Advanced Engineering Informatics 30 (2016) 390–405

Contents lists available at ScienceDirect

Advanced Engineering Informatics

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

Full length article

Berth allocation and quay crane assignment in a container terminal for

the trade-off between time-saving and energy-saving
Junliang He ⇑
Engineering Research Center of Container Supply Chain Technology, Ministry of Education, Shanghai Maritime University, Shanghai 201306, PR China

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

Article history: This paper addresses the problem of integrated berth allocation and quay crane (QC) assignment for the
Received 9 October 2015 trade-off between time-saving and energy-saving. This problem is formulated as a mixed integer
Received in revised form 15 April 2016 programming model (MIP), in order to minimize the total departure delay of all vessels and the total
Accepted 19 April 2016
handling energy consumption of all vessels by QCs. Furthermore, an integrated simulation and
Available online 31 May 2016
optimization method is developed, where the simulation is designed for evaluation and optimization
algorithm is designed for searching solution space. Finally, numerical experiments are conducted to
Keywords:
verify the effectiveness of the proposed method.
Container terminal
Berth allocation
Ó 2016 Elsevier Ltd. All rights reserved.
Quay crane assignment
Mixed integer programming
Energy consumption

1. Introduction allocation, and the scheduling of handling equipment. We can save

energy by using clean energy or energy-saving devices. However,
The competition among container terminals has been getting due to high cost of the energy-saving devices, this paper focuses
fiercer and fiercer. Almost all container terminals are bearing the on how to realize energy-saving at the operational level without
pressure of attracting more customers. Moreover, with the increase additional equipment investment.
of sizes of container vessels, container terminals are encountering As the most important space resource and handling equipment,
another challenge, i.e., the rapid handling of containers for mega- berth and quay crane (QC) play an important role in maintaining
vessels. Thus, container terminals must shorten the vessel turn- the service level and controlling energy consumption of container
around time, which is an impact factor of the improvement of their terminals. In particular, the integrated berth allocation and QC
service level. However, due to high cost of the handling equipment assignment problem (B&QCAP) secure a crucial position in the
and scarcity of land resources, container terminals can hardly pur- operation of container terminal. An effective berth allocation and
chase additional handling equipment or expand their territory to QC assignment can significantly shorten the total vessels’ turn-
promote the efficiency and service level. Therefore, the reasonable around time and save total energy consumption of all vessels.
space resource allocation and scheduling of handling equipment However, most studies on port operations solely aim to improve
are critical to enhancing the service level of container terminals. the efficiency of container terminals, and do not consider energy-
Green port recently has been becoming the mainstream in the saving. Therefore, it’s imperative to develop an effective approach
sustainable development of global economy. Therefore, energy- for the B&QCAP by considering the trade-off between efficiency
saving is another goal of container terminals. We should save and energy consumption.
energy of container terminals as much as possible, but it cannot This paper is organized as follows. Relevant literature is
be implemented at the expense of the cost of the service level. reviewed in Section 2. In Section 3, the B&QCAP is described. In
All container terminals must seek appropriate approaches to save Section 4, the relationships among energy consumption, QC hours
energy without reducing the throughput or hampering the service demand, berth allocation and QC assignment are firstly analyzed,
level. In this respect, the main energy consumption of container and then, a mixed integer programming model (MIP) is formulated
terminals is significantly impacted by berth allocation, yard for the B&QCAP. An integrated simulation and optimization
method is proposed for solving the MIP in Section 5. Numerical
experiments are conducted in Section 6 to evaluate the effective-
⇑ Address: 1550 Haigang Avenue, Lingang New Port City, Logistics Engineering
School, Shanghai Maritime University, Shanghai 201306, PR China.
ness of the proposed solution methods. Conclusions and future
E-mail address: [email protected] research are given in the last section.

http://dx.doi.org/10.1016/j.aei.2016.04.006
1474-0346/Ó 2016 Elsevier Ltd. All rights reserved.
 J. He / Advanced Engineering Informatics 30 (2016) 390–405 391

2. Literature review as decision variables instead of previously-known parameters

when formulating the berth allocation problem in their works.
Up to present, intensive research were attempted in the area of Golias et al. [21] tried to reduce fuel consumption and vessel emis-
port operations, such as the berth allocation, storage space alloca- sions by minimizing the total waiting time of vessels, based on the
tion, QC assignment and internal truck scheduling. Also the litera- assumption that the shorter the waiting time is, the less the fuel
ture of case reports were almost exploding [1,2]. For a consumption and vessel emissions. Lang and Veenstra [22] pro-
comprehensive overview on B&QCAP, see the review work given vided a direct quantitative analysis on fuel consumption of a vessel
by [3,4]. However, most of related works have only referred to and minimized the fuel consumption for sailing with a customized
the handling efficiency of container terminals. Only a few studies simulation tool. The optimization approaches of vessel sailing
addressed the energy-saving of container terminals at the opera- speed for minimizing the fuel consumption were proposed by Du
tional level, not to speak of the modeling the trade-off between et al. [23] and Wang et al. [24]. Furthermore, Du et al. [23] con-
handling efficiency and energy consumption. In this section, only ducted the vessel emission (in sailing periods) calculation with
literature review of studies directly related to B&QCAP and the widely-used emission factors. Using policies and sailing speed
energy-saving of container terminals at operational level is optimization of vessels to improve harbor congestion and marine
conducted. fuel consumption were addressed by Alvarez et al. [25]. Aiming
For B&QCAP, there are many related works. In order to consider at the energy-saving of terminals, Chang et al. [26] developed a
handling priorities and preferred berth for vessels, the B&QCAP MIP model for B&QCAP considering total energy consumption of
was extended and priorities were addressed by adding penalty all vessels. To reduce emissions from idling truck engines at con-
costs for violating a vessel’s planning departure time ([5–7]. Con- tainer terminals, Chen et al. [27] developed a bi-objective model
sidering interference among QCs, marginal productivity of QCs to minimizing both truck waiting times and truck arrival pattern
was modeled and was included in the B&QCAP model by Meisel change. He et al. [28] addressed the problem of sharing internal
and Bierwirth [8]. Cost structure of the B&QCAP was also proposed trucks among multiple container terminals considering transporta-
by them. To solve the problem a construction heuristic, local tion energy consumption. In order to reduce energy consumption
refinement procedures, and two meta-heuristics were developed of yard crane, He et al. [29] revealed negatively correlated relation-
by them. Furthermore, the model of Meisel and Bierwirth [8] has ship between the energy-saving and high efficiency of yard crane
been extended considering periodic balancing utilization of QCs scheduling problem, and provided an appropriate approach for bal-
by Hu [9], and the rolling-horizon heuristics; neighborhood search ancing the two objectives. Similarly, the integrated handling
heuristics and parallel computing were proposed for obtaining the equipment scheduling model for minimizing energy consumption
optimize solution by him. Zhang et al. [10] addressed B&QCAP con- was formulated by He et al. [30].
sidering the coverage ranges of quay cranes and allows for limited Overall, B&QCAP is a well researched domain. However, to the
adjustments of quay cranes during loading and discharging, and best of our knowledge, the efficient model and method at the oper-
proposed a sub-gradient optimization technique to solve this prob- ational level for the trade-off between improving efficiency and
lem. Yang et al. [11] developed an evolutionary algorithm with energy-saving in B&QCAP are lacking. Our study thus focuses on
nested loops for B&QCAP, where two inner loops are proposed the new and practical research topic.
for resolving berth allocation and quay crane assignment respec-
tively, and an outer loop is used to search approximate solution.
Regarding the number of cranes assigned to a vessel during the 3. Problem description
time the vessel is at the berth, there are two main policies: time-
invariant and time-variant QC assignment. For time-variant QC B&QCAP includes three interrelated aspects of berthing position,
assignment policy, the assumption that the number of cranes berthing time and the number of cranes to be allocated to each ves-
may change dynamically between the minimum and maximum sel during a given planning horizon. Fig. 1 illustrates B&QCAP prob-
number of QCs were proposed in many related works [7,12,13]. lem in the two-dimensional space. The horizontal axis represents
Almost all of them suggested that time-variant QC assignment the berthing position along the quayside, while the vertical one rep-
can promote the total handling efficiency. Time-invariant QC resents the berthing time. Each rectangle represents a vessel to be
assignment policy also was considered in some related works served. The height of a rectangle denotes the handling time of the
[14,15]. Most of them argued that time-invariant QC assignment corresponding vessel, while the length of a rectangle represents
policy is too restrictive. However, this policy can either reduce the vessel length. The horizontal and vertical coordinates of the
large setup losses due to reallocation of quay cranes, or reduce lower left corner of a rectangle represent the berthing position
computational complexity. The comparison between the two poli- and the berthing time of the corresponding vessel, respectively.
cies was addressed by Iris et al. [16]. Their conclusion is that time- The berth allocation problem and the QC assignment problem
invariant QC assignment policy may result in an additional cost. are strongly related. The turnaround time of a vessel at berth
Similarly, Rodriguez-Molins et al. [17] presented the time- depends on the number of QCs allocated to a vessel, and the QC
invariant QC assignment policy and the time-variant QC assign- assignment for the vessel depends on its berthing position. There-
ment policy for modeling QCAP in the B&QCAP, and proposed a fore, the two problems should be integrated. In general, the pri-
GRASP-based meta-heuristic for solving this problem. Since ves- mary objective of B&QCAP is the minimization of the total
sel’s arrival and handling time is uncertain, some studies addressed departure delay of all vessels in a planning horizon. With the
the B&QCAP under uncertainties and considered the robustness of embracing of green port concept, almost all container terminals
model and algorithm [18–20]. seek energy-saving and emission reduction. Therefore, the mini-
In energy-saving of container terminals at operational level, mization of total handling energy consumption of QCs for all ves-
only a few literatures can be found. Most of works are related to sels in a planning horizon is another objective of B&QCAP.
the energy-saving of vessels in sailing period. In order to minimize For berth allocation, all vessels should be arranged for available
the fuel consumption of vessels, Golias et al. [21] and Lang and berths without any collisions. Both berthing position and berthing
Veenstra [22] proposed that it can be realized by the way of poten- time are two important decision factors. If they are reasonably
tial coordination opportunity between terminal operators and planned, more vessels can be arranged for available berths in
shipping line. Therefore, the arrival time of vessels was regarded the same time period, and the total departure delay can also be
392 J. He / Advanced Engineering Informatics 30 (2016) 390–405

Time

V6, B6', T6'

V5, B5, T5

V6, B6, T6
Planning period

E2'
V4, B4', T4 V4,, B4, T4
V
V4 T V3, B3, T3

E2

T4
V2, B2, T2

V1, B1, T1
T2

Berth
B4' B4

Fig. 1. Illustration of B&QCAP.

shortened. As shown in Fig. 1, if the berthing position and time of position is specified for each vessel, which is closest to the yard
Vessel 4 respectively are B4 and T4, Vessel 6 only can be berthed areas allocated to all containers of the vessel. If the real berthing
after Vessel 4 departs. However, if Vessel 4 is located at B4’ at position of the vessel deviates from the expected best position,
T4, Vessel 6 can be berthed immediately after arriving. It is con- the total horizontal transportation distance for all containers of
cluded the departure delay using the latter schedule is less than this vessel between yard and berth position increases. Thus the idle
that using the former schedule. time of QC’s waiting for trucks is extended, and the total handling
QC assignment is an effective method to shorten the departure efficiency of QCs assigned to the vessel decreases. Therefore, we
delay of each vessel and reduce energy consumption of QCs for should analysis the relationship between the handling efficiency
each vessel. In general, the QC’s energy consumption of a vessel of QC and the berthing position deviation.
is related to the number of QCs assigned to the vessel. Since the Meisel and Bierwirth [8] proposed a berth deviation factor b to
handling process of a QC consists of working and non-working calculate the relative increase of QC hours demand per unit of
time (such as waiting time), the QC’s energy consumption of a ves- berthing deviation. But it is difficult to determine the deviation fac-
sel also consists of working and non-working energy consumption. tor b. Therefore, we analyze the quantitative relationship between
Generally, due to QCs’ interference and internal truck congestion in a vessel’s real QC hours demand and berthing deviation according
the yard, the more QCs are assigned to a vessel, the greater the to plenty of historical data. In this study it is proposed that the
non-working time of the vessel is experiencing, and the greater berth is split into a number of segments. Let Db denote the number
the QC’s energy consumption of the vessel is suffering. Thereby, of berth deviation segments of a vessel. If the length of each berth
we should determine the most suitable number of QCs assigned segment is too long, the quantitative relationship between a ves-
to each vessel. As shown in Fig. 1, there are two QC assignments sel’s real QC hours demand and berthing deviation cannot be
for Vessel 2: (1) assigning more QCs and Vessel 2 can depart at reflect. It is found that a vessel’s relative increase of QC hours
E2; (2) assigning less QCs and Vessel 2 can depart at E2’. It is appar- demand is very small if its berth deviation is less than 20 m. There-
ent that the former assignment can make Vessel 2 depart earlier fore, the length of each berth segment is set as 20 m for the quan-
than the latter assignment, but the QC’s energy consumption using titative relationship between a vessel’s real QC hours demand and
the former assignment more than that using the latter assignment. berthing deviation. In our study, the vessel’s QC hours demand is
Therefore, container terminals should select the former QC assign- measured as a number of QC-hours.
ment if only the time-saving objective is considered, or prefer the We acquire 30,000 historical data referring to the number of
latter assignment if it only considers the energy-saving objective. It berth deviation segments (Db), QC hours demand without berthing
is obvious that there may be conflict between time-saving and deviation (w), real QC hours demand with berthing deviation (wr).
energy-saving. Hence, this paper aims to propose an appropriate Let pc ¼ wrw
denote the proportion of QC real hours demand to QC
approach for the trade-off between the two objectives, not only hours demand without berthing deviation. Fig. 2 shows the rela-
considering one objective. tionship between pc and Db. From this figure it can be seen that
the proportion pc increases with the increase of the number of
4. Efficiency and energy consumption analysis berthing deviation segments, and the increasing velocity is getting
slower and slower. There is no linear correlation between a vessel’s
4.1. Efficiency analysis real QC hours demand and its berthing deviation segments. It is
difficult to directly acquire the quantitative relationship from his-
For B&QCAP, the real berthing position significantly affects the torical data. Nevertheless, the regression analysis is a statistical
handling efficiency of QCs. Generally, an expected best berthing process for estimating the relationships between a dependent
 J. He / Advanced Engineering Informatics 30 (2016) 390–405 393

Fig. 2. The relationship between vessels’ berthing deviations and their real QC hours demand.

variable and one or more independent variables. Therefore, we use the lighting energy consumption and other auxiliary equipment’s
regression analysis to fit the relationship curve between pc and Db. energy consumption, which is mainly determined by the QCs’ idle
The regression process is as follows. time, and the QCs’ idle time is directly related to the number of QCs
assigned to the vessel. Schonfeld and Sharafeldien [31] suggested
Step 1. Set b ¼ lnðDbÞ and j ¼ lnðpcÞ. that the work hours of n QCs assigned to a vessel was increased
Step 2. Estimate the relation function of j to b by linear regres- to na times, where a denotes interference factor and 0 < a 6 1.
sion. Let b independent variable and j is dependent variable. But it is difficult to determine the deviation factor a. Therefore,
We use the regression analysis tool of a commercial software we analyze the quantitative relationship between the idle time of
Tecnomatix Plant Simulation TM to obtain the regression equa- QCs assigned to a vessel and the number of QCs assigned to the
tion. The regression coefficient is 0.14285714, and the intercept vessel according to historical data, and then develop the vessel’s
is 0.0002. The intercept can be ignored. The coefficient of deter- non-working energy consumption model. There are approximately
mination R2 is 0.99835. Therefore, the linear regression equa- 30,000 historical data collected from different container terminals
tion is as follow. in China. As shown in Fig. 3, it is found that the proportion (r) of
non-working time to working time of a QC increases with increas-
j ¼ 0:14  b ð1Þ ing the number (n) of QCs assigned to a vessel. Moreover, the
2 increasing velocity is getting slower and slower. There is no linear
Since the coefficient of determination R is 0.99835, the linear
correlation between r and n. Therefore, we use regression analysis
regression equation is credible. Accordingly, we can deduce the
to fit the relationship curve between r and n. The regression pro-
equation of the proportion of QC real hours demand to QC hours
cess is as follows.
demand without berthing deviation as follow.

pc ¼ Db
0:14
ð2Þ Step 1. Set d ¼ lnðrÞ and u ¼ n1 .
Step 2. Estimate the relation function of d to u by linear regres-
Therefore, a vessel’s real QC hours demand is formulated as sion. Let u independent variable and d is dependent variable.
wr ¼ w  Db
0:14
ð3Þ We also use the regression analysis tool of a commercial soft-
ware Tecnomatix Plant Simulation TM to obtain the regression
equation. The regression coefficient is 0.22713626, and the
4.2. Energy consumption analysis intercept is 0.00338315. The intercept can be ignored. The
coefficient of determination R2 is 0.99997. Therefore, the linear
For B&QCAP, the energy consumption of a vessel handled by regression equation is as follow. The linear regression equation
QCs consists of two parts: (a) working energy consumption; (b) is as follow.
non-working energy consumption.
Generally, the working energy consumption of a vessel is deter- d ¼ 0:227u ð5Þ
mined by the working energy consumption of a QC per move, total 2
Since the coefficient of determination R is 0.99997, the linear
handling volume of the vessel and handling volume of a QC per
regression equation is credible. Accordingly, we can deduce the
move. The working energy consumption equation of a vessel is for-
equation of the non-working time proportion as
mulated as
W
r ¼ 0:79641=n ð6Þ
Ew ¼ l  ð4Þ
p Thus, the proportion of non-working time to total handling time
0:79641=n
w of each QC assigned to a vessel is .
where E the working energy consumption of a vessel, l the work- 1þ0:79641=n

ing energy consumption of a QC per move (unit: kW h/move), W the Therefore, the non-working energy consumption equation of a
total handling volume of a vessel (unit: TEU), p the unit handling vessel is formulated as
volume of a QC per move (unit: TEU/move). n  tc  0:79641=n
The non-working energy consumption of a vessel handled by Enw ¼ n  ð7Þ
QCs mainly refers to the air conditioning energy consumption,
ð1 þ 0:79641=n Þ
394 J. He / Advanced Engineering Informatics 30 (2016) 390–405

(5) We assume that QCs can be assigned to different vessels at

the beginning of each planning period. This means that each
QC can take over each other’s workload and can interrupt its
service operation to one vessel and move along the railway
to serve another vessel. This assumption may cause more
setup time because more frequent changes of QC positions,
but brings into the system more flexibility.
(6) The working hours of each QC is the same.
(7) The working energy consumption of each QC per move is the
same.

5.3. Notation

Parameters
V The set of all vessels which have not been at berth at
Fig. 3. The relationship between the non-working time and the number of QCs
assigned to a vessel. the beginning of the planning horizon (indexed by i). |
V| = N
L The length of the quay (unit: m)
Q The number of QCs
where Enw the non-working energy consumption of a vessel, B The total number of berth segments
n the non-working energy consumption of a QC per hour G The set of all vessels which have been at berth at the
(unit: kW h/h), tc total handling time of QCs assigned to a vessel beginning of the planning horizon (indexed by i). |G|
(unit: h). =S
The total handling energy consumption equation of a vessel is NT The number of periods in the planning horizon
formulated as (indexed by t)
sp The time span of each period (unit: h)
W n  tc  0:79641=n The start time of the planning horizon
E¼l þn ð8Þ T bps
p ð1 þ 0:79641=n Þ li The length of Vessel i, i 2 V [ G (unit: m)
where E the total handling energy consumption of a vessel. T ba The arriving time of Vessel i, i 2 V [ G. If
i
i 2 G; T ba
i ¼ T
bps

5. Model formulation T bpd The planned departure time of Vessel i, i 2 V [ G

i
Wi The handling volume of Vessel i, i 2 V [ G (unit: TEU)
5.2. Modeling approach N bqmax The maximum number of QCs that can be assigned to
i
Vessel i at any time, i 2 V [ G
In most studies, the number of assigned QCs to each vessel is bai The left-most berthing position of Vessel i, i 2 G
kept unchanged. This may result in lower utilization of the QCs’ Pbi The expected best berthing position of Vessel i,
capacity. It is proposed in this study that the number of QCs i 2 V [ G. If i 2 G, P bi ¼ bai
assigned to each vessel should vary within the planning horizon. l The working energy consumption of a QC per move
As shown in Fig. 6, after Vessel B departs, the QCs assigned to it (unit: kW h/move)
can be assigned to Vessel A, only if the maximum number of QCs p The unit handling volume of a QC per move (unit:
assigned to the vessel simultaneously is not exceeded. However, TEU/move)
if the number of QCs assigned to each vessel dynamically varies, n The non-working energy consumption of a QC per
it is difficult to develop linear programming model. Thus, it is hour (unit: kW h/h)
proposed here that the planning horizon should be divided into s The handling efficiency of a QC (unit: h/move)
several time segments and the number of assigned QCs to each cn The handling volume of a QC per unit time when n
vessel dynamically varies at all period. For example, there are 10 QCs simultaneously serve the same vessel (unit:
QCs, and the numbers of QCs assigned to Vessels D, E and G are move/h)
3, 2 and 5 within Period T15 (see Fig. 4). Since Vessel E has Cpi The unit penalty cost for delayed departure time of
departed within Period T16, the numbers of QCs assigned to Vessel i, i 2 V [ G (unit: $/h)
Vessels D and G can be increased to 4 and 6. CE The unit energy cost (unit: $/kW h)
# Berthing safety factor
5.2. Assumptions M A large positive number

The model is formulated based on the following assumptions.

(1) The berthing position of each vessel is kept unchanged in its Since the proportion of working time to total handling time of each
1
entire handling process. QC assigned to a vessel is 1þ0:79641=n , Parameter cn handling volume of

(2) There are no physical or technical restrictions such as vessel a QC per unit time when n QCs simultaneously serve the same
draft and water depth. vessel, can be defined by the following equation.
(3) Each vessel has a maximum number of assigned QCs.
1
(4) Every vessel has a best-expected berthing position. cn ¼ =s ð9Þ
1 þ 0:79641=n
 J. He / Advanced Engineering Informatics 30 (2016) 390–405 395

hij þ uij þ hji þ uji 6 2; 8i; j 2 V [ G; i < j ð19Þ

Decision variables
T bb The berthing time of Vessel i hij þ hji 6 1; 8i; j 2 V [ G; i < j ð20Þ
i
P rb The left-most berthing position of Vessel i
i
Dbim Dbim ¼ 1, if the number of berth deviation segments of uij þ uji 6 1; 8i; j 2 V [ G; i < j ð21Þ
Vessel i is m; Dbim ¼ 0, otherwise & ’
v itn vitn = 1, if n QCs are assigned to Vessel i in Period t; X
B
Prb
i  Pi
pb

vitn = 0, otherwise Dbim  m P ; 8i 2 V [ G ð22Þ

m¼1
20
dit dit = 1, if at least one QC is assigned to Vessel i in Period
t; dit = 0, otherwise & ’
X
B
Ppb rb
T brd The departure time of Vessel i i  Pi
i Dbim  m P ; 8i 2 V [ G ð23Þ
hij hij = 1, if Vessel i is located in the left of Vessel j in the m¼1
20
2-dimensional berth-time plane as shown in Fig. 1;
hij = 0, otherwise X
B
Dbim 6 1; 8i 2 V [ G ð24Þ
uij uij = 1, if Vessel i is located below Vessel j in the m¼1
2-dimensional berth-time plane; uij = 0, otherwise
T X
X Q XB  
W i  m0:14
v itn  ðn  cn  spÞ P Dbim  ; 8i 2 V [ G ð25Þ
t¼1 n¼1 m¼1
p

5.4. Mathematical model

XX
Q

X   n  v itn 6 Q ; 8t ¼ 1; 2; . . . ; NT ð26Þ
Objective 1 : min f 1 ¼ max T brd
i  T bpd
i ; 0  Cpi ð10Þ i2V[G n¼1
i2V[G
0 !1 X
Q
v itn ¼ dit ; 8i 2 V [ G; t ¼ 1; 2; . . . ; NT
bqmax
X XNT NX
n  sp  0:79641=n A ð27Þ
@l  W i þ
i

Objective 2 : min f 2 ¼ v itn  n   CE n¼1

i2V[G
p t¼1 n¼1 ð1 þ 0:79641=n Þ
ð11Þ X
Q
n  v itn 6 Nbqmax
i ; 8i 2 V [ G; t ¼ 1; 2; . . . ; NT ð28Þ
S.T. n¼1

i P T i ; 8i 2 V
T bb ba
ð12Þ
i ; 8i 2 V [ G; 8t ¼ 1; 2; . . . ; NT
dit  ðt  1Þ  sp 6 T brd ð29Þ
T bb
i ¼T bps
; 8i 2 G ð13Þ
i ; 8i 2 V [ G; 8t ¼ 1; 2; . . . ; NT
dit  t  sp P T bb ð30Þ
Prb
i þ ð1 þ #Þ  li < L; 8i 2 V [ G ð14Þ
Prb bb brd
i ; Ti ; Ti P 0; 8i 2 V [ G ð31Þ
Prb ¼ bai ; 8i 2 G ð15Þ
hij ; uij ; v itn ; dit ; Dbim 2 f0; 1g; 8i 2 V [ G; i–j; t
i

T brd
i 6 T bb
j þ M  ð1  uij Þ; 8i; j 2 V [ G; i–j ð16Þ ¼ 1; 2; . . . ; T; m ¼ 1; 2; . . . ; B; n
¼ 1; 2; . . . ; Q ð32Þ
i þ ð1 þ #Þ  li 6 P j þ M  ð1  hij Þ; 8i; j 2 V [ G; i–j
Prb rb
ð17Þ
The objective function (10) of the model is to minimize the cost
of total departure delay of all vessels, i.e., the sum of differences
hij þ uij þ hji þ uji P 1; 8i; j 2 V [ G; i < j ð18Þ
between the real departure times and the planned departure times

Time

T19 6
T18 6
T17 4 6
T16 4 6
T15 3 2 G 5
D 3 2 5
3 2 5
3 2 5
Planning period

E
4 4
4 3
4 F 3
5 2 3
5 3
5
A 3 4 3
3 4 3
T3 3 B 4 C 3
T2 3 4 3
T1 5
Berth

Fig. 4. Illustration of time segments for B&QCAP.

396 J. He / Advanced Engineering Informatics 30 (2016) 390–405

Memetic algorithm
Mutation
n

Generating Initial swarm

GA Control variables Control variables SA

module module

Evaluation indices Evaluation indices

Simulation module

Fig. 5. General framework of the optimization method.

of all vessels. This equation indicates the departure delay of a ves- beginning of the planning horizon. Constraints (14) imply that
sel is zero when it departs ahead of its planned departure time. The the positions of all vessels are restricted by the length of the wharf.
objective function (11) is to minimize the cost of total handling Constraints (15) define the berthing positions of all vessels which
energy consumption of all vessels by QCs. In this objective func- have been at berth at the beginning of the planning horizon.
tion, the number of QCs n assigned to Vessel i is no decision vari- Constraints (16)–(21) ensure that there is no overlap among
able in order to ensure the linearity of corresponding objective all vessels in the 2-dimensional berth-time plane. Constraints
function. Instead, we utilize the proposed approach by Meisel (16) ensure that Vessel j must be moored at the berth after Vessel
and Bierwirth [8], i.e., the number of QCs assigned to Vessel i is i departs if Vessel i is located below Vessel j in the berth-time
described by binary variable vitn, which indicates whether exactly plane. Constraints (17) ensure that the left side position of Vessel
n QCs are assigned to Vessel i in Period t. j must be greater than the right side position Vessel i if Vessel i
However, the cost for delay departure time ($/h) is in a different is located in the left of Vessel j in the berth-time plane. Constraints
scale than the unit energy cost ($/kW h). In order to avoid the opti- (18) and (19) ensure that there are one or two position relation-
mization of both parameters focusing only in one of the parame- ships between any two vessels in the 2-dimensional berth-time
ters, we must normalize them in order to obtain a fair value. The plane. Constraints (20) ensure that a vessel cannot be located in
normalized objective function is as follows. the left and right of another vessel in the berth-time plane.
min min
Constraints (21) ensure that a vessel cannot be located below
f1  f1 f2  f2 and above another vessel in the berth-time plane. Constraints
min f ¼ max min
þ max min
ð33Þ
f1  f1 f2  f2 (22) and (23) define the number of berth deviation segments of
Vessel i. Constraints (24) ensure that the number of berth deviation
where segments of Vessel i takes only one specific value. Constraints (25)
min ensure that QC assignments for a vessel must satisfy the vessel’s
f1 ¼0 ð34Þ
real QC hours demand considering the QCs’ idle times as the result
0   0:14 0 11 of berthing deviation and QC interference. In Constraints (22)–(25),
b b
X BW i  s  max Pi ; L  Pi W i =p the number of berth deviation segments of a vessel m is no
max
f1 ¼ @ @ AC
A  Cpi decision variable in order to ensure the linearity of corresponding
p bq max
i2V[G c bq max  N
Ni
i constraints. Instead, the number of berth deviation segments of a
ð35Þ vessel is described by binary variables Dbim , indicating whether
the number of berth deviation segments of Vessel i is m.
min
X Wi Wi  s
 Constraints (26) ensure that at most Q QCs can be utilized in any
f2 ¼ l þn  1  0:79641=1  CE ð36Þ period. Constraints (27) define the relationship between variables
p p
i2V[G vitn and dit , which ensure that the number of QCs assigned to a
0 1 vessel takes only one value or zero. Constraints (28) ensure that
0 1
X
bqmax
the number of QCs assigned to a vessel must not be greater than
B Wi W i =p N bqmax 0:79641=Ni C
@l  þn @ A 
max
f2 ¼ i
bqmax
 A CE its maximum number of QCs allowed to serve simultaneously.
i2V[G
p c bqmax
Ni
N bqmax
i 1þ0:79641=Ni Constraints (29) ensure that no QC is assigned to a vessel after it
departs. Constraints (30) ensure that no QC is assigned to a vessel
ð37Þ
before it is moored at the berth. Constraints (31) define the integer
Formula (35) denotes the difference between the total handling variables, and Constraints (32) define the binary variables.
times of two solutions: the one is that the berthing deviation value Since the objective function (10) contains maximum value forms,
of each vessel is maximum and the number of QC assigned each the model can be linearized by substituting maxðT brd  T bpd
i i ; 0Þ with
vessel is one; the other is that the berthing deviation value of each auxiliary variable f1i. The rewritten model is as follows.
vessel is zero and the number of QC assigned each vessel is the P min min
maximum. Formula (36) means the non-working energy consump- i2V[G f 1i  f 1 f2  f2
½ILP min f ¼ max min
þ max min
ð38Þ
tion of each vessel is the minimum when the number of QC f1  f1 f2  f2
assigned each vessel is one. Formula (36) means the non-
working energy consumption of each vessel is the maximum when S.T. Constraints (12)–(32) and
the number of QC assigned each vessel is the maximum. f 1i P ðT brd
i i Þ  Cpi ; 8i 2 V [ G
 T bpd ð39Þ
Constraints (12) ensure that the berthing time of each vessel
cannot be earlier than its arrival time. Constraints (13) define the f 1i P 0; 8i 2 V [ G ð40Þ
berthing time of all vessels which have been at berth at the
 J. He / Advanced Engineering Informatics 30 (2016) 390–405 397

6. Solution method time and berthing position. Therefore, we should decode a

sequence of vessels into a solution using a constructive algorithm.
6.1. General framework This algorithm aims to insert all vessels which have not been at
berth at the beginning of the planning horizon (i 2 V) into the
From Fig. 4, we can find that the B&QCAP is similar to a 2-dimensional berth-time plane one by one according to the berth-
2-dimensional bin packing problem. It is well known that the ing sequence. In the constructive algorithm, we take one vessel at a
2-dimensional bin packing problem is NP-hard [32]. Therefore, time and look for its least-cost assignment at each available posi-
the vessel scheduling problem also is NP-hard. Typically, the length tion and each available berthing time. For each vessel, we firstly
and height of each rectangle in 2-dimensional bin packing problem check whether berthing the vessel at its expected best berthing
are deterministic. However, since the QC assignments for all ves- position and its arrival time overlaps any other previously assigned
sels dynamically vary in the planning horizon, the handling time vessel. If it does not, we take its expected best berthing position
for each vessel is not deterministic. Therefore, the height of each and its arrival time as its berthing position and berthing time;
rectangle in this study is uncertain. The exiting methods for solving otherwise, new feasible solution with least cost should be identi-
the 2-dimensional bin packing problem cannot be applied to the fied. Fig. 7 shows an example to illustrate the detail of the con-
vessel scheduling problem. Thus, it’s necessary to develop an structive algorithm. In this example, we should insert Vessel i
appropriate approach for solving the B&QCAP. An integrated simu- into the 2-dimensional berth-time plane, where the B&QCAP solu-
lation and optimization method is proposed for solving the tion of Vessels 1–4 has been given. The expected best berthing
0
B&QCAP, where the simulation is designed for evaluation and gene position and the arrival time of Vessel i are P rb bb
i and T i . Firstly,
repair, and optimization algorithm is designed for searching solu- we want insert Vessel i into the berth-time plane at its desire posi-
tion space. In order to improve the quality of solution, memetic 0
tion (P rb bb
i , T i ), given by its expected best berthing position and its
algorithm (MA) as the optimization algorithm is proposed, where
arrival time. As the insertion point overlaps with Vessel 3, another
the genetic algorithm (GA) is used for global search, and the simu-
insertion point must be found for Vessel i. Subsequently, the con-
lated annealing (SA) algorithm is used as local search algorithm for
structive algorithm looks for a set of feasible candidate points in
mutation operation in GA. The general framework of the integrated
its searching space, and selects the best insertion point according
simulation and optimization method is illustrated in Fig. 5.
to the penalty cost of departure delay and the energy cost which
are calculated using Eq. (33). For the example in Fig. 7, the points
6.2. Memetic algorithm 0 0
(Prb bb rb bb
i , T i ) and (P i , T i ) are feasible candidate points for Vessel i,
and we should select the least-cost point as the insertion point.
6.2.1. Solution encoding and decoding
To reduce the computational complexity, we should avoid search
By considering the main decision variables of the proposed
all feasible points.
model, the B&QCAP is represented as a two-vector chromosome,
The decoding procedure for berthing position and berthing time
which denotes berthing sequence of vessels and the number of
is as follows:
QCs assigned to a vessel within each period. For the berthing time
and berthing position of each vessel, we use ‘‘berthing sequence”
representation in which the chromosome is an ordered list of ves- If Vessel i can be inserted into the berth-time plane at its
sels, instead of using layout representation in the berth-time plane. desire insertion point, given by its expected best berthing
Therefore, we should decode a sequence of vessels into the berth- position and its arrival time Then
ing time and left-most berthing position of each vessel. Each gene The berthing position and the berthing time of Vessel i are
in the ‘‘berthing sequence” vector is encoded by an integer number. its best expected berthing position and its arrival time,
For all vessels which have been at berth at the beginning of the respectively; //Inserting the vessel at its desire point;
planning horizon (i 2 G), we set their berthing sequence as zero, Execute the QC assigning procedure; // the QC assigning
and their berthing sequence, berthing time, berthing position are procedure is described later;
kept unchanged and are not evolved. The ‘‘QC assignment” vector Else
consists of N + S parts, and each part includes NT genes, and each //looking for feasible candidates.
gene represents the QC assignment of a vessel within a period. If there exists feasible insertion points at the arrival time of
The ‘‘QC assignment” gene is encoded by an integer number. An Vessel i Then
example of solution representation is shown in Fig. 6. For instance, The point, given by the nearest position to the best expected
the berthing sequence of Vessel 5 is 2, and the number of QCs berthing position and the arrival time of Vessel i, is set as a
assigned to it within the Period 1 is n51. candidate;
Based on the representation, the berthing sequence and QC End if;
assignment of each vessel can be obtained, but not the berthing For each Vessel i0 which solution has been built and
departure time is greater than the arrival time of Vessel i
Vessel No. Sequence QC assignment Loop
n11 n12 n1T If there exists feasible insertion points for Vessel i at the
1 0 ……
Vessels which departure time of Vessel i0 Then
2 0 n21 n22 …… n2T
have been at
The point, given by the nearest position to the best
3 0 n31 n32 …… n3T expected berthing position of Vessel i and the departure
terminal
4 0 n41 n42 …… n4T time of Vessel i0 , is set as a candidate;
5 2 n51 n52 …… n5T End if;
Vessels which 6 1 n61 n62 …… n6T End for;
n71 n72 n7T Execute the QC assigning procedure for all candidates;
will arrive at 7 3 ……
Calculate the costs of all candidates and selecting the
terminal 8 4 n81 n82 …… n8T
candidate with least-cost solution as the insertion point of
9 2 n91 n92 …… n9T Vessel i;
End if;
Fig. 6. Chromosome representation.
398 J. He / Advanced Engineering Informatics 30 (2016) 390–405

The decoding procedure for QC assignment is as follows: 6.2.2. Initialization

In the initialization process, with respect to the berthing
Obtain the first period t c in which the number of QCs assigned sequence, chromosomes are generated to construct the initial pop-
to Vessel i is not zero from the current chromosome; ulation by using a set of priority rules. Subsequently, we use the
decoding procedure (described in Section 6.2.1) to decode a
tb ¼ bT bbi =spc;
sequence of vessels into the berthing positions and the berthing
nit : the number of QCs assigned to Vessel i at the beginning of
times of vessels. With respect to QC assignment initialization, the
Period t;
decoding procedure for QC assignment in the decoding process
QSt: the number of surplus QCs within Period t;
(described in Section 6.2.1) is substituted by the QC assignment
VSt: the surplus handling volume of Vessel i at the beginning
initialization procedure (described later in this section).
of Period t;
The priority rules for generating the berthing sequence in the
For t = tb to NT Loop
initialization process are listed in Table 1. We can randomly select
If QSt ¼ 0 Then
1–8 rules, and combine them into more complex rules from per-
t = t + 1;
mutations of the selected rules. Furthermore, we randomly select
Go to the next loop;
the rules from permutations to form all the berthing sequences
End if;
of the initial population.
If VSt > 0 Then
The QC assignment initialization procedure is as follows:
If 1 6 niðtðtb tc ÞÞ 6 minðN bq
i
max
; QSt Þ Then
If niðtðtb tc ÞÞ  cn  sp 6 VSt Then For t = t b to NT Loop
iðtðt b t c ÞÞ

nit ¼ niðtðtb tc ÞÞ ; If QSt ¼ 0 Then

Else t = t + 1;
  Go to the next loop;
VSt
nit ¼ c sp End if;
n
iðtðt b t c ÞÞ

End if; If VSt > 0 Then

Else if niðtðtb tc ÞÞ P minðN bq max

; QSt Þ or niðtðtb tc ÞÞ ¼ 0 nit ¼ randomð0; minðN bqmax
i ; QSt ÞÞ;
i
Else
Then
Exit Loop;
If minðN bq
i
max
; QSt Þ  cminðNbq max ;QS Þ  sp 6 VSt Then End if;
i t

nit ¼ minðN bq max

; QSt Þ; End for;
i
Else
& ’
VSt
nit ¼ c sp ;
minðN
bqmax
i
;QS Þ t 6.2.3. Individual improving
End if; We should check all chromosomes if any constraints presented
End if; in Section 5.4 are violated. If none of the constraints is violated, the
Else chromosome is feasible and can be held; otherwise, the unfeasible
Exit Loop; genes of the chromosome should be repaired through the gene
End if; repair procedure. However, based on our proposed procedure of
End for; solution encoding and decoding, there is no unfeasible chromo-
some. But we should improve some low quality chromosomes,
such as adding the number of QCs assigned to a vessel if there exist
surplus QCs. The individual improving procedure is as follows:

Time
The expected best berthing
position of Vessel i

Vessel i

( Pi rb , Ti bb ) The arrival
Vessel 4
time of
( Pi rb ' , Ti bb ') Vessel i

( Pi rb , Ti bb ')
Vessel 3
Vessel 2
Vessel 1

Berth

Fig. 7. Illustrate the detail of the constructive algorithm.

 J. He / Advanced Engineering Informatics 30 (2016) 390–405 399

Table 1 6.2.4. Fitness evaluation

Priority rules for generating the initialization berthing sequence. After the completion of the gene repair process, a feasible
No. Rule Sorting B&QCAP schedule is created. Afterwards, the schedule is inputted
1 Non-
into the simulation module for obtaining the departure delay and
The arrival time of each vessel ðT ba
i Þ
decreasing the handling energy consumption of each vessel. Finally, the fitness
2 The planned departure time of each vessel ðT bpd
i
Þ Non- value of the chromosome is calculated by fitness evaluation func-
decreasing tion. As the B&QCAP is a minimization problem, the fitness func-
3 The handling volume of each vessel ðW i Þ Non-
tion is defined as follows.
increasing
4 The unit penalty cost for delayed departure time of each Non- F ¼ 1=ð1 þ expðf =aÞÞ ð41Þ
vessel ðCpi Þ increasing
f 1 f min
5 Vessel length ðli Þ Non- where a is a problem scale related parameter. Since 1
6 1 and
f max
1 f min
1
increasing
f 2 f min
6 The planned stayed time at terminal of each vessel Non- 2
6 1, the parameter a is taken the value 2.
f max
2 f min
2
ðT bpd  T ba increasing
i i Þ
7 The maximum number of QCs that can be assigned to Non-
increasing 6.2.5. Parent selection
each vessel ðN bqmax
i Þ
8 Non- The roulette wheel sampling is used for parent selection mech-
The maximum slack ðT bpd
i  ðT ba
i þ W i =cN bqmax ÞÞ
i
decreasing anism [33]. The details are as presented as follows:

(1). Calculate the sum of fitness in population, i.e.,

P
SUM ¼ Pc¼1 F c , where P is the population size.
P
Step 1: For t ¼ 1; 2; . . . ; NT, if i2V [G N qit 6 Q , the surplus QCs can (2). Calculate the selection probability of each individual, i.e.,
be assigned to other vessels at berth and of which the maxi- Sc ¼ F c =SUM; ðc ¼ 1; 2; . . . ; PÞ.
mum number of QCs have been not reached and QC hours (3). Calculate the cumulative probability of all individuals.
P
demands are not satisfied. The detail of the procedure is as mc ¼ cj¼1 Sj ; c ¼ 1; 2; . . . ; P.
follows: (4). Generate a random number, i.e. r, between 0 and 1. If
mc1 < r 6 mc , the individual c is selected.

6.2.6. Crossover operation

QSt: the number of surplus QCs within Period t; In this paper, two crossover operators are employed. For the
VSt: the surplus handling volume of Vessel i at the beginning ‘‘berthing sequence” gene, in order not to generate unfeasible off-
of Period t; spring, such as some sequence may be missed while some
For t = 1 to NT Loop sequence may appear repeatedly in the offspring, the order cross-
P
If i2V[G N qit < Q Then over [34] is used as the crossover operator. For the ‘‘QC assign-
Sort all vessels at berth within Period t according to the ment” gene, the arithmetic crossover is employed.
ascending order of the vessels’ planned departure time; The berthing sequences of all vessels at berth at the beginning of
Assign the surplus QCs to each vessel according to the the planning horizon (i 2 G) are not participated in the crossover
arranged order. If the maximum number of QCs assigned to operation. The details of the order crossover are as follows (see
the current vessel has been reached or QC hours demand of Fig. 8).
the current vessel is satisfied, we should assign the
additional QCs to other vessels until the surplus QCs is zero; Step 1: Select a substring from Parent 1 randomly.
End if; Step 2: Generate a protochild by copying the substring into the
End for; corresponding positions in the protochild.
Step 3: Delete those genes in the substring from Parent 2.
Step 4: Insert the genes into the unfilled positions of the pro-
tochild from left to right according to the sequence of the
remained genes of Parent 2.
Step 2: If Vessel i is moored within Period t and T bb
i > ðt  1Þ  sp, Step 5: Repeat Steps 1–4 to generate another offspring by
the QCs assigned to it should be assigned to other vessels at exchanging the two parents.
berth within time-window ððt  1Þ  sp; T bb
i Þ and of which the
maximum number of QCs have been not reached and departure The arithmetic crossover for the ‘‘QC assignment” gene is repre-
delay can be shortened. The assigning order is ascending order sented as follows.
of the vessels’ planned departure time. Offspring 1 ¼ k  Parent 1 þ ð1  kÞ  Parent 2 ð42Þ
Step 3: If Vessel i departs within Period t and T brd
i < t  sp, the
Offspring 2 ¼ k  Parent 2 þ ð1  kÞ  Parent 1 ð43Þ
QCs assigned to it should be assigned to other vessels at berth
within time-window ðT brd
i ;t
 spÞ and of which the maximum where k is a randomly-generated number within an interval from 0
number of QCs have been not reached and departure delay to 1. Parameter k is not a constant, which should be changed
can be shortened. The assigning order also is ascending order dynamically with the evolved generation increasing. When the cur-
of the vessels’ planned departure time. rent generation is closer to the maximum generation, the k should
be closer to 0. Upon completion of crossover, the gene should be
The individual improving procedure is very complex and encoded as an integer and rounded.
dynamic for multiple planning periods via modeling analytically.
However, as a result of good dynamic characteristics, simulation 6.2.7. Simulated annealing algorithm for mutation operation
can easily handle all constraints and execute gene repair strategy. Generally, local search algorithm is used as mutation operator in
Therefore, individual improving procedure is modeled in the simu- MA. In this paper, replacing traditional mutation operators, we use
lation module. the SA algorithm as local search algorithm for mutation operation.
400 J. He / Advanced Engineering Informatics 30 (2016) 390–405

Vessel No. Sequence Vessel No. Sequence Vessel No. Sequence

1 2 1 3 1 3
2 1 2 8 2 5
3 3 3 2 3 6
4 4 4 4 4 8
5 6 5 6 5 2
6 7 6 7 6 1
7 5 7 5 7 4
8 9 8 1 8 7
9 8 9 9 9 9
Parent 1 Offspring 1 Parent 2

Fig. 8. An example of order crossover operation.

6.2.7.1. The framework of the SA. The SA algorithm involves initial Step 3: Select Vessel Vi with the largest Gap(i), and swap the
temperature T, cooling rate a (0 < a < 1), and the number of nested sequences of Vi and Vi+1 to generating a new inserting sequence.
loops R, which represents the size of neighborhood of a solution. Step 4: Obtain other neighbors with the second largest, . . . , the
The framework of the SA is as follows. Rth largest Gap(i). If there exist some neighbors which have
been generated in evolving process, they will be ignored and
M: the size of the selected individuals for mutation, where Sm more neighbors with the (R + 1)th largest Gap(i) will be gener-
represents the mth selected individual;Set SA parameters T, ated until all neighbors are not duplicate.
r, threshold;
For m = 1 to M Loop It is noted that we swap berthing sequence of two vessels, but
While T > threshold Do not swap their QC assignments. The QC assignments of all vessels
Generate R neighbor sequences of the individual Sm using can be rearranged by decoding procedure.
the later proposed method for neighborhood generation,
where Si represents the ith neighbor sequence; 6.2.8. Offspring acceptance
For i = 1 to R Loop In particular, the population size is kept unchanged. As such, all
Obtain the schedule of the neighbor sequence Si using parents and offspring are sorted via fitness in a descending
the decoding procedure described in Section 6.2.1; sequence. If the sequential number of an individual is more than
Calculate the objective value f(Si) of the neighbor the population size, the individual is discarded as the new
sequence Si using Eq. (33); generation.
Let D ¼ f ðSi Þ  f ðSm Þ;
If D < 0 Then 6.2.9. Stopping criterion
Set Sm = Si; The algorithm is terminated based on the experimentally-
Else determined maximum number of elapsed generations.
Generate a random number r between 0 and 1;
If r < eD=T Then 6.3. Simulation module
Set Sm = Si;
End if; According to the modeling approach proposed by He et al. [28],
End if; we construct the simulation module of the B&QCAP five parts: (1)
End for; Input module; (2) Control module; (3) Object module; (4) Evalua-
Decrease T = a  T; tion and output module; and (5) Optimization module.
End while;
End for; 6.3.1. Input module
Input module is used to initialize all parameters for simulation,
which includes all user-defined parameters and some decision
6.2.7.2. Neighborhood generation. Specifically for the neighborhood variables presented in Section 5.3, such as berthing time, berthing
generation at each iteration step, the proposed SA selects two ves- position and QC assignment. Some user-defined parameters are
sels from the currently proposed solution and swaps their berthing used in the entire simulation process, such as the handling effi-
sequences to generate new neighboring solutions. In order to ciency of a QC and berthing safety factor. The other user-defined
strength the solution, the SA does not swap two randomly selected parameters are used for schedule evaluation, such as the unit pen-
vessels in the sequence, but use the method proposed by Zhen alty cost and the unit energy cost. Decision variables are dynamic,
et al. [35]. Regarding to their idea, the vessels in the front of and are generated by Optimization module and adjusted by Con-
sequence will have more opportunities to obtain better B&QCAP trol module.
schedule than the ones in the rear, and a better schedule means
that the vessel would have lower operation cost. The details of 6.3.2. Control module
the neighborhood generation are illustrated as follows: The functions of control module are: (1) time control, i.e., con-
trol the pushing of the simulation events; and (2) scheduling con-
Step 1: Calculate the cost of each vessel using Eq. (33). trol, i.e., check and improve schedules. The time control can (1)
Step 2: Calculate cost gap between adjacent vessels in the start and end running simulation; and (2) monitor the processing
sequence: Gap(i) = Cost(Vi+1)–Cost(Vi), where Vi represents the time of each event, such as the arriving, berthing, departure time
ith vessel in the sequence. of a vessel. The scheduling control is very important, which
 J. He / Advanced Engineering Informatics 30 (2016) 390–405 401

includes two functions: (1) control each vessel berthing, handling TM i7-2820QM @ 2.3 GHz processors and 16 GB RAM and 64 bit
by QCs and departing according its schedule, (2) improve low qual- operation system. The constant initial parameters are set as follows
ity solutions. (Table 2):

6.3.3. Object module

7.1. Determining the span of time segment
The object module is used to simulate all movements and
behaviors, e.g., arriving, berthing, handling and departing. This
Four sets of experiments with different vessel arriving time
module consists of two parts: (1) static object, (2) dynamic object.
intervals and with different handling volumes are conducted,
Static object is the quayside. Before performing a simulation, a map
where the arriving time intervals follow the negative exponential
of the quayside with coordinate plane should be defined for a con-
distribution with expected value Ati, and the vessel length and
tainer terminal. Dynamic objects are QCs and vessels. All QC
handling volume follow an empirical distribution as shown in
objects should be created before starting simulation. However, to
Table 3. The number of quay cranes and quay length are 12 and
reduce the simulation scale, each vessel object should be dynami-
1100 m, respectively.
cally created when it arrives at terminal and be dynamically
In order to validate the reliability of the span of time segment,
deleted when it departs.
for each set of experiments, ten instances are randomly generated
according to the corresponding statistical distribution. For the pro-
6.3.4. Evaluation and output module posed method, the span of time segment takes the value of 0.5 h,
The evaluation and output module is used to obtain all statisti-
1 h, 2 h, 4 h and 6 h, respectively. The population size, the probabil-
cal indices related to efficiency and energy consumption, and cal- ity of crossover, the probability of mutation, and the maximum
culate the fitness value of each schedule.
generations of the GA are set as 60, 0.6, 0.1 and 80, respectively.
The parameters of the SA, i.e., the initial temperature T, cooling rate
6.3.5. Optimization module a, the number of nested loops R, and the threshold are set as 40,
The optimization module is used to generate and evolve all 0.6, 40 and 1, respectively. Table 4 illustrates the comparisons
individuals by the GA and the SA, and then converts all individuals among different spans of time segment. The comparative indica-
into the B&QCAP schedules, and feed back the evaluation indices tors include the objective values of Model ILP and CPU time.
into the GA or the SA for calculating objective values. From Table 4, it can be found that the differences among the
objective values of all schedules are very small when the length
6.4. Implementation of algorithm of time segment takes the value of 0.5 h, 1 h and 2 h respectively.

The flowchart of the integrated simulation and optimization

Table 2
method is presented in Fig. 9.
Some constant parameters of the B&QCAP.

The working energy consumption of a QC per move 4

7. Computational experiments (unit: kW h/move)
The unit handling volume of a QC per move (unit: TEU/move) 2
The performance test and analysis of the proposed integrated The non-working energy consumption of a QC per hour 36
simulation and optimization method for the B&QCAP are (unit: kW h/h)
The handling efficiency of a QC (unit: h/move) 0.033
conducted in this section. This section includes three parts: (1)
The unit penalty cost for delayed departure time of each vessel 800
the determination of the span of time segment; (2) the computa- (unit: $/h)
tional performance analysis of the proposed MA by a series of The unit energy cost (unit: $/kW h) 0.12
experiments with different sizes; and (3) the energy consumption Berthing safety factor 0.1
analysis. All experiments are run on a workstation with Intel Core

Initialization Start

Offspring acceptance

Decoding
Simulation

Individual improving SA-based Mutation

Fitness value calculation Crossover

Stopping Criterion N Parent selection

Output the best solution End

Fig. 9. Flowchart of the integrated simulation and optimization method.

402 J. He / Advanced Engineering Informatics 30 (2016) 390–405

Table 3 Table 5
Length and handling volume of vessels. Computational results of all instances.

Length Handling volume (TEU) Proportion (%) Problem The proposed method CPLEX
155 100–500 10.10 N L (m) Q f CPU (s) Gap (%) fc CPU (s)
155 500–1000 38.80 a
8 600 6 3742 94 0 3742 47
200 1000–1800 26.40 a
8 600 6 4286 97 0 4286 51
272 1800–2800 9.21 a
8 600 6 3936 101 0 3936 61
282 2800–4000 8.95 a
8 600 6 3952 96 0 3952 44
300 4000–6000 7.17 a
8 600 6 4202 100 0 4202 54
a
10 800 8 5525 133 0 5525 103
a
10 800 8 4946 141 0 4946 116
a
10 800 8 5237 144 0 5237 95
Table 4 10 800 8 4854 137 0 4854 a
104
Comparisons of the span of time segment. 10 800 8 5613 134 0 5613 a
102
a
12 1000 10 6016 169 0 6016 202
Arriving interval, Ati Span of time segment Objective value CPU time a
12 1000 10 5822 178 0 5822 205
(h) (h) ($) (s) a
12 1000 10 6526 192 0 6526 202
a
3 0.5 3032 377 12 1000 10 6663 185 0 6663 187
a
1 3036 253 12 1000 10 5752 174 0 5752 193
a
2 3040 159 8 1100 12 3092 161 0 3092 133
a
4 3146 125 8 1100 12 4209 161 0 4209 134
a
6 3171 96 8 1100 12 3863 158 0 3863 127
a
8 1100 12 3546 154 0 3546 143
2 0.5 5776 415 a
8 1100 12 4331 151 0 4331 122
1 5836 278 a
12 1100 12 5966 189 0 5966 203
2 5814 187 a
12 1100 12 5496 186 0 5496 215
4 5950 149 a
12 1100 12 6162 204 1.12 6094 216
6 5990 133 a
12 1100 12 6200 190 0.00 6200 220
1.5 0.5 7056 486 12 1100 12 5259 200 0.98 5208 a
208
1 7095 324 16 1100 12 7345 213 0.56 7304 a
1150
2 7109 217 16 1100 12 8379 225 1.07 8291 a
1200
4 7225 174 16 1100 12 7338 211 0.75 7283 a
1057
6 7361 162 16 1100 12 7259 221 0.38 7231 a
1461
a
1.2 0.5 8586 572 16 1100 12 8077 222 1.30 7973 1483
a
1 8536 344 20 1100 12 8680 248 0.56 8631 1686
a
2 8574 238 20 1100 12 9588 242 0.55 9535 1671
a
4 8715 187 20 1100 12 8980 243 1.23 8871 1701
a
6 8801 174 20 1100 12 9418 255 1.01 9324 1691
a
20 1100 12 9935 253 1.72 9767 1683
a
24 1600 16 11,300 302 0.51 11,243 3581
a
24 1600 16 12,061 309 0.18 12,039 4535
L
However, the objective values of all schedules are very different 24 1600 16 11,728 328 1.05 11,606 –
L
when the span of time segment takes the value of 4 and 6 h respec- 24 1600 16 12,018 321 0.97 11,902 –
L
24 1600 16 12,231 330 0.51 12,169 –
tively. Therefore, based on the computational experiments, the L
28 1600 16 13,486 370 0.63 13,401 –
span of time segment should take the value of 2 h. 28 1600 16 13,882 386 0.94 13,753 L
–
L
28 1600 16 13,947 387 0.39 13,893 –
L
7.2. Performance analysis 28 1600 16 14,807 373 0.96 14,666 –
L
28 1600 16 13,371 398 0.37 13,321 –
L
30 1600 16 14,558 479 0.81 14,442 –
In this section, 13 sets of random experiments with different 30 1600 16 15,615 499 1.25 15,423 L
–
sizes are tested to compare the results and CPU time obtained by 30 1600 16 14,662 480 1.05 14,509 L
–
L
the proposed method and CPLEX. The sizes of all experiments are 30 1600 16 15,407 497 0.91 15,268 –
L
directly related to the number of vessels, number of QCs and quay 30 1600 16 14,752 492 0.06 14,743 –
L
32 2000 20 15,302 560 3.67 14,761 –
length. The lengths and handling volumes of all vessels also follow L
32 2000 20 15,999 585 0.39 15,937 –
an empirical distribution as shown in Table 3. The arriving time 32 2000 20 15,560 561 0.38 15,501 L
–
intervals of all vessels follow the negative exponential distribution. 32 2000 20 15,264 559 0.54 15,181 L
–
L
For each sets of experiments, five instances are randomly gener- 32 2000 20 16,370 559 0.88 16,228 –
L
ated. For comparison, we need to find the optimal solution by 36 2000 20 18,432 698 4.42 17,651 –
L
36 2000 20 18,432 685 0.54 18,334 –
CPLEX. However, it is difficult to obtain optimal solutions by CPLEX 36 2000 20 19,172 683 0.99 18,985 L
–
for large size instances. If the optimal objective values of some 36 2000 20 19,139 689 0.04 19,131 L
–
L
large-size instances are not obtained by CPLEX directly until out 36 2000 20 18,531 679 1.11 18,328 –
L
of memory, we use the lower bounds corresponding to these 40 2000 20 25,972 835 2.00 25,463 –
L
40 2000 20 20,159 815 0.65 20,028 –
instances for comparison. However, the lower bounds directly gen- L
40 2000 20 22,086 843 0.72 21,929 –
erated by CPLEX are very loose. Thus, we calculate the lower 40 2000 20 24,624 825 1.65 24,223 L
–
bounds corresponding to the instances by relaxing some complex 40 2000 20 25,120 823 0.80 24,920 L
–
constraints. Since Constraints (16) and (17) include big-M and sig- Average gap (%) 0.63
nificantly impacts the CPU time based on the preliminary tests,
Notes: ‘–’ denotes out of memory, Gap (%) = (f  fc)  100/fc.
two new valid inequalities are set to substitute them: a
The optimal result.
! L
The lower bound.
Wi
T ba
i þ j ; 8i; j 2 V [ G; i–j
 uij 6 T bb ð44Þ  
cNbqmax
i þ ð1 þ #Þ  li  hij 6 P j ; 8i; j 2 V [ G; i–j
Ppb pb
i ð45Þ
 J. He / Advanced Engineering Informatics 30 (2016) 390–405 403

Table 6
Comparison between ‘‘Energy-saving strategy” and ‘‘Time-saving strategy”.

Problem Energy-saving strategy Time-saving strategy

N L (m) Q f f1 f2 Gap (%) f f1 f2
8 600 6 3742 69 3673 2.31 3831 65 3766
8 600 6 4286 66 4219 1.26 4340 63 4278
8 600 6 3936 58 3878 1.50 3996 54 3942
8 600 6 3952 59 3893 1.54 4014 57 3957
8 600 6 4202 71 4131 2.21 4297 68 4229
10 800 8 5525 85 5440 1.64 5617 81 5536
10 800 8 4946 78 4869 2.68 5083 75 5008
10 800 8 5237 93 5144 2.65 5380 90 5291
10 800 8 4854 73 4781 1.65 4935 69 4866
10 800 8 5613 68 5545 2.61 5764 66 5698
12 1000 10 6016 91 5926 1.78 6125 87 6039
12 1000 10 5822 114 5709 1.14 5890 107 5782
12 1000 10 6526 94 6432 1.38 6618 91 6527
12 1000 10 6663 104 6559 2.97 6867 98 6769
12 1000 10 5752 121 5631 1.50 5840 114 5725
8 1100 12 3092 88 3004 1.97 3154 83 3071
8 1100 12 4209 121 4089 1.92 4292 115 4177
8 1100 12 3863 123 3740 0.64 3888 118 3770
8 1100 12 3546 116 3430 0.37 3560 111 3449
8 1100 12 4331 116 4215 0.62 4358 110 4247
12 1100 12 5966 71 5895 2.51 6120 67 6053
12 1100 12 5496 165 5330 0.84 5542 156 5386
12 1100 12 6162 132 6030 1.31 6244 125 6118
12 1100 12 6200 148 6052 0.15 6209 141 6069
12 1100 12 5259 102 5156 0.95 5309 95 5214
16 1100 12 7345 96 7249 0.69 7396 92 7304
16 1100 12 8379 124 8255 0.89 8455 118 8336
16 1100 12 7338 60 7278 3.01 7566 58 7509
16 1100 12 7259 126 7134 1.84 7395 118 7277
16 1100 12 8077 188 7889 0.54 8121 180 7941
20 1100 12 8680 275 8405 2.68 8919 263 8656
20 1100 12 9588 203 9385 1.38 9723 190 9532
20 1100 12 8980 258 8721 1.74 9138 245 8893
20 1100 12 9418 240 9178 0.86 9499 227 9272
20 1100 12 9935 291 9645 1.80 10,117 273 9844
24 1600 16 11,300 198 11,102 0.05 11,306 187 11,119
24 1600 16 12,061 195 11,866 2.24 12,337 183 12,154
24 1600 16 11,728 121 11,607 3.15 12,110 115 11,994
24 1600 16 12,018 104 11,914 2.33 12,304 99 12,205
24 1600 16 12,231 196 12,035 3.14 12,628 187 12,441
28 1600 16 13,486 369 13,116 3.19 13,930 358 13,571
28 1600 16 13,882 294 13,588 3.41 14,372 283 14,090
28 1600 16 13,947 290 13,658 3.12 14,396 275 14,122
28 1600 16 14,807 335 14,472 2.74 15,224 318 14,906
28 1600 16 13,371 313 13,058 3.83 13,903 300 13,602
30 1600 16 14,558 357 14,201 2.71 14,964 341 14,622
30 1600 16 15,615 430 15,185 2.78 16,062 407 15,655
30 1600 16 14,662 364 14,298 3.54 15,200 347 14,853
30 1600 16 15,407 580 14,827 2.51 15,803 562 15,241
30 1600 16 14,752 444 14,308 1.97 15,048 430 14,618
32 2000 20 15,302 285 15,017 1.60 15,551 271 15,280
32 2000 20 15,999 143 15,856 1.41 16,227 137 16,090
32 2000 20 15,560 287 15,273 1.32 15,768 271 15,496
32 2000 20 15,264 201 15,064 1.83 15,548 196 15,352
32 2000 20 16,370 255 16,115 1.80 16,670 242 16,428
36 2000 20 18,432 1548 16,885 2.15 18,838 1473 17,365
36 2000 20 18,432 1352 17,080 3.01 19,004 1308 17,696
36 2000 20 19,172 1385 17,787 2.13 19,589 1347 18,242
36 2000 20 19,139 1259 17,880 2.96 19,724 1198 18,525
36 2000 20 18,531 1444 17,086 2.17 18,942 1394 17,548
40 2000 20 25,972 7657 18,315 5.36 27,443 7305 20,138
40 2000 20 20,159 5886 14,273 4.79 21,173 5530 15,643
40 2000 20 22,086 6347 15,738 4.27 23,071 6046 17,025
40 2000 20 24,624 7192 17,432 4.73 25,847 6914 18,933
40 2000 20 25,120 7473 17,647 5.16 26,485 7175 19,310
Average gap (%) 2.17

Constraints (16) is substituted by Inequality (44), and tests, the parameters of the SA, i.e., the initial temperature T, cool-
Constraints (18) is substituted by Inequality (45). ing rate a, the number of nested loops R, and the threshold are set
Based on the preliminary tests, the probability of crossover, the as 40, 0.6, 40 and 1, respectively.
probability of mutation, and the maximum generations of the GA The computational results of all instances are shown in Table 5.
are set as 60, 0.6, 0.1 and 80, respectively. Based on the preliminary Columns 1–3 depict the sizes of all instances. Columns 4 and 5
404 J. He / Advanced Engineering Informatics 30 (2016) 390–405

represent best solution and the computational time obtained from (1) Most of the related works in B&QCAP concern how to opti-
the proposed method. Column 6 shows the gap between the result mize the vessels’ departure delay. In this paper energy con-
obtained from the proposed method and obtained from CPLEX for sumption is given proper consideration. Energy
each instance. Column 7 lists the best solution or lower bound consumption cost is a part of optimized objective. The direct
obtained from CPLEX. Column 8 presents the computational time quantitative calculation equations on energy consumption
of CPLEX. of QCs are also developed according to the relationship
As observed in Table 5, the gaps between the results obtained between energy consumption and QC assignment.
from the proposed method and the optimal results or lower (2) In this paper an enriched mathematical model for B&QCAP is
bounds obtained from CPLEX are very small (for example, the aver- developed aiming to minimize total departure delay and
age gap is only 0.63%, the minimum gap is 0, and the maximum total energy consumption of all vessels, while capturing all
gap is only 4.42%). The CPU times of the proposed method for all the essential constraints in practice. Due to the NP-hard with
the instances are acceptable in practice (for example, the longest regard to B&QCAP, we propose an integrated simulation and
CPU time is less than 15 min). Based on these 65 computational optimization method to efficiently solve the problem, where
experiments, it is clear that near optimal solutions obtained from memetic algorithm (MA) is used as the optimization algo-
the proposed method are of high quality. The performance of the rithm. In the MA, we use the SA algorithm as local search
proposed method is thus satisfactory in solving different size algorithm for mutation operation.
instances. (3) Two kinds of scheduling strategies are analyzed in this
paper: one is referred to energy consumption objective,
7.3. Energy consumption analysis i.e., ‘‘Energy-saving strategy”, and the other is not referred
to energy consumption objective, i.e., ‘‘Time-saving strat-
In this section, the 13 sets of instances presented in Section 7.2 egy”. The experimental results illustrate ‘‘Energy-saving
are conducted to compare two scheduling strategies: (i) the pro- strategy” is more suitable for the tendency of green trans-
posed method considering the energy consumption objective portation and green port. The weight coefficient is proposed
(namely ‘‘Energy-saving strategy”) and (ii) the proposed method to satisfy the preference to the energy-saving and the time-
not considering the energy consumption objective (namely saving of different container terminals.
‘‘Time-saving strategy”). Two scheduling strategies are measured
in terms of departure delay costs, energy consumption costs and However, improvement potential is still remaining, e.g. it might
total berthing costs. be a future trend on the determination of the optimal handling effi-
Table 6 shows the scheduling results obtained from these two ciency of QCs for minimization of total berthing costs. In the future
strategies, where ‘f’ denotes the total berthing cost, ‘f1’ denotes research, the relationship between energy consumption and the
departure delay cost, ‘f2’ denotes energy consumption cost, Gap handling efficiency of QCs will be investigated. The handling effi-
(%) = (f from ‘‘Energy-saving strategy”  f from ‘‘Time-saving strat- ciency of QCs will be as decision variable for optimization.
egy”)  100/f from ‘‘Time-saving strategy”. As observed in Table 6,
total scheduling costs obtained from ‘‘Energy-saving strategy” are Acknowledgements
smaller than those obtained from ‘‘Time-saving strategy” (for
example, the maximum gap among the 13 sets instances is This work is sponsored by ‘‘Chenguang Program” supported by
5.36%, the minimum gap is 0.05%, and the average gap is Shanghai Education Development Foundation and Shanghai
2.17%). Furthermore, the delay costs obtained from ‘‘Energy- Municipal Education Commission (14CG48), Shanghai Sailing Pro-
saving strategy” are larger than those obtained from ‘‘Time-saving gram (14YF1411200), Doctoral Fund of the Ministry of Education
strategy”. However, the energy consumption costs obtained from (20133121110001), Shanghai Municipal Education Commission
‘‘Energy-saving strategy” are evidently smaller than those obtained Project (14YZ112), Shanghai Science & Technology Committee
from ‘‘Time-saving strategy”. Furthermore, since green transporta- Research Project (15590501700). We also thank anonymous refer-
tion recently has been the mainstream in the development of global ees and the editor-in-chief.
economy, the ‘‘Energy-saving strategy” is more reasonable.
Based on the assumptions that the total departure delay and the References
total energy consumption are different in different container ter-
minals, the two objectives’ weights can be set at different values. [1] S. Dirk, V. Stefan, S. Robert, Container terminal operation and operations
research-a classification and literature review, OR Spectrum 26 (1) (2004) 3–
We use coefficient k to combine the two objectives: 49.
min min [2] S. Vob, R. Stahlbock, Operations research at container terminals: a literature
f1  f1 f2  f2 update, OR Spectrum 30 (1) (2008) 1–52.
min f ¼ k  max min
þ ð1  kÞ  max min
ð46Þ [3] C. Bierwirth, F. Meisel, A survey of berth allocation and quay crane scheduling
f1  f1 f2  f2 problems in container terminals, Eur. J. Oper. Res. 202 (3) (2009) 615–627.
[4] C. Bierwirth, F. Meisel, A follow-up survey of berth allocation and quay crane
In the objective function (46), the coefficient k could be set as scheduling problems in container terminals, Eur. J. Oper. Res. 244 (3) (2015)
0.1, 0.2, . . . , 1.0, with a step size of 0.1. This step could be adjusted 675–689.
manually according to the problem size and terminal operator’s [5] Y.M. Park, K.H. Kim, A scheduling method for berth and quay cranes, OR
Spectrum 25 (2003) 1–23.
preference. The solving procedure is conducted with k from 0.1 to
[6] K. Kim, K. Moon, Berth scheduling by simulated annealing, Transport. Res. Part
0.9, and the obtained solutions are from energy-saving oriented B: Methodol. 37 (2003) 541–560.
to time-saving oriented. In other words, when k is more approach- [7] B. Raa, W. Dullaert, R.V. Schaeren, An enriched model for the integrated berth
allocation and quay crane assignment problem, Expert Syst. Appl. 38 (2011)
ing 0.1, the solutions are more energy-saving. On the contrary, with
14136–14147.
k approaching 1.0, the solutions are more departure time-saving. [8] F. Meisel, C. Bierwirth, Heuristics for the integration of crane productivity in
the berth allocation problem, Transport. Res. Part E: Logist. Transport. Rev. 45
(1) (2009) 196–209.
8. Conclusions [9] Z.-H. Hu, Heuristics for solving continuous berth allocation problem
considering periodic balancing utilization of cranes, Comput. Ind. Eng. 85
By comparing with other research in this area, the major contri- (2015) 216–226.
[10] C.R. Zhang, L. Zheng, Z.H. Zhang, L.Y. Shi, A.J. Armstrong, The allocation of
bution of this paper is mainly reflected in the following three berths and quay cranes by using a sub-gradient optimization technique,
aspects: Comput. Ind. Eng. 58 (1) (2010) 40–50.
 J. He / Advanced Engineering Informatics 30 (2016) 390–405 405

[11] C.X. Yang, X.J. Wang, Z.F. Li, An optimization approach for coupling problem of [23] Y.Q. Du, Q.S. Chen, X.W. Quan, L. Long, R.Y.K. Fung, Berth allocation considering
berth allocation and quay crane assignment in container terminal, Comput. fuel consumption and vessel emissions, Transport. Res. Part E: Logist.
Ind. Eng. 63 (1) (2012) 243–253. Transport. Rev. 47 (6) (2011) 1021–1037.
[12] Y.-M. Fu, A. Diabat, I.-T. Tsai, A multi-vessel quay crane assignment and [24] S. Wang, Q. Meng, Z. Liu, A note on ‘‘Berth allocation considering fuel
scheduling problem: Formulation and heuristic solution approach, Expert Syst. consumption and vessel emissions”, Transport. Res. Part E: Logist. Transport.
Appl. 41 (15) (2014) 6959–6965. Rev. 49 (1) (2013) 48–54.
[13] Y.-M. Fu, A. Diabat, A Lagrangian relaxation approach for solving the integrated [25] J.F. Alvarez, T. Longva, E.S. Engebrethsen, A methodology to assess vessel
quay crane assignment and scheduling problem, Appl. Math. Model. 39 (2015) berthing and speed optimization policies, Marit. Econ. Logist. 12 (4) (2010)
1194–1201. 327–346.
[14] G. Giallombardo, L. Moccia, M. Salani, I. Vacca, Modeling and solving the [26] D.F. Chang, Z.H. Jiang, W. Yan, J.L. He, Integrating berth allocation and quay
tactical berth allocation problem, Transport. Res. Part B: Methodol. 44 (2) crane assignments, Transport. Res. Part E: Logist. Transport. Rev. 46 (6) (2010)
(2010) 232–245. 975–990.
_ Altınel, Optimal berth allocation and
[15] Y.B. Türkoğulları, Z.C. Tasßkın, N. Aras, I.K. [27] G. Chen, K. Govindan, M.M. Golias, Reducing truck emissions at container
time-invariant quay crane assignment in container terminals, Eur. J. Oper. Res. terminals in a low carbon economy: proposal of a queueing-based bi-objective
235 (1) (2014) 88–101. model for optimizing truck arrival pattern, Transport. Res. Part E: Logist.
[16] C. Iris, D. Pacino, S. Ropke, A. Larsen, Integrated berth allocation and quay crane Transport. Rev. 55 (2013) 3–22.
assignment problem: set partitioning models and computational results, [28] J.L. He, W.M. Zhang, Y.F. Huang, W. Yan, A simulation optimization method for
Transport. Res. Part E Logist. Transport. Rev. 81 (2015) 75–97. internal trucks sharing assignment among multiple container terminals, Adv.
[17] M. Rodriguez-Molins, M.A. Salido, F. Barber, A GRASP-based metaheuristic for Eng. Inform. 27 (4) (2013) 598–614.
the berth allocation problem and the quay crane assignment problem by [29] J.L. He, Y.F. Huang, W. Yan, Yard crane scheduling in a container terminal for
managing vessel cargo holds, Appl. Intell. 40 (2) (2014) 273–290. the trade-off between efficiency and energy consumption, Adv. Eng. Inform. 29
[18] P.F. Zhou, H.G. Kang, Study on berth and quay-crane allocation under (1) (2015) 59–75.
stochastic environments in container terminal, Syst. Eng. – Theory Pract. 28 [30] J.L. He, Y.F. Huang, W. Yan, Integrated internal truck, yard crane and quay
(1) (2008) 161–169. crane scheduling in a container terminal considering energy consumption,
[19] X.-L. Han, Z.-Q. Lu, L.-F. Xi, A proactive approach for simultaneous berth and Expert Syst. Appl. 42 (5) (2015) 2464–2487.
quay crane scheduling problem with stochastic arrival and handling time, Eur. [31] P. Schonfeld, O. Sharafeldien, Optimal berth and crane combinations in
J. Oper. Res. 207 (2010) 1327–1340. container ports, J. Waterway Port Coast. Ocean Eng. 111 (6) (1985) 1060–1072.
[20] M. Rodriguez-Molins, L. Ingolotti, F. Barber, M.A. Salido, M.R. Sierra, J. Puente, A [32] M.R. Garey, D.S. Johnson, Computers and Intractability: A Guide to the Theory
genetic algorithm for robust berth allocation and quay crane assignment, Prog. of NP-Completeness, Freeman & Co., New York, 1979.
Artif. Intell. 2 (4) (2014) 177–192. [33] D.E. Goldberg, Genetic Algorithms in Search, Optimization and Machine
[21] M.M. Golias, G.K. Saharidis, M. Boile, S. Theofanis, M.G. Ierapetritou, The berth Learning, Addison-Wesley, New York, 1989.
allocation problem: optimizing vessel arrival time, Marit. Econ. Logist. 11 (4) [34] M. Gen, R. Cheng, Genetic Algorithms and Engineering Design, John Wiley &
(2009) 358–377. Sons Inc., New York, 1997.
[22] N. Lang, A. Veenstra, A quantitative analysis of container vessel arrival [35] L. Zhen, L.H. Lee, E.P. Chew, A decision model for berth allocation under
planning strategies, OR Spectrum 32 (3) (2010) 477–499. uncertainty, Eur. J. Oper. Res. 212 (2011) 54–68.