Optimal Train Schedule With Headway And

Passenger Flow Dynamic Models

Xubin Sun Shaobo Zhang, Hairong Dong* and Hainan Zhu
School of Electronic and Information Engineering State Key Laboratory of Rail Traffic Control and Safety
Beijing Jiaotong University Beijing Jiaotong University
Beijing,China,100044 Beijing,China,100044
Email: [email protected] Email: [email protected]

Abstract—This paper proposes an optimization method of to solve the optimization problem. Lagrangian relaxation al-
train schedule for metro line, which will make the timetable, gorithm, graph theory and branch-and-bound approach are
including train dwell time, in line with passenger demand. This applied to solve the train scheduling problem as presented in
method is also provide space for timetable improvement in
robustness and energy-saving. Train dwell time is modeled to [11-14].
calculate an optimal dwell time at each station for boarding Dwell time has a direct effect on the passengers’ travel and
and alighting passengers, except headway equation and passenger the load of the train, which is affected by on-off passenger
equation. Train operation schedule includes the train dwell time flows according to Lam [15]. Cao et al. [16] established
at stations, running time between adjacent stations and headway, a piecewise mathematical model for studying the rules of
where dwell time at stations determines the process of passenger
exchanging between stations and trains. A train schedule model passengers boarding time in urban rail transit. Andre [17]
is established with constrains of headway equations, passenger presented dwell time as a function of passenger alighting
equations, and train dwell time equations, where the train and boarding flows, and of the on-board crowding level.
dwell time is modeled as a function of the passenger boarding The models in [17] showed the linear effects in passenger
and alighting volumes. The aim of the optimal problem is to boardings and alightings.
minimize the waiting time of passengers and operation cost
directly or indirectly. Lagrangian duality theory is adopted to Specially, dwell time at stations contributes to much of the
solve this optimal problem. Simulation results illustrate that runningtime and headway variations. The number of passen-
this method is efficient to generate the train schedule, which gers exchanging between trains and platforms is actually deter-
meets the passengers exchanging requirements between trains mined by headway and passengers’ origin-destination matrix
and platforms. (OD matrix). So there is an internal relationship between dwell
Index Terms—train scheduling, dwell time, passenger flow,
lagrangian duality theory, decomposition. time and headway for satisfying the passengers’ demand. In
this paper, a developed analytical traffic model is proposed
due to Assis et al. [6], in which dwell time is presented as
I. I NTRODUCTION
a function of the headway and OD matrix. The Lagrangian
Metro line system plays an important role in the urban duality theory is used to solve this optimization problem with
rail transit as it provides efficient, safe and comfortable travel the safety and operational constraints.
services for passengers in the large cities. One of the important The rest of the paper is as follows. In section II, an
problems for metro line system is the train scheduling prob- analytical traffic model is proposed which represents the train
lem. This is the optimization problem of obtaining the dwell and passenger movements along the metro line. In section III,
time at stations, running time between adjacent stations and a dynamic optimization problem is established, and duality
headway between trains according to the passenger demand theory is used to solve the optimization problem with the
along with the metro line [1-3]. developed traffic models. Numerical simulations are presented
Generally, the techniques used to solve the optimization in section IV.
problem can be mainly classified into three methods: math-
ematical programming, simulation techniques and expert sys- II. ANALYTICAL MODEL OF METRO LINE SYSTEM
tems. In recent years, the mathematical programming has been A. Headway equation
widely used. Cury et al. [4] proposed an analytical traffic Considering the metro line in Fig. 1, it has two terminal
model, and a multilevel hierarchical optimization method was stations and n-2 through stations. The headway equation and
used to solve the optimization problem. Assis et al. [6] devel- passenger equation will be present to describe trains and
oped the analytical traffic model and used the model predictive passenger movements subject to the safety and operational
control method to solve the train scheduling problem in metro constraints [4][6].
lines. Wong et al. [8] presented a mix-integer-programming The headway equation can be presented as
optimization model for train scheduling problem. Li et al.
[9] and Chang et al. [10] used fuzzy mathematical approach xi (k) = ti (k) − ti−1 (k) (1)

978-1-4673-5277-2/13/$31.00 ©2013 IEEE

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on April 28,2024 at 09:06:14 UTC from IEEE Xplore. Restrictions apply.
 7KURXJK
7HUPLQDO VWDWLRQ
7HUPLQDO
where ST is the total number of stations of the metro line.
ĂĂ i (k, j) is obtained from OD matrices. For the metro line, the
     Q Q
passenger equation can be presented as
Fig. 1. Metro Line. k−1
p(k + 1) = p(k) + x(k)∇(k) − ∇ε(k − j, k)x(k − j)
j=1
(9)
xi (k + 1) − xi (k) = di (k) + ri (k) − di−1 (k) − ri−1 (k) (2) where
c(k) = −t0 (k + 1) + t0 (k) (3) Δ T
∇(k) = diag α1 (k) α2 (k) α3 (k) ··· αT R (k)
where t0 (k) is the open time of the kth station, ti (k) is the
arrival time of train i at station k, di (k) is the dwell time of Δ
train i at station k, ri (k) is the running time of train i between ∇ε(k − j, k) = diag ε1 (k − j, k) ε2 (k − j, k)
T
the adjacent stations k and k + 1. Then (2) can be rewritten ε3 (k − j, k) · · · εT R (k − j, k)
as
C. Dwell time determined by passengers exchanged in the
x(k + 1) = x(k) + Ld(k) + Lr(k) + C(k + 1) (4) stations
where Dwell time is defined as the time elapsed between the door
Δ opening and the door closing of a train stopping at a station.
x(k) = [ x1 (k) x2 (k) x3 (k) ··· xT R (k) ] T
Dwell time should guarantee the passengers can exchange
Δ successfully between platforms and trains. Therefore, board-
d(k) = [ d1 (k) d2 (k) d3 (k) ··· dT R (k) ]T
ing and alighting at stations are the most significant factors
Δ
r(k) = [ r1 (k) r2 (k) r3 (k) ··· rT R (k) ]T affecting dwell time, and dwell time can be expressed as
where TR is the total number of trains in service on the di (k) = 4.074 + 1.214Boi (k) + 1.155Ali (k) (10)
line. L and C(k) is defined as
⎡ ⎤ where Boi (k) is the number of passengers boarding train
1 0 0 0 ··· 0 0
⎢ −1 1 0 0 · · · 0 0 ⎥ i at station k, Ali (k) is the number of passengers alighting
Δ ⎢ ⎥ from train i at station k, 4.074 is the empirical value of the
L=⎢ . .. .. .. .. .. .. ⎥
⎣ .. . . . . . . ⎦ time that trains’ doors open and close when there are no
0 0 0 0 · · · −1 1 passengers boarding and alighting. 1.214 and 1.155 are the
Δ time which passengers boarding and alighting need for per
C(k) = [ c(k) 0 0 ]T person. Considering equation (6) and (7), (11) can be
B. Passengers Equation ST
εi (k,j)
j=k+1
The passengers in the ith train when it arrives at the k + 1th di (k) = 4.074 + 1.214xi (k) 2Ndoor
k−1 (11)
station can be presented as εi (k−j,k)xi (k−j)
j=1
+ 1.155 Ndoor

pi (k + 1) = pi (k) + pei (k) − psi (k) (5) where Ndoor is the number of doors of a train.
where pi (k) is the number of passengers on the train i when D. Safety and Operational Constraints
it arrives at the station k, pei (k) is the number of passengers
boarding when the train i is in the station k, psi (k) is the The headway, dwell time, running time and the number of
number of passengers alighting when the train i is in the passengers in a train all have lower and upper bounds for the
station k. And pei (k) and psi (k) can be defined as safety and operational rules as following.
ST
pei (k) = xi (k)αi (k) = xi (k) εi (k, j) (6) x ≤ xi (k) ≤ x̄ k = 1, 2, · · · , ST ; i = 1, 2, · · · , T R
j=k+1


k−1
psi (k) = εi (k − j, k)xi (k − j) (7) d ≤ di (k) ≤ d¯ k = 1, 2, · · · , ST ; i = 1, 2, · · · , T R
j=1

where αi (k) is the number of passengers per second that enter r ≤ ri (k) ≤ r̄ k = 1, 2, · · · , ST ; i = 1, 2, · · · , T R
the station k and i (k − j, k) is the number of passengers/per
second entering the station k − j whose destination is station
k. Equation (5) can be rewritten as p ≤ pi (k) ≤ p̄ k = 1, 2, · · · , ST ; i = 1, 2, · · · , T R
ST
The x, d, r and p are the minimum values of headway, dwell
pi (k + 1) = pi (k) + xi (k) εi (k, j) ¯ r̄ and p̄ are the maximum of
j=k+1 time and running time. The x̄, d,
k−1 (8) Δ
headway, dwell time, running time and passenger load, p̄ =
− εi (k − j, k)xi (k − j)
j=1
1.2Pr , where Pr is the rated passenger load of a train.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on April 28,2024 at 09:06:14 UTC from IEEE Xplore. Restrictions apply.
 III. SOLUTION WITH THE DECOMPOSITION and the Lagrangian function is defined as
METHOD AND DUAL THEORY
L(x, p, r, d, λ1 , λ2 , λ3 )
ST −1
A. Optimal Schedule Problem = k=0 {F (k)− λT1 (k) [x(k + 1) − x(k) − Ld(k)
An efficient cost function should consider the passenger − Lr(k) − C(k) − λT2 (k) [p(k + 1) − p(k)

k−1
waiting time, passengers load, headway and train operation −∇(k)x(k) + j=1 ∇ε(k − j, k)x(k − j)
cost. Then the cost function can be presented as
− λT3 (k) [d(k) − 4.074Ic − 1.214∇(k)x(k)

2Ndoor
ST 
− N1.155
k−1
J= 1 2 γi (k+1) ∇ε(k − j, k)x(k − j)
2 Ri (k + 1)αi (k + 1)xi (k + 1) + xi (k+1) door j=1
k=1
2 2
(18)
1 1
2 p(k + 1) − p (k +1)T (k) d(k) − dr (k)D(k) where the F (k) is the kth term of (12). If the values of
r
+ + 2
2
+ 12 r(k) − rr (k)Q(k) φ(λ1 , λ2 , λ3 ) and ∇φ(λ1 , λ2 , λ3 ) for fixed λ1 (k) = λ∗1 ,
∗ ∗
(12) λ2 (k) = λ2 and λ3 (k) = λ3 can be obtained, the dual
Then the optimization problem of train schedule for a metro problem (15) can be solved with some gradient techniques
line can be formulated as: such as Newton-method, conjugate gradient, etc [4][7][18].
⎧ Then (18) can be rewritten as
⎪ min J(x, p, r, d)
⎨ x,p,d,r
L(·)
s.t. (4), (9), (11) (13)
⎪
⎩
ST −1
= k=0 F (k) − λT1 (k)x(k + 1) + λT1 (k + 1)x(k + 1)
boundary constraints
+ λT1 (k)Ld(k) + λT1 (k)Lr(k) + λT1 (k)C(k + 1)
where (4), (9) and (11) are linear equality constraints, and − λT2 (k)p(k + 1)+λT2 (k+1)p(k + 1)
boundary constrains are listed in the Section II. As the cost + λT2 (k+1)∇(k+1)x(k + 1)
function is convex function, the optimal solution exists for 
ST −1 T
optimization problem (13). − j=k+2 λ 2 (j)∇ε(k + 1, j)x(k + 1)
− λT3 (k)d(k) + 4.074λT3 (k)Ic
B. Solution for Optimal Schedule Problem with Dwell Time 1.214
+ λT3 (k + 1) 2N door
∇(k+1)x(k + 1)
Model 1.155 ST −1
+ Ndoor j=k+1 λT3 (j)∇ε(k + 1, j)x(k + 1)
The optimization problem (13) is of high dimensionality (19)
and its cost function is discrete stage-varying nonlinear, so where Ic is column vector whose T R elements are all 1, and
duality theory [4][5][7][19] is used here to deal with high
dimensionality of the optimization problem. The Lagrange λ1 (k) = 0 λ2 (k) = 0 λ3 (k) = 0 k ≥ ST
dual optimization method is used in this paper, where decom- To get the value of the dual function (17) for fixed λ1 (k) =
position technique is used to obtain a set of small independent λ∗ , λ2 (k) = λ∗ and λ3 (k) = λ∗ , it is needed to solve
1 2 3
minimization problems. The optimization problem based on
Lagrange method can be presented as min L(x, p, r, d, λ∗1 , λ∗2 , λ∗3 ) (20)
x,p,d,r
⎧
⎪ min J(x, p, r, d, λ1 , λ2 , λ3 )
⎨ x,p,d,r subject to (4), (9) and (11).
s.t. (4), (9), (11) (14)
⎪
⎩ 
boundary constraints &DOFXODWH I O O O ’I O O O      

&RQMXJDWH JUDGLHQW WHFKQLTXH 

7KH)LUVW/HYHO 
and the dual of the primal problem can be presented as: O L 
O D
L
G L

GL 9 O L  EL N G L 
⎧
⎪
⎨ λ max J(λ1 , λ2 , λ3 , x, p, r, d, )
1 ,λ2 ,λ3
[ N
[  [ Q
G N
s.t. (4), (9), (11) (15) 7KH6HFRQG/HYHO G  G Q
⎪
⎩ O S 
U  O
U N
S N
U Q
S Q
O
boundary constraints 0LQ 0LQ 0LQ
/ [  S U G O O O
 ĂĂ  >/[ [SSUUG@G O O O   
ĂĂ / [  S U G O O O
where λ1 , λ2 , λ3 are the dual variables of Lagrangian func- >[  S U G @  >[  S U G @
V W     V W    V W    
tion. The two problems (14) and (15) will have the the same
1VXESUREOHPV
 
solution when
min L(x, p, r, d, λ∗1 , λ∗2 , λ∗3 )
x,p,d,r Fig. 2. Decomposition Structure.
= max J(λ1 , λ2 , λ3 , x∗ , p∗ , r∗ , d∗ ) (16)
λ1 ,λ2 ,λ3
As the additive separability of the Lagrangian function, the
where λ∗1 , λ∗2 , λ∗3 , x∗ ,p∗ ,d∗ ,r∗ are the optimal values. The minimization problem is decomposed into the ST independent
dual function is presented as minimization problems according to duality theory as showed
in Fig. 2. In the second level, T(k + 1),D(k) and Q(k)
φ(λ1 , λ2 , λ3 ) = min L(x, p, r, d, λ1 , λ2 , λ3 ) (17) are assumed to be positive definite diagonal matrices, the
x,p,d,r

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on April 28,2024 at 09:06:14 UTC from IEEE Xplore. Restrictions apply.
 ⎡  ⎤

minimization problem (14) is a set of 4*ST independent 1- ∇λ1 (0) φ(λ)
⎢ λ1 =λ
⎥
i
1
variable minimization problems as showed in Fig.2. In the ⎢ .. ⎥
second level, the optimal value of x, p, d, r will be calculated ⎢ . ⎥
⎢  ⎥
with fixed value of λ∗1 , λ∗2 , λ∗3 provided by the first level. In the ⎢  ⎥
⎢ ∇λ1 (ST −1) φ(λ) ⎥
first level, the optimal value of λ1 , λ2 , λ3 will be calculated ⎢  λ1 =λ1 ⎥
i
⎢  ⎥
with fixed value of x∗ , p∗ , d∗ , r∗ provided by the second level. ⎢ ∇λ2 (0) φ(λ) ⎥
⎢ λ2 =λ2 i ⎥
⎢ ⎥
The iteration will continue until stop requirement is met. In i ⎢ .. ⎥
τ (λ ) = ∇φ(λ)|λ=λi = ⎢ . ⎥
the second level, minimization conditions for (14) are gained ⎢  ⎥
⎢  ⎥
by setting partial derivatives equal to zero. ⎢ ∇λ2 (ST −1) φ(λ) ⎥
The optimal values x∗i (k + 1), d∗i (k), ri∗ (k) and p∗i (k + 1) ⎢  λ2 =λi2 ⎥
⎢  ⎥
are obtained with fixed λ∗1 , λ∗2 and λ∗3 at the second level in ⎢ ∇λ3 (0) φ(λ) ⎥
⎢ λ3 =λ3 i ⎥
⎢ ⎥
Fig. 2. If either the upper or the lower bound of a constraint ⎢ .. ⎥
⎢ . ⎥
is violated, the variable is set equal to the bound. Therefore, ⎣  ⎦

(17) is presented as ∇λ3 (ST −1) φ(λ)
λ3 =λ3i

ST −1 (27)
φ(λ1 , λ2 , λ3 ) = {F [x∗ (k + 1), p∗ (k + 1), r∗ (k), d(k)] where βi is the length step, and the initial value of λ is set to
k=0
0. In the iterative computation, if the condition of convergence
+λT1 (k) [x∗ (k + 1)−x∗ (k)−Ld∗ (k) − Lr∗ (k)−C(k + 1)]
is satisfied, the procedure of computation will stop, and the
−λT2 (k) [p∗ (k + 1)−p∗ (k) − ∇(k)x∗ (k)
k−1
solution will be obtained. Otherwise the λi+1 calculated in the
∗ T ∗
+ ∇ε(k − j, k)x (k − j) + λ3 (k) [d (k) − 4.074Ic ith iteration is set to the initial value of the i + 1th iteration,
j=1  the values λi+1 i+1 i+1
1 , λ 2 , λ3 , x i
i+1
(k + 1), di+1
i (k), ri∗ (k) and
1.214 ∗ 1.155 k−1 ∗ ∗
− 2Ndoor x (k)∇(k) − Ndoor j=1 ∇ε(k − j, k)x (k − j) pi (k + 1) are calculated according to the two levels in Fig. 2.
(21)
IV. COMPUTATIONAL RESULTS
then
 The values of the headway, passenger load, running time

∇λ1 φ(λ1 , λ2 , λ3 ) ∗ and the parameters are considered as TABLE I.
λ1 (22)
= −x(k + 1) + x(k) + Ld(k) + Lr(k) + C(k + 1)
TABLE I
THE VALUES OF HEADWAY,PASSENGER LOAD AND RUNNING
 TIME

∇λ2 φ(λ1 , λ2 ) = −p(k + 1) + p(k)
λ∗2 (23) Variable x x̄ p p̄ r r̄ d d¯
k−1
+ ∇(k)x(k) − j=1 ∇ε(k − j, k)x(k − j) Value(sec) 90 300 0 1200 100 300 20 60

 ⎧
 ⎪
⎪ R(k + 1) = 10I
∇λ3 φ(λ1 , λ2 , λ3 ) = −d(k) + 4.074 ⎪
⎪
λ∗3 ⎪
⎪ T(k + 1) = I
⎪
⎪
1.214 1.155
k−1
⎨ b(k) = I
+ 2N ∇(k)x(k) + Ndoor ∇ε(k − j, k)x(k − j)
door c(k) = I
j=1
⎪
⎪
(24) ⎪
⎪ b(0) = 0.1I
⎪
⎪
These are dual function and its gradient calculated with ⎪
⎪ c(0) = 0.1I
⎩
x∗i (k + 1), d∗i (k), r∗i (k) and p∗i (k + 1). It’s considered to find Ndoor = 8
the maximum of φ(λ1 , λ2 , λ3 ) using a gradient technique in The I is the identity matrix. b(0)and c(0) associate to
the first level of Fig. 2. Here the FR conjugate gradient method the dispatches time from the yard to the first station. In the
is used. For the ith iteration, the current dual variables λi1 , λi2 , simulation, we just consider that trains run from the 1st station
λi3 are updated by to the ST th station. The value of γi (k +1) is chosen to restrict
⎡ ⎤ ⎡ i ⎤ xi (k) and pi (k + 1) in a reasonable initial scope. In the first
λi1 d1 interaction, the dual variable λ1 , λ2 , λ3 are always chosen as
λi+1 = λi + adi = ⎣ λi2 ⎦ +a ⎣ di2 ⎦ (25) zero, the initial solution of x∗i (k + 1) is
λi3 di3 
γi (k + 1)
where a is the length step and di is the search direction x∗i (k + 1) = 3 (28)
Ri (k + 1)αi (k + 1)
presented as
di = τ (λi ) + βi−1 di−1 (26) γi (k + 1) determines the initial value of xi (k + 1), x0i is
determined by
 k−1 
λ = [ λT1 λT2 λT3 ]T pi (k + 1) = pi (k) + x0i αi (k) − εi (j, k) (29)
j

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on April 28,2024 at 09:06:14 UTC from IEEE Xplore. Restrictions apply.
 where x0i is the initial constant headway that make the actual operation. The dwell time matches efficiently to the
maximum of passenger load pi (k +1) lower than Pr , the rated variation trend of on-off passenger flows in the station just
passenger load of a train, then γi (k + 1) is as it showed in the Fig 4. So the dwell time of the proposed
model can satisfy the demand of passengers along the line,
γi (k + 1) = Ri (k + 1)αi (k + 1) + x03i (30) and reduce the needless dwell time.
In the computational section, the number of stations is set to
9, the number of trains is set to 15, and the OD matrix is 100

considered as a constant parameter over a period of time. The 90

model M1
profile of the passenger load with constant headway and the 80 Reference
reference profiles of passenger load pr (k + 1) are shown in

Dwell time(sec)
70
Fig. 3. When trains are in actual operation, the passenger load
60
is permitted to be higher than the rated passenger load, as
50
shown in Fig. 3.
40

30
1200
20
model M1
1000 10
constant headway 1 2 3 4 5 6 7 8
Station NO.
Passenger Load /person

800

Fig. 5. Simulation result dwell time about the model M1 and the reference
600 data

400

V. C ONCLUSIONS
200
This paper proposes a traffic model of the metro line,
0
1 2 3 4 5 6 7 8 9
where dwell time is presented as a function of the passen-
Station NO. ger boarding and alighting volumes. The Lagrangian duality
theory is applied to solve the optimization problem, which
Fig. 3. Passenger Load Of Constant Headway And Two Kinds Models. has three equality constraints about passenger load and dwell
time. Finally, simulation results are presented to show that
Fig. 4 shows the number of passengers exchanging in the the proposed model is efficient in matching the passengers’
station. Its value is the sum of pei (k) and psi (k). It can be seen demand.
that the passengers exchanging between trains and stations
changes in different stations as the influence of the headway VI. ACKNOWLEDGEMENT
and the OD matrix.
The authors would like to thank NSFC projects
(61304196, 61322307 and 61233001), National High
380 Technology Research and Development Program of China
360 model M1
(2012AA041701 and 2011AA110502), and Fundamental
Research Funds for the Central Universities (2013JBM003
Passengers exchanged(person)

340
and 2013JBZ007).
320

300

280
R EFERENCES
260
[1] J. F. Cordeau, P. Toth and D. Vigo, “A survey of optimization models for
240
train routing and scheduling,” Transportation Science, vol. 32, no. 4, pp.
1 2 3 4 5 6 7 8 380-404, Nov. 1998.
Station NO.
[2] J. Törnquist, “Computer-based decision support for railway traffic
scheduling and dispatching: A review of models and algorithms,” in
Fig. 4. Passengers exchanging between the platforms and the trains Proc. Fifth Workshop on Algorithmic Methods and Models for Optimiza-
tion of Railways, 2006, pp. 1-23.
[3] K. Ghoseiri, F. Szidarovszky and M. J. Asgharpour, “A multi-objective
Fig. 5 shows the dwell time and reference data. The black train scheduling model and solution,” Transportation Research B, vol.
line is the reference dwell time according to the passengers 38, no. 10, pp. 927-952, Dec. 2004.
exchanged in the Fig. 4. The green line is the optimal dwell [4] J. E. Cury, F. A. C. Gomide, and M. J. Mendes, “A methodology for
generation of optimal schedules for an underground rail system,” IEEE
time which connects to the passengers exchanging in the Transactions on Automatic Control, vol. l25, no. 2, pp. 217-222, Apr.
station. The result shows that dwell time is suitable for the 1980.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on April 28,2024 at 09:06:14 UTC from IEEE Xplore. Restrictions apply.
 [5] W. Grega, “Decomposition approach to the public transport scheduling
problem,” Automatic, vol. 29, no. 3, pp.745-50, May 1993.
[6] W. O. Assis, B. E. A. Milani. “Generation of optimal schedules for metro
lines using model predictive control,” Automatica, vol. 40, no. 8, pp.
1397-1404, Aug. 2004.
[7] L. S. Lasdon, Optimization theory for large systems, New York: Macmil-
lan, 1970.
[8] R. C. W. Wong, T. W. Y. Yue, K. W. Fung and J. M. Y. Leung,“Optimizing
timetable synchronization for rail mass transit,” Transportation Science,
vol. 42, no. 1, pp. 57-69, Feb. 2008.
[9] X. Li, D. Wang, K. Li and Z. Gao, “A green train scheduling model
and fuzzy multi-objective optimization algorithm” Applied Mathematical
Modeling, vol. 37, no. 4, pp. 2063-2073, Feb. 2012.
[10] Y. H. Chang, C. H. Yeh and C. C. Shen, “A multiobjective model for
passenger train services planing: application to Taiwan’s high-speedrail
line,” Transportation Research B, vol. 34, no. 2, pp. 91-106, Feb. 2000.
[11] A. Caprara, M. Monaci, P. Toth and P. L. Guida, “A Lagrangian heuristic
algorithm for a real-world train timetabling problem,” Discrete Applied
Mathematics, vol. 154, no. 5, pp. 738-753, Apr. 2006.
[12] A. Caprara, M. Fischetti and P. Toth, “Modeling and solving the train
timetabling problem,” Operations Research, vol. 50, no. 5, pp. 851-861,
Sept./Oct. 2002.
[13] X. Zhou, M. Zhong, “Single-track train timetabling with guaranteed
optimality: Branch-and-bound algorithms with enhanced lower bounds,”
Transportation Research Part B, vol. 41, no. 3, pp. 320-341, Mar. 2007.
[14] A. D’Ariano, D. Pacciarelli, and M. Pranzo, “A branch and bound
algorithm for scheduling trains in a railway network,” European Journal
of Operational Research, vol. 183, no. 2, pp. 643-647, Dec. 2007.
[15] W. H. K. Lam, C. Y. Cheung and Y. F. Poon, “A study of train dwelling
time at Hong Kong mass transit Railway system,” Journal of Advanced
Transportation, vol. 32, no. 3, pp. 280-295, Jan. 1988.
[16] S. H. Cao, Z. Z. Yuan, D. Zhao, “Characteristics analysis and model
establishment for passenger boarding time in urban rail transit,” Journal
of the China Railway Society, vol. 31, no. 3, pp. 89-93, Mar. 2009.
[17] A. Andre, “Dwell time model and analysis for the MBTA Red Line,”
[Online]. Available: http://ocw.nur.ac.rw.
[18] B. C. Zhang, Y. Lu, “Research on dwelling time modeling of urban rail
transit,” Traffic & Transportation, no. 2, pp. 48-52, Feb. 2011.
[19] H. Tamura, “Decentralized optimization for distributed-lag models of
discrete systems,” Automatica, vol.11, no. 6, pp. 593-602, Nov. 1975.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on April 28,2024 at 09:06:14 UTC from IEEE Xplore. Restrictions apply.