You are on page 1of 9

Expert Systems with Applications 39 (2012) 4213–4221

Contents lists available at SciVerse ScienceDirect

Expert Systems with Applications


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

A modified genetic algorithm for quay crane scheduling operations


S.H. Chung ⇑, K.L. Choy
Department of Industrial & Systems Engineering, Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong

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

Keywords: The demand for the maritime transportation has significantly increased over the past 20 years due to the
Container terminal rapid pace of globalization. Terminal managers confront the challenge in establishing the appropriate
Quay crane quay crane schedule to achieve the earliest departure time of ship and provide efficient service. In gen-
Scheduling eral, quay crane schedule problems include two main issues (1) the allocation of quay cranes to handle
Genetic algorithm
the discharging and loading operations, and (2) the service sequence of ship bays in a vessel of each quay
crane. Traditionally, the terminal planners determine the quay crane schedule based on their experience
and own judgment. In addition, the interference among cranes and the increased in ship size further mag-
nify its difficulty dramatically. Accordingly, this paper proposed a modified genetic algorithm to deal with
the problem. To test the optimization reliability of the proposed algorithm, a set of well known bench-
marking problem is solved, and the results obtained are being compared with other well known existing
algorithms. The comparison demonstrates that the proposed algorithm performs as good as many exist-
ing algorithms and obtains better solutions than the best known ones in certain instances. In addition, the
computational time(s) required are significantly much lesser, allowing it to be more applicable in prac-
tical situation.
Crown Copyright Ó 2011 Published by Elsevier Ltd. All rights reserved.

1. Introduction operations (Bierwirth & Meisel 2010; Meisel & Bierwirth 2006;
Steenken, Vos, & Stahlbock, 2004). It includes the berth allocation
The rapid pace of globalization has driven the trade among plan, the loading and discharging plan of vessels, the yard plan of
countries significantly, and thereupon, boosted the economics the container stowage and import and export container schedule
around the world. Container ports play an indispensable role in through gate (Lim, Rodrigues, Xiao, & Zhu, 2004; Park & Kim,
nowadays global logistics network. Global container throughput 2003). Since penalty cost will be incurred if the vessels cannot
was 534 million TEUs (twenty-foot equivalent unit) in 2008 departure on time, the main objective of all these activities is to en-
(Drewry Shipping Consultants Limited, 2010). Although the global sure the vessels can departure on time by minimizing the turn-
container throughput has been turned down to 464 million TEUs around time of the vessels. All these plans had a close linkage
due to the financial tsunami, it is still optimistic that the global and integrated as terminal operation system. If one of the plans
economy will rebound and the sustainable growth in container failed to achieve its goal, it may induce a negative impact on the
throughputs in 2010–2014 is foreseen. Besides, the geographically operation of the other plans.
closeness within the ports, especially in Europe and Asia, results in Quay Crane Scheduling Problem (QCSP) was firstly described by
the immediate challenge on the competition among ports. The ter- Daganzo (1989). Daganzo (1990) and also in Daganzo and Peterkof-
minal managers are devoted to deliver high quality of services con- sky (1990) defined that the vessel can be partitioned longitudinally
sistently and improve the efficiency of available facilities, in order into a number of ship bays based on the size of the vessel. The turn-
to keep the good relationship with existing customers, attract the around time of a vessel is determined by how to arrange each quay
potential customers and enhance the operational productivity. crane to serve which ship bays and the processing time of each ship
In the last decade, the major development in port business fo- bay in a stowage plan. A stowage plan consists of the planning of
cused on advanced technology development (Günther & Kim loading and discharging. The workload of a ship bay depends on
2005; Zaffalon, Rizzoli, Gambardella, & Mastrolilli, 1998). Ports the number of containers need to be handled, including loading
put much effort on developing various decisions supporting tech- and discharging operations on the deck and in the hold. The prece-
nologies to enhance their resources planning system since resource dence relationship among tasks concerns in the discharging and
allocation directly determines the success and efficiency of the port loading operations. Although they can be handled at different time
and by different cranes, discharging operation must precede the
⇑ Corresponding author. loading operation. At the same ship bay, the tasks on the deck must
E-mail address: mfnick@inet.polyu.edu.hk (S.H. Chung). be finished before the tasks in the hold to start its discharging

0957-4174/$ - see front matter Crown Copyright Ó 2011 Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2011.09.113
4214 S.H. Chung, K.L. Choy / Expert Systems with Applications 39 (2012) 4213–4221

operation. Oppositely, the task in the hold must be performed be- bound for the same yard have to be perform separately in time
fore the tasks on the deck start in loading operation at the same ship to avoid trailer congestion in lanes. A probabilistic tabu search
bay. Therefore, some tasks must be performed before others or can- and a ‘‘squeaky wheel’’ optimization heuristic were proposed to
not be performed simultaneously even by other cranes (Bierwirth & solve the problem, while dynamic programing algorithms were
Meisel 2009). used to solve simplified variants.
As the size of the quay crane is enormous, each ship bay can In solving QCSP, Lee, Wang, and Miao (2006) first proved that
only be served by one quay crane at a time. Furthermore, since QCSP is NP-complete. Many researchers proposed many different
all the quay cranes travel on the same track, a quay crane can only heuristic approaches (Gambardella, Rizzoli, & Zaffalon 1998; Vacca,
perform the next ship bay without crossing the others. If a crane is Bierlaire, & Salani 2007; Wang, Chen, & Wang 2009). Kim and Park
needed to cross others, then it must wait until the other crane fin- (2004) formulated the problem in mixed integer linear model and
ished its tasks and moved away. The objective of this paper is to solved by using branch-and-bound (B&B) method and greedy ran-
propose a genetic algorithm to deal with the QCSP. The paper will domized adaptive search procedure (GRASP). They assumed that
be divided into the followings. Section 2 will be a literature review. the vessel was partitioned into number of ship bays and the dis-
Section 3 will describe the problem description and formulation. charging and loading operation associated with the same ship bays
Section 4 will present the proposed methodology. Section 5 will are divided into different tasks. The objective is to minimize the
test the optimization reliability of the proposed algorithm and dis- makespan of the operation. The result in the study indicated that
cuss the results obtained. The paper will be concluded with a con- the B&B method can only be practical in the small sized problem
clusion in Section 6. and once the number of task and the number of quay crane in-
creased to 20 and 3 respectively, the computational time of B&B
2. Literature review method increase rapidly. The GRASP identified the feasible solu-
tions, which did not exceed the optimal more than 10%, in med-
In general, QCSP refers to (1) the allocation of quay cranes to ium-size problem with reasonable time. Therefore the study
handle the discharging and loading operations, and (2) the service recommended the GRASP in solving the practical QCSP rather than
sequence of ship bays in a vessel of each quay crane (Daganzo B&B method.
1989). The stowage plan refers to (1) the plan indicates the loca- Later on, Moccia et al. (2006) have studied the QCSP with the
tion of the discharged containers in each ship bay of a vessel, same assumption given by Kim and Park (2004) and made modifi-
and (2) the preferred plan about the location of loaded containers cation to the model in order to consolidate the lower bound and
in each ship bay of a vessel, which is based on the destination, the upper bound for the completion time. The valid inequalities
weight, and type of containers. are provided to process the cutting planes method. They proposed
Operations research in the area of terminal operations is known a Branch and Cut (B&C) algorithm to solve the problem. A prepro-
as one of the most challenging topics because of its complexity and cessing technique is introduced to reduce the problem size and the
practical applicability. In the last decades, numerous researches separation heuristic method is also introduced to identify the vio-
have been carried out. The first QCSP studied by Daganzo (1989) lated inequalities. In the study, the problem set studied by Kim and
is to determine the departure time of multiple vessels. Later on, Park (2004) are used to test the solution and the results obtained
Daganzo and Peterkofsky (1990) proposed a branch and bound by using the B&B method and B&C algorithm are compared. The re-
algorithm to deal with the problem under the similar assumption, sults showed that the B&C method outperformed the B&B method
in which one task is performed at one ship bay. The study was to and the proposed algorithm is able to handle the practical prob-
determine the departure times of multiple vessels and the number lems size.
of cranes to be assigned to individual vessels at a specific time seg- Recently, Sammarra et al. (2007) further studied the QCSP mod-
ment, and the objective of the study is to minimize the sum of the el, which have been studied by Kim and Park (2004) and Moccia
delay costs. et al. (2006) under the same assumption made. They solved the
Some researchers stated that QCSP can be modeled as m-paral- QCSP by separating the problem into two parts, routing problem
lel machine scheduling problems (Guinet, 1993; Lee & Pinedo, and the scheduling problem. The routing problem is defined to
1997) and vehicle routing problems (Sammarra, Cordeau, Laporte, determine the sequence of tasks on each crane and the scheduling
& Monaco, 2007). Sammarra et al. (2007) solved the QCSP by sep- problem is defined to determine the starting handling time of the
arating the problem into two parts, routing problem and the sched- tasks belonging to the route for each sequence. They proposed a lo-
uling problem. The routing problem was defined to determine the cal search algorithm to solve the scheduling sub-problem and a
sequence of tasks on each crane and the scheduling problem was Tabu Search (TS) algorithm for routing problem. By using the same
defined to determine the starting handling time of the tasks objective as the one studied by Kim and Park (2004) and Moccia
belonging to the route for each sequence. The significance of the et al. (2006), the results of the TS are compared with them. The
studies is that the precedence constraint among tasks and the performance of the TS is outstanding compared to GRASP, but infe-
non-interference constraint of quay cranes have been defined. rior to the B&C method. TS can obtain the almost optimal solution
Kim and Kim (1999) pointed out that since the quay cranes travel- in the medium-size problem with the reasonable computation
ling on the same track, blocking situation among the cranes may time.
occur. Cranes can only travel to the ship bays and perform the task
without interference. The objective of the study is to address the
problems in scheduling multiple quay cranes to perform discharg- 3. Problem description and formulation
ing and loading operations and to minimize the makespan of the
operation. Later on, Kim and Park (2004) and Moccia, Cordeau, The studied Quay Crane Scheduling Problem (QCSP) in this pa-
Gaudioso, and Laporte (2006) also studied the QCSP with the pre- per consists of K quay cranes to be assigned to serve |X| number of
cedence constraint of the operation and the non-interference con- tasks, including loading and discharging operation on the deck or
straint among cranes. While Lim et al. (2004) assumed that the in the hold. The studied model is based on the model developed
QCSP was task-to-crane assignment problem with integrated spa- by Kim and Park (2004) and the modification made by Moccia
tial constraints, included non-interference constraint among et al. (2006).
cranes and neighborhood constraints. The authors also assumed The following notations are used to describe the problem stud-
that some tasks cannot be perform simultaneously since tasks ied throughout the paper:
S.H. Chung, K.L. Choy / Expert Systems with Applications 39 (2012) 4213–4221 4215

is the main concern. In addition, it aims to find the solution with


Indices the minimum sum of the quay cranes completion time among a
i,j Index of task, arranging in an ascending order of their pool of solutions with the same makespan. A quay crane comple-
relative locations in the direction of increasing ship tion time (Yk) consists of the idle times, traveling times and the
bay numbers. handling times. The task handling time is fixed but the idle times
k Index of quay crane, where k = 1 . . . K, arranging in an and traveling times can affect the completion time of the quay
ascending order of their relative locations in the crane. Hence, the schedule which has the minimum sum of the
direction of increasing ship bay numbers. quay crane completion time is expected to be chosen since more
Problem data quay cranes can become available to be assigned to work on other
pi Processing time required by task i. ships as soon as possible.
rk The earliest available time of quay crane k. Problem Constraints
li The location of task i which expressed by a ship bay The problem is subject to the following constraints:
number.
Yk  W 8k 2 K; ð2Þ
lk The starting location of quay crane k expressed by a
0

ship bay number. X


T The final location of quay crane k expressed by a ship xk0j ¼ 1 8k 2 K; ð3Þ
lk
j2X
bay number.
T Traveling time between two consecutive bays. X
tij Traveling time required by a quay crane from location xkiT ¼ 1 8k 2 K; ð4Þ
i2X
of task i (li) to the location of task j (lj), tij = t  |lj  lj|.
t k0j Traveling time required by quay crane k from the XX
initial location to the first task location lj, xkij ¼ 1 8j 2 X; ð5Þ
0 k2K i2X
tk0j
¼ t   lj j.
jlk
X X
t kiT Traveling time required by quay crane k from the xkij  xkji ¼ 0 8i 2 X; 8k 2 K; ð6Þ
initial location to the first task location lj, j2XT j2X0
T
tkiT ¼ t  jli  lk j (tkiT ¼ 0; 8i 2 X;, meaning that
travelling time after completion of the final task is not Di þ tij þ pj  Dj  Mð1  xkij Þ 8i; j 2 X; 8k 2 K; ð7Þ
considered).
M A sufficiently large constant. Di þ pj  Dj 8i; j 2 U; ð8Þ
a1 Weighting assigned to the makespan (the maximum
completion time) in the objective function.
Di  Dj þ pj  Mð1  zij Þ 8i; j 2 X; ð9Þ
a2 Weighting assigned to the sum of the completion time
in the objective function.
Set of indices zij þ zji ¼ 1 8i; j 2 W; ð10Þ
X The set of all tasks.
W The set of pairs of tasks that cannot be performed X
k X X
k X
xvuj  xvui  Mðzij þ zji Þ 8i 2 X; li < lj ; 8 k 2 K
simultaneously. Where tasks i and j cannot be v ¼1 0 v ¼1 0
u2X u2X
performed simultaneously, ði; jÞ 2 W .
ð11Þ
U The set of ordered pairs of tasks between which there
is a precedence relationship. Where task i must
precede task j, ði; jÞ 2 U . Dj þ tkjT  Y k  Mð1  xkjT Þ 8j 2 X; 8k 2 K; ð12Þ
Decision variables
xkij =1, if quay crane k performs task j immediately after rk  Dj þ tk0j þ pj  Mð1  xk0j Þ 8j 2 X; 8k 2 K; ð13Þ
performing task i = 0, otherwise.
Yk The completion time of quay crane k. xkij 2 f0; 1g 8i; j 2 A; 8 k 2 K; where A # V  V and V
Di The completion time of task i.
¼ X [ f0; Tg; ð14Þ
zij =1, if task j starts later than the completion time of task
i = 0, otherwise.
W The complete the final task assigned. zij 2 f0; 1g 8i; j 2 X; ð15Þ

Y k ; Di  0 8i 2 X; 8k 2 K; ð16Þ
Constraints (2) ensure the makespan is equal to the largest comple-
The problem is to determine how to assign crane to serve the
tion time among the quay cranes. Constraints (3) ensure each quay
task aiming to minimize the makespan, which is the largest com-
crane will be assigned with an initial task. Constraints (4) ensure each
pletion time among all the quay cranes. The ship can only depart
quay crane will be assigned with a final task. Constraints (5) ensure
after all quay cranes finish the jobs assigned to them. The objective
each task will be completed by one quay crane only. Constraints (6)
function is shown as in Eq. (1). The problem formulation is similar
ensure the flow balance for each task in each crane so that the task
to the one modeled by Kim and Park (2004) and as shown in the
is performed in well-defined sequence. Constraints (7) ensure the
followings.
" # completion time of task j should be larger than the task i plus its
X processing time and travelling time in each crane. Constraints (8) en-
Objectiv e Z : MIN a1 W þ a2 Yk ð1Þ sure task i should be completed before task j. Constraints (9) ensure
k
that the completion time of task j should be larger than the task i plus
The components include the makespan and the sum of the quay its processing time. Constraints (10) ensure the flow balance for each
cranes completion time. As commonly adopted by researchers, a1 task and each crane. Constraints (11) ensure to avoid the interference
is generally set to be much larger than a2, meaning the makespan among quay cranes. Constraints (12) ensure the completion time of
4216 S.H. Chung, K.L. Choy / Expert Systems with Applications 39 (2012) 4213–4221

Generation (G) = 0 Roulette Wheel Selection G=G+1

Formation of Initial Pool Crossover Operation


G= Output
Termination Best Solution
Chromosome Validation Mutation Operation

END
Fitness Evaluation Chromosome Validation

Insert Best Solution


Fitness Evaluation Recorded

Fig. 1. Outline of the proposed modified genetic algorithm.

Encoding of Chromosome

Segment 1 Segment 2
4 10 5 1 6 7 3 9 8 2 1 2 2 1 1 2 1 1 2 2

Task Sequence Crane assignment

Decoding of Chromosome

Quay Crane Sequence


1 4, 1, 6, 3, 9
2 10, 5, 7, 8, 2
Fig. 2. Sample of encoding of chromosome.

each crane is equal to completion time the final task plus the traveling is randomly generated and cannot duplicate with other genes. In
time to the final location. Constraints (13) ensure the completion the second segment, each gene represents a quay crane being as-
time of the second task is larger than the completion time of the ini- signed to perform the task which is the corresponding position in
tial task plus its processing time and travelling time in each quay the first segment. The value of the gene is randomly generated in
crane. Constraints (14) and (15) define the variables to be the binary this part based on the number of quay cranes assigned. For the
number. Constraints (16) define the non-negativity constraint. An example in Fig. 2, there are 10 tasks being assigned on 2 quay
excellent explanation of the constraints can be found in Kim and Park cranes. One of the quay cranes is assigned with 5 tasks in the se-
(2004) and Moccia et al. (2006). quence of tasks 4, 1, 6, 3, and 9.

4.2. Validation of chromosome


4. Methodology
As stated by Kim and Park (2004), precedence constraints and
This paper proposed a modified genetic algorithm to solve the
non-interference constraints must not be violated. For precedence
QCSP as outline in Fig. 1.
constraint, some of the tasks cannot be performed before others
due to the operation or the position of the task (located in hold or
4.1. Encoding and decoding of chromosome on deck). For non-interference constraint, quay cranes cannot per-
form the task when the route between the quay crane and the loca-
The chromosome is designed in permutation encoding (real tion of target task is blocked by another quay crane(s). Because of
number) as shown in Fig. 2. Each chromosome represents a possi- these constraints, a chromosome may become invalid. Therefore,
ble solution for the assignment of the quay cranes and the se- modification will be carried out to make the chromosome valid.
quence of the tasks on each quay cranes. Each chromosome
consists of 2|X| number of genes, and consists of 2 segments with 4.2.1. Precedence constraint
different encoding schemes. In the first segment, each gene repre- Discharging operation must precede the loading operation in
sents a task and the sequence of gene represents the sequence of the same ship bay. Containers on deck must be discharged before
the task performed from the left to the right. The value of the gene those in hold. Oppositely, the containers in hold must be loaded
S.H. Chung, K.L. Choy / Expert Systems with Applications 39 (2012) 4213–4221 4217

Task Bay Operation


Before Validation 1 2 Loading
2 2 Discharging

Segment 1 Segment 2

1 10 5 4 6 7 3 9 8 2 1 2 2 1 1 2 1 1 2 2

After Validation
Segment 1 Segment 2
4 10 5 1 6 7 3 9 8 2 1 2 2 1 1 2 1 1 2 2

Fig. 3. Sample of the chromosome validation.

Table 1
Relationship between quay crane and location of task. from other scheduling problem is that the quay cranes travel on
Number of quay crane = 2 Location of task Possible quay crane assigned the same truck. As a result, some of the quay cranes cannot per-
Ship bay 1 QC 1
form those tasks located in the end of the ship bay. Table 1 shows
Ship bay 10 QC2 the relationship between quay crane and location of task. In the
Number of quay crane = 3 Location of task Possible quay crane assigned case of 2 quay cranes, the task in the first ship bay can only be per-
Ship bay 1 QC 1 formed by the first quay crane, where the task in the last ship bay
Ship bay 2 QC 1 & 2 can only be performed by the last quay crane, while in the case of 3
Ship bay 9 QC2 & 3 quay cranes, the task located in the ship bay 2 can be performed by
Ship bay 10 QC3
Quay Cranes 1 and 2 only but not all 3 cranes, as shown in Fig. 4.
For example in the case of 2 quay cranes as in Fig. 5, Task 2 located
before those on deck in the loading operation. Therefore, the se- in Bay 10 should not be performed by Quay Crane 1.
quence of the genes must follow this constraint. Each gene in the
first segment of the chromosome must be checked to identify the
4.3. Evaluation of fitness value
position and the operation of the tasks. When there are two tasks
are located in the same ship bay, then the operation of the tasks is
The objective is to minimize the makespan and the total com-
compared first and the position of the task in the ship bay is
pletion time of all the cranes. Thus, the fitness value of chromo-
compared later.
some will be calculated by Eq. (17).
If a chromosome is invalid as shown in Fig. 3, swapping will be
carried out to correct the sequence of tasks. For example, Task 4 " #
X
should precede Task 1. In the second segment, the genes which Fitness Value ¼ 1= a1 W þ a2 Yk ð17Þ
are in the corresponding position must also swap to the position k
referred to the first segment, since those genes represent the num-
ber of quay crane assigned to the tasks. It is essential to keep the
consistency with the original chromosome. 4.4. Genetic Operations

4.2.2. Non-interference constraint In the proposed modified genetic algorithm, the initial pool will
The non-interference constraint restricts the movement of the be randomly generated. In addition, for the selection operator, the
quay cranes. The distinguishing characteristic of QCSP which differ widely used Roulette wheel selection will be adopted.

Fig. 4. The accessibility of the quay cranes.


4218 S.H. Chung, K.L. Choy / Expert Systems with Applications 39 (2012) 4213–4221

Before Validation Task: 2


Bay: 10
Crane Assigned: 1

Segment 1 Segment 2
4 10 5 1 6 7 3 9 8 2 1 2 2 1 1 2 1 1 2 1

After Validation

Segment 1 Segment 2

4 10 5 1 6 7 3 9 8 2 1 2 2 1 1 2 1 1 2 2

Fig. 5. The example of the chromosome validation in second part of chromosome.

4.4.1. Crossover operation Step 5: Repeat step 1–4 to produce the second offspring.
In this paper, the sequence of the genes in the first part of chro-
mosome represents the sequence of the tasks to be performed. The 4.4.2. Mutation Operation
ordered form of chromosome will reproduce illegal offspring. For The purpose of mutation is to prevent the population trapped in
this reason, some of the gene values may be missed and some of the local optimal and enable the genetic search in new solution
them may be duplicated in the offspring. To avoid the duplication space by changing one or more gene(s). In the proposed algorithm,
or loss of gene value, order crossover proposed by Gen and Cheng since the chromosome is divided into two segments, the mutation
(1996) is applied and repairing procedure is essential to prevent methods used are different. In the first segment, two genes in the
the illegitimacy of offspring as shown in the following steps. same chromosome are randomly selected and then swap each other
Fig. 6 shows a sample of crossover. because of the prevention of duplicating gene values. The order form
of chromosome can be retained as shown in Fig. 7. Its purpose is to
Step 1: Select the segment randomly in the first part of chromo- change the sequence of gene and the sequence of the tasks.
some in one parent. In the second segment, a gene is randomly selected and mu-
Step 2: Select the corresponding segment in the second part of tated. As the selected gene represents the particular quay crane
chromosome. being assigned to perform a certain task, the value changed will
Step 3: Copy the segment into the corresponding positions in the be based on the total number of quay crane assigned, ranging from
offspring. 1 to K, as shown in Fig. 8. Its purpose is to ensure the task can be
Step 4: Delete the genes which exist in the segment of the off- performed by different quay crane after reproducing the popula-
spring from the second parent. The remained genes are tion of new offspring. Its significance is to prevent trapping in
then copied to the offspring from left to right according the local optimal, increasing the diversity of the solution popula-
to the sequence in the second parent. tion by assigning tasks to new crane.

Fig. 6. Sample of crossover operation.


S.H. Chung, K.L. Choy / Expert Systems with Applications 39 (2012) 4213–4221 4219

Before Mutation Table 4


Summary of the best results obtained in Instances 23–32.

9 2 5 8 3 10 1 4 7 6 Problem no. (Proposed) (2007) (2006) (2004) (2004)


GA TS B&C GRASP B&B
After Mutation 23 576 582 576 591 576
24 669 669 666 675 669
25 744 741 738 741 738
9 2 7 8 3 10 1 4 5 6 26 645 639 639 651 639
27 660 657 660 687 657
Fig. 7. Sample of mutation in Segment 1 of a chromosome. 28 531 531 531 549 537
29 810 810 807 819 807
30 897 891 891 906 891
31 570 570 570 570 570
32 594 591 591 597 591
Before Mutation
Problem no. Difference between GA and different approaches

1 2 1 1 1 2 1 2 2 1 TS B&C GRASP B&B


23 1.03% 0.00% 2.54% 0.00%
24 0.00% 0.45% 0.89% 0.00%
After Mutation 25 0.40% 0.81% 0.40% 0.81%
26 0.94% 0.94% 0.92% 0.94%
1 2 2 1 1 2 1 2 2 1 27 0.46% 0.00% 3.93% 0.46%
28 0.00% 0.00% 3.28% 1.12%
Fig. 8. Sample of mutation in Segment 2 of a chromosome. 29 0.00% 0.37% 1.10% 0.37%
30 0.67% 0.67% 0.99% 0.67%
31 0.00% 0.00% 0.00% 0.00%
Table 2 32 0.51% 0.51% 0.50% 0.51%
Outlines of the instances.

Instance set Number of QCs Number of tasks


Table 5
13–22 2 10
Summary of the best results obtained in Instances 33–42.
23–32 2 15
33–42 3 20 Problem no. (Proposed) (2007) (2006) (2004) (2004)
43–49 3 25 GA TS B&C GRASP B&B
33 603 603 603 666 603
34 717 735 717 762 717
35 690 690 684 699 678
Table 3 36 636 681 678 708 678
Summary of the best results obtained in Instances 13–22. 37 522 519 510 540 510
38 618 618 618 660 609
Problem no. (Proposed) (2007) (2006) (2004) (2004)
39 519 519 513 579 513
GA TS B&C GRASP B&B
40 567 567 564 597 561
13 453 453 453 453 453 41 588 594 588 642 585
14 546 546 546 546 546 42 576 576 570 666 561
15 513 513 513 516 513 Problem no. Difference between GA and different approaches
16 312 312 312 321 321
17 453 453 453 456 456 TS B&C GRASP B&B
18 375 375 375 375 375 33 0.00% 0.00% 9.46% 0.00%
19 543 543 543 552 552 34 2.45% 0.00% 5.91% 0.00%
20 399 399 399 480 480 35 0.00% 0.88% 1.29% 1.77%
21 465 465 465 465 465 36 6.61% 6.19% 10.17% 6.19%
22 537 537 537 720 720 37 0.58% 2.35% 3.33% 2.35%
Problem no. Difference between GA and different approaches 38 0.00% 0.00% 6.36% 1.48%
39 0.00% 1.17% 10.36% 1.17%
TS B&C GRASP B&B
40 0.00% 0.53% 5.03% 1.07%
13 0.00% 0.00% 0.00% 0.00% 41 1.01% 0.00% 8.41% 0.51%
14 0.00% 0.00% 0.00% 0.00% 42 0.00% 1.05% 13.51% 2.67%
15 0.00% 0.00% 0.58% 0.00%
16 0.00% 0.00% 2.80% 2.80%
17 0.00% 0.00% 0.66% 0.66%
18 0.00% 0.00% 0.00% 0.00%
19 0.00% 0.00% 1.63% 1.63%
20 0.00% 0.00% 16.88% 16.88% Procedure (GRASP), later on Moccia et al. (2006) using branch-and-
21 0.00% 0.00% 0.00% 0.00% cut (B&C), and Sammarra et al. (2007) using Tabu search (TS). The
22 0.00% 0.00% 25.42% 25.42% genetic parameters used in the proposed algorithm are: population
size = 50 individuals, crossover rate = 0.9 and the mutation
rate = 0.05. The number of evolution is 30,000. The algorithm is
5. Numerical examples programed by using Java Language and performed on an Intel Core
2 Quad 2.0 GHz PC with 2 GB RAM.
In order to test the optimization reliability of the proposed ge- The optimization results are summary as shown in Tables 3–6.
netic algorithm, a set of well known benchmarking data obtained Table 3 summarizes the results for Instances 13–22, regarded as
from Kim and Park (2004) will be used. The data set consists of the small size problem set with 2 quay cranes and 10 tasks. The re-
43 instances as outlined in Table 2. It has been used by many fa- sults show that the proposed GA can obtain the same best known
mous researchers including Kim and Park (2004) originally using solutions as obtained by other existing optimization approaches in
branch-and-bound (B&B) and Greedy Randomized Adaptive Search all the instances as the ones obtained by TS and B&C.
4220 S.H. Chung, K.L. Choy / Expert Systems with Applications 39 (2012) 4213–4221

Table 6 performance of GA deteriorates in the largest size of problem set,


Summary of the best results obtained in Instances 43–49. the computational time of GA is more attractive and practical.
Problem no. (Proposed) (2007) (2006) (2004) (2004) Table 7 shows the computational time required by B&C (Moccia
GA TS B&C GRASP B&B et al., 2006) and TS (Sammarra et al., 2007) which computed by
43 897 879 897 942 864 the 2.5 GHz P4 machine equipped with 1 GB of RAM. Under going
44 855 834 822 858 813 30,000 evolutions for all the problem sets, the computational
45 864 852 840 873 825 time(s) are within 2 min. While TS required from the shortest
46 723 690 690 735 687
47 819 792 792 807 789
1.52 min to the longest 48.29 min for 1,000 iterations. For B&C,
48 663 663 645 669 627 Moccia et al. (2006) limited the computational time to be within
49 915 912 927 972 888 120 min. It is found that the computational time increases sharply
Problem no. Difference between GA and different approaches when the problem size increases from the shortest 0.97 min to the
longest 120 min. Therefore, GA is advantaged with respect to both
TS B&C GRASP B&B
heuristic approaches and optimization approaches in terms of
43 2.05% 0.00% 4.78% 3.82%
computational time.
44 2.52% 4.01% 0.35% 5.17%
45 1.41% 2.86% 1.03% 4.73%
46 4.78% 4.78% 1.63% 5.24%
6. Conclusion
47 3.41% 3.41% 1.49% 3.80%
48 0.00% 2.79% 0.90% 5.74%
49 0.33% 1.29% 5.86% 3.04% QCSP is one of the core operations in container terminal, con-
sisting of the sequence of the discharging and loading operations
of container and the assignment of quay cranes. In the past two
Table 7 decades, the sharp increase in the container throughput establishes
Computational time of GA, TS and B&C. the indispensable role of maritime transportation in the global
trade. Most of the container terminal starts focusing not only the
Instances Computational Time (mins)
customer service, but also its productivity. Million dollars has been
(Proposed) (2007) (2006) invested in each terminal to improve the system in order to inte-
GA TS B&C
grate the operations, assist in decision making and achieve the
13–22 0.52 1.52 0.97 shorter turnaround time of the ships. This paper proposed a genetic
23–32 0.75 5.86 8.87
33–42 1.18 21.34 72.17
algorithm to deal with the QCSP. The design and the implementa-
43–49 1.58 48.29 120 tion of algorithm have been developed and described. The optimi-
zation reliability of proposed GA was test by solving a set of well
known benchmarking problems. By comparing the results ob-
Table 4 summarizes the results for Instances 23–32 with 2 quay tained by the proposed GA and the ones obtained by other well
cranes and 15 tasks, regarded as medium size problem set. The re- known existing approaches included GRASP, Tabu Search, B&B
sults show that the proposed GA obtained the same best known re- method and B&C method, it is found that the proposed GA outper-
sults found by other approaches, in Instances 23, 27, 28 and 31. formed in small sized instances and obtain better results in some
Compare with other heuristic approaches, the performance of GA medium sized instances as well. Besides, the computational time
is better than GRASP about 0.50%–3.93% and the performance of required by the proposed GA is much faster than the existing
GA is quite similar to the TS. On the side of comparison with opti- approaches.
mization approaches, the differences between GA and the optimi-
zation approaches B&B and B&C are not more than 1%. References
Table 5 summarizes the results for the Instances 33–42 with 3
Bierwirth, C., & Meisel, F. (2009). A fast heuristic for quay crane scheduling with
quay cranes and 20 tasks, regarded as the second large problem
interference constraints. Journal of Scheduling, 12, 345–360.
set. The results show that in Instances 35, 37–42, the results ob- Bierwirth, C., & Meisel, F. (2010). A survey of berth allocation and quay crane
tained by the proposed GA are not as good as the best known ones scheduling problems in container terminals. European Journal of Operational
by 1.77%, 2.35%, 1.48%, 1.17%, 1.07%, 0.51% and 2.67% respectively. Research, 202, 615–627.
Daganzo, C. F. (1989). The crane scheduling problem. Transportation Research Part B,
However, the proposed GA obtained the best known results in In- 23, 159–175.
stances 33 and 34. In addition, it obtains a better result than the Daganzo, C. F. (1990). The productivity of multipurpose seaport terminals. Transport
best known one in Instance 36 by 6.19%. Science, 24, 205–216.
Daganzo, C. F., & Peterkofsky, R. I. (1990). A branch and bound solution method for
Table 6 summarizes the results obtained for the last data set the crane scheduling problem. Transportation Research Part B, 24, 159–172.
with 3 quay cranes and 25 tasks, regarded as the largest problem Drewry Shipping Consultants Limited (2010), Available from http://
set for Instances 43–49. The results show that in general, the re- www.drewry.co.uk/publications/list_publications.php?market=26.
Gambardella, L. M., Rizzoli, A., & Zaffalon, M. (1998). Simulation and planning for
sults obtained by the proposed GA is close to the best known ones intermodal container terminal. Simulation, 71, 107–116.
in about 5%. Comparing the results with the other heuristic ap- Gen, M., & Cheng, R. (1996). Genetic algorithms and engineering design. New York:
proaches, the performance of GA is better than GRASP with the dif- John Wiley.
Guinet, A. (1993). Scheduling sequence-dependent jobs on identical parallel
ference up to 5.86% in 6 out of 7 instances. From instance 43-47,
machines to minimize completion time criteria. International Journal of
the performance of GA is inferior to TS around 1.41% to 4.78%, Production Research, 31(7), 1579–1594.
but only obtain similar results in Instances 48 and 49. Comparing Günther, H. O., & Kim, K. H. (2005). Container terminals and automated transport
systems: logistics control issues and quantitative decision support. Springer.
with optimization approaches, the performance of GA deteriorates
Kim, K. H., & Kim, K. Y. (1999). An optimal routing algorithm for a transfer crane in
in this set of instances. port container terminals. Transportation Science, 33(1), 17–33.
In overall view, the proposed GA can obtain the reasonable Kim, K. H., & Park, Y. M. (2004). A crane scheduling method for port container
quality of solution, and it especially performs well in the small size terminals. European Journal of Operational Research, 156, 752–768.
Lee, Y. H., & Pinedo, M. (1997). Scheduling jobs on parallel machines with sequence-
problem set such as in Instance 13–22. As the problem become dependent setup times. European Journal of Operational Research, 100(3),
more complex, the performance of GA can still work well by com- 464–474.
paring with other heuristic approaches, resulting from GA do not Lee, D. H., Wang, H. Q., & Miao, L. (2006). Quay crane scheduling with non-
interference constraints in port container terminals. Transportation Research
exceed the best known solutions by more than 6%. Although the Part E, 44, 124–135.
S.H. Chung, K.L. Choy / Expert Systems with Applications 39 (2012) 4213–4221 4221

Lim, A., Rodrigues, B., Xiao, F., & Zhu, Y. (2004). Crane scheduling with spatial Steenken, D., Vos, S., & Stahlbock, R. (2004). Container terminal operation and
constraints. Naval Research Logistics, 51, 386–406. operations research–a classification and literature review. OR Spectrum, 26,
Meisel, F., & Bierwirth, C. (2006). Integration of berth allocation and crane 3–49.
assignment to improve the resource utilization at a seaport container Vacca, I., Bierlaire, M. & Salani, M. (2007). Optimization at container terminals:
terminal. In Operations Research Proceedings 2005 (pp. 105–110). Springer. Status, trends and perspectives. In Swiss Transport Research Conference 7th (pp.
Moccia, L., Cordeau, J. F., Gaudioso, M., & Laporte, G. (2006). A branch-and-cut 1–21).
algorithm for the quay crane scheduling problem in a container terminal. Naval Wang, Y., Chen, Y., & Wang, K. (2009). A case study of genetic algorithms for quay
Research Logistics, 53, 45–59. crane scheduling. Opportunities and Challenges, SCI, 214, 119–125.
Park, Y. M., & Kim, K. H. (2003). A scheduling method for berth and quay cranes. OR Zaffalon, M., Rizzoli, A. E., Gambardella, L. M., & Mastrolilli, M. (1998). Resource
Spectrum, 25, 1–23. allocation and scheduling of operation in an intermodal terminal. European
Sammarra, M., Cordeau, J. F., Laporte, G., & Monaco, M. F. (2007). A tabu search Simulation Symposium and Exhibition, Simulation in Industry, 10, 520–528.
heuristic for the quay crane scheduling problem. Journal of Scheduling, 10,
327–336.

You might also like