test railway line, the range for the downstream
time of a train on a segment track is calculated based on the over-night trains and the upstream over-night trains should
practical timetable, and the maximum running time equals to be 19:30-21:15 and 20:00-21:40 respectively, so that a
the associated minimum one plus 5 minutes. The additional feasible solution exists. In this context, to maintain the
accelerating time and decelerating time are set to 2 and 3 feasibility and the service quality of over-night trains, we
minutes, respectively. The values of ∆1 and ∆2 equal to 0 assume that downstream and upstream over-night trains

Prepared using sagej.cls


Wang, Zhan, Peng and Zhou 9

1 (5, 4, 180)
(2, 2, 120) 2
3 (4, 4, 120)
(3, 3, 60) 4
5 (2, 2, 120)
(2, 2, 120) 6
Beijing 7 (5, 5, 115)
Shijiazhuang (2, 2, 120) 8
Zhengzhou 9 (3, 3, 110)
Wuhan (4, 3, 120) 10

Changsha 11 (4, 3, 60)


(3, 3, 120) 12
Guangzhou 13 (4, 3, 120)
(8, 8, 120) 14

15 (3, 2, 120)
(3, 3, 120) 16

17 (3, 3, 120)

Station ID Station Name Station ID Station Name (2, 2, 120) 18


19 (3, 4, 60)
1 Beijing West 20 Xiaogang North
(3, 3, 120) 20
2 Zhuozhou East 21 Hengdian East
3 Gaobeidian East 22 Wuhan 21 (2, 2, 60)
4 Baoding East 23 Xianning North (8, 7, 180) 22

5 Dingzhou East 24 Chibi North 23 (3, 3, 120)


(2, 2, 120) 24
6 Zhengding Airport 25 Yueyang East 25 (3, 4, 120)
7 Shijiazhuang 26 Miluo East (2, 2, 120) 26
8 Gaoyi West 27 Changsha South
27 (7, 8, 180)
9 Xingtai East 28 Zhuzhou West (4, 3, 120) 28
10 Handan East 29 Hengshan West
29 (2, 2, 120)
11 Anyang East 30 Hengyang East (3, 4, 60) 30
12 Hebi East 31 Leiyang West 31 (2, 2, 120)
13 Xinxiang East 32 Chengzhou West (3, 3, 120) 32
14 Zhengzhou East 33 Lechang East
33 (2, 2, 60)
15 Xuchang East 34 Shaoguan (3, 3, 120) 34
16 Luohe East 35 Yingde West 35 (2, 2, 120)
17 Zhumadian West 36 Qingyuan (2, 2, 120) 36
18 Minggang East 37 Guangzhou North 37 (2, 2, 120)
19 Xinxiang East 38 Guangzhou South (9, 10, 180) 38

Figure 3. The Beijing-Guangzhou high-speed railway line

should depart from their origin stations within 19:30-21:15 allow to compress the maintenance time, and the mainte-
and 20:00-21:40, respectively, and arrive at their destination nance time of each segment track equals to that adopted in
station within 06:00-08:00 of the next day. We extend our practice (see Figure 4).
investigated time horizon to 19:00-09:00(+1d) to select the
regular trains which might be affected by over-night trains. The reduced model by CPLEX has in total 32,430 vari-
Thus, in total 326 (regular, over-night, and inspection) trains ables (including 26,175 binary variables) and 84,597 con-
are considered, including 161 downstream and 165 upstream straints when solving the whole problem. The computation
trains, in our model. Besides, the actual arrival (departure) terminates after 4 hours with a relative gap of 1.53%, and
time of regular train i ∈ T2 at station m ∈ Ai is fixed to the obtained train timetable and maintenance plan are shown
the planned time if asch sch in Figure 5. The meanings of the lines in this figure are the
im (dim ) locates within 19:00-19:30
or 08:30-09:00(+1d), so that other regular trains are fixed same as those in Figure 4. The objective value is 167,548 sec,
and are not influenced by the over-night trains. Therefore, where the total travel time of over-night trains, total deviation
the obtained train timetable is also feasible in the period out of regular trains and total absolute difference between the
of our considered horizon. maintenance start time of consecutive segments are 473,168
sec, 432 sec and 29,045 sec, respectively. Even though
relative long computational time is needed to solve the test
Optimization solution allowing clustering. In this subsec- instance to optimality, our problem is a tactical one and the
tion, we allow to cluster the over-night trains in our model, computational time is still acceptable. Besides, a feasible
because this situation is acceptable by the railway operators. solution with a relative gap less than 4.2% can be found by
Besides, we try to find the train timetable and maintenance CPLEX within 10 minutes.
plan where the maintenance does not need to be compressed.
Unfortunately, from our computational results, no feasible As observed from Figure 5, though the over-night trains
solution can be found within the given computational time are still clustered, and the optimized solution may not fully
under this situation (coincides with the practice). Thus, we meet the passengers’ demand on over-night trains arising

Prepared using sagej.cls


10 Journal Title XX(X)

Figure 4. Practical train timetable and maintenance plan

Figure 5. Optimized train timetable and maintenance plan with compressed maintenance and clustering

from some time periods (e.g., during 20:00-20:40 and 21:20- that in the practical plan. Third, the influence of the over-
21:40), our optimized solution still outperforms the practical night trains on regular trains can be effectively controlled
solutions, and the reasons are as follows. First, most of the to reduce the negative influence to regular trains, because
over-night trains do not need to stop at intermediate stations the total deviation of the regular trains is 432 sec. Only
where no passenger needs to aboard and alight the train. the timetables of trains 135 and 136 are slightly modified
Hence, the total travel time of these trains reduces by 8.29%. at some stations, and the maximum deviation of a train at
Second, the whole high-speed railway line is only divided a station does not exceed 60 sec. Fourth, since there is an
into 8 maintenance sections instead of 21 sections in practice, available time slot (from station 38 to station 27) during
which makes the maintenance tasks more convenient. The 20:00-20:40 in our optimized solution, the railway operators
total absolute difference between the maintenance start time can insert extra short-distance trains if necessary based on
of consecutive segments reduces by 33.69% compared to actual passengers’ demand. Thus, conclusion can be made

Prepared using sagej.cls


Wang, Zhan, Peng and Zhou 11

that, our optimization model is capable to obtain high- from the current 300km/h to 350km/h (actually it has been
quality over-night train timetable and maintenance plan for realized on the Beijing-Shanghai high-speed railway line
practical-size problems, which outperform the practical ones in China), and a new type of sleeper EMUs capable for
made manually by the railway operators. Our optimization 350km/h are designed. Thus, we investigate the effect of
model can provide a better solution to the railway operators the new EMUs on the over-night train scheduling and
and reduce their solution burden. maintenance planning. We suppose that the new sleeper
EMUs are adopted such that the running time of over-
Optimization solution without clustering. Considering that night trains on a segment track reduces. Therefore, the
the over-night trains in the above optimized solution (shown maintenance time of each segment track may increase to
in Figure 5) are still clustered. In this subsection, we try to no less than 4 hours to finish the maintenance tasks every
distribute the over-night trains more evenly according to their evening. We set the maintenance duration time of each
departure time at their origin stations, so that the passenger’s segment track to 4 hours. Meanwhile, we also evenly
demand arising from different time periods can be met. distribute the over-night trains based on their departure time
Specifically, we first divide the ranges 19:30-21:15 and at their origin station (as done in the above subsection).
20:00-21:40 into 7 and 6 continuous time slots respectively The other parameters keep unchanged. The computation
with a nearly equal length. Then, we assume that one and terminates within 4 hours with a relative gap of 0.73%,
only one over-night train should depart from its origin and the obtained train timetable and maintenance plan
station within each time slot via constraints (6). The reduced are illustrated in Figure 7. The total travel time of over-
model by CPLEX has in total 30,005 variables (including night trains, total deviation of regular trains, and total
23,750 binary variables) and 76,357 constraints. Optimality absolute difference between the maintenance start time of
is proved by CPLEX with a computational time of 3,561 sec, consecutive segments are 498,054 sec, 1,265 sec and 26,004
and the obtained train timetable and maintenance plan are sec, respectively.
shown in Figure 6. The meanings of the lines in this figure As seen from Figure 7, the over-night trains are evenly
are the same as those in Figure 4. The objective value is distributed in the predefined ranges of the departure time to
173,724 sec, where the total travel time of over-night trains, satisfy the passengers’ demand arising from different time
total deviation of regular trains, and total absolute difference periods in the evening. Maintenance staffs can finish all the
between the maintenance start time of consecutive segments required maintenance tasks during the evening, because no
are 489,171 sec, 2,066 sec and 29,935 sec, respectively. maintenance is compressed after adopting the new sleeper
Compared with the optimized solution shown in Figure 5, EMUs. Meanwhile, the total travel time of over-night trains
the total travel time of over-night trains and the total absolute and the total absolute difference between the maintenance
difference between the maintenance start time of consecutive start time of consecutive segments still reduce compared
segments become larger, and the railway line is divided to the practical ones, and the influence of the over-night
into 10 maintenance sections in the optimized solution trains on regular trains is small. Besides, the test railway line
illustrated in Figure 6. However, the over-night trains is only divided into 7 maintenance sections, which makes
are evenly distributed and the corresponding passengers’ maintenance tasks more convenient. However, each over-
demands arising from different periods are met. Meanwhile, night train needs to stop at a certain station to wait for the
the values of all the objective components in our optimized maintenance due to the long duration of the maintenance.
solution shown in Figure 6 are still less than those in the Because there are only 6 downstream platform tracks in
practical solution. The total travel time of over-night trains station 27, a downstream over-night train must occupy an
and the total absolute difference between the maintenance upstream platform track in this station. Finally, because all
start time of consecutive segments are reduced by 5.19% maintenance tasks are finished every night, over-night trains
and 31.66%, respectively. Only two regular trains 136 and can operate every day and more over-night trains might
307 need to be slightly modified, while the modification be scheduled under the waiting strategy after adopting the
does not exceed 6 min. Meanwhile, the railway operators new sleeper EMUs. To this end, the increasing passengers’
can also insert extra short-distance trains to the optimized demand on over-night trains can be met.
solution (e.g., between over-night trains 306 and 308) if
necessary, so that the demands of the passengers travelling
for a short distance can be met. Thus, we conclude that Conclusions
our model can consider different passengers’ demands
arising from different time periods, and obtain an acceptable Over-night high-speed trains are very popular and convenient
train timetable and maintenance plan to provide managerial for passengers in countries with a large territory. The
insights to the railway operators. For more passenger- passenger demand on the over-night high-speed trains has
oriented timetabling, we refer to 28 . been increasing recently in China. However, the operation
of these trains is inevitably conflict with the regular evening
maintenance and parts of the day-time regular trains. To
Optimization for new over-night EMUs satisfy the increasing passenger demand and ensure the
As explained above, no solution can be found if maintenance service quality of high-speed trains, it is of great significance
time is not allowed to be compressed under the restriction to operate more over-night trains. Unfortunately, the over-
of the narrow scope of the trains’ arrival and departure night train schedule and maintenance plan on Chinese high-
time at their origin and destination stations. Fortunately, the speed railway lines are still made manually by experienced
operators of the Chinese high-speed railway are planning operators in practice, and the quality of the obtained
to increase the high-speed trains’ maximum running speed train timetable and maintenance plan cannot always be

Prepared using sagej.cls


12 Journal Title XX(X)

Figure 6. Optimized train timetable and maintenance plan with compressed maintenance and without clustering

Figure 7. Optimized train timetable and maintenance plan without compressed maintenance for new EMUs

guaranteed. Thus, this paper investigates the integrated over- total absolute arrival and departure deviation of day-time
night train scheduling and maintenance planning problem regular trains from their planned timetable, and the total
on high-speed railway lines to provide decision support for absolute difference between the maintenance start time of
the operators in practice. The influence of the over-night consecutive segments. The model can be efficiently solved
trains and the comprehensive inspection trains on the day- by state-of-the-art commercial solvers due to its linear
time regular trains, which run early in the morning and late feature, to find high quality feasible solutions for practical-
in the evening, is taken into account to maintain the practical size problems. The effectiveness of the model is tested
feasibility of the obtained train timetable and maintenance on a large real-world instance, construct from the long
plan in a whole day. and busy Beijing-Guangzhou high-speed railway line in
A mixed integer linear programming model is formulated China. The computational results demonstrate that our model
to minimize the total travel time of over-night trains, the can find high quality train timetable and maintenance plan

Prepared using sagej.cls


Wang, Zhan, Peng and Zhou 13

which outperform the practical ones within an acceptable 4. Cacchiani V., L. Galli, and P. Toth. A tutorial on non-periodic
computational time. More segments can start maintenance train timetabling and platforming problems. Euro Jorunal on
tasks simultaneously, which makes the maintenance more Transportation and Logistics. 2015, 4:285320.
convenient. Besides, by adopting a new type of sleeper 5. Cacchiani V., D. Huisman, M. Kidd, L. Kroon, and P. Toth.
electric multiple units, the maintenance of all segments does An overview of recovery models and algorithms for real-
not need to be compressed and the maintenance staffs can time railway rescheduling. Transportation Research Part B:
finish all the maintenance tasks in every evening. More over- Methodological, 2014, 63:15-37.
night trains may be scheduled every evening to satisfy the 6. Fang W., S. Yang, and X. Yao. A survey on problem models and
increasing passenger demand on the over-night trains. solution approaches to rescheduling in railway networks. IEEE
We assume that the number and paths of over-night trains Transactions on Intelligent Transportation Systems, 2015,
are given in advance based on the current passenger demand. 16:2997-3016.
However, it is interesting to investigate where and how many 7. Cacchiani V., A. Caprara, and P. Toth. Scheduling extra freight
over-night trains should be operated with the change of trains on railway networks. Transportation Research Part B:
passenger demand. Besides, the line plan of comprehensive Methodological, 2010, 44: 215231.
inspection trains has great influence on the over-night 8. Burdett R.L., and E. Kozan. Techniques for inserting additional
train scheduling and maintenance planning. However, it is trains into existing timetables. Transportation Research Part B:
assumed to be fixed in this research. It is worthy to explore Methodological, 2009, 43:824-836.
the effect of the comprehensive inspection trains on the 9. Tan Y., and Z. Jiang. A branch and bound algorithm and
over-night train scheduling and maintenance planning by iterative reordering strategies for inserting additional trains in
properly modifying their line plan. Finally, we focus on real time: A case study in germany. Mathematical Problems in
the integrated over-night train scheduling and maintenance Engineering, 2015, 2015:1-12.
planning on a high-speed railway line. However, extending 10. Jiang F., V. Cacchiani, and P. Toth. Train timetabling by
our model to a complicated high-speed railway network and skip-stop planning in highly congested lines. Transportation
developing more efficient solution algorithms need to be Research Part B: Methodological, 2017, 104:149-174.
further considered in the future. 11. Gao Y., L. Kroon, L. Yang, and Z. Gao. Three-stage
optimization method for the problem of scheduling additional
Acknowledgements trains on a high-speed rail corridor. Omega, 2018, 80:175-191.
The authors thank the editor and five anonymous reviewers for 12. Lidn T. Railway infrastructure maintenance-a survey of
their insightful comments and suggestions. This paper is partially planning problems and conducted research. Presented at
supported by the National Natural Science Foundation of China 18th Euro Working Group on Transportation, Delft, The
(No. U1834209), the National Natural Science Foundation of China Netherlands, 2015.
(No.71701174), the Fundamental Research Funds for the Central 13. Higgins A. Scheduling of railway track maintenance activities
Universities (No.2682017CX020), and the Fundamental Research and crews. Journal of the Operational Research Society, 1998,
Funds for the Central Universities (No. A1920502051905-005). 49:10261033.
14. Budai G., D. Huisman, and R. Dekker. Scheduling preventive
railway maintenance activities. Journal of the Operational
Author contributions
Research Society, 2006, 57:1035-1044.
The authors confirm contributions to the paper as follows: 15. Peng F., S. Kang, X. Li, and Y. Ouyang. A heuristic approach
problem analysis and definition: Dian Wang, Shuguang to the railroad track maintenance scheduling problem. Comput-
Zhan, Qiyuan Peng; model formulation: Dian Wang, Aided Civil Infrastructure Engineering, 2011, 26:129145.
Shuguang Zhan; case study: Dian Wang, Shuguang 16. Aken S.V., N. Beinovic, and R.M.P. Goverde. Designing
Zhan, Wentao Zhou; manuscript preparation: Dian Wang, alternative railway timetables under infrastructure maintenance
Shuguang Zhan, Wentao Zhou. All authors reviewed the possessions. Transportation Research Part B: Methodological,
results and approved the final version of the manuscript. 2017, 98:224-238.
17. Aken S.V., N. Beinovic, and R.M.P. Goverde. Solving large-
Declaration of Conflicting Interests scale train timetable adjustment problems under infrastructure
maintenance possessions. Journal of Rail Transport Planning
The authors declare no potential conflicts of interest with
& Management, 2017, 7:141-156.
respect to the research, authorship, and/or publication of this
18. Arenas D., P. Pellegrini, S. Hanafi, and J. Rodriguez. Timetable
article.
rearrangement to cope with railway maintenance activities.
Computers & Operations Research, 2018, 95:123-138.
References 19. Caprara A., M. Monaci, P. Toth, and P.L. Guida. A Lagrangian
1. Zhang C., Y. Gao, L. Yang, U. Kumar, and Z. Gao. Integrated heuristic algorithm for a real-world train timetabling problem.
optimization of train scheduling and maintenance planning on Discrete Applied Mathematics, 2006, 154:738753.
high-speed railway corridors. Omega, 2019, 87:86-104. 20. Vansteenwegen P., T. Dewilde, S. Burggraeve, and D.
2. Lusby R., J. Larsen, M. Ehrgott, and D. Ryan. Railway Cattrysse. An iterative approach for reducing the impact of
track allocation: models and methods. OR Spectrum, 2011, infrastructure maintenance on the performance of railway
33:843883. systems. European Journal of Operational Research, 2016,
3. Cacchiani V., and P. Toth. Nominal and robust train timetabling 252:39-53.
problems. European Journal of Operation Research, 2012, 21. Albrecht A.R., D.M. Panton, and D.H. Lee. Rescheduling rail
219:727737. networks with maintenance disruptions using Problem Space

Prepared using sagej.cls


14 Journal Title XX(X)

Search. Computers & Operations Research, 2013, 40:703-712.


22. Forsgren M., M. Aronsson, and S. Gestrelius. Maintaining
tracks and traffic flow at the same time. Journal of Rail
Transport Planning & Management, 2013, 3:111-123.
23. Lidn T., and M. Joborn. An optimization model for
integrated planning of railway traffic and network maintenance.
Transportation Research Part C: Emerging Technologies,
2017, 74:327-347.
24. Lidna T., T. Kalinowskib, and H. Watererb. Resource
considerations for integrated planning of railway traffic and
maintenance windows. Journal of Rail Transport Planning &
Management, 2018, 8:1-15.
25. Luan X., J. Miao, L. Meng, F. Corman, and G. Lodewijks.
Integrated optimization on train scheduling and preventive
maintenance time slots planning. Transportation Research Part
C: Emerging Technologies, 2017, 80:329-359.
26. DAriano A., L. Meng, G. Centulio, and F. Corman. Integrated
stochastic optimization approaches for tactical scheduling of
trains and railway infrastructure maintenance. Computers &
Industrial Engineering, 2019, 127:1315-1335.
27. Lima R. IBM ILOG CPLEX-What is inside of the box? In
Proc. 2010 EWO Seminar, 2010:1-72.
28. Zhan S., S.C. Wong, and S.M. Lo. Social equity-based
timetabling and ticket pricing for high-speed railways.
Transportation Research Part A: Policy and Practice, 2020,
137:165-186.

Prepared using sagej.cls

View publication stats

You might also like