Journal of Cleaner Production 214 (2019) 1024e1037

Contents lists available at ScienceDirect

Journal of Cleaner Production

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

Energy consumption optimization of train operation for railway

systems: Algorithm development and real-world case study
Huiru Zhang a, Limin Jia b, c, Li Wang a, b, *, Xinyue Xu b, **
a
School of Traffic and Transportation, Beijing Jiaotong University, Beijing, 100044, China
b
State Key Laboratory of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing, 100044, China
c
Beijing Engineering Research Center of Urban Traffic Information Intelligent Sensing and Service Technologies, Beijing, 100044, China

a r t i c l e i n f o a b s t r a c t

Article history: Traction energy is the main component of railway operation energy, and a timetable that predefines the
Received 31 August 2018 running time of train operation can be used to determine the traction energy consumption. This study
Received in revised form proposes a bi-level model that optimizes timetables to achieve the energy-saving control of railway
28 December 2018
systems. The upper level of the model ensures the relative stability of the timetable while maintaining
Accepted 3 January 2019
Available online 4 January 2019
railway safety constraints, which makes train operations more convenient for the railway sector as well
as passengers; while the lower level of the model optimizes the arrival and departure time among in-
termediate stations to minimize the energy consumption of each train. Then, a unified iterative opti-
Keywords:
Railway
mization algorithm combining particle swarm is developed to solve the model, and a timetable that
Energy-efficient operation ensures energy consumption optimizations is thus obtained. A case study using actual operation data
Timetable from the Beijing-Shanghai high-speed railway is developed to illustrate the proposed method. Results
Particle swarm optimization show that the total energy consumption is reduced by more than 7.6%, and the average adjustment time
Bi-level optimization for each distance interval is approximately 1 min, which maintains the stability of the original timetable.
© 2019 Elsevier Ltd. All rights reserved.

1. Introduction train operations is the main consumer of energy in railway trans-

portation (Gonza lez-Gil et al., 2014). Moreover, train operations are
Owing to the advantages of rapid speed, high frequency, and scheduled by a timetable that specifies the departure time and
good service, railway transportation has considerable appeal for arrival time for trains (i.e., the running time between stations and
passengers in China. According to the 2017 Statistic Bulletin of the dwell time in stations). Thus, timetable optimization is an
China Railway Corporation, the number of passengers who trav- effective means to achieve energy-efficient operations and reduce
elled by the national railway in 2017 was 3.084 billion, and the train energy consumption (Su et al., 2013). Besides, timetable
energy consumption of the railways, as expressed in terms of should be maintained relatively stable to consider the convenience
standard coal, reached 162.165 million tons (National Railway of the railway managers and passengers, and only a certain range of
Administration, 2018). Owing to environmental concerns and timetable change is permitted.
high energy costs, the energy consumption of railways needs to be The main objective of this study is to introduce a new method to
studied further with the aim of saving energy. Through an assess- solve the energy consumption of a single train, and ensure the
ment of the whole life-cycle of railways and vehicles, it can be seen security constraints among multiple trains while maintaining the
that the operation phase of the Electric Multiple Unit (EMU) is the stability of the timetable. A bi-level model for energy consumption
main contributor to the total environmental impact (Del Pero et al., optimization of train operation is discussed, and a unified iterative
2015; Shinde et al., 2018). Train traction that provides the power for optimization algorithm is proposed. The upper level of the model
maintains the relative stability of the timetable under the condition
of safety constraints among the trains, while the lower level of the
model optimizes the energy consumption of each train. The model
* Corresponding author. State Key Laboratory of Rail Traffic Control and Safety,
Beijing Jiaotong University, Beijing, 100044, China.
and algorithm are then applied to optimize the energy-efficient
** Corresponding author. timetable in Beijing-Shanghai high-speed railway, which can be
E-mail addresses: [email protected] (H. Zhang), [email protected] (L. Jia), extended to other lines.
[email protected] (L. Wang), [email protected] (X. Xu).

https://doi.org/10.1016/j.jclepro.2019.01.023
0959-6526/© 2019 Elsevier Ltd. All rights reserved.
 H. Zhang et al. / Journal of Cleaner Production 214 (2019) 1024e1037 1025

The rest of this paper is structured as follows: a literature review

Primary train timetable
is provided in section 2. Section 3 describes a bi-level programming
model of timetable optimization for a high-speed railway based on
energy-efficient train operation. Section 4 describes the solving
algorithm. A case study is described in section 5 to verify the Basic timetable Quarterly train timetable
effectiveness of the model and algorithm. Finally, conclusions are
provided in section 6.
Monthly train timetable

2. A-state-of-the-art

Train operation between two adjacent stops includes traction, Daily train timetable
coasting, and braking (Powell and Palacín, 2015), which involves Timetable
four steps: accelerating, cruising, coasting, and braking, as shown in rescheduling
Fig. 1. On the one hand, energy consumption during the cruising Three hour狭s dispatching timetable
and braking phases is less than that during the accelerating phase.
On the other hand, there is no need for energy during the coasting Fig. 2. The classification of train timetables.
phase. Therefore, the key to achieving energy-savings during train
operation lies in making the correct choice of transition points for
different states, that is, to find the optimal strategies of train called timetable rescheduling. The purpose of this study is to
operation to minimize energy consumption. Jiaxin and Howlett optimize the energy consumption by adjusting the basic timetable,
(1992) showed that energy consumption is related to the speed at hereinafter referred to as timetable. As the timetable creation
the switch point under a given sequence of throttle settings. On process is based on the predefined running times between stations,
lines without steep grades, a power-speedhold-coast-brake strat- the definition of running times will directly impact not only the
egy was the optimal driving strategy for trains (Albrecht et al., traffic planning but also the entire energy consumption (Chevrier
2013). Actually, the optimal strategies can be derived from the et al., 2013). Generally, the running time can be adjusted within a
Hamiltonian function and the Pontryagin principle (Albrecht et al., certain range by using the running time supplements, which are
2016; Howlett et al., 2009). According to the above two mathe- extra running time on top of the technical minimum running time
matical methods, the optimal strategies were further summarized between every two stations; the longer the running time is, the
as follows: maximum acceleration and braking during the begin- lower the energy consumption for the same distance (Scheepmaker
ning and end of a train journey, particle acceleration or braking et al., 2017; Scheepmaker and Goverde, 2015). However, the longer
during cruising, and starting to coast as early as possible (Liu and the running time is, the higher the time cost of the railway sector
Golovitcher, 2003). In addition, the operation strategy regarding and passengers; thus, a trade-off between the running time and
the upper and lower bound of the subway speed was confirmed for energy consumption is necessary. In addition, the train order may
lines with different distances, from the perspective of energy saving change after adjusting the running time based on energy
as well as transport efficiency improvement (Feng et al., 2011). consumed, and then the train operation needs to be adjusted to
The impact of train operations on energy consumption depends maintain safety constraints, and the adjustment of train operation
on the timetable. In general, there are five kinds of timetables in will result in a further change in energy consumption.
China's railway department (see Fig. 2). The first three timetables In fact, train operation adjustment is a large-scale combination
are referred to as the basic timetable. The other two timetables are optimization problem that needs to consider several factors.
added to adjust train operations in time, and research in this field is Corman et al. (2012) considered the balance between the mini-
mized train delays and missed connections. D'Ariano (2007) and
Youshi and Weidong (2005) took the minimized deviation from the
original timetable as the objective function. Shen et al. (2017) took
into consideration of the passenger travel time and congestion, and
Meng et al. (2010) aimed to improve the stability of the timetable.
cruising
Vac Previous models of train operation adjustment problems involved
mathematical programming with linear or nonlinear targets, and
coasting the solving algorithms included exact and heuristic algorithms,
Speed (km/h)

such as branch and bound algorithms, genetic algorithms, and

Speed(km/h)

particle swarm algorithms (D'Ariano et al., 2007; Meng et al., 2010).

However, energy-saving train operation and timetable optimi-
zation adjustment are not concurrently considered in the above
acceleration studies. On the one hand, if the former is individually considered,
Vcb safety constraints among trains and adjustment degrees will be
difficult to control because the operation is optimized primarily for
braking a single train. On the other hand, if only the latter is considered, the
vehicle cannot determine the most appropriate energy strategy to
adopt to achieve the goal of energy conservation. Therefore, the
0 S1 Distance(km) S2 S3 S establishment of an energy-saving timetable, which combines
energy-saving operation and timetable optimization, has become a
Distance (km) hot topic in energy research in the field of railway transportation.
Fig. 1. The optimal speed-distance curve of a train. The train accelerates to S1 , with the Lindner and Zimmermann (2003) postulated that the optimal train
speed increasing from 0 to Vac ; then, it runs at a constant speed Vac to S2 and then timetable can be obtained by studying the start time of train
coasts to S3 , with the speed falling from Vac to Vcb ; finally, it brakes to S, where the coasting. Wong and Ho (2007) thought it necessary to optimize the
speed becomes zero.
1026 H. Zhang et al. / Journal of Cleaner Production 214 (2019) 1024e1037

stop time at the station. Su et al. (2013) created a comprehensive (4) Only basic resistance is considered during train operation.
timetable by optimizing the timetable and the speed curve be- (5) The stop strategy at intermediate stations and the departure
tween adjacent stations simultaneously. Further, for an in-depth and arrival times of the original and terminal stations remain
study of the speed curve, Li and Lo (2014) considered frequent unchanged.
changes in passenger demands on the subway, and Yang et al. (6) The object of this study is the basic timetable mentioned in
(2016) considered the time and space uncertainty of train quality. section 2, so it is reasonable and acceptable that the depar-
Gong et al. (2014) mentioned the adaption problem of the energy- ture time of trains can be up to 3 min earlier than the time
efficient timetable under different disturbance conditions and specified in the original timetable.
designed an algorithm to recover the system to energy-saving
status. Feng et al. (2017) focused on the uncertain passenger de- For a line with n stations and m trains, the notations are defined
mand during off-peak hours and assigned the dwelling time margin in the Nomenclature list in Appendix A.
of this period to the running time. Nonetheless, there are still some
open issues to be addressed in the study of energy-saving 3.2. The upper level of the model
timetable.
On the one hand, the current application of energy-efficient 3.2.1. Objective function
timetable is mainly in the field of urban rail transit, while studies The objective function of the upper level of the model is pre-
that consider the high-speed railway field with more complicated sented as following:
conditions are rare. On the other hand, in the design of energy-
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
efficient timetable, studies mostly aim at reducing the energy  2  2
XX
consumption of trains without considering arrival and departure zstability ¼ min a aupper
i;j
 a lower
i;j
þ d upper
i;j
 d lower
i;j
times in the existing timetable (Li and Lo, 2014; Lindner and i2I j2J
Zimmermann, 2003; Su et al., 2013; Wong and Ho, 2007; Yang sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
 2  2
et al., 2016). Therefore, considering the above problems compre-
þb aupper
i;j
 aorigin
i;j
þ dupper i;j
 dorigin
i:j
hensively, a bi-level model to design an energy-efficient timetable
is proposed in this paper.
(1)

3. Model description where zstability is the overall stability of all the trains on the line. The
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2
upper upper
In this section, a bi-level programming model is presented. The first part on the right side, ðai;j  alower
i;j
Þ þ ðdi;j  dlower
i;j Þ ,
lower level adjusts the arrival and departure time of each train in
represents the stability of timetable between the upper and lower
order to save energy. The upper level is constructed to maintain the
level. The second part on the right side,
security constraints among trains and stability of timetable, whose rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
output is the input of the lower level (see Fig. 3). upper origin 2 upper origin
ðai;j  ai;j Þ þ ðdi;j  di:j Þ , represents the stability of
timetable that is the difference between the adjusted timetable by
3.1. Assumptions and notations
the upper level and the original timetable. a and b represent the
proportion of the two parts in the comprehensive target, respec-
To facilitate problem formulation, the following six assumptions
tively. Herein, both a and b are set to 0.5, which means that the two
are made in this study.
aspects are equally important in this study.

(1) In the bi-level model, stations with side lines are abstracted
as nodes, and block sections between railway signals are 3.2.2. Constraints
represented by arcs that are used to connect those nodes.
(2) The train is modeled as a mass point, which has been suc-
cessfully used in the field of train optimal control (Franke aupper
i;jþ1
 dupper
i;j
 t sec;min
i;j
(2)
et al., 2000; Wang et al., 2011).
(3) Trains run on ideal straight rail lines, and gradients, curves, upper upper
di;j  ai;j  t sta;min
i;j
(3)
and tunnels are not considered.

T lower upper
i;i0 ;j þ T i;i0 ;j  1; i; i0 2I and isi0 ; j2J (4)
Original timetable
Xn
 1; i; i0 2I and isi0 ; j2J
upper
T
j¼1 i;i0 ;j
(5)
Energy-saving model
Lower level Xn
ailower
,j , dilower
,j , and energy T lower
j¼1 i;i ;j
0  1; i; i0 2I and isi0 ; j2J (6)
of each train
Xuj
aiupper
,j , diupper
, j of each train,
ll
l¼1 i;j
¼ 1; i2I; j2J (7)
and stability of timetables
Upper level
Timetable adjustment model dorigin
i;j
 dupper
i;j
 s; ci ¼ 1; 2; :::; m; j ¼ 1; 2; :::; n (8)

Constraints (2) and (3) ensure that the running and the dwell
Optimal timetable time of train i are not less than the minimum time for the section
and the station, respectively. Constraint (4) specifies that the order
Fig. 3. Overview of the bi-level model for optimizing timetables. between trains should be kept the same at both levels. Constraints
 H. Zhang et al. / Journal of Cleaner Production 214 (2019) 1024e1037 1027

(5) and (6) ensure that the order of any train pairs can be changed
( X
no more than one time. Constraint (7) guarantees that a train can ti;k ¼ t *i;k n o
X * acc cruise coast brake
only occupy one track at a time. Constraint (8) ensures that the t i;k ¼ Ti ; t i;k 2 t i;k ; t i;k ; t i;k ; t i;k
* (15)
departure time adjustment of train i at station j is not higher than k2K
the upper bound s. The headway constraints are as follows.
8
   Xn  Xn > t *i;k
ð
upper

upper
þN 1 T lower  tsta T lower >
> X X
ai0 ;j ai;j k¼j i;i ;k
0 j , j¼k i;i ;k
0 ; cj >
< Si;k ¼ S*i;k ¼ vðtÞdt n o
¼ 2; 3; :::; n  1 ; S*i;k 2 Sacc cruise coast brake
i;k ; Si;k ; Si;k ; Si;k
>
> X 0
>
>
(9) : Si ¼ Si;k
k2K
   Xn  Xn upper (16)
upper upper upper
ai;j  ai0 ;j þN 1 T
k¼j i;i0 ;k
 tsta
j , T
k¼j i;i0 ;k
; cj
The movement of train i includes accelerating, cruising, coasting
¼ 2; 3; :::; n  1 and braking within each distance interval k; thus, the time and
(10) distance are equal to the sum of the four processes as shown in the
first equation in constraint (15) and constraint (16).
   Xn  Xn
dupper  upper
þ  lower
 tsec T lower 
i ;j
0 di;j
N 1 k¼j
T i;i0 ;k j , k¼j i;i ;k
0 ; cj V0 ¼ VT ¼ 0
vðtÞ  Vmax (17)
¼ 2; 3; :::; n  1 j

(11) Xdi;k sec;min

þ t  ti;k (18)
   Xn  Xn upper di;k i;j
dupper
i;j
 dupper
i0 ;j
þN 1 T upper
k¼j i;i0 ;k
 tsec
j , T
k¼j i;i0 ;k
; cj
Constraint (17) ensures that the initial speed and final speed of
¼ 2; 3; :::; n  1 all trains at each distance interval is 0 km/h, and that the speed is
(12) not higher than the maximum speed on section [j, jþ1]. In
constraint (18), the lower limit of optimal running time is given at
where N is a very large integer, e.g., 100 000. Constraints (9) to (12) distance interval k.
ensure that the headway between two adjacent trains is not less Remark 1. The overall goal of the bi-level model is to optimize
than the headway for station j, and is not less than the headway for timetables with a trade-off between energy efficiency and time-
section [j, jþ1]. table stability. The objective function of the bi-level model is pre-
sented as follows:
3.3. The lower level of the model
 
3.3.1. Objective function z ¼ min zstability ; zenergy (19)
The objective function of the lower level is presented as
following:
XX   4. An iterative-loop algorithm
zenergy ¼ min E ti;k
i2I k2K
tði;k In this section, a unified iterative optimization algorithm is
  (13)
designed to solve the trains’ security constraints and optimal time
E ti;k ¼ kf ðtÞ,vðtÞ,FðtÞdt
distribution, and the main steps are as follows.
0

(1) First, the shortest running time T min

i , which is obtained by
where zenergy is the total traction energy consumption of all the
train traction calculation, is subtracted from the original
trains at the total distance; F is the traction force that varies over
timetable to get the spare time T spare
i
. Then, an iterative al-
time, and the value can be obtained based on the traction charac-
gorithm is used to find the optimal speed-distance curve, and
teristic curve.
energy-saving running time is obtained after the distance
interval with the largest energy reduction is selected to add
3.3.2. Constraints the spare time.
The relationship among force, mass, and acceleration is as (2) A particle swarm algorithm is used to ensure the security
follows. constraints among trains and the stability between original
 . timetable and energy-saving timetable, and then a new
kf F  kb B  W M ¼ dvðtÞ=dt (14) stable timetable is obtained.
(3) The time difference between the stable timetable and
where the left side denotes the ratio of traction force to train mass, energy-saving timetable is compared, and if the time differ-
and the right side denotes the change in velocity, i.e., the acceler- ence is less than 30 s, the final energy-saving and stable
ation. According to the Davies equation, W ¼ m þ k,vðtÞ þ g,vðtÞ2 is timetable is output; otherwise, make the stable timetable be
obtained. Note that the values of F and braking force B are dynamic the “original timetable”, and then return to next cycle.
in the four states according to kf and kb . Relationships among train
speed, time, and distance are as follows. Next, the exact calculation process will be introduced.
1028 H. Zhang et al. / Journal of Cleaner Production 214 (2019) 1024e1037

4.1. Optimize the train energy consumption (the lower level of the and energy consumption in the four stages of train operation,
model) respectively. Eqs. (24) and (25) are the shortest running time and
the energy consumption under this operation. Howlett, Milroy, and
4.1.1. Calculate the shortest running time and spare time of trains Pudney (1994) suggested that there was a numerical relationship
The spare time here refers to the running time supplements between the cruising speed and the speed at the end of coasting on
mentioned in the Introduction, which is different from the defini- an ideal straight line. Therefore, the distance in each stage and the
tion of Scheepmaker and Goverde (2015). Due to uncertain factors energy consumed in the accelerating and cruising phases can be
in practice, almost no trains can operate using the sequential pro- calculated.
cess of accelerating, cruising at the maximum speed or braking. Thus, the spare time of train i is:
Therefore, in this study, the shortest running time additionally in- X
spare
cludes coasting phase. Ti ¼ Ti  t min
i;k (26)
k2K
8
> F W
> acc
> ai;k ¼ M
>
>
>
>
>
>
>
> V max V max ,M
>
> t acc
¼
j
¼
j 4.1.2. Find the optimal speed curve and distribute spare time
>
> i;k
a acc
ðF  WÞ
>
> i;k On actual railway lines, the schedule plan of train stopping is not
<
 2 (20) exactly the same. Consequently, the values of energy consumption
>
> 1  2 V max ,M are different in different intervals with the same spare time (see
>
> acc acc acc j
> Si;k ¼ 2,ai;k , t i;k
>
>
¼
2,ðF  WÞ Fig. 4).
>
>
>
>  2 Obviously, the spare time should be distributed to the intervals
>
>
>
>   F, V max ,M where energy consumption reduction is larger. The spare time is
>
> E t acc ¼ F,Sacc ¼ j
: i;k distributed at a smaller granularity using the expression:
i;k 2,ðF  WÞ
.
spare spare
ti ¼ Ti Z (27)
8 cruise
>
> ai;k ¼0
>
>
>
> where Z is an integer used to determine the granularity of distri-
>
> cruise
> cruise Si;k
>
spare
bution of the spare time. If Z ¼ T i
spare
, then t i ¼ 1, and the spare
< t i;k ¼ max
Vj (21) time is allocated by a second.
>
> For each distance interval, the distance Si;k can be obtained from
>
> cruise
>
> S ¼ S  Sacc brake
i;k  Si;k  Scoast
> i;k
>
i;k the original timetable, assuming that the running time is ti;k and
>
>    
: E t cruise ¼ W,S ¼ W, S  Sacc  Sbrake  Scoast initial speed is v, and then the optimal velocity can be found
i;k y i;k i;k i;k
through iteration as shown in Fig. 5.
Therefore, the adjusted train running time is obtained by
combining the stop strategies with the result of spare time
8 distribution.
>
> W
>
> acoast ¼
>
>
i;k M
>
>   4.2. Maintain relative stability of the timetable (the upper level of
>
>
>
> V max
 V ,M the model)
> coast
> j cb
>
< t i;k ¼ W
 2 4.2.1. Define the position and speed vector
>
>
>
>   M, V max
 V 2
The numbers of trains, stations and particle swarm are m, n, and
>
> 1 2 j cb
>
> Scoast ¼ V max ,t coast þ ,acoast coast
>
> i;k j i;k i;k , t i;k ¼ 2,m,n, respectively. The running time on the section is randomly
>
> 2 2,W
>
>   generated under the constraints (2)e(12), and thus an initial po-
>
: coast
E t i;k ¼0 sition vector is obtained (see Fig. 6). The speed vector of a particle is
randomly generated within the distance interval average speed and
(22)

8
> ðB þ WÞ Interval 1
>
> abrake ¼
>
> i;k M Interval 2
>
>
Energy consumption (kw·h)

>
>  V ,M
>
> 0 V
>
> t brake ¼ brakecb ¼ cb
< i;k ai;k ðB þ WÞ E1 E2
(23)
>
>  2
>
> 1 V 2cb M
>
> Sbrake ¼ Vcb ,t brake þ ,ab , t brake ¼ T
>
> i;k i;k 2 i;k 2,ðB þ WÞ
>
> T
> 
> 
>
: E t brake ¼ 0
i;k

t min acc cruise

i;k ¼ t i;k þ t i;k þ t coast
i;k þ t brake
i;k (24)

          Running time (s)

E ti;k ¼ E t acc
i;k þ E t cruise
i;k þ E t coast
i;k þ E t brake
i;k (25)
Fig. 4. Comparison of the reduction of energy consumption in different intervals with
Eqs. 20e23 calculate the acceleration, running time, distance the same spare time.
 H. Zhang et al. / Journal of Cleaner Production 214 (2019) 1024e1037 1029

Knowing distance Si ,k , let time ti ,k timin

,k tispare
,k ,
and let initial speed v Si ,k ti ,k
The optimal cruising speed is obtained,
Find the optimal speed curve with different step and then Vcb can be calculated so that the
length. Initial step length dv 1 km h speed curve of interval k and the change
of energy Ei ,k are obtained.
Let v ' v dv, t Si ,k v '
if k K Y k k 1
if t ti , k Y Let v v'
N N
Let v v, dv 0.1 dv Choose the largest Ei ,k , let ti ,k timin
,k tispare
,k

N if dv Y
0.0001

Fig. 5. Flowchart of optimal velocity iteration.

Fixed dwelling time at stations

sec,min
Minimum running time on sections ti , j
Random running time on sections i , j

Original station Station a Station b Station j Terminal station

…
0 …
Ti
Train i

0 Ti spare i ,a Ti spare i ,b Ti spare

spare spare
[0, Ti ] Generate randomly [ i ,a , Ti spare
] Generate randomly i, j T i i ,b

Fig. 6. Diagram of initial position vector generation.

Vmax
j , and the former is equal to the ratio of the section distance to evolutionary algebra. FunRandðÞ is a function that generates a
tsec;min
i;j
. number between 0 and 1 randomly. wini ; wend , and Gk are the initial
After the initial solution is generated, the global and individual inertia factor, final inertia factor and maximum evolution algebra,
optimum are calculated according to the fitness function, that is,
the objective function of the upper level (see Eq. (1)).

 
X ¼ xx ¼ xij ðtÞ; ci2I; j2J
Bei J i ngNan
Pbest ¼ fitnessðXÞ ¼ zstability ðXÞ (28) 59km
10_BJN-LF
Gbest ¼ min Pbest 11_LF-TJN 72km
LangFang
Ti anJ i nNan
12_TJN-CZX 88km
where xij ðtÞ is the position vectors of particle i at station j at time t,
CangZhouXi
X is set of position vectors. Pbest is set of optimal positions for each 13_CZX-DZD 108km

particle, and Gbest is the one with the smallest fitness in Pbest . 14_DZD-JNX 79km DeZhouDong
15_JNX-TAX 56km
16_TAX-QFD 71km J i NanXi
4.2.2. Update the speed and position
17_QFD-TZD 58km
The speed and position vectors are updated by the following Tai AnXi
18_TZD-ZZX 36km
QuFuDong
Eqs. 19_ZZX-XZD 65km
TengZhouDong

  
20_XZD-SZD 75km ZaoZhuangXi
vij ðt þ 1Þ ¼ k, wðtÞ,vij ðtÞ þ c1 ,FunRandðÞ, Pbest  xij ðtÞ 21_SZD-BBN 77km XuZhouDong
  22_BBN-DY 53km
þ c2 ,FunRandðÞ, Gbest  xij ðtÞ 23_DY-CZ 62km SuZhouDong
(29) 24_CZ-NJN 64km
25_NJN-ZJN 64km BengBuNan
ZhenJ i angNan
26_ZJN-DYB 25km Di ngYuan DanYangBei
xij ðt þ 1Þ ¼ xij ðtÞ þ vij ðt þ 1Þ (30) 27_DYB-CZB 32km ChuZhou
NanJ i ngNan
ChangZhouBei
WuXi Dong
28_CZB-WXD 57km
SuZhouBei KunShanNan
29_WXD-SZB 26km HongQi ao
30_SZB-KSN 41km
wðtÞ ¼ ðwini  wend Þ,ðGk  tÞ=Gk þ wend (31) 31_KSN-HQ 50km

 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
  
k¼2 2  f  f2  4f; f ¼ c1 þ c2 (32)
  Fig. 7. Railway network structure of the Beijing-Shanghai high-speed railway.
Note that on the left of the figure is the name and length of each section. A section
name is a combination of numbers and characters, where the characters are the initials
where vij ðtÞ is the speed vectors of particle i at station j at time t. of the origin and destination station. For example, the first section connecting Bei-
wðtÞ is an inertia factor at time t that linear decreases with JingNan Station to LangFang Station is denoted by 10_BJN-LF.
1030 H. Zhang et al. / Journal of Cleaner Production 214 (2019) 1024e1037

Table 1
Related parameters of the bi-level model including information on the Beijing-Shanghai high-speed railway, the EMU train, and iterations.

Parameter Value Parameter Value

Total distance (km) 1318 Air resistance parameter 0.000 895

Number of stations 23 Distribution granularity Z ¼ T spare
i
Number of trains 37 Learning factor c1 2
Train mass (kg) 890 000 Learning factor c2 2.1
Train braking force (N) 400 000 Maximum number of iterations 450
Reference resistance parameter 0.007 4 Number of particles 95
Rolling resistance parameter 0.053 5 e e

25000
Shortest running me (s)

20000

15000

10000

5000

G151

G153
G157
G101

G105

G107
G111
G113

G115
G117

G119
G121

G125
G411
G129
G131
G133
G135
G137
G139

G141
G143
G145

G147
G21

G23
G11

G41

G13

G15

G43

G17

G7
G9
G5

G1

G3
Original Time Shortest Time
Fig. 8. Diagram of original and shortest running times.

respectively. A large wðtÞ guarantees the global convergence ability Because of the rapid speed (the maximum running speed is
of the algorithm, while a smaller wðtÞ guarantees the ability of local 350 km/h) and high frequency (approximately 90 trains per day),
search (Shi and Eberhart, 1998). k is the convergence factor, and c1 the benefits of using an energy-efficient timetable on the Beijing-
and c2 are two constant learning factors. Note that convergence of Shanghai high-speed railway will be considerable (People's Daily
the particle swarm algorithm was rigorously proved by Clerc and People's Daily Online, 2011).
Kennedy (2002).
After the update, the particles are put into the constraints 5.1. Data and parameters
(2)e(12), and, if they are satisfied, the fitness values are calculated;
otherwise, the particles are reinitialized, and then the fitness values In this case, only the data in the direction from Beijing to
are calculated. Shanghai is considered for convenience without loss of generality.
The all-day railway schedule on January 17, 2018 is selected, and
4.2.3. Update global and individual optimum the relevant parameters in the example are listed in Table 1. Data
of the line, such as the distance and the stations, are collected

       from the Construction Documents of Beijing to Shanghai High-speed
X ¼ xx ¼ min fitness xij ðt þ 1Þ ; fitness xij ðtÞ ; i2I; j2J
Railway (China Railway Design Group Corporate Limited, 2007).
Pbest ¼ fitnessðXÞ ¼ zstability ðXÞ
The original timetable is captured from the tickets ordering
Gbest ¼ min Pbest
website (i.e., 12306 China Railway), and the train parameters are
(33) obtained from a book edited by Huang and Wang (2014). The code
At time t and time tþ1, the individual optimum always chooses is implemented in MATLAB R2013a and run on a 64-bit Windows
the one whose fitness is smaller, and the smallest fitness in the operating system.
individual position set is taken as the global optimum (see Eq. (33)).
The procedure will be executed repeatedly until the termination 5.2. Results
condition (maximum value of iteration) is fulfilled.
The shortest running time is calculated in seconds, and the
5. Case study difference with the original timetable is the spare time (see Fig. 8).
The average spare time per train is approximately 49 min; the
The Beijing-Shanghai high-speed railway (see Fig. 7) that con- maximum spare time is 58.5 min (G111), and the minimum spare
nects two megacities, Beijing and Shanghai, is one of the busiest time being 20 min (G7).
lines in China, with an average of more than 0.29 million passen- The spare time is distributed based on the iterative algorithm, and
gers travelling per day. This high-speed railway began to generate the energy-saving speed-distance curve is obtained. The stopping
profits after three years of operation, and the total profit in strategies of the 37 trains and the train distribution over time are not
2014e2017 was 31.17 billion yuan (Daily economic news, 2018). the same. For example, the number of stops for G5, which overtakes
 H. Zhang et al. / Journal of Cleaner Production 214 (2019) 1024e1037 1031

Shortest running time Original timetable Energy-efficient timetable and running time of the original timetable.
Speed-distance curve of train G5 Speed-distance curve of train G41
400 400 G5 is taken as an example here. From the perspective of the
shortest running time, the speed of train G5 is the highest, and the
300 300
energy consumption is the largest. From the perspective of an
Speed(km/h)

Speed(km/h)
200 200
energy-efficient timetable, more spare time is given to long dis-
tance intervals so that the cruising speed decrease and the distance
100 100 is enough for full coasting.
The train's new running time is formed after combining the
0 0
0 500 1000 1500 0 500 1000 1500 unchanged information in the original timetable with the speed-
Distance(km) Distance(km)
distance relationship adjusted by the lower level and will be
Speed-distance curve of train G117 Speed-distance curve of train G135
400 400 adjusted by the upper level. Given that the algorithm used in the
upper level is heuristic, a comparison experiment is needed on the
300 300
iterations and the particle swarm size.
Speed(km/h)

Speed(km/h)

200 200 Fig. 10 (a) shows the trend of fitness along with the increase in
iterations and particle swarm size, where the little red circle marks
100 100 the global best. Fig. 10 (b) shows the fitness curve under the con-
0 0
dition of optimal iterations and particle swarm size. As the number
0 500 1000 1500 0 500 1000 1500 of iterations increases, the fitness value gradually becomes stable.
Distance(km) Distance(km)
The optimal value of fitness is 7.49,105s when the number of iter-
Fig. 9. Speed-distance curve of train G5, G41, G117 and G135 under three conditions. ations is 450 and the particle swarm size is 95.
A total of 287 distance intervals are changed (see Table 2). The
overall changed time is 19 320 s, and the average adjustment time
its former train, is small. The distribution over time between G41 and for each interval is 67 s. There are 52 time durations during which
its adjacent trains is tight, while it is slack between G117 and its 37 trains can potentially arrive at the intermediate stations earlier
neighboring trains. The number of stops for G135 is large, which than the time in the original timetable. The total time earlier than
impacts its latter train significantly. Speed-distance curves of these the original timetable is 4320 s, and the average time for each
four trains are drawn as representatives, as shown in Fig. 9. earlier arrival is 83 s. Specially, when train G41 arrives at JiNanXi
Because the detailed speed-distance data of each train is Station, the earlier arrival time is the longest 180 s, which has little
currently unavailable, the speed-distance of the proposed energy- impact on the operation of the station.
efficient operation strategies is drawn based on the stop strategy In Fig. 11, some rules for energy-saving optimization adjustment

5
x 10
8.4
5
x 10
Fitness value (s)

8.2 8.3

8 8.2
Fitness value (s)

7.8 8.1

7.6 8
Pa

7.4 7.9
100
r
tic

7.8
le

80
7.7
sw

60
ar

7.6
m

40 500
450
siz

7.5
400
350
300
e

20 250

rations
200 7.4

Number of ite
0 50 100 150 200 250 300 350 400 450

Number of iterations
(a) Fitness value for different iterations and particle swarm size (b) Optimal individual fitness

Fig. 10. Diagram of particle swarm optimization.

Table 2
Time adjustment result of running distance intervals.

Train number G101 G5 G105 G11 G107 G111 G113 G1

Number of time-changed stations 10 4 8 4 8 8 7 3

Total changed time(s) 660 300 540 300 540 540 480 240
Train number G41 G115 G117 G13 G119 G121 G15 G125
Number of time-changed stations 7 9 10 3 9 11 3 10
Total changed time(s) 480 600 660 240 600 720 240 660
Train number G411 G129 G131 G133 G135 G137 G139 G3
Number of time-changed stations 10 10 9 10 10 10 10 3
Total changed time(s) 660 660 600 660 660 660 660 240
Train number G43 G141 G143 G145 G17 G147 G21 G151
Number of time-changed stations 8 10 10 11 3 12 5 11
Total changed time(s) 540 660 660 720 240 780 360 720
Train number G23 G153 G157 G7 G9 e e e
Number of time-changed stations 9 8 8 2 4 e e e
Total changed time(s) 600 480 540 120 300 e e e
1032 H. Zhang et al. / Journal of Cleaner Production 214 (2019) 1024e1037

+RQJ4LDR
.XQ6KDQ1DQ 2ULJLQDO WLPHWDEOH
6X=KRX%HL (QHUJ\VDYLQJ WLPHWDEOH
:X;L'RQJ
&KDQJ=KRX%HL
'DQ<DQJ%HL
=KHQ-LDQJ;L
1DQ-LQJ1DQ

&KX]KRX

'LQJ\XDQ

%HQJ%X1DQ

6X=KRX'RQJ

;X=KRX'RQJ

=DR=KXDQJ;L
7HQJ=KRX'RQJ
4X)X'RQJ

7DL$Q;L

-L1DQ;L

'H=KRX'RQJ

&DQJ=KRX;L

7LDQ-LQ1DQ
/DQJ)DQJ
%HL-LQJ1DQ

                  

Fig. 11. Comparison diagram of the original timetable and energy-efficient timetable. Note that the connection time between partial trains is short for each station with multiple
tracks.

All Energy Train Class Train Section Acceleration & Cruise Energy

Fig. 12. The Sankey diagram for the energy flow in trains and sections.
 H. Zhang et al. / Journal of Cleaner Production 214 (2019) 1024e1037 1033

are summarized as follows: brake to stop at the following stations (14_DZD-JNX, 24_CZh-
NJN), and the distance of the sections are small (18_TZD-ZZ,
(1) When a train stops at intermediate stations for a few times, 26_ZJN-DYB, 27_DYB-CZB, 29_WXD-SZB).
or when overtaking takes place, the changed time for the (3) Because of the long running distance, the energy consumed
train is low, for example, G5, G11, G1, G13, G15, G17, G7 and for cruising is 3.9 times than the energy consumed at
G9. Furthermore, time adjustment of the train whose dis- acceleration.
tribution over time with adjacent trains is tight is also low,
such as G41, G23 and G43.
(2) When the time distribution of the trains is slack, trains run at
5.3. Comparison analysis
a lower speed than the original timetable to reduce energy
consumption (G115, G117, G119, G125, G133, G129 and G147).
Experience indicates that the electric energy consumption for
When the distribution over time between two adjacent
eight groups of EMU trains in China Railway High-speed series to
trains is small, the two trains will simultaneously operate at a
run 1318 km is 9885 kW,h. After energy-efficient optimization,
speed less than that in the original timetable to save energy
there was an energy savings of 7.6% (see Fig. 13).
(G135 and G137, G143 and G145).
The energy consumption of all the trains in each section is
analyzed before and after optimization, as shown in Fig. 14. Note
The energy consumption of running equals the energy con-
that the second column is the virtual distribution connecting the
sumption of traction zenergy divided by h, which is the energy
energy flows. The energy flowing from the first column to the last
conversion efficiency of train drive system, and the value of h is 0.9.
rectangle of the second column represents the saved energy in all
Therefore, the optimal solution for the Beijing-Shanghai high-
sections, while part of energy flows from the last rectangle to
speed railway is obtained when the energy consumption is
sections in the third column to compensate for excessive energy
337 968.61 kW,h and the stability of the new timetable is consumption after optimization.
7.49,105 s. Energy consumption is illustrated by different views, for
After optimization, the energy consumption is increased in six
example, from the perspective of trains and sections, the Sankey sections, while the total energy consumption is reduced. Further-
diagram for energy flow after optimization is shown in Fig. 12. The
more, the origin station of the section consuming more energy is
first column represents the total energy consumed by all trains. The usually a hub station, at which many trains will stop due to the
second column is the energy consumed by different grades of
large passenger demand. For example, BeiJingNan Station, the first
trains. The definitions of different grades of trains are as follows: if section BJN-LF, is the departure station, and JiNanXi Station, the
the number of stops of the train is less than five, the train belongs to
origin station of sixth section, is a hub station where all the trains
the first-class; otherwise the second-class. The third and four col- stop. Therefore, it is energy-saving to accelerate trains with
umns are the trains and sections, respectively, and the last column
maximum acceleration, which also confirms the valid of the
is the energy consumed by acceleration and cruising. Some results
energy-efficient strategy mentioned in section 2 (Liu and
are found:
Golovitcher, 2003).
The overall goal of the bi-level model is to optimize timetables
(1) The first-class train accounts for about a quarter of the total
with a trade-off between energy efficiency and timetable stability.
number of trains, whose passengers are long travel distance.
The following is the Pareto analysis of the reduced energy con-
Compared with the second-class train, the first-class train
sumption with different time distribution of train G111 at DZD-JNX
has a shorter number of stops and a shorter total running
section.
time. So, the average energy consumption of the first-class
In Fig. 15 (a), the first rectangle on the left means the reduced
train is 1.2 times than that of the second-class one.
energy will account for 28.70% of the total reduced energy if 90 s
(2) If the energy consumption of a certain section is small, two
are allocated to this section. According to the proposed algorithm,
main reasons may be addressed: many trains will coast and
the final optimal running time is 4440 s (see Fig. 15 (b)), which

12000
Train energy consump on (kw·h)

10000

8000

6000

4000

2000

0
G115
G117

G119
G121

G125
G411
G129
G131
G133
G135
G137
G139

G141
G143
G145

G147

G151

G153
G157
G101

G105

G107
G111
G113
G11

G41

G13

G15

G43

G17

G21

G23
G5

G1

G3

G7
G9

Energy-efficient metable Experienced energy consump on

Fig. 13. Comparison diagram of train energy consumption.
1034 H. Zhang et al. / Journal of Cleaner Production 214 (2019) 1024e1037

accounts for about 90% of the total reduced energy, as shown in Comparison with performance of the proposed method, the
Fig. 15 (a). Energy saving, timetable stability, and multi-train energy saving using the previous methods is shown in Table 3.
tracking constraints are all considered in the calculation of the According to the above results, some conclusions can be sum-
final optimal time, which also proves the effectiveness of the pro- marized as follows.
posed algorithm.

Fig. 14. The Sankey diagram for the energy flow in sections before and after optimization.

94.06% 100.00%
87.78% 100.00%
1200 81.01%
Reduced energy consumption

Cumulative percentage (%)

73.79% 80.00%
1000 66.02%
57.69%
800 60.00%
48.76%
(kW·h)

600 39.11%
40.00%
28.70%
400
20.00%
200
0.00%
0 0.00%
90 180 270 360 450 540 630 720 810 900
time (s) Sections %

(a) The Pareto analysis of reduced energy consumption with different time distribution
Optimal time: 4440s
DZD JNX
Station Station

Minimum time 3645s Spare time 900s

(b) The optimal, minimum and spare time
Fig. 15. Analysis of reduced energy consumption and different time relation of G111 at DZD-JNX section.
 H. Zhang et al. / Journal of Cleaner Production 214 (2019) 1024e1037 1035

Table 3
Literature on energy efficient timetable.

Publication Objective(s) Transport mode Energy saving

Su et al. (2013) Energy consumption Metro trains 14.5% (practical timetable)

Yang et al. (2016) Energy consumption Metro trains 10.66% (peak hours)
9.94% (off-peak hours)
(practical timetable)
Cucala et al. (2012) Delay, energy consumption High-speed train 6.67% (with delay)
5.25% (no delay)
(commercial timetable)
Binder and Albrecht (2013) Expected delays, energy consumption Regional trains Between 4.3% and 12.9% (technical min running time)

(1) In general, energy consumption optimization of train oper- According to the 2017 Statistic Bulletin of China Railway Cor-
ation contains two levels: timetable and energy-efficient poration, the national railway mileage has reached 127 000 km by
speed curve (Su et al., 2013). The effect of energy consump- the end of 2017. If the energy-efficient timetable proposed in this
tion optimization in the railway sector is generally lower paper is used in the whole railway network, the saved energy
than that of the subway, but the security constraints and consumption, as expressed in terms of standard coal, is approxi-
energy efficient train operation are more complicated mately 120,000 tons per year. The total mileage of the high-speed
(Binder and Albrecht, 2013; Yang et al., 2016). Therefore, the railway will reach about 380 000 km by 2025, and a high-speed
ratio of energy saved in these two fields is somewhat railway network with a main framework of “eight vertical eight
different and needs to be studied separately. horizontal” will be formed by 2030 (National Railway
(2) With the improvement of transportation organizations, en- Administration, 2016). With the rapid development of the rail-
ergy consumption can even be achieved in the case of delay way, the benefits of using energy-efficient timetable will be
(Cucala et al., 2012). Once the high-speed train on a very busy remarkable.
railway is delayed, energy consumption will be ignored to The revealing of the relationship between energy consumption
make up for the delay in most cases. Moreover, considering and train energy-efficient timetable contributes to promoting the
the convenience of station staff and passengers, it is more railway train control towards more energy-saving modes. The
practical to consider energy consumption optimization of environmental effectiveness of train operation strategy such as
trains with no delay within a certain range. maximum acceleration can be identified. Moreover, the proposed
method can be important for awakening the environmental
awareness of the railway system, and the sector connecting with
the train control can more effectively craft the strategic decisions
5.4. Limitations
for providing more energy-saving and environmentally friendly
service.
The proposed method works best when the assumptions
mentioned in section 3.1 are satisfied, which is proved by the In future work, the authors will add a case comparison to verify
the validity of the model and algorithm and study the gradients,
illustrated case study. Nonetheless, when the line condition is so
poor that the gradients and curves cannot be ignored. As a result, curves and tunnels that affect the optimal speed profile. If the de-
parture time of trains is allowed to be earlier than the original
the effectiveness of this method will be reduced.
timetable, it is necessary to increase the efficiency of train stations
accordingly.
6. Conclusions
Acknowledgement
This paper presents a bi-level programming model of timetable
optimization for high-speed railways based on energy-efficient
This study was funded by the National Natural Science Foun-
train control. In the lower level, an energy-efficient model is
dation of China (71701010).
established under the constraints of force, velocity, distance and
time, where train energy consumption is optimized by the
adjustment of arrival and departure times in intermediate stations. Nomenclature list
In the upper level, a timetable rescheduling model is developed
that considers timetable stability and the safety constraints of Sets
running, stopping, tracking, track occupation and timetable sta-
bility. The most feature of this approach is that the timetable of the I set of trains
railway is optimized while also considering energy consumption J set of stations
and timetable stability. Taking the actual operation data of the L set of tracks
Beijing-Shanghai high-speed railway as a case study, 37 trains are K set of distance intervals, which represent the distance
simulated. The results show that the average adjustment time for between every two stops
each distance interval is 67s after the optimization in the bi-level
programming model. As the result, the timetable is relatively sta- Parameters
ble, and it is convenient for station staff to implement relevant
measures. Moreover, the timetable also facilitates passengers to aorigin
i;j
arrival time of train i at station j defined in the original
buy tickets according to the original schedule. A total of 27 776 kW, timetable
h of energy is saved, accounting for a savings of 7.6% compared with dorigin
i;j
departure time of train i from station j defined in the
the experienced energy consumption. In addition, it is found that original timetable
the algorithm proposed in the study has the tendency to distribute alower
i;j arrival time of train i at station j after being adjusted by
more spare time to longer distance interval. the lower level of the model
1036 H. Zhang et al. / Journal of Cleaner Production 214 (2019) 1024e1037

dlower
i;j departure time of train i from station j after being Eðti;k Þ energy consumption of train i at distance interval k.
adjusted by the lower level of the model Note that, the energy consumption calculated in the
tsec;min minimum running time of train i on section [j, jþ1] study is the traction energy consumption, unless
i;j
otherwise stated
t sta;min
i;j
minimum dwell time of train i at station j
tsec
j headway of section [j, jþ1], i.e., departure time of train References
i’- departure time of train i, i’2I
tsta
j headway between two adjacent trains at station j, i.e., Albrecht, A., Howlett, P.G., Pudney, P.J., Vu, X., 2013. Energy-efficient train control:
from local convexity to global optimization and uniqueness. Automatica 49,
arrival time of train i’ - departure time of train i 3072e3078. https://doi.org/10.1016/j.automatica.2013.07.008.
t acc
i;k
accelerating time of train i at distance interval k Albrecht, A.R., Howlett, P., Pudney, P., Vu, X., Zhou, P., 2016. The key principles of
optimal train controldPart 1: formulation of the model, strategies of optimal
t cruise
i;k
cruising time of train i at distance interval k type, evolutionary lines, location of optimal switching points. Transp. Res. Part B
Methodol. 94, 482e508. https://doi.org/10.1016/j.trb.2015.07.023.
t coast
i;k
coasting time of train i at distance interval k
Binder, A., Albrecht, T., 2013. Timetable evaluation and optimization under
t brake
i;k
braking time of train i at distance interval k consideration of the stochastic influence of the dwell times. In: Proceedings of
the 3rd International Conference on Models and Technologies for Intelligent
Ti total running time of train i in the original timetable Transportation Systems 2013, pp. 471e481.
Si total distance of the line Chevrier, R., Pellegrini, P., Rodriguez, J., 2013. Energy saving in railway timetabling:
a bi-objective evolutionary approach for computing alternative running times.
Sacc
i;k accelerating distance of train i at interval k Transport. Res. C Emerg. Technol. 37, 20e41. https://doi.org/10.1016/j.trc.2013.
Scruise
i;k cruising distance of train i at distance interval k 09.007.
China Railway Design Group Corporate Limited, 2007. Construction Documents of
Scoast
i;k coasting distance of train i at distance interval k Beijing to Shanghai High-speed Railway.
Clerc, M., Kennedy, J., 2002. The particle swarm - explosion, stability, and conver-
Sbrake
i;k braking distance of train i at distance interval k gence in a\nmultidimensional complex space. IEEE Trans. Evol. Comput. 6,
V0 ; VT initial and final speed, which are the same in one 58e73. https://doi.org/10.1109/4235.985692.
Corman, F., D'Ariano, A., Pacciarelli, D., Pranzo, M., 2012. Bi-objective conflict
interval for every train detection and resolution in railway traffic management. Transport. Res. C
Vmax
j maximum speed on section [j, jþ1] Emerg. Technol. 20, 79e94. https://doi.org/10.1016/j.trc.2010.09.009.
W running resistance, W ¼ m þ k,vðtÞ þ g,vðtÞ2 , vðtÞis the Cucala, A.P., Sicre, C., Domínguez, M., 2012. Fuzzy optimal schedule of high speed
train operation to minimize energy consumption with uncertain delays and
real-time speed driver ’ s behavioral response. Eng. Appl. Artif. Intell. 25, 1548e1557.
m reference resistance parameter D'Ariano, A., Pacciarelli, D., Pranzo, M., 2007. A branch and bound algorithm for
k rolling resistance parameter scheduling trains in a railway network. Eur. J. Oper. Res. 183, 643e657. https://
doi.org/10.1016/j.ejor.2006.10.034.
g air resistance parameter Daily economic news, 2018. Beijing-Shanghai high-speed rail 4 years profit excee-
uj number of tracks at station j ded 30 billion into the “world's most profitable” high-speed rail [WWW
s an upper bound that allows trains to depart earlier than Document]. Phoenix Netw. Inf. URL. http://news.ifeng.com/a/20180718/
the original timetable, s ¼ 180s 59258036_0.shtml.
Del Pero, F., Delogu, M., Pierini, M., Bonaffini, D., 2015. Life Cycle Assessment of a
kb coefficient of braking force, kb 2f0; 1g heavy metro train. J. Clean. Prod. 87, 787e799. https://doi.org/10.1016/j.jclepro.
kf ðtÞ coefficient of traction, kf 2½0; 1 2014.09.023.
M train mass Feng, J., Li, X., Liu, H., Gao, X., Mao, B., 2017. Optimizing the energy-efficient metro
train timetable & control strategy in off-peak hours with uncertain passenger
dþ 
i;k ; di;k sign for train i arrive and depart distance interval k. An demands. Energies 10. https://doi.org/10.3390/en10040436.
interval may contain multiple sections, which are Feng, X., Mao, B., Feng, X., Feng, J., 2011. Study on the maximum operation speeds of
determined by the stop strategy metro trains for energy saving as well as transport efficiency improvement.
Energy 36, 6577e6582. https://doi.org/10.1016/j.energy.2011.09.004.
Franke, R., Terwiesch, P., Meyer, M., 2000. An algorithm for the optimal control of
Decision variables the driving of trains. In: Proceedings of the 39th IEEE Conference on Decision
and Control, vol. 3, pp. 2123e2128. https://doi.org/10.1109/CDC.2000.914108.
upper Gong, C., Zhang, S., Zhang, F., Jiang, J., Wang, X., 2014. An integrated energy-efficient
ai;j arrival time of train i at station j after being adjusted by operation methodology for metro systems based on a real case of Shanghai
the upper level of the model Metro Line One. Energies 7, 7305e7329. https://doi.org/10.3390/en7117305.
upper
di;j departure time of train i from station j after being Gonz alez-Gil, A., Palacin, R., Batty, P., Powell, J.P., 2014. A systems approach to
reduce urban rail energy consumption A. Energy Convers. Manag. 80, 509e524.
adjusted by the upper level of the model https://doi.org/10.1126/science.1228806.Link.
Howlett, P.G., Milroy, I.P., Pudney, P.J., 1994. Energy-efficient train control. Contr.
8 1; if the departure order Eng. Pract. 2, 193e200. https://doi.org/10.1016/0967-0661(94)90198-8.
<
between train i and i on station j Howlett, P.G., Pudney, P.J., Vu, X., 2009. Local energy minimization in optimal train
T lower
i;i0 ;j ¼ ;
: is changed after adjustment: control. Automatica 45, 2692e2698. https://doi.org/10.1016/j.automatica.2009.
07.028.
0; otherwise:
Huang, X., Wang, F., 2014. Electric Multiple Unit Traction and Control System.
c i; i0 2I and isi0 ; j2J Southwest Jiaotong University Press.
Jiaxin, C., Howlett, P., 1992. Application of critical velocities to the minimisation of
8 1; if the departure order
< fuel consumption in the control of trains. Automatica 28, 165e169. https://doi.
between train i and i0 on station j org/10.1016/0005-1098(92)90017-A.
T upper
i;i0 ;j
¼ ;
: is changed after adjustment: Li, X., Lo, H.K., 2014. Energy minimization in dynamic train scheduling and control
0; otherwise: for metro rail operations. Transp. Res. Part B Methodol. 70, 269e284. https://
doi.org/10.1016/j.trb.2014.09.009.
c i; i0 2I and isi0 ; j2J Lindner, T., Zimmermann, U.T., 2003. Train schedule optimization in public rail
transport. Math. Technol. Futur. Jt. Proj. Between Univ. Ind. 703e716.
 Liu, R., Golovitcher, I.M., 2003. Energy-efficient operation of rail vehicles. Transport.
1; if train i occupies track l of station j:
lli;j lli;j ¼ ; Res. Part A Policy Pract. 37, 917e932. https://doi.org/10.1016/j.tra.2003.07.001.
0; otherwise: Meng, X., Jia, L., Qin, Y., 2010. Train timetable optimizing and rescheduling based on
improved particle swarm algorithm. Transp. Res. Rec. J. Transp. Res. Board 2197,
71e79. https://doi.org/10.3141/2197-09.
c i2I; j2J; l2L
National Railway Administration, 2018. The 2017 Statistic Bulletin of China Railway
Corporation [WWW Document]. http://www.nra.gov.cn/xwzx/zlzx/hytj/
201804/t20180412_55248.shtml.
ti;k optimal running time of train i at distance interval k National Railway Administration, 2016. The medium and long term railway
vcruise
i;k
cruising speed of train i at distance interval k network planning [WWW Document]. URL. http://www.nra.gov.cn/jgzf/flfg/
 H. Zhang et al. / Journal of Cleaner Production 214 (2019) 1024e1037 1037

gfxwj/zt/other/201607/t20160721_26055.shtml. Railway, India. J. Clean. Prod. 188, 989e1003. https://doi.org/10.1016/j.jclepro.

People’s Daily Online, 2011. Some Nomal-speed Trains on the Beijing-shanghai 2018.04.022.
High-speed Railway Will Stop. Running [WWW Document]. URL. http://news. Su, S., Li, X., Tang, T., Gao, Z., 2013. A subway train timetable optimization approach
163.com/11/0615/09/76J3OIE500014JB6.html. based on energy-efficient operation strategy. IEEE Trans. Intell. Transport. Syst.
Powell, J.P., Palacín, R., 2015. A comparison of modelled and real-life driving profiles 14, 883e893. https://doi.org/10.1109/TITS.2013.2244885.
for the simulation of railway vehicle operation. Transport. Plann. Technol. 38, Wang, Y., Schutter, B. De, Ning, B., Groot, N., Boom, T.J.J. Van Den, 2011. Optimal
78e93. https://doi.org/10.1080/03081060.2014.976984. trajectory planning for trains using mixed integer linear programming. In:
Scheepmaker, G.M., Goverde, R.M.P., 2015. The interplay between energy-efficient Intelligent Transportation Systems (ITSC), 2011 14th International IEEE Con-
train control and scheduled running time supplements. J. Rail Transp. Plan. ference on. IEEE, pp. 1598e1604.
Manag. 5, 225e239. https://doi.org/10.1016/j.jrtpm.2015.10.003. Wong, K.K., Ho, T.K., 2007. Dwell-time and run-time control for DC mass rapid
Scheepmaker, G.M., Goverde, R.M.P., Kroon, L.G., 2017. Review of energy-efficient transit railways. IET Electr. Power Appl. 1, 956e966. https://doi.org/10.1049/iet-
train control and timetabling. Eur. J. Oper. Res. 257, 355e376. https://doi.org/ epa.
10.1016/j.ejor.2016.09.044. Yang, X., Chen, A., Ning, B., Tang, T., 2016. A stochastic model for the integrated
Shen, Y., Ren, G., Liu, Y., 2017. Timetable design for minimizing passenger travel time optimization on metro timetable and speed profile with uncertain train mass.
and congestion for a single metro line. Promet - Traffic - Traffico 1, 21e33. Transp. Res. Part B Methodol. 91, 424e445. https://doi.org/10.1016/j.trb.2016.06.
Shi, Y., Eberhart, R.C., 1998. Parameter Selection in Particle Swarm Optimization, 006.
pp. 591e600. https://doi.org/10.1007/BFb0040810. Youshi, Z., Weidong, J., 2005. Model and algorithm for train operation adjustment
Shinde, A.M., Dikshit, A.K., Singh, R.K., Campana, P.E., 2018. Life cycle analysis based on single-track railways based on genetic algorithm. J. Southwest Jiao Tong
comprehensive environmental performance evaluation of Mumbai Suburban Univ. 40, 147e152.