Professional Documents
Culture Documents
A Discrete Time Model For Optimizing The Rail Por 2012 IFAC Proceedings Volu
A Discrete Time Model For Optimizing The Rail Por 2012 IFAC Proceedings Volu
Abstract: The main objective of this work is to model and optimize the rail cycle in seaport
terminals. The model proposed in the paper represents the transfer of import containers from
their storage in the yard until their exit from the terminal by train. At this purpose, the standing
of containers and their movements inside the terminal are modelled by a set of queues, whose
dynamic evolution is described by discrete-time equations, where the state variables represent
the queue lengths and the control variables take into account the utilization of terminal resources
(handling systems as well as tracks) and the timing of train movements from the terminal
towards their destinations. On the basis of this model, an optimization problem is defined that
consists in minimizing the transfer delays of containers in the terminal while satisfying specific
real constraints. This model has been tested on real data regarding an Italian container terminal,
as reported in the paper.
devoted to report some experimental results tested on real that are known in advance to continue by rail. Containers
data. Finally, some concluding remarks are reported in stored in the import area are lifted up by Rubber Tyred
Section 5. Gantry cranes (RTG), then loaded with reach stackers on
trailers that bring them to the rail domestic park (where
2. THE RAIL PORT CYCLE AND THE RMG cranes load them on the train). On the contrary,
DISCRETE-TIME MODEL containers located in the rail yard area are directly loaded
by reach stackers on trailers and transported close to the
In this work, the port rail cycle is studied and, in par- domestic rail park. The model proposed in the following
ticular, the import flow of goods from the port to the will represent this terminal layout (with two import areas)
hinterland is considered. Moreover, the present research but, of course, this model can be easily generalized in case
limits the analysis to containerized cargo and dry goods, the number of import areas or the sequence of handling
excluding refrigerated ones (which require a different man- operations is different.
agement and equipment as well as much shorter forwarding
The objective of this paper is to define an optimization
times, given their perishable nature).
approach in order to determine the optimal system con-
The model boundaries are shown in Fig. 1, i.e. containers figuration in relation to the import port railway cycle
will be modeled from the moment in which, once unloaded in terms of number of handling resources and timing of
from ships, they are stored in the yard until when they handling operations. The final goal is to decrease the total
leave the terminal by rail. More specifically, the phase of time spent to compute the whole cycle, which corresponds
handling (loading and unloading) on the quay is neglected to the maximization of the number of trains leaving the
assuming that, once they arrive in the port, the containers terminal. Even if the problem could appear not so hard,
are automatically placed in the yard. Containers stored in many complexities arise in reality. First of all, this system
the yard are transported by terminal resources to the rail is very rigid, starting from the fact that the railway tracks
park inside the terminal where they are loaded onto trains, represent a resource difficult to be varied; besides, the
usually through Rail Mounted Gantry (RMG) cranes. railway transportation implies a strict and well defined
Then, through a shunting operation by a diesel locomotive, planning that cannot be changed easily.
trains are brought to an interchange park outside the
In the model that we propose, the considered system is
terminal from which they will depart by using the electric
represented with a set of queues, modelling the presence
traction. In the following of the paper, with domestic (or
of containers in specific areas of the terminal. Fig. 2
internal) rail park we refer to the beam of tracks inside
provides a logic representation of the queues considered in
the terminal area, where loading/unloading operations of
this system, whose dynamics is represented with discrete-
containers from/to the train are carried out and where
time equations with sample time equal to ∆T . At a
there is a diesel traction. On the other hand, the external
generic time step t, the arrival rates of containers (that
rail park is the park in which the change of traction occurs
will be forwarded by rail) in the two different import
(from diesel to electric and vice versa).
areas of the terminal are given by the quantities a1 (t)
and a2 (t) (expressed in containers per hour). Analogously,
d(t) models the demand of containers in import (again
expressed in containers per hour), i.e. the pattern of
railway slots scheduled from the terminal towards the
inland. Such processes can equivalently be modelled either
as deterministic sequences or as random sequences, but in
this paper they are assumed to be deterministic.
q5
q4+N +1
q1 q2
a1 q6 q4+N +2
q4
. d
q3 . .
a2 . .
.
q4+N +M
Fig. 1. The considered system boundaries. q4+N
84
CTS 2012
September 12-14, 2012. Sofia, Bulgaria
buffer represents the cycle described before: containers external rail network. Hence, the effective departure of a
from the yard are lifted by a crane, then loaded with a train from qi , i = 4 + N + 1, . . . , 4 + N + M , is then verified
reach stacker on a trailer which transports them to the when d(t)zi (t)∆t > 0, i.e. d(t)zi (t)∆t = C since only full
internal rail yard. On the contrary, containers present in trains (with C containers) can leave the terminal.
the second import yard have a different handling cycle
and they are directly brought under the crane in the The dynamics of the overall transfer activities in the
terminal can be described with conservation equations
rail park (hence, containers exiting from q3 go into q4 ).
that, at each time step t + 1, update the queue lengths
The productivity of the handling resources dedicated to
work in the import terminal yard, at time t, is denoted according to their length at the previous time step t
and the number of entering and exiting containers in
as ui (t), i = 1, . . . , 3, (this productivity is expressed in
the time interval [t, t + 1), with length ∆t. The discrete-
containers per hour); in particular, ui (t) represents the
rate at which containers are taken from qi at time t. time equations representing the system dynamics are the
following:
Containers wait in queue q4 , i.e. in the area close to q1 (t+1) = q1 (t)+[a1 (t)−u1 (t)]∆t t = 0, . . . , T −1 (1)
the domestic rail park, till they are loaded on the rail
q2 (t+1) = q2 (t)+[u1 (t)−u2 (t)]∆t t = 0, . . . , T −1 (2)
cars available in the internal tracks. The domestic rail
yard is composed of N tracks represented by queues qi (t), q3 (t+1) = q3 (t)+[a2 (t)−u3 (t)]∆t t = 0, . . . , T −1 (3)
i = 5, . . . 4 + N ; we suppose that, when a train is fully 4+N
X
loaded with C containers, it must leave the internal park. q4 (t + 1) = q4 (t) + u2 (t) + u3 (t) − u4,i (t) ∆t
Moreover, it is worth underlining that rail cars are assumed i=5
to be always available in the internal rail park. Let us t = 0, . . . , T − 1 (4)
denote with u4,i (t), i = 5, . . . , 4 + N , the productivity at
time t of the rail cranes used to load the containers on 4+N
X +M
trains in the related track (again, this productivity is given qi (t + 1) = qi (t) + u4,i (t)∆t − C yi,j (t)
in containers per hour). j=4+N +1
85
CTS 2012
September 12-14, 2012. Sofia, Bulgaria
matical programming problem must be defined. This prob- The objective function (7) is a weighted sum of the queues
lem must take into account the system dynamics repre- over the whole time horizon, in which each queue qi (t),
sented by the discrete-time queue-based model described i = 1, . . . , 4 + N + M , is appropriately weighted with the
in Section 2 and some other operative constraints, over a coefficient ωi . In this way, the total transfer delay in the
time horizon given by T time steps. terminal is minimized, as done by Alessandri et al. (2007).
It is worth noting that, by suitably tuning the weights
The optimal control problem for planning the port rail
associated with the different queue lengths, it is possible to
cycle can be then stated with the following mixed-integer
provide the defined cost function with different objectives,
programming formulation.
privileging the presence of containers in given areas of the
Problem 1. Given the arrival and departure rates, i.e. terminal or limiting them in areas where there are space
ai (t), i = 1, 2, and d(t), t = 0, . . . , T − 1, given the limitations to be taken into account. In this model, in order
initial conditions, i.e. qi (0), i = 1, . . . , 4 + N + M and to plan correctly the departure of trains from the internal
yi,j (t), i = 5, . . . , 4 + N , j = 4 + N + 1, . . . , 4 + N + M , and external yards according to the scheduled demand,
t = −τ, . . . , −1, given the maximum queue lengths qimax , the highest weights will be associated with the queues of
i = 1, . . . , 4 + N + M , the maximum handling rates umax i , the external tracks, medium weights will be related to the
i = 1, . . . , 3 and umax 4,i , i = 5, . . . , 4 + N , given the cost queues of internal tracks, while the other queues will be
weighting parameters ωi , i = 1, . . . , 4 + N + M , given C given the smallest weights.
and τ , find the state variables qi (t), i = 1, . . . , 4 + N + M ,
t = 1, . . . , T , and the control variables ui (t), i = 1, . . . , 3, Constraints (8) impose that, in the external rail park, one
t = 0, . . . , T − 1, u4,i (t), i = 5, . . . , 4 + N , t = 0, . . . , T − 1, track must be always free, for the reasons explained before
yi,j (t), i = 5, . . . , 4 + N , j = 4 + N + 1, . . . , 4 + N + M , in the paper. Moreover, constraints (9) model the presence
t = 0, . . . , T − 1, zi (t), i = 4 + N + 1, . . . , 4 + N + M , of only one siding track connecting the domestic with the
t = 0, . . . , T − 1 that minimize external rail park and the fact that τ time steps are needed
T 4+N +M to cross it; for this reason, in the τ time steps in which a
X X
ωi qi (t) (7) train moves from the internal to the external park, the sum
of the yi,j (t) must be lower or equal to 1, i.e. only one train
t=1 i=1
can be transiting on the siding track. Constraints (10) are
subject to the model dynamics given by (1)-(6), and needed to model the fact that, at each time step t, only
4+N
X +M one external track can be available to satisfy the external
qi (t) ≤ (M − 1)C t = 1, . . . , T (8) demand.
i=4+N +1
Constraints (11) and (12) impose that a train can leave
4+N
X 4+N +M τ −1
X X the domestic rail park only when it has been loaded with
yi,j (t + k) ≤ 1 t = 0, . . . , T − τ (9) C containers. When a train leaves the domestic rail park,
i=5 j=4+N +1 k=0 it can go to one of the different external tracks; therefore
4+N
X +M it is necessary to impose that, when qi (t) = C, it must be
zi (t) = 1 t = 0, . . . , T − 1 (10) P4+N +M
verified that j=4+N +1 yi,j (t) = 1, i.e. the train from qi
j=4+N +1
must be sent to a queue qj , j = 4 + N + 1, . . . , 4 + N + M ,
4+N +M at time step t. If instead qi (t) < C, then it must be
X P4+N +M
qi (t) − C + G 1 − yi,j (t) ≥ 0 j=4+N +1 yi,j (t) = 0.
j=4+N +1
Constraints (13) and (14) indicate that the quantity ex-
i = 5, . . . , 4 + N t = 0, . . . , T − 1 (11) iting each queue qi , i = 1, . . . , 4, at time t, must be
4+N +M lower than or equal to the queue length at the same time
step. Constraints (15) impose that each queue qi at each
X
C − qi (t) + G yi,j (t) > 0
time step cannot be larger than its maximum value qimax ;
j=4+N +1
analogously, constraints (16) and (17) impose that the
i = 5, . . . , 4 + N t = 0, . . . , T − 1 (12) handling rates cannot exceed the maximum values of the
ui (t)∆t ≤ qi (t) i = 1, . . . , 3 t = 0, . . . , T − 1 (13) capacity of the terminal resources, i.e. umax i , i = 1, . . . , 3,
4+N
X and umax
4,i , i = 5, . . . , 4 + N . Finally, constraints (18) and
u4,i (t)∆t ≤ q4 (t) t = 0, . . . , T − 1 (14) (19) impose that yi,j (t) and zi (t) are binary variables.
i=5
Hence, the resulting control problem is a mixed-integer
0 ≤ qi (t) ≤ qimax i = 1, . . . , 4 + N + M t = 1, . . . , T
linear programming problem that can be solved with com-
(15)
mercial solvers (which apply standard solution methods
0 ≤ ui (t) ≤ umax
i i = 1, . . . , 3 t = 0, . . . , T − 1 (16)
providing the global optimal solution). Anyway, as known,
0 ≤ u4,i (t) ≤ umax
4,i i = 5, . . . , 4 + N t = 0, . . . , T − 1 for mixed-integer linear programming problems, in the
(17) worst case the solution time depends exponentially on the
yi,j (t) ∈ {0, 1} i = 5, . . . , 4 + N number of integer variables. These computational issues
must be taken into account when choosing the planning
j = 4 + N + 1, . . . , 4 + N + M t = 0, . . . , T − 1 (18)
horizon for the problem. As a matter of fact, a short plan-
zi (t)(t) ∈ {0, 1} ning horizon does not allow to correctly predict the system
i = 4 + N + 1, . . . , 4 + N + M t = 0, . . . , T − 1 (19) state but it corresponds to a smaller problem instance, to
be solved in short computational times; on the other hand,
where G is a large quantity arbitrarily chosen.
86
CTS 2012
September 12-14, 2012. Sofia, Bulgaria
87
CTS 2012
September 12-14, 2012. Sofia, Bulgaria
160 containers/hour. Hence, in order to model 4 train In Figs. 3-5 the optimal patterns of q6 (t), q8 (t) and q9 (t)
departures in a day, the demand d(t) is set equal to 0 are shown, indicating that the external demand (4 trains)
∀t = 0, . . . , T − 1, except for the following cases: is met by 4 trains present in the internal tracks, respec-
tively in q6 , q9 , q8 and q9 . Analysing the optimal values
• d(20) = 160 containers/hour, corresponding to a train
of the handling rates, it can be stated that the maximum
scheduled at 5 a.m.;
capacity is reached by u2 (t) and u3 (t) (9 containers/hour)
• d(32) = 160 containers/hour, corresponding to a train and by u4,8 (t) and u4,9 (t) (30 containers/hour); this means
scheduled at 8 a.m.;
that the corresponding handling resources (reach stacker
• d(80) = 160 containers/hour, corresponding to a train
and trailer, on one side, and RMG crane, on the other)
scheduled at 8 p.m.; are the bottlenecks of the system, i.e. by increasing their
• d(92) = 160 containers/hour, corresponding to a train
productivity it is possible to schedule more trains and to
scheduled at 11 p.m..
increase the total terminal rail throughput.
With these data, Problem 1 has been solved and the
optimal values of the state and control variables have 5. CONCLUSION
been found. The resulting problem is characterized by
5088 decision variables (2592 are binary and the remaining In this paper, a discrete-time dynamic model has been
2496 are continuous), 1536 equality constraints and 2112 proposed for representing the import rail cycle in a seaport
inequality constraints. The solution is found by Cplex in container terminal. Based on this model, a control problem
few seconds. has been formalized in order to optimally plan the use of
handling resources in the system and the timing of train
45
departures from the terminal. This control problem has
40
been applied to a real case study and some results have
35 been shown in the paper.
30
Present and future research is devoted to extend the
Containers in q6
25
present model to consider also the export flow of containers
20
which, of course, partially shares the resources (tracks,
15 cranes, and so on) used for the import cycle. Moreover, a
10 discrete-event simulation framework is being implemented
5
in order to model the port rail cycle in a more detailed
0
way and to be used to validate the results obtained by the
10 20 30 40 50 60
Time steps
70 80 90
optimization framework.
Fig. 3. The pattern of q6 (t). REFERENCES
Alessandri, A., Sacone, S., and Siri, S. (2007). Modelling
45 and optimal receding-horizon control of maritime con-
40
tainer terminals. Journal of Mathematical Modelling
35
and Algorithms, 6, 109–133.
Ambrosino, D., Bramardi, A., Pucciano, M., Sacone, S.,
30
and Siri, S. (2011). Modeling and solving the train load
Containers in q8
25
planning problem in seaport container terminals. In 7th
20
annual IEEE Conference on Automation Science and
15 Engineering.
10 Bemporad, A. and Morari, M. (1999). Control of systems
5
integrating logic, dynamics, and constraints. Automat-
0
ica, 35, 407–427.
10 20 30 40 50 60
Time steps
70 80 90
Bostel, N. and Dejax, P. (1998). Models and algorithms
for container allocation problems on trains in a rapid
Fig. 4. The pattern of q8 (t). transshipment shunting yard. Transportation Science,
32, 370–379.
45
Crainic, T. (2009). Service design models for rail inter-
40
modal transportation. In B. et al. (ed.), Innovations in
Distribution Logistics, Lecture Notes in Economics and
35
Mathematical Systems 619. Springer Verlag.
30
Stahlbock, R. and Voss, S. (2008). Operations research at
Containers in q9
25
container terminals: a literature update. OR Spectrum,
20 30, 1–52.
15 Steenken, D., Voss, S., and Stahlbock, R. (2004). Con-
10
tainer terminal operation and operations research - a
5
classification and literature review. OR Spectrum, 26,
3–49.
0
10 20 30 40 50 60
Time steps
70 80 90
Vis, I. and Koster, R.D. (2003). Transshipment of con-
tainers at a container terminal: an overview. European
Fig. 5. The pattern of q9 (t). Journal of Operational Research, 147, 1–16.
88