You are on page 1of 11

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 14, NO.

2, JUNE 2013 883

A Subway Train Timetable Optimization Approach


Based on Energy-Efficient Operation Strategy
Shuai Su, Xiang Li, Tao Tang, and Ziyou Gao

Abstract—Given rising energy prices and environmental con-


cerns, train energy-efficient operation techniques are paid more
attention as one of the effective methods to reduce operation
costs and energy consumption. Generally speaking, the energy–
efficient operation technique includes two levels, which optimize
the timetable and the speed profiles among successive stations,
respectively. To achieve better performance, this paper proposes
to optimize the integrated timetable, which includes both the
timetable and the speed profiles. First, we provide an analytical
formulation to calculate the optimal speed profile with fixed trip
time for each section. Second, we design a numerical algorithm to
distribute the total trip time among different sections and prove
the optimality of the distribution algorithm. Furthermore, we
extend the algorithm to generate the integrated timetable. Finally,
we present some numerical examples based on the operation data
from the Beijing Yizhuang subway line. The simulation results
show that energy reduction for the entire route is 14.5%. The
computation time for finding the optimal solution is 0.15 s, which
implies that the algorithm is fast enough to be used in the auto-
matic train operation (ATO) system for real-time control. Fig. 1. Two-level approach to energy-efficient operation.
Index Terms—Energy-efficient operation, Pontryagin maxi-
mum principle, subway, timetable. a timetable that regulates the departure time and arrival time
for trains, i.e., the trip time between stations and the dwell time
I. I NTRODUCTION in stations. The second level is to obtain the energy-efficient
speed profile with the given trip time. With the guidance of
R AILWAY transportation, as one of the most energy-
intensive industries, consumes a large amount of energy
each year. For example, more than 290 billion tons of standard
the automatic train operation (ATO) system, there are many
feasible speed profiles for the train to finish the transportation
task, among which we should select the one that minimizes the
coal is consumed in the Chinese railway industry in 2010.
energy consumption under the safety constraints. The conven-
Given the rising energy prices and environmental concerns,
tional optimal train control studies optimize the speed profile at
train energy-efficient operation techniques are being paid more
each interstation separately. Generally speaking, it is difficult to
attention as one of the effective methods to reduce operation
obtain the global optimal solution for the entire route because
costs and energy consumption.
the distribution of the total trip time among interstations may
Generally speaking, train energy-efficient operation contains
not be optimal. On the other hand, the studies on timetable
two levels (see Fig. 1). At the first level, the dispatchers design
optimization are based on the assumption that trains have
constant speeds (see literature [1]), and energy consumption is
used to overcome the rolling resistance during the trip. This
Manuscript received May 7, 2012; revised October 28, 2012 and
January 22, 2013; accepted January 30, 2013. Date of publication March 15,
is unreasonable and should not be applied for short distance
2013; date of current version May 29, 2013. This work was supported in part railway, particularly the subway system. Hence, the obtained
by the National Natural Science Foundation of China under Grant 71101007, timetable may not be optimal due to the imprecise estimation
by the National High Technology Research and Development Program of China
under Grant 2011AA110502, by the National Basic Research Program of China
for the energy consumption.
under Grant 2012CB725401, by the State Key Laboratory of Rail Traffic Con- The optimizations on speed profiles and timetable are closely
trol and Safety under Grant RCS2012ZT002, and by the Fundamental Research dependent. On one hand, the interstations’ trip time scheduled
Funds for the Central Universities under Grant 2011JBZ014, 2013YJS019.
The Associate Editor for this paper was M. Chowdhury. (Corresponding
by the timetable is an important constraint for the energy-
author: X. Li). efficient train control model. On the other hand, the energy
S. Su, T. Tang, and Z. Gao are with the State Key Laboratory of Rail Traffic consumption determined by the speed profile is the foundation
Control and Safety, Beijing Jiaotong University, Beijing 100044, China.
X. Li is with the School of Economics and Management, Beijing Univer- for optimizing the timetable. Hence, the studies, which stud-
sity of Chemical Technology, Beijing 100029, China (e-mail: xiang-li04@ ied the timetable and speed profile separately, cannot get the
mail.tsinghua.edu.cn). global optimal solution for energy-saving. To achieve a better
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. performance on energy efficiency, this paper presents a bilevel
Digital Object Identifier 10.1109/TITS.2013.2244885 programming model and designs an integrated algorithm to
1524-9050/$31.00 © 2013 IEEE
884 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 14, NO. 2, JUNE 2013

optimize the speed profile between successive stations and the pute delay-resistant cyclic timetables. In addition, applying he
timetable for an entire route, which is named as the integrated timetable theories, coordination among different modes through
timetable. cooperative timetables [18]–[21], and equally headway control
The rest of this paper is organized as follows. In Section II, [22] have been studied by researchers.
we review some important literature on energy-efficient opera- On the other hand, there are many studies focusing on how
tion. In Section III, we formulate a bilevel programming model to obtain the energy-efficient speed profile between successive
for the energy-efficient problem and design an algorithm to dis- stations. Literature on this problem can date back to 1960s. For
tribute the trip time to different sections, and the optimality of example, Ishikawa [23] proposed an optimal control model on
the distribution algorithm is proven. Furthermore, we extend the the assumption that the train runs on a flat track with constant
algorithm to calculate the integrated timetable. In Section IV, gradient and traction efficiency. By using the Pontryagin maxi-
we present some numerical examples based on the infrastruc- mum principle, the optimal driving strategy is proven to consist
ture data and the operation data from Beijing Yizhuang subway of maximum acceleration, cruising, coasting, and maximum
line in China. The results illustrate that the proposed algorithm braking, which lays the foundation for the optimal train control
can get a good performance on energy saving. theory. For taking the theory into practice, variable gradients,
variable speed limits, and traction efficiency were gradually
II. L ITERATURE R EVIEW considered [23]–[29]. For example, Cheng and Howlett [24]
considered the traction efficiency varying with train speed.
In practice, the transportation demand is not uniform for the Khmelnitsky [26] presented a complete study on the optimal
whole day. It will always be higher during peak hours and train control problem, in which variable gradients, variable trac-
lower during low-traffic hours. To meet the variable operation tion efficiency, and arbitrary speed limits have all been consid-
demand, the operation company traditionally schedules a few ered. Liu and Golovitcher [27] gave an analytical solution to the
timetables in advance and then choose one based on the pas- problem with variable gradients for finding driving strategies
senger flow. For a chosen timetable, the trains servicing on for each part of the route. In addition, Howlett et al. [30] pro-
the same line will follow each other with a fixed cycle time vided an analytical method for the problem with more than one
[2]. Solving the timetable that contains fixed cycle time is steep slope and which first divided the route into small parts,
named the cyclic railway timetabling problem (CRTP). Most such that each part contains one steep slope and then solved
studies for solving the CRTP are based on the Periodic Event the precise switching strategy for each part by using a local
Scheduling Problem (PESP) model, which is introduced by optimal principle. In addition to the analytical methods, there
Serafini and Ukovich [3], which considers the problem of are also some studies on evolutionary algorithms. For example,
scheduling a set of periodically recurring events under periodic Chang and Sim [31] applied a genetic algorithm on the train
time window constraints. Nachtigall [4] introduced the Cycle control problem to generate an optimal coast control based on a
Periodicity Formulation (CPF) model, which used the PESP joint evaluation of the punctuality, riding comfort, and energy.
constraints to formulate the cyclic behavior of the railway Ke et al. [32] proposed an ant colony optimization algorithm
timetables, such that the passengers’ waiting time and rolling to find the energy-efficient speed profile and designed a combi-
stock circulation are minimized. Based on the CPF model, a natorial optimization model to reduce the computation time. In
cutting plane algorithm was proposed by Odijk [5] to generate addition, Rémy [33] designed a genetic algorithm to minimize
quickly several feasible timetables, which aims to compose the travel time, delays, and energy consumption, in which the
specifications for the extension of the infrastructure within and concept of the Pareto optimization is used during the selection
around a station by studying a large number of possible future process. In addition, the optimal control theory is also used to
timetables. In addition, Nachtigall and Voget [6] generated sub- assist the driver for energy-efficient driving [34].
optimal solutions using a hybrid genetic algorithm to minimize
the passengers’ waiting time. In the following years, the CPF
was extended to consider different periods [5], variable travel III. M ODEL F ORMULATION AND A LGORITHM
times [7], safety, and frequency constraints [8]. Other recent Here, we formulate a bilevel programming model for the
developments in this field have been mainly concentrated on energy-efficient problem and design a fast algorithm to gen-
the design of robust cyclic timetables [9]–[13] to cope with erate the integrated timetable, which cannot only calculate
stochastic delays. For example, Odijk et al. [14] defined a new the optimal driving strategy for interstations but can generate
probability distribution of feasible robust timetable classes and the timetable by optimizing the distribution of trip time for the
solved the problem with a heuristic sampling method, which entire route as well. First, the energy-efficient driving strategy
is proven to favor classes containing robust timetables over is obtained based on the Pontryagin maximum principle. Then,
others. Kroon et al. [15] designed a stochastic optimization we calculate the minimum trip time and the reserve time for
model that can be used to allocate the time supplements and the train traveling between the successive stations. Furthermore,
the buffer times in a given timetable to reduce the average according to the principle of distribution of the reserve time,
delays. In addition, Fischetti and Monaci [16] proposed a Light we obtain the trip time for each section with speed limits so
Robustness model by coupling robust optimization with a sim- that we can calculate the energy-efficient speed profile for each
plified stochastic two-stage programming approach, which can section. Finally, we extend the algorithm to the entire route for
produce comparable solutions with the stochastic optimization optimizing the integrated timetable.
model. As an extension of Light Robustness, Liebchen et al. For a better understanding of this paper, the assumptions,
[17] provided the first computational study that aims to com- parameters, and variables are introduced first.
SU et al.: TRAIN TIMETABLE OPTIMIZATION APPROACH BASED ON ENERGY-EFFICIENT OPERATION STRATEGY 885

Fig. 3. Different driving strategies between two stations.

t3 coasting time for a given section;


t4 braking time for a given section.
Fig. 2. Curve of running resistance of the Dalian subway line.

A. Assumptions and Symbols


B. Model
Assumptions:
Generally speaking, train energy-efficient operation contains
• The train runs on a flat track, and its mass is fixed. two levels. At the first level, the dispatchers design a timetable
• The unit traction force and the unit braking force are that regulates the trip time for interstations and the dwell time
constants according to the literature [35]. in stations. The second level is to obtain the energy-efficient
• The unit running resistance is assumed as a constant based speed profile for interstations, which minimizes the energy
on the data measured in the Dalian subway line in China consumption with the given trip time. Hence, the energy-
(see Fig. 2) because the running resistance is much smaller efficient operation can be formulated as the following bilevel
than the traction force and the braking force. programming model:
Decision Variables: ⎧
Ti trip time for the ith section; ⎪
⎪ d

⎪ (U ) min E(Ti )
u cruising speed for a given section; ⎪



i=1
w braking speed for a given section; ⎪
⎨ d
s.t. Tt = Ti
s1 cruising points for a given section; (1)


i=1
s2 coasting points for a given section; ⎪
⎪ T min
≤ T ≤ T max


i i i
s3 braking points for a given section. ⎪
⎪ Ti

⎩ E(Ti ) = kt (t)v(t)F (t)dt
Parameters:
0
v speed of the train;
V speed limits; where E(Ti ) is implicitly defined as
v0 initial speed for the given section; ⎧
vT final speed for the given section; ⎪
⎪ Ti

⎪ (L) min E(T ) = kt (t)v(t)F (t)dt
s position of the train; ⎪

i


0

⎨ dv(t)
dt = kt F − kb B − r
Si trip distance for a given section i; s.t.
T trip time for a given section;
⎪ Ti (2)
E energy consumption per unit mass; ⎪
⎪ Si = v(t)dt


Tt total trip time; ⎪

0

⎪ v(0) = v0 , v(T ) = vT
Tr reserve time; ⎩
kt ∈ [0, 1], kb ∈ [0, 1], v ≤ V .
Tmin minimum trip time;
Tdwell total dwell time during the trip;
F maximum traction force per unit mass;
C. Optimal Strategy for Single Section
B maximum braking force per unit mass;
r running resistance per unit mass; Assume that there is a train moving from one station to the
d number of sections. next with given trip time. There are many feasible speed profiles
Intermediate Variables: (see the speed profiles 1, 2, and 3 in Fig. 3), whereas the energy-
kt relative traction force; efficient speed profile is the one that requires the minimum
kb relative braking force; energy.
t1 accelerating time for a given section; The optimization model of the optimal strategy for a single
t2 cruising time for a given section; section has been presented in (2), in which the first equation
886 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 14, NO. 2, JUNE 2013

denotes the optimality criterion, the following two equations


denote the kinematic equation of the train and the constraint on
the trip distance, and the last two equations denote the boundary
conditions and the constraints on the relative traction force, the
relative braking force, and the train speed.
According to the Pontryagin maximum principle [36], [37],
the optimal solution should maximize the Hamiltonian function
p1
H= × (kt F − kb B − r) + p2 v − kt vF (3)
v
where p1 should satisfy the differential equation Fig. 4. Optimal driving strategy for interstations.
dp1 ∂H
=− . (4) acceleration phase can be described as follows:
ds ∂v
dv(t)
It is easy to prove that the Hamiltonian reaches the maximum = F − r. (7)
dt
with respect to the control variables kt and kb as follows:
⎧ Note that the traction force is assumed as a constant. Then,
⎪ kt = 1, kb = 0, if p1 > v 2 the distance s1 , the accelerating time t1 , and the energy con-


⎨ kt ∈ [0, 1], kb = 0, if p1 = v 2 sumption per unit mass E1 of the acceleration phase can be
kt = 0, kb ∈ [0, 1], if p1 = 0 (5) calculated as follows:


⎪ k
⎩ t = 0, k b = 0, if 0 < p 1 < v 2
⎧ 
kt = 0, kb = 1, if p1 < 0. ⎨ s1 = u2 − v02 /2(F − r)
⎩ 1
t = (u − v0 )/(F
− r) (8)
As a consequence, there are five possible phases to be applied E1 = u2 − v02 F/2(F − r).
in the following sequence:
⎧ Since the total energy for traveling between the stations is E,
⎪ Maximum acceleration with kt = 1, kb = 0 which is used for tracting the train in the acceleration and


⎨ Cruising with kt ∈ [0, 1], kb = 0 cruising phases, it is easy to calculate the energy consumed in
Cruising with kt = 0, kb ∈ [0, 1] (6) the cruising phase, which is



⎩ Coasting with k t = 0, k b = 0
Maximum braking with kt = 0, kb = 1. E2 = E − E1 = r × (s2 − s1 ). (9)

Thomas [38] explained the optimal phases as follows. Then, the cruising time becomes
• Maximum acceleration and braking. The slower a train

accelerates or brakes, the more time it needs to come to t2 = (s2 − s1 )/u = E − u2 − v02 F/2(F − r) /(r × u).
a standstill. To obtain the same trip time with a lower (10)
acceleration or braking rate, the train should accelerate to For the braking phase, the deceleration rate is constant. Then,
a higher speed, which consumes more energy. Therefore, we have
the maximum acceleration and braking must be the most 
energy efficient. s3 = S − w2 − vT2 /2(B + r). (11)
• Cruising. Application of the two different cruising strate-
gies (one obtained with partial acceleration, one obtained Due to the law of conservation of energy, the energy applied
with partial braking) depends on the running resistance. for accelerating the train is equal to the energy that is applied to
Under most conditions, running resistance is positive, so overcome the resistance and consumed for braking force, i.e.,
that partial acceleration needs to be applied. The running
E = r × S + B × (S − s3 ) + 0.5vT2 − 0.5v02 . (12)
resistance becomes negative only with very steep descents.
• Coasting. During coasting, when no traction force and Based on (11) and (12), the braking speed is calculated as
braking force are applied, the train only rolls forward and follows:
consumes no energy. Thus, the earlier coasting can start,
the more energy can be saved. w = 2(B + r) (E − rS + 0.5v02 − 0.5vT2 ) /B + vT2 . (13)
Focusing on the subway systems, the optimal sequence of the
driving strategy is described in Fig. 4, where s1 denotes the Note that the braking speed has no relation with the cruising
switching position from acceleration to cruising, s2 denotes speed. Since the total energy consumed is a constant, the
the switching position from cruising to coasting, s3 denotes the energy consumed for running resistance r · S is a constant.
switching position from coasting to braking, u is the cruising Consequently, the energy consumed for overcoming the braking
speed, and w is the braking speed. force and the distance of braking are also constant.
We first calculate the minimum trip time T with the given Since the train moves at a constant deceleration rate in the
energy consumption E. The motion equation of the train in the coasting and braking phases, it is easy to get the coasting time
SU et al.: TRAIN TIMETABLE OPTIMIZATION APPROACH BASED ON ENERGY-EFFICIENT OPERATION STRATEGY 887

Fig. 6. Minimum trip time speed profile.


Fig. 5. Characteristic dependence of energy consumption on trip time.

and braking time as follows: traveling time is T1 , the feasible area for energy consumption
is in A2 . The minimum energy needed for tracking the train is
t3 = (u − w)/r E1 . In other words, if we get the minimum trip time T with
(14)
t4 = (w − vT )/(B + r). the energy constraint E, then E is also the minimum energy
consumption to travel between stations with the given trip
According to the trip time constraint, we have time T .

u − v0 E − u2 − v02 F/2(F − r) Drawing all these in conclusion, we can analytically solve
T = + the optimal switching points. We first calculate the energy
F −r r×u
consumption by (18) with the trip time. Then, we obtain cruis-
u − w w − vT
+ + . (15) ing speed u and braking speed w according to (17) and (13).
r B+r Finally, the optimal switching points can be calculated based
The trip time is uniquely determined by the cruising speed. on (2). In general, the parameters for energy-efficient operation
For obtaining the minimum trip time, according to the are listed as follows:
Kuhn–Tucker conditions, the optimal cruising speed u should ⎧ 

⎪ u = 2E(F − r)/F + v02
satisfy the following equation: ⎪


⎨ w = 2(B + r)(E − rS + 0.5v0 − 0.5vT )/B + vT
2 2 2

E/r + F v02 / (2r(F − r)) s = F u − v0 /2(F  2 − r)2


dT F 2 2 (19)
= − = 0 (16) ⎪ 1

du 2r(F − r) u2 ⎪
⎪ s = s + E/r − F u − v /2r(F − r)
⎩ 2 1 0
s3 = S − (E − Sr + 0.5v02 − 0.5vT2 )/B.
which implies that

u= 2E(F − r)/F + v02 . (17)
Algorithm 3.1: The algorithm for calculating the energy-
efficient speed sequence is summarized as follows.
According to (10) and (17), it is easy to prove that t2 = 0.
Step 1. Initialize v0 = a, vT = b, the distance between sta-
It means that there is no cruising phase in the subway sys-
tions S, and trip time T ;
tems, which coincides to the conclusion from Howlett and
Step 2. Calculate the energy consumption E according to
Pudney [39]: The energy-efficient strategy for short journey,
(18), and calculate the switching points s1 , s2 , and
such as subway systems, consists of acceleration, coasting,
s3 according to (19);
and braking.
Step 3. Calculate the speed sequence for different phases
Substituting (17) into (15), we can find that
according to the motion equation of the train.
  2
2F E F v0 v0 vT
T = + /r − −
F −r F −r F −r B+r D. Minimum Trip Time
  2
2B(E − rS) BvT Here, we present an algorithm to calculate the minimum trip
− + (18)
r2 (B + r) (B + r)r time for each interstation or the entire route. First, from the
right end of each speed limit section (besides the start), we
which denotes the relationship between the energy consump- draw the maximum traction speed profiles. Then, from the left
tion and the trip time for the given section (see Fig. 5). The end of each speed limit section, we draw maximum braking
convexity of the E−T function is proven in the Appendix. speed profiles. Among the maximum traction speed profiles,
Given the total energy E1 , A1 is the area of the feasible maximum braking speed profiles, and the speed limits, we
solution for the trip time T . Since the function is strictly choose the minimum one for the minimum trip time speed
decreasing, T1 is the minimum trip time. On the contrary, if the profile with respect to the displacement axis (see Fig. 6).
888 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 14, NO. 2, JUNE 2013

The speed profile defined must cost the minimum trip time.
The reasons are listed as follows.
1) The speed in the right ends of the speed limit sections
is the maximum value that the train can get in the posi-
tion, and every traction speed profile is calculated with
the maximum acceleration from right ends of the speed
limit sections; therefore, the speed sequence contains
the maximum speed value that the train can get in the
corresponding position.
2) The speed in the left ends of the speed limit sections is
the maximum value that the train can reach, and every
braking speed profile is calculated with the maximum
deceleration from the left ends of the speed limit sections.
If there is a speed value beyond any value of the braking
speed profiles, the speed of the train must exceed the Fig. 7. Distribution of reserve time.
speed limits at some time, and emergency braking will
bring into effect, which is not expected for both the
passengers and the operation company. the first section to reduce more energy consumption. In other
3) The train cannot exceed the speed limits, which is against words, we should keep to the principle that the reserve trip time
the operation requirement. T must be added to the section whichever has the largest ratio
between E and T .
The speed values that we choose meet all the given demands,
and they are the maximum speed values in the corresponding
position. Therefore, the speed sequences constitute the speed Algorithm 3.3: The algorithm for distributing the reserve
profile with the minimum trip time. time among different sections is described as follows.
Step 1. Initialize the maximum trip time for each section Ti ;
Step 2. Calculate the minimum trip time on each section ti
Algorithm 3.2: The algorithm for calculating the minimum
and the total reserve time.
trip time is described as follows.
Step 3. Initialize m = 0 and a large integer number n, and
Step 1. Initialize the speed limit section with (Vi , xi , yi ), in
set T = Tr /n.
which Vi is the speed limit and xi and yi are the
Step 4. Add a short time T to section i and calculate the
initial and final positions of the section, respectively.
energy decreased Ei through (18).
Step 2. Calculate the acceleration speed sequences vit (s) for
Step 5. Make a comparison among the energy reduction, and
each section according to the motion equation of the
add the reserve time T to the ith section with the
train.
largest energy reduction; then, set ti = ti + T .
Step 3. Calculate the braking speed sequences vib (s) for each
Step 6. If ti ≥ Ti , we no longer distribute the reserve time to
section according to the motion equation of the train.
section i.
Step 4. Divide the section into N parts.
Step 7. Set m = m + 1. If m < n, go to step 4.
Step 5. Calculate the minimum trip time
Step 8. Return the trip time sequence {ti } after distribution

N
   for each section.
(yi − xi )/ N × min Vi , vib (s), vit (s) . (20)
i=1 After distributing the reserve time to different sections, we
can optimize the speed profile for each section with the trip
time ti according to Algorithm 3.1. Thus, we get the optimal
E. Distribution of the Reserve Time speed profile with different speed limits.

The reserve time, which is defined as the difference between


the total trip time and the minimum trip time [39], should be F. Optimality Analysis
distributed to different sections for safety constraints, compen-
In the following, we proved the optimality of the allocation
sation of the delay, and efficiency improvement. Considering
of the reserve time with the proofs by contradiction.
to reduce the energy consumption, the same amount of reserve
First, we assume that the assignment on the time units
time added to different sections may lead to a very different
obtained by Algorithm 3.3 is K = [k1 , k2 , . . . kn ], where ki
energy reduction [38].
denotes the number of time units assigned to the ith section.
Assume that there are two sections with the minimum trip
If K is not the optimal assignment, we assume that M =
time T1 and T2 , and the energy consumptions E1 and E2 (see
[m1 , m2 , . . . mn ] is the optimal solution, which implies that
Fig. 7). Distributing the reserve time T to the first section, the
energy consumption will decrease by T · E1 /T1 , com- 
n 
ki 
n 
mi
pared with T · E2 /T2 for the second section. Apparently, Eij < Eij . (21)
if E1 /T1 is larger, we should add the reserve time T to i=1 j=1 i=1 j=1
SU et al.: TRAIN TIMETABLE OPTIMIZATION APPROACH BASED ON ENERGY-EFFICIENT OPERATION STRATEGY 889

Fig. 9. Flowchart of Algorithm 3.4.

which is in contradiction with (26). Therefore, the assignment


of reserve time based on Algorithm 3.3 is the optimal distribu-
tion strategy.

G. Optimization on the Timetable


The energy-efficient timetable aims at optimizing the trip
time for interstations with the constraints of the total trip time
Fig. 8. Optimal distribution of the reserve time. and the maximum and the minimum interstation trip times,
such that the total energy consumption is minimized [see (1)].
It follows that the total trip time constraint is The timetable optimization is based on the energy-efficient

n 
n speed profile. Here, we regard the scheduled stops at stations
ki = mi . (22) as sections with a speed limit of 0 m/s and integrate the reserve
i=1 i=1 time distribution algorithm (see Algorithm 3.3) and the energy-
efficient speed profile algorithm (see Algorithm 3.1) to optimize
Without loss of generality, we assume that x and y (see Fig. 8) the integrated timetable.
satisfy We first calculate the minimum trip time and the reserve time
for the entire route and then distribute the reserve time to the
k x < mx , k y > my . (23)
sections of interstations, which can generate the timetable of the
train by optimizing the running time. Finally, we calculate
Since M is the optimal solution, we have
the optimal speed sequences for each section based on the
Exmx > Eymy +1 . (24) obtained trip time. Thus, we get the optimal timetable and the
energy-efficient speed sequence for the entire route, which is
Otherwise, there is another assignment M  = [m1 , m2 , . . . named as the integrated timetable.
mx − 1, . . . my + 1, . . . mn ], satisfying
Algorithm 3.4: The algorithm for generating the integrated

n 
ki 
mx 
my
timetable is designed as follows.
Eij + Eij + Eij Step 1. Initialize the number of the sections and calculate the
i=1,i=x,y j=1 j=1 j=1
minimum trip time for each section ti and for the

n 
ki 
m x −1 
my +1 entire route Tmin according to Algorithm 3.2.
< Eij + Eij + Eij . (25) Step 2. Initialize the total trip time T and scheduled dwell
i=1,i=x,y j=1 j=1 j=1 time Tdwell and then calculate the reserve time

Then, according to the convexity of the E−T function and Treserve = Tt − Tmin − Tdwell . (29)
my + 1 ≤ ky , kx + 1 ≤ mx , we can further get
Step 3. Distribute the reserve time to each section according
 Exmx > Ey(my +1) ≥ Eyky (26) to Algorithm 3.3 and return the optimal trip time ti .
Step 4. Calculate the energy-efficient speed profile for each
 Ex(kx +1) ≥ Exmx . (27)
section based on Algorithm 3.4.
In addition, kx , ky ∈ K. According to the Algorithm 3.3, there
must exist an index 0 < j ≤ kx + 1, which satisfies The flowchart of Algorithm 3.4 is shown by Fig. 9.
Remark 3.1: When we design a timetable, we get the sched-
Exmx ≤ Ex(kx +1) ≤ Exj < Eyky (28) uled trip time at first, which consists of pure trip time for the
890 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 14, NO. 2, JUNE 2013

TABLE I
O PERATION R EQUIREMENT OF THE Y IZHANG S UBWAY L INE

train to travel between stations and dwell time at scheduled


stops. The pure trip time between stations is the sum of the
minimum trip time from calculation and the reserve time added Fig. 10. Energy-efficient speed profiles with different trip times, where the
to the minimum trip time to enable a train to compensate some five profiles correspond to the trip time as the minimum trip time, 139, 145,
delays. According to the given conclusion, we only distribute 152, and 166 s.
the pure trip time to interstations without considering the dwell
time in the timetable, and we should predict the dwell time at time for the train to travel the interstation is 138 s, as shown
stations if we use the algorithm for the entire route. Fortunately, in the blue speed profile (see Fig. 10). In addition given the
the automatic train supervision system or operation control trip time, we can get different energy-efficient speed profiles by
center in the subway system can send the signal of the dwell Algorithm 3.4. The calculation time is about 0.01 s on average.
time to the ATO system, which guarantees that the algorithm When the trip time is 139 s, there is only 1 s reserved for
can be applied to the ATO system for real-time control. the interstation. For arriving in the station on time, the train
Remark 3.2: Here, we take a period of time, beginning with has to keep the maximum speed for a long distance and coast
the departure at the first station and ending with the arrival at for a little, as shown by the red speed profile in Fig. 10. If
the last station, as the scheduled trip time, excluding the time a longer trip time is given, the train can coast to the braking
for train movement in depot. profile without accelerating to the maximum speed, which can
reduce a lot of energy, as shown by the black and green speed
IV. E XAMPLES profiles. When the trip time is greater than 164 s, the train can
coast to the third section without braking, as shown by the pink
To illustrate the proposed model and numerical algorithm, speed profile. The train only needs to accelerate to the suggested
we present three numerical examples based on the data from speed and then coast until the train has to brake to the station;
the Beijing Yizhuang subway line in China. The operation therefore, the energy is only needed in the first accelerating
requirement and basic infrastructure data about the Yizhuang phase. We can easily conclude from the analysis given that
subway line are described in Table I. In Example 1, we apply the more trip time the train is given, the more energy that we
Algorithm 3.4 to optimize the energy-efficient speed profile for can save.
a single interstation with different trip times. In Example 2, we Example 2: In this example, we apply Algorithm 3.4 to
apply Algorithm 3.4 to the interstations of the Yizhuang subway each section of the Yizhuang subway line to generate the
line to optimize the speed profiles and to compare the energy optimal driving strategy with the same trip time as the practical
consumption with that of practical operation. In Example 3, we operation and make a comparison between them on energy
apply Algorithm 3.4 to generate the integrated timetable of the consumption. The reduction of the energy for each interstation
entire route for energy saving, which can achieve better energy- is calculated as
saving performance.
The parameters in the simulation are defined as follows: The Re = (Ep − Es )/Ep (30)
traction force per unit mass F = 0.8 m/s2 , the braking force
per unit mass B = 0.4 m/s2 , and the running resistance per unit in which Ep denotes the practical energy consumption, Es
mass r = 0.02m/s2 . The simulation is performed on a personal denotes the energy consumption of tracking the optimal driving
computer with a processor speed of 2.6 GHz and a memory size strategy, and Re denotes the efficiency of energy reduction.
of 2 GB. The results, which are shown by Table II, show that the
Example 1: Taking the interstation between Jiugong and energy reduction for interstations is 10.3% on average. Note
Yizhuangqiao for example, there are three sections with speed that for the interstations of Yizhuangqiao–Wenhuayuan and
limits:13 m/s for section {0 m ≤ s ≤ 180 m}; 20 m/s for sec- Ciqu–Yizhuang, the distance is longer than others. According
tion {180 m ≤ s ≤ 2074 m}; and 13 m/s for section {2074 m ≤ to the optimal driving strategy, the train can coast to the
s ≤ 2265 m}. According to Algorithm 3.2, the minimum trip braking speed profile, but the practical driving strategy switches
SU et al.: TRAIN TIMETABLE OPTIMIZATION APPROACH BASED ON ENERGY-EFFICIENT OPERATION STRATEGY 891

TABLE II
I NFRASTRUCTURE DATA OF THE Y IZHUANG S UBWAY L INE AND C OMPARISON OF E NERGY C ONSUMPTION B ETWEEN THE O PTIMAL S TRATEGY AND
THE P RACTICAL S TRATEGY. T HE U NITS OF THE S PEED L IMIT, I NTERVAL , T RIP T IME , AND E NERGY C ONSUMPTION A RE IN M ETERS
PER S ECOND , IN M ETERS , IN S ECONDS , AND IN K ILOWATTHOURS , R ESPECTIVELY

TABLE III
C OMPARISON OF E NERGY C ONSUMPTION B ETWEEN THE O PTIMAL
S TRATEGY AND THE P RACTICAL O PERATION

Fig. 11. Optimal timetable for the Yizhuang subway line. Curves 1 and 2
correspond to the trip time as 1800 and 1900 s.
between acceleration and coasting, which causes increasing
energy consumption. Therefore, we obtain a large percentage of
V. C ONCLUSION
energy saving in these two interstations. In addition, we obtain
a small percentage of energy saving for some interstations, such This paper has proposed an analytical algorithm for solving
as Jiugong–Yizhuangqiao and Rongchang–Tongjinan, because the switching strategy of the energy-efficient speed profile and
the practical driving strategy is very close to the energy-efficient generalized it to solve the timetable of the entire route by dis-
speed profile. tributing the reserve time among different sections. In addition,
Example 3: In this example, we apply Algorithm 3.4 to the the distribution algorithm has been theoretically proven to be
entire route of the Beijing Yizhuang subway line for generating optimal.
the integrated timetable. It follows from Table III that we can Based on the operation data from the Beijing Yizhuang
obtain a better energy saving performance (14.5% energy re- subway line, we have performed numerical examples to prove
duction) through optimizing the timetable and the speed profiles that the proposed algorithm can reduce the energy consumption
as a whole, although the energy consumed Eo is larger than by 10.3% on average for a single interstation and 14.5% for the
Es in some interstations. In Table III, Tp denotes the practical entire route. In addition, the computation time is short enough
trip time, To denotes the optimal trip time, and Eo denotes to apply the algorithm to the onboard control system for a real-
the energy consumption for each interstation in the integrated time adjustment of the timetable.
timetable. Although the algorithm is proven to be efficient on numerical
In addition, when we increase the total trip time to 1800 examples, we need to further conduct empirical studies with
and 1900 s, we use Algorithm 3.4 to generate the integrated variable gradients and variable traction force and braking force,
timetable for the entire route (see Fig. 11). The computation which are always existing in intercity railways and will have
time is about 0.15 s on average. Therefore, the algorithm is an influence on the optimal speed profile. For example, the
quick enough to be applied to the real-time control system. driving strategies may include partial braking when there is a
892 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 14, NO. 2, JUNE 2013

gradient [39]. In addition, the traction force and the braking which implies that
force also vary with speed, which also causes changes of the
speed profile [26]. In these cases, the proposed model may not [E(T1 )−E(T1 +T )] > [E(T1 + T ) − E(T1 + 2  T )]
be valid, and new methods of generating the speed profile and (38)
measuring the energy consumption for interstations are needed.
[E(T1 +T )−E(T1 )] [E(T1 + 2  T ) − E(T1 + T )]
< .
T T
A PPENDIX
(39)
P ROOF OF C ONVEXITY OF THE E−T F UNCTION
In Section III, we have proven that the cruising phase is no When T → 0, (39) denotes that the first-order derivative of
longer an optimal phase for subway systems. Assume that a the E−T function is increasing. Therefore, the convexity of the
train accelerates to coasting speed v, then coasts for a while, and E−T function is proven.
finally brakes to a standstill with trip time T . Travel distance S
can be uniquely determined as R EFERENCES
[1] K. Ghoseiri, F. Szidarovszky, and M. J. Asgharpour, “A multi-objective
S(v, T ) = M1 v 2 + M2 T v + M3 T 2 (31) train scheduling model and solution,” Transp. Res. B, Methodol., vol. 38,
no. 10, pp. 927–952, Dec. 2004.
[2] M. Wardman, J. Shires, W. Lythgoe, and J. Tyler, “Consumer benefits and
in which demand impacts of regular train timetables,” Int. J. Trans. Manage., vol. 2,
⎧ −F (F +B)
no. 1, pp. 39–49, Jan. 2004.

⎨ M1 = 2B(F −r)2 < 0
[3] P. Serafini and W. Ukovich, “A mathematical model for periodic schedul-
ing problems,” SIAM J. Discrete Math., vol. 2, no. 4, pp. 550–581,
2F B 2 +B 2 r+Br 2 +F r 2 +3F Br (32)
M2 = >0 Nov. 1989.


2B 2 (F −r)
[4] K. Nachtigall, “A branch and cut approach for periodic network program-
−(B+r)r
M3 = 2B < 0. ming,” Hildesheimer Inf.-Berichte, Hildesheim, Germany, Tech. Rep. 29,
1994.
Assume that there are three driving strategies with trip time [5] M. A. Odijk, “A constraint generation algorithm for the construction of
T1 , T2 , and T3 , satisfying T1 < T2 < T3 and T2 − T1 = T3 − periodic railway timetables,” Transp. Res. B, Methodol., vol. 30, no. 6,
pp. 455–464, Dec. 1996.
T2 = T > 0. It is obvious that v1 > v2 > v3 and that [6] K. Nachtigall and S. Voget, “Minimizing waiting times in integrated fixed
interval timetables by up-grading railway tracks,” Eur. J. Oper. Res.,
S(v1 , T1 ) = S(v2 , T2 ) = S(v3 , T3 ). (33) vol. 103, no. 3, pp. 610–627, Dec. 1997.
[7] L. Kroon and L. Peeters, “A variable trip time model for cyclic railway
timetabling,” Transp. Sci., vol. 37, no. 2, pp. 198–212, May 2003.
Based on (31), it is easy to calculate that [8] L. Peeters, “Cyclic railway timetable optimization,” Ph.D. dissertation,
Erasmus Univ. Rotterdam, Rotterdam, The Netherlands, 2003.

⎪ [S(v2 , T2 ) − S(v2 , T1 )] − [S(v3 , T2 ) − S(v3 , T1 )] [9] W. S. Lin and J. W. Sheu, “Metro traffic regulation by adaptive optimal

⎪ control,” IEEE Trans. Intell. Transp. Syst., vol. 12, no. 4, pp. 1064–1073,

⎪ = M2 (v2 − v3 )  T > 0
⎨ Dec. 2011.
[S(v2 , T2 ) − S(v3 , T2 )] − [S(v2 , T1 ) − S(v3 , T1 )] [10] Z. H. Zhan, J. Zhang, Y. Li, O. Liu, S. K. Kwok, W. H. Ip, and O. Kaynak,
(34)

⎪ = M2 (v2 − v3 )  T > 0 “An efficient ant colony system based on receding horizon control for the

⎪ aircraft arrival sequencing and scheduling problem,” IEEE Trans. Intell.

⎩ [S(v 3 , T2 ) − S(v3 , T1 )] − [S(v3 , T3 ) − S(v3 , T2 )] Transp. Syst., vol. 11, no. 2, pp. 399–412, Jun. 2010.
= −2M3  T 2 > 0. [11] A. D’Ariano, M. Pranzo, and A. Hansen, “Conflict resolution and train
speed coordination for solving real-time timetable perturbations,” IEEE
By combining (33) and (34), we can get Trans. Intell. Transp. Syst., vol. 8, no. 2, pp. 208–222, Jun. 2007.
[12] M. A. Shafia, M. P. Aghaee, S. J. Sadjadi, and A. Jamili, “Robust train
timetabling problem: mathematical model and branch and bound algo-
S(v1 , T1 ) − S(v2 , T1 ) > S(v2 , T1 ) − S(v3 , T1 ) (35) rithm,” IEEE Trans. Intell. Transp. Syst., vol. 13, no. 1, pp. 307–317,
Mar. 2012.
[13] M. B. Khan and X. S. Zhou, “Stochastic optimization model and solu-
which can be reformulated as tion algorithm for robust double-track train-timetabling problem,” IEEE
Trans. Intell. Transp. Syst., vol. 11, no. 1, pp. 81–89, Mar. 2010.
(v1 − v2 ) [M1 (v1 + v2 ) + M2 T1 ] [14] M. A. Odijk, H. E. Romeijn, and M. H. Van, “Generation of classes of
robust periodic railway timetables,” Comput. Oper. Res., vol. 33, no. 8,
pp. 2283–2299, Aug. 2006.
> (v2 − v3 ) [M1 (v2 + v3 ) + M2 T1 ] . (36) [15] L. Kroon, G. Maróti, M. R. Helmrich, M. Vromans, and R. Dekker,
“Stochastic improvement of cyclic railway timetables,” Transp. Res. B,
It follows from v1 + v2 > v2 + v3 and M1 < 0 that M1 (v1 + Methodol., vol. 42, no. 6, pp. 553–570, Jul. 2008.
[16] M. Fischetti and M. Monaci, “Light robustness,” in Robust and On-
v2 ) + M2 T1 < M1 (v2 + v3 ) + M2 T1 , which implies that v1 − line Large-scale Optimization of Lecture Notes in Computer Science,
v2 > v2 − v3 . R. K. Ahuja, R. Moehring, and C. Zaroliagis, Eds. Berlin, Germany:
Furthermore, we use Ei to denote the energy consumption Springer-Verlag, 2009, pp. 61–84.
[17] C. Liebchen, M. Schachtebeck, A. Schobel, S. Stiller, and A. Prigge,
of the optimal strategy with trip time Ti . Since the energy “Computing delay-resistant railway timetables,” Comput. Oper. Res.,
consumed only in the acceleration phase, we have vol. 37, no. 5, pp. 857–868, May 2010.
[18] X. Yang, X. Li, Z. Y. Gao, H. W. Wang, and T. Tang, “A cooperative
[E(T1 ) − E(T2 )] − [E(T2 ) − E(T3 )] scheduling model for timetable optimization in subway system,” IEEE
Trans. Intell. Transp. Syst., vol. 14, no. 1, pp. 438–447, Mar. 2013.
F  2 [19] Y. H. Wang and Z. Y. Fan, “Dynamic behavior of multi-agent systems
= v1 + v32 − 2v22 > 0 (37) with distributed sampled control,” Acta Autom. Sin., vol. 38, no. 3,
2(F − r) pp. 357–365, Mar. 2011.
SU et al.: TRAIN TIMETABLE OPTIMIZATION APPROACH BASED ON ENERGY-EFFICIENT OPERATION STRATEGY 893

[20] K. Lu, J. M. Xu, S. J. Zheng, and S. M. Wang, “Research on fast dy- Xiang Li received the B.S. degree from Jilin Uni-
namic division method of coordinated control subarea,” Acta Autom. Sin., versity, Changchun, China, in 2004 and the Ph.D.
vol. 38, no. 2, pp. 279–287, Feb. 2011. degree in operations research and cybernetics from
[21] L. F. Li, H. Zhang, X. F. Wang, W. Lu, and Z. P. Mu, “Urban transit Tsinghua University, Beijing, China, in 2008.
coordination using an artificial transportation system,” IEEE Trans. Intell. From August 2008 to March 2009, he was a Visit-
Transp. Syst., vol. 12, no. 2, pp. 374–383, Jun. 2011. ing Scholar with the City University of Hong Kong,
[22] L. Y. Xiao and F. Cao, “Practical string stability of platoon of adaptive Kowloon, Hong Kong. He is currently a Professor
cruise control vehicles,” IEEE Trans. Intell. Transp. Syst., vol. 12, no. 4, with the Beijing University of Chemical Technol-
pp. 1184–1194, Dec. 2011. ogy. He is the author or coauthor of more than 40
[23] K. Ishikawa, “Application of optimization theory for bounded state vari- papers in international journals, including Informa-
able problems to the operation of trains,” Bull. JSME, vol. 11, no. 47, tion Sciences, the European Journal of Operational
pp. 857–865, Oct. 1968. Research, the IEEE T RANSACTIONS ON F UZZY S YSTEMS, etc., which have
[24] J. Cheng and P. G. Howlett, “A note on the calculation of optimal strate- been cited more than 200 times in Web of Science. His current research
gies for the minimization of fuel consumption in the control of trains,” interests include uncertainty theory, fuzzy programming, and fuzzy logic and
IEEE Trans. Autom. Control, vol. 38, no. 11, pp. 1730–1734, Nov. 1993. its applications in sustainable railway transport.
[25] I. Golovitcher, “Control algorithms for automatic operation of rail vehi- Dr. Li served as a member of the Editorial Board for Information: An
cles,” Autom. Remote Control, vol. 11, pp. 118–126, 1986. International Interdisciplinary Journal. His current research was supported by
[26] E. Khmelnitsky, “On an optimal control problem of train operation,” IEEE the National Science Foundation of China, Doctoral Fund of the Ministry of
Trans. Autom. Control, vol. 45, no. 7, pp. 1257–1266, Jul. 2000. Education of China, etc.
[27] R. Liu and I. Golovitcher, “Energy-efficient operation of rail vehi-
cles,” Transport. Res. A, Policy Practice, vol. 37, no. 10, pp. 917–932,
Dec. 2003.
[28] P. G. Howlett, “Optimal strategies for the control of a train,” Automatica,
vol. 32, no. 4, pp. 519–532, Apr. 1996.
[29] P. G. Howlett, “The optimal control of a train,” Ann. Oper. Res., vol. 98,
no. 1–4, pp. 65–87, Dec. 2000.
[30] P. G. Howlett, P. J. Pudney, and X. Vu, “Local energy minimization
in optimal train control,” Automatica, vol. 45, no. 11, pp. 2692–2698,
Nov. 2009. Tao Tang received the Ph.D. degree from the Chi-
[31] C. S. Chang and S. S. Sim, “Optimising train movements through coast nese Academy of Sciences, Beijing, China, in 1991.
control using genetic algorithms,” Proc. Inst. Elect. Eng.––Elect. Power He is the Academic Pacesetter with the National
Appl., vol. 144, no. 1, pp. 65–73, Jan. 2008. Key Subject Traffic Information Engineering and
[32] B. R. Ke, M. C. Chen, and C. L. Lin, “Block-layout design using Control and the Director of the State Key Laboratory
MAX-MIN ant system for saving energy on mass rapid transit systems,” of Rail Traffic Control and Safety, Beijing Jiaotong
IEEE Trans. Intell. Transp. Syst., vol. 10, no. 2, pp. 226–235, Jun. 2009. University. His research interests include both high-
[33] C. Rémy, “Multi-objective evolutionary algorithm for speed tuning opti- speed and urban railway train control systems and
mization with energy saving in railway management,” in Proc. RailRome intelligent control theory. He a Specialist with the
Conf., 2010, pp. 1–19. National Development and Reform Commission and
[34] H. Liimatainen, “Utilization of fuel consumption data in an ecodriving the Beijing Urban Traffic Construction Committee.
incentive system for heavy-duty vehicle drivers,” IEEE Trans. Intell.
Transp. Syst., vol. 12, no. 4, pp. 1087–1095, Dec. 2011.
[35] M. Kondo, “Traction and Braking Performance of Beijing Subway Line,”
Dongyang City Dongzheng Motor Co., Ltd., Dongyang, China, Tech.
Rep., 2009.
[36] I. M. Ross, A Primer on Pontryagin’s Principle in Optimal Control.
Ames, IA, USA: Collegiate, 2009.
[37] H. P. Geering, Optimal Control with Engineering Applications. Berlin,
Germany: Springer-Verlag, 2007.
[38] A. Thomas, Railway Timetable and Traffic. Hamburg, Germany:
Eurailpress, 2008. Ziyou Gao was born in 1963. He received the Ph.D.
[39] P. G. Howlett and P. J. Pudney, “Energy-efficient train control,” in Ad- degree in operations research and control theory
vances in Industrial Control. Berlin, Germany: Springer-Verlag, 1995. from the Institute of Applied Mathematics, Chinese
Academy of Sciences, Beijing, China, in 1994.
From October 1993 to September 1995, he was
a Postdoctoral Researcher in transportation man-
Shuai Su received the B.S. degree from Beijing
agement with the Transportation Simulating Center,
Jiaotong University, Beijing, China, in 2010. He is
Beijing Jiaotong University, where he was previously
currently working toward the Ph.D. degree with the a Professor, later a Doctor Advisor, and currently the
State Key Laboratory of Rail Traffic Control and
Director of the State Key Laboratory of Rail Traffic
Safety, Beijing Jiaotong University.
Control and Safety.
His current research interests include energy-
Dr. Gao was elected foreign member of the Russian Academy of Natural
efficient operation and control in railway systems,
Sciences in 2003. He is also the Vice Chair of the China Transportation
such as timetable optimization, optimal driving
Systems Engineering Society and an Advanced Member of the China Railway
strategies, and rescheduling.
Academy. He is the Coeditor of Traffic and Transportation Systems Engineer
and Information and a member of the Editorial Board for Transportmetrica.

You might also like