Professional Documents
Culture Documents
handle containers and scheduling operations on the cranes. For instance, Henesey et al. (2006) developed an agent
Simulation approaches were also presented in the paper based simulator to evaluate operational policies in trans-
of Vis et al. (2005) where an Integer linear programming shipments containers. A simulation platform to evaluate
model and a Simulation model are suggested to minimize the efficiency of different port designs with agent based
the size of the vehicle fleet under the constraints of time models is introduced in the paper of Sun et al. (2012). Feng
windows and limited capacity. et al. (2015) presented an integrated multi agent planning
in hinterland transport with a design, implementation and
Stochastic models were carried out as well in the work
evaluation.
of Vidovic and Kim (2006) who proposed a continuous
Markov chain model and Mathematical models based on We opted for this particular approach on the basis of its
probability theory to estimate the productivity and cycle ability to deal with stochastic perturbations that occur
time of material handling equipment. An increase in re- during the execution of deterministic schedules already
search work about scheduling quay cranes and yard trucks prepared. Then, we propose a multi agent system (MAS)
is noted recently. As examples: Lee et al. (2008) presented to maximize the productivity of the terminal, while assign-
a MIP and genetic algorithm to minimize the makespan ing simultaneously vessels, cranes, trucks and workers. To
of cranes operations while scheduling quay cranes and evaluate the quality of the solutions given by MAS, we
yard trucks in an integrated way. A similar problem was proposed a constraint programming model.
treated in the paper of Cao et al. (2010) where the author
presented an integrated model for the yard truck and yard 2.1 Problem data
crane scheduling to minimize the makespan of operations.
Yuan et al. (2011) presented a job grouping strategy To generate the integrated plan, the following data must
for the integration of quay crane QC and yard truck be available. We denote:
YT operations with straddle carrier path planning and
task allocation in the yard to minimize the makespan C Number of quay cranes.
of yard jobs and straddle carrier waiting time. Chen P Number of bays along the quay.
et al. (2013) presented a constraint programming model T Number of time periods in the scheduling horizon.
for the simultaneous scheduling of QCs, YTs, and YCs V Number of vessels to serve during the scheduling
to minimize the makespan to serve vessels. Integrated horizon.
scheduling of quay cranes, automated guided vehicles and M Sufficiently large positive number.
handling platforms were also reported in the work of O Number of quay crane operators.
Homayouni et al. (2014). G Maximum number of yard trucks assigned to each
quay crane.
As far as we know, worker scheduling was not integrated Kv Total moves on vessel v.
with the problems stated above and the productivity of nv Priority of vessel v.
quay cranes is considered fix which is not always the case. Dvp Number of containers to be moved in bay p of vessel
In our paper, we consider the productivity rate of quay v.
cranes a variable depending on the performance of the Jv Expected arrival time of vessel v.
operator allocated to the quay crane, the number of yard Ev Right berth position of vessel v.
trucks deployed to serve that quay crane. According to the Fv Left berth position of vessel v.
theory of constraints, the performance of a chain is the one eot 1: if operator o is available at time period t. 0:
of its weakest link. Then, the productivity of quay cranes otherwise.
will be the smallest value between the two parameters βo Productivity rate of operator o.
mentioned above. fct 1: if crane c is available at time period t. 0: otherwise.
gt Number of available trucks at time period t.
That issue imposes an integration of the yard truck de-
α Productivity rate of yard trucks.
ployment problem with the worker scheduling problem
ic Failure rate of quay crane c.
as well as the well-known QCAP and QCSP. Real time
jo Failure rate of operator o.
stochastic perturbations are also carried out in our work
ku Failure rate of truck u.
with a reactive approach proposed to online scheduling and
ac Time from last repair of quay crane c.
rescheduling vessel, quay crane, operator and truck simul-
bo Time from last recovery of operator o.
taneously in container terminals. The method is based on
cu Time from last repair of truck u.
multi agent systems known for their reactivity.
Decision variables are:
2. MULTI AGENT BASED APPROACH Qvtpc 1: if quay crane c is assigned to bay p to serve the
vessel v at time period t. 0: otherwise.
An agent is a physical or a software entity able to par- Utp Number of trucks assigned to the quay crane in bay
ticipate in an organized activity by interacting with its p at time period t.
environments and other agents. A multi agent system is mtp Productivity rate in bay p at time period t.
a set of agents interacting with each other in a common Sitc 1: if operator i is assigned to crane c at time period
environment on real time. In port container terminals, t. 0: otherwise.
multi agents systems are highly solicited and showed great yv Start time of vessel v.
outcomes. Many studies were conducted in order to build Hvtp Number of containers remained to move from vessel
multi agents architecture to be implemented in ports. Sim- v in the bay p at the beginning of time period t.
ulation studies are the most carried out in the literature. Iv Number of time periods spent to serve the vessel v.
127
2019 IFAC MIM
126 Germany, August 28-30, 2019
Berlin, Kaoutar Chargui et al. / IFAC PapersOnLine 52-13 (2019) 124–129
svt 1: if vessel v is being served at time period t. 0: beginning of the scheduling horizon. An updated schedule
otherwise is a new schedule built after a real time perturbation has
Xvtp The amount of workload handled on vessel v at bay occurred starting from the perturbation instant freezing
p at time period t-1. the past. Algorithm 1 describes the heuristic steps.
Zvtp The amount of workload remained on vessel v at bay
p from time period t-1. Algorithm 1 general structure of the heuristic
Deterministic stage: Schedule generated before the beginning of
2.2 Multi agent architecture scheduling horizon.
128
2019 IFAC MIM
Berlin, Germany, August 28-30, 2019 Kaoutar Chargui et al. / IFAC PapersOnLine 52-13 (2019) 124–129 127
the objective function linear, we considered the inverse of Hvtp = 0 (∀v = 1...V, ∀p = 1...P, ∀t = 1...T : t < Jv ) (10)
the PMPH. Each term is assigned a weight depending on Hvtp ≤ Dvp (t == yv ) + Zvtp − Xvtp
the priority of the vessel. (11)
(∀v = 1...V, ∀p = 1...P, ∀t = 1...T : t ≥ Jv )
M inimize :
V T
Iv
nv K (1) t=1
Xvtp ≥ Dvp (∀v = 1...V, ∀p = 1...P ) (12)
v=1 v
C
Xvtp ≤ M (( c=1
Qv(t−1)pc = 1)&(t ≥ yv + 1))
3.2 Constraints (13)
(∀v = 1...V, ∀p = 1...P, ∀t = 2...T : t ≥ Jv)
Xvtp ≤ m(t−1)p (∀v = 1...V, ∀p = 1...P, ∀t = 2...T : t ≥ Jv ) (14)
O P V
i=1
Sitc = p=1 v=1
Qvtpc (∀c = 1...C, ∀t = 1...T ) (2) C
Xvtp ≥ m(t−1)p − M (1 − ( Qv(t−1)pc = 1)&(t ≥ yv + 1))
C c=1
c=1
Sitc ≤ eit M (∀i = 1...O, ∀t = 1...T ) (3) (∀v = 1...V, ∀p = 1...P, ∀t = 2...T : t ≥ Jv )
P V (15)
p=1 v=1
Qvtpc ≤ fct (∀c = 1...C, ∀t = 1...T ) (4)
C V Zvtp ≤ M (t ≥ (yv + 1))
c=1
Q
v=1 vtpc
≤1 (∀p = 1...P, ∀t = 1...T ) (5) (16)
(∀v = 1...V, ∀p = 1...P, ∀t = 2...T : t ≥ Jv )
C V
mtp ≤ M c=1 v=1
Qvtpc (∀p = 1...P, ∀t = 1...T ) (6)
Zvtp ≤ Hv(t−1)p (∀v = 1...V, ∀p = 1...P, ∀t = 2...T : t ≥ Jv ) (17)
C O V
mtp ≤ min( (βi ((Sitc = 1)& Qvtpc = 1)), Utp α) Zvtp ≥ Hv(t−1)p − M (1 − (t ≥ (yv + 1))
c=1 i=1 v=1 (18)
(∀p = 1...P, ∀t = 1...T ) (∀v = 1...V, ∀p = 1...P, ∀t = 2...T : t ≥ Jv )
(7)
C P
Qvtp1c1 ≤ M (1 − Qvtpc )
C V c1=1,c1<c p1=1,p1<p (19)
Utp ≤ G c=1
Q
v=1 vtpc
(∀p = 1...P, ∀t = 1...T ) (8) (∀v = 1...V, ∀p = 1...P, ∀t = 1...T, ∀c = 1...C)
C
c=1
Qvtpc ≤ Hvtp M (∀v = 1...V, ∀p = 1...P, ∀t = 1...T ) (9) Iv ≥ maxT
t=1 (t(svt == 1)) − Jv + 1 (∀v = 1...V ) (20)
129
2019 IFAC MIM
128 Germany, August 28-30, 2019
Berlin, Kaoutar Chargui et al. / IFAC PapersOnLine 52-13 (2019) 124–129
P
Hvtp ≤ M svt (∀v = 1...V, ∀t = 1...T ) (21) 4. COMPUTATIONAL RESULTS
p=1
P
svt ≤ p=1
Hvtp (∀v = 1...V, ∀t = 1...T ) (22) For performance validation purposes, the heuristic pre-
sented in this paper is developed with Java. The multi
sv1t ≤ M (1 − (Jv1 ≥ Jv2 )&(sv2t == 1)& agent system approach is implemented in Java Agent De-
((Ev1 ≥ Ev2 &Ev1 ≤ Fv2 )|(Fv1 ≥ Ev2 &Fv1 ≤ Fv2 ))) (23) velopment Framework (JADE) and executed on a com-
(∀v1 = 1...V, ∀v2 = 1...V : v1 = v2, ∀t = 1...T : t ≥ Jv1 ) puter with an Intel core i3, 1.7 GHZ, and 4-GB RAM.
P First, we generated 5 small instances (A1..A5) to compare
Utp ≤ gt (∀t = 1...T ) (24)
p=1 the result given by the heuristic to the one obtained
Zv1p = 0 (∀v = 1...V, ∀p = 1...P ) (25)
by CP optimizer CPLEX solver dedicated to constraint
programming models. For each small instance, we have
Xv1p = 0 (∀v = 1...V, ∀p = 1...P ) (26) 4 cranes and 8 time periods. Table 1 proves that the
heuristic provides the optimal solution for small instances.
Sitc ∈ {0, 1} (∀i = 1...O, ∀t = 1...t, ∀c = 1...C) (27) ’Obj’ for heuristic is the average found after 10 executions
with ’Sdv’ the standard deviation. For real based instances
Hvtp , Xvtp , Zvtp ≥ 0 (∀v = 1...V, ∀t = 1...T, ∀p = 1...P ) (28)
(B1..B5) having larger size, the solver runs out of memory
Constraint (2) ensures that a quay crane, c, is assigned (*). For each large instance, we have 10 cranes and 24 time
to only one operator, i. The quay crane, c, is assigned periods.
to an operator only when it is needed at time period, t.
Constraint (3) guarantees that a quay crane is assigned Table 1. Comparison of the heuristic to
to an operator only when he is available. Constraint (4) CPLEX
ensures that a quay crane is allocated to a maximum of Heuristic Cplex
one bay and one vessel at a time period if it’s available. Instance Obj Sdv Time(s) Obj Time(s)
Constraint (5) ensures that a bay is assigned to a max- A1 0.06 0 1.8 0.06 7
imum of one quay crane. Constraint(6) sets the value of A2 0.053 0 1.6 0.053 28
the productivity on a bay p to zero if it’s not assigned A3 0.054 0 3 0.054 29
to any quay crane. Constraint (7) fixes the value of the A4 0.065 0 2.1 0.065 14
productivity rate of the quay crane assigned to bay, p. This A5 0.083 0 1.9 0.083 26
B1 0.031 0.004 1.1 * *
rate is the smallest value between the performance of the B2 0.037 0.005 2.6 * *
operator allocated to that quay crane and the productivity B3 0.047 0.002 1.2 * *
that can be achieved by assigning a number Utp of trucks B4 0.034 0.002 1.4 * *
to the quay crane. Constraint (8) ensures that the number B5 0.051 0.004 2.8 * *
of trucks deployed to serve the quay crane assigned to a
specific bay, should not exceed the target G. Constraint (9) Next, we compare two solutions. Planners prepare a sched-
and (10) ensure that if a vessel has not arrived yet to the ule using the solution given by heuristic ”Planned solu-
terminal, the remaining workload on bays is null, and no tion”, then, rescheduling is done after perturbations occur-
crane will be assigned to it. Constraint (11) ensures that rence, resulting in ”Actual solution” at the end. Planners
the remaining workload for a vessel, v, on bay, p, at time could also use the robust schedule provided after simu-
period t is equal to the its initial load minus the amount of lating the previous approach. Then, beginning with this
containers handled during the current period if this period solution, rescheduling is done as well resulting at the end
is the starting period, otherwise, the remaining amount is in ”Actual solution” as described in Fig. 3.
equal to remaining workload of the previous period minus
the number of containers handled during that period. Con-
straint (12) ensures that the number of containers moved
on bay p of vessel v should not be less that the load on
that bay. Constraints (13), (14) and (15) ensure that the
variable X is equal to the number of containers handled
during the period t, that period should be bigger that the
starting time to serve the vessel. Constraint (16), (17) and
(18) ensure that the variable Z is the remaining workload
from the previous time period. Constraint (19) is added to
avoid interference between quay cranes. Constraint (20) is
used to calculate the port stay of a vessel. Constraint (21)
and (22) ensure that remaining load on a vessel is null if
it not yet served. Constraint (23) avoids physical contact Fig. 3. Reactive and proactive approaches
between vessels. If two vessels are physically overlapped,
the recently arrived one should wait. Constraint (24) en- To measure the ability of the multi agent system to react
sures that the number of deployed trucks doesn’t exceed against real time events, we evaluate the gap between the
the total number of available trucks. Constraints (25) and ”Planned solution” and ”Actual solution” for the reactive
(26) ensure that the amount of workload remained are approach. We investigate how much the actual solution
null at the first time period. Constraints (27) ensures that differs from the first planned one for the reactive approach.
S is binary. Constraint (28) ensures the non-negativity of Table 2 reports the results of simulations. Each instance
variables. is evaluated for 100 replications. ”Reactive Gap (Re)” is
130
2019 IFAC MIM
Berlin, Germany, August 28-30, 2019 Kaoutar Chargui et al. / IFAC PapersOnLine 52-13 (2019) 124–129 129
the average of gaps found during the simulation. The gap Henesey, L., Davidsson, P., and Persson, J.A. (2006).
is calculated as in (Eq. 29) Agent based simulation architecture for evaluating oper-
P lannedObjective−ActualObjective
(29) ational policies in transshipping containers. In German
ActualObjective
Conference on Multiagent System Technologies, 73–85.
Table 2. Simulation results Springer.
Homayouni, S.M., Tang, S.H., and Motlagh, O. (2014). A
Instances Reactive Gap(Re) genetic algorithm for optimization of integrated schedul-
B1 2.64%
ing of cranes, vehicles, and storage platforms at auto-
B2 3.54%
B3 -0.43%
mated container terminals. Journal of Computational
B4 -0.58% and Applied Mathematics, 270, 545–556.
B5 -1.79% Lee, D.H., Cao, J.X., and Shi, Q.X. (2008). Integrated
quay crane and yard truck schedule for inbound contain-
Simulation results prove that reactive approach is the ers. In Industrial Engineering and Engineering Man-
suitable for planning stability with small gaps, and it can agement, 2008. IEEM 2008. IEEE International Con-
react to real time perturbations without deviating too ference on, 1219–1223. IEEE.
much from the initial schedule. Liu, J., Wan, Y.w., and Wang, L. (2006). Quay crane
scheduling at container terminals to minimize the max-
5. CONCLUSION imum relative tardiness of vessel departures. Naval
Research Logistics (NRL), 53(1), 60–74.
Integrated scheduling is a vital issue encountered by the Meier, L. and Schumann, R. (2007). Coordination of
majority of port container terminals in the world. Re- interdependent planning systems, a case study. In GI
sources utilization should be managed effectively to in- Jahrestagung (1), 389–396.
teract with different agents in the entire system. In our Meisel, F. (2009). Seaside operations planning in container
research, we proposed a scheduling method for vessel, terminals. Springer.
quay crane, worker and trucks deployment. The heuristic Sun, Z., Lee, L.H., Chew, E.P., and Tan, K.C. (2012).
is integrated into a multi agent system as a scheduling Microport: A general simulation platform for seaport
agent and validated by a constraint programming model. container terminals. Advanced Engineering Informatics,
The MAS is able to react to real time perturbations. This 26(1), 80–89.
reactive approach reveals good results when comparing Tavakkoli-Moghaddam, R., Makui, A., Salahi, S., Bazzazi,
planned and actual solutions. The architecture described M., and Taheri, F. (2009). An efficient algorithm for
in this paper is for implementation in embedded systems to solving a new mathematical model for a quay crane
establish a concrete interaction between different agents. scheduling problem in container ports. Computers &
Further studies on this topic are therefore recommended in Industrial Engineering, 56(1), 241–248.
order to guarantee a near global optimization of the entire Vidovic, M. and Kim, K.H. (2006). Estimating the cycle
process. time of three-stage material handling systems. Annals
of operations research, 144(1), 181–200.
ACKNOWLEDGEMENTS
Vis, I.F., de Koster, R.M.B., and Savelsbergh, M.W.
We gratefully acknowledge our colleagues at APM Termi- (2005). Minimum vehicle fleet size under time-window
nals Tangier, Tangier Med Port, Morocco, for their ongoing constraints at a container terminal. Transportation
collaboration with our research team. We also thank them science, 39(2), 249–260.
for their recommendations to accomplish this work. Yuan, S., Skinner, B.T., Huang, S., Liu, D., Dissanayake,
G., Lau, H., and Pagac, D. (2011). A job grouping
REFERENCES approach for planning container transfers at automated
seaport container terminals. Advanced Engineering
Ak, A. and Erera, A. (2006). Simultaneous berth and quay Informatics, 25(3), 413–426.
crane scheduling for container ports. Atlanta: Working
paper, H. Milton Stewart School of Industrial and Sys-
tems Engineering, Georgia Institute of Technology.
Bish, E.K. (2003). A multiple-crane-constrained schedul-
ing problem in a container terminal. European Journal
of Operational Research, 144(1), 83–107.
Cao, J.X., Lee, D.H., Chen, J.H., and Shi, Q. (2010).
The integrated yard truck and yard crane scheduling
problem: Benders decomposition-based methods. Trans-
portation Research Part E: Logistics and Transportation
Review, 46(3), 344–353.
Chen, L., Langevin, A., and Lu, Z. (2013). Integrated
scheduling of crane handling and truck transportation
in a maritime container terminal. European Journal of
Operational Research, 225(1), 142–152.
Feng, F., Pang, Y., and Lodewijks, G. (2015). Integrate
multi-agent planning in hinterland transport: design,
implementation and evaluation. Advanced Engineering
Informatics, 29(4), 1055–1071.
131