You are on page 1of 16

Received October 10, 2020, accepted November 8, 2020, date of publication November 16, 2020,

date of current version November 27, 2020.


Digital Object Identifier 10.1109/ACCESS.2020.3037675

Optimizing the Stowage Planning and Block


Relocation Problem in Inland Container Shipping
JUN LI 1, YU ZHANG2 , ZHIXIONG LIU3 , AND XIAOLEI LIANG1 , (Member, IEEE)
1 School of Automobile and Traffic Engineering, Wuhan University of Science and Technology, Wuhan 430065, China
2 School of Logistics Engineering, Wuhan University of Technology, Wuhan 430063, China
3 School of Machinery and Automation, Wuhan University of Science and Technology, Wuhan 430065, China

Corresponding author: Jun Li (lj_whut1989@163.com)


This work was supported by the National Natural Science Foundation of China under Grant 71372202 and Grant 61603280.

ABSTRACT The stowage planning and block relocation problem (SPBRP) consider the relations between
stowage planning and block relocation comprehensively, in which the retrieval sequence of containers in
yard and the stowage plan of a ship are concerned. In this paper, an integer programming formulation
called SPBRP-I is presented for the proposed problem in inland container liner shipping to minimize the
total number of relocations in yard and re-handles in ship. To deal with the utilization preference in inland
shipping, the minimization of ship stack occupancy number is introduced into SPBRP-I to develop a new
model called SPBRP-II. In view of the complexity of problem-solving, a heuristic algorithm (HA) based on
priority sorting, retrieval and local search strategies is proposed to achieve the efficient optimization. The
algorithm consists of three parts including the yard container retrieving, yard container relocation and ship
stowage planning. Finally, the container liner shipping on the Yangtze River is chosen with some typical ships
for the case study. Numerical results show that SPBRP-II performs better than the original SPBRP-I, and
it’s more suitable for solving the small-sized cases. Compared with the models, the algorithm with retrieval
strategy RS1 (HA_RS1) can achieve the efficient problem solving with high-quantity solutions and very
short computational time.

INDEX TERMS Inland container shipping, stowage planning, block relocation, integer programming,
heuristic algorithm.

I. INTRODUCTION the container transport demands between different ports of


The national strategy of building China’s strength in trans- the fixed shipping lines.
portation highlights the status of inland waterway transporta- Nowadays, the duplicate construction and economic com-
tion system in the country. As the important link in connecting petitions between the inland container ports highlight the
the country’s inland and coastal economical trade, the inland ship owners’ dominant positions in container shipping along
waterway transportation plays an important role for the exter- the Yangtze River. Due to the surplus transport capacity in
nal trade of the urban agglomeration along the river. The fast inland shipping, ship owners prefer the capacity utilization in
developments of urban economies and cargo containeriza- stowage planning for enabling a lowering of transport costs.
tion have enabled the rapid development of inland container Meanwhile, typical inland container ships on the Yangtze
transport. Figure 1 shows the container shipping network of River are refitted by bulk-cargo ships. The physical charac-
the Yangtze River, where Chongqing, Wuhan, Nanjing, and teristics of the ship (i.e., small size and no hatch covers) leave
Shanghai are hub ports belonging to the trunk line of the river it fewer options for using ballast water to modify stability. It’s
(Li et al. [1]). The container transport between different ports more difficult to achieve the coordinating stability adjustment
of the Yangtze River is mainly undertaken by inland container and efficiently using full ship capacity.
liner shipping. It is organized by the liner company to meet Comparing with the seaports, the informatization construc-
tion of the terminals along the Yangtze River is in a relatively
lower level. The information of containers is mainly declared
The associate editor coordinating the review of this manuscript and by the cargo owners, which results in the inconsisten-
approving it for publication was Sun-Yuan Hsieh . cies between the actual and reported container information.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 8, 2020 207499
J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

FIGURE 1. The container shipping network of the Yangtze River.

The inconsistencies often make the manually made pre- in Section 6; details of computational experiments performed
stowage plan becoming unreasonable. Then the re-handles to assess optimization of models and algorithm are described
will be necessary for the ship’s stability. Re-handle is an extra in Section 7; and finally, conclusions and future work are
operation required to move a container or temporarily unload discussed in Section 8.
a container to gain access to the one stowed below it, which
is both time and money consuming. Also, the unreasonable II. LITERATURE REVIEW
plan may cause relocations in yard during the loading process. A. CONTAINER STOWAGE PLANNING PROBLEM
In order to retrieve a certain container in yard, all the contain- The container stowage planning problem (CSPP) consists
ers on its top in one stack must be relocated. This operation is of determining how to assign the containers in yard into
called the relocation which is an expensive and unproductive a ship. In general, the ship’s constraints of stability and
movement. capacity should be fulfilled to minimize its berthing time or
To meet the practical operation requirements of inland the number of re-handles in ship. State-of-the-art research
container shipping along the Yangtze River, the existing on this problem generally concerns the maritime container
pre-stowage planning problem could be overcome by making shipping and applies a two-phase approach. Due to the large
a stowage plan with the actual container information. Once size of ship in maritime shipping, the existing research which
all the containers have arrived in the yard of a terminal, their studied CSSP wholly formulates grossly abstracted model of
actual information becomes known. With the given containers the ship spaces (Dubrovsky et al. [2]; Ding and Chou [3]).
in the yard, how to assign these containers into the ship while Comparing with these monolithic models, the two-phase
considering their retrieval sequences becomes the key of approach could achieve better performing solutions, in which
stowage planning. It reveals that there is a close relationship the CSSP is divided into two sub-problems: master bay
between the stowage planning and block relocation prob- planning problem (MBPP) and slot planning problem (SPP)
lem in inland container shipping. Thus, the decision-making (Parreño et al. [4]). The two-phase approach originates from
problem which concerns the two problems comprehensively Wilson and Roach [5], [6]. Kang and Kim [7] designed a
should be considered. We refer to it as the stowage planning greedy heuristic for the MBPP and a tree search method for
and block relocation problem (SPBRP). the SPP.
In this paper, the following four aspects have been accom- Dealing with the MBPP, Ambrosino et al. [8], [9] for-
plished for optimizing the proposed problem: (a) An integer mulate a basic 0-1 linear programming model, and present
programming model (SPBRP-I) is constructed; (b) Consider- a heuristic procedure and a three-step heuristic tabu-based
ing the ship capacity utilization, a modified model SPBRP-II search. Sciomachen and Tanfani [10] relate the problem to
is proposed; (c) A heuristic algorithm (HA) is developed to a 3D bin packing problem (3D-BPP) and propose a heuris-
solve the SPBRP efficiently; (d) The effectiveness of mod- tic method inspired by the branch-and-bound algorithm.
els and HA is validated by a considerable of computational Moura et al. [11] study the MBPP with the ship routing
experiments based on realistic scenarios. problem to present a mixed integer programming model. The
The rest of the article is organized as follows: in Section 2, MBPP has been expanded to the multi-port master planning
the existing literature and related works are reviewed; The problem (MP-MBPP). Zhang et al. [12] study the CSSP based
details of SPBRP are described in Section 3; the mathematical on bin-packing problem and design the tabu search heuristic
models for SPBRP based on integer programming formula- algorithms for solving MP-MBPP and SPP. Pacino et al. [13]
tion is constructed in Section 4; the model solving method formulate an integer programming model of MB-MBPP, and
is introduced in Section 5; a heuristic algorithm is designed a combination of constraint programming model and local

207500 VOLUME 8, 2020


J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

search for the SPP. Ambrosino et al. [14]–[16] investigate As an NP-hard problem (Caserta et al. [25]), the basic
MP-MBPP with the mixed integer programming and heuristic version of BRP originates from Kim and Hong [26], in which
algorithms. a branch-and-bound algorithm for exact solving and a greedy
As for the SPP, Delgado et al. [17] focus on the con- heuristic for large-instances are developed to solve the prob-
tainer stowage problem for below deck locations (CSPBDL), lem. In Caserta et al. [25], two mixed integer programming
and the constraint programming and integer programming models called BRP-I and BRP-II are formulated separately.
are utilized to formulate the models. Parreño et al. [4] pro- The BRP-II follows assumption A1 corresponding to the
posed a novel integer programming formulation and a Greedy R-BRP and the BRP-I allows all possible relocations corre-
Randomized Adaptive Search Procedure (GRASP). The new sponding to the U-BRP.
algorithm can obtain high-quality solutions within 1 sec- For solving the R-BRP, Petering et al. [27] present a new
ond. With the constructive heuristic introduced in Parreño mixed integer programming formulation called BRP-III. A
et al. [4], Korach et al. [18] present an efficient ruin-and- look-ahead algorithm is designed to obtain better solutions
recreate matheuristic based on an integer programming model with shorter CPU time. Expósito-Izquierdo et al. [28] correct
and a heuristic search framework. the BRP-II from Caserta et al. [25] and present a branch-
Besides the CSSP in maritime shipping, only few works and-bound algorithm. Zehendner et al. [29] also present an
concern this problem in inland shipping. The CSSP in improved mathematical model with less variables to correct
inland container shipping involves different problems and the BRP-II from Caserta et al. [25]. Galle et al. [30] for-
objectives than maritime shipping (Li et al. [1]; Fazi [19]; mulate a binary integer programming model which could
Li et al. [20]). In Li et al. [1], the multi-port stowage planning significantly reduce the number of variables and constraints
problem along the Yangtze River is formulated as a stochastic compared to existing formulations. Based on the branch and
programming model considering the weight of the containers bound algorithm, Quispe et al. [31] implement an exact iter-
to be uncertain. They propose three solution methods includ- ative deepening A∗ algorithm using two new lower bounds.
ing two exact approaches and a hybrid neighborhood search Jin [32] investigates the relaxed restricted CRP and present
algorithm based on heuristics. Fazi [19] develop two novel an integer programming formulation to correct the model in
mathematical formulations in inland shipping considering the Galle et al. [30]. Bacci et al. [33] propose a branch-and-cut
ship’s stability. The first one aims to maximize the number algorithm with an integer linear programming formulation.
of containers on board, and the other one obtains a feasible Considering the difficulty in solving large-scale instance of
handling sequence of containers to ensure the stability during R-BRP, besides these exact methods (i.e., the mathematical
the loading/unloading process. Li et al. [20] investigate the formulation and exact algorithm), many studies focus on the
multi-port stowage planning problem through a multi-stage heuristic algorithm to obtain high-quality solutions with short
hierarchical decomposition approach, in which the problem computational time. Ku et al. [23] propose a search-space
is divided into the MP-MBPP and SPP. The mathematical reduction technique called the abstraction method. A search-
formulations and algorithms are proposed separately for the based algorithm for the solution procedure and the bidirec-
two sub-problems. tional search using the abstraction method are developed.
Zehendner et al. [34] investigate the online restricted CRP,
in which the information of container departures is partly
B. BLOCK RELOCATION PROBLEM known and the retrieval sequence of containers is revealed
The block relocation problem (BRP), also known as the con- over time. They introduce a leveling heuristic algorithm
tainer relocation problem (CRP), aims at finding an optimal based on the look-ahead horizon. In Ting et al. [35], a new
retrieval sequence of the containers according to their given novel heuristic is embedded in a beam search algorithm.
priorities, with the objective of minimizing the total number Bacci et al. [36] develop a bounded beam search algorithm
of relocations. Considering the relocatable containers, BRP with a new lower bound. Zhang et al. [22] design a novel
has been classified into two versions: one is the restricted method which integrates optimization methods and machine
block relocation problem (R-BRP), where the relocations learning techniques. Both the exact branch-and-bound algo-
only concern those containers on top of the container to rithm and the beam search method are injected with the
be retrieved next; Other is the unrestricted block relocation machine learning-driven upper bounds to enhance the perfor-
problem (U-BRP), where the relocations may concern con- mance for solving different scale instances. Zweers et al. [37]
tainers on top of any stack in yard (Feillet et al. [21]). The present a new problem called the stochastic container reloca-
BRP can also be divided by the priorities of containers: tion problem with pre-processing (SCRPPP) which is seen
the problem with distinct priorities and the one with dupli- as a combination of the restricted CRP and container pre-
cate priorities (Zhang et al. [22]). For solving the BRP, the marshalling problem. Both a heuristic method and a branch-
existing research mainly contains three methods: the math- and-bound algorithm are proposed for the SCRPPP.
ematical formulation, the exact algorithm, and the heuristic For solving the U-BRP, Tricoire et al. [38] design a
algorithm. For a more detailed classification and research of new constructive metaheuristic framework with some new
the BRP in the literature, the interested readers are referred to heuristic operations to quickly find high quality solutions.
Ku et al. [23] and Tanaka et al. [24]. Tanaka et al. [24] consider improving the efficiency of the

VOLUME 8, 2020 207501


J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

FIGURE 2. The Stowage planning and block relocation problem in inland shipping.

branch-and-bound algorithm with several dominance prop- To the best of our knowledge, this is the first time to present
erties to propose an exact algorithm. Feillet et al. [21] this specific problem in inland shipping.
propose the first local search type improvement heuris-
tic. Jovanovic et al. [39] present an ant colony optimiza-
tion (ACO) algorithm for U-BRP and R-BRP with the related III. PROBLEM DESCRIPTION
greedy algorithms. In this section, a formal description of SPBRP addressed in
To the best of our knowledge, there is only limited work this paper is presented. The SPBRP is inspired by the problem
dealing with the relationships between CSSP and BRP. Based faced by inland ports along the Yangtze River. In Figure 2,
on the given stowage plan of container ship, Ji et al. [40] the containers in yard are stacked into bays after arriving
develop a mathematical model and a genetic algorithm with the yard. Each bay contains some stacks, and each stack
the loading sequence and rehandling strategy considering the contains several tiers with a limited height. Similarly, the ship
multi-quay crane parallel operations during loading process. is divided into several sections called bays. Within each bay,
Tanaka and Voß [41] propose a branch-and-bound algorithm there are several stacks for stowing containers. Each stack
with iterative deepening for solving R-BRP and U-BRP. has several tiers and each tier contains one slot, with each
However, they both concern how to obtain the feasible slot representing a 1 TEU (Twenty-Foot Container Equivalent
retrieval order of containers with respect to the given stowage Unit) holding capacity.
plan, and lack the consideration of making stowage planning For the containers in yard, as shown in Figure 2, they are
decisions. retrieved one by one based on their retrieval sequences during
In conclusion, the related works mainly focus on the CSPP the loading process. If the current container is on the top of
or BRP separately and usually concerns maritime container its stack in yard, it can be retrieved directly to load aboard.
shipping. The inland container liner shipping has its particu- Otherwise, it can only be retrieved after the containers upon
larities, and the CSSP for it contains some different problems it being relocated. Therefore, SPBRP contains two types of
and objectives. Furthermore, the stowage plan of the ship in operations: a retrieval operation and a relocation operation.
inland shipping needs to be re-made with the actual container Specifically, the relocation operation corresponds to a R-BRP,
information in the yard. The CSSP and BRP should be con- and the retrieval operation corresponds to a CSSP. All the
sidered simultaneously to meet the practical requirements. retrieval and relocation operations could only be achieved
Thus, the existing methods for CSSP or BRP respectively on the top of stack either in yard or in ship. The relocation
lack the consideration of their integrated optimization and operation is permitted only within the given bays in yard,
the particularity in inland shipping, which makes them cannot in which those containers to be loaded aboard are stowed.
be directly utilized for this specific problem. Different from This article explores SPBRP in inland container shipping
the work in Ji et al. [40] and Tanaka and Voß [41], this along the Yangtze River. This problem concerns assigning
paper focuses on the stowage planning problem with block containers in yard into the stacks of ship at current port. The
relocations which needs to determine the retrieval sequence constraints of BRP in yard and ship’s constraints of stability,
of containers in yard and the stowage plan of a ship simul- strength, and capacity in CSSP must be fulfilled. Considering
taneously. This problem is referred as SPBRP, and it’s more the operation efficiency in container shipping, the objective
complex and suitable for the practical operations in inland aims to minimize the total number of relocations in yard
shipping. Our main aim is to formalize this problem with and re-handles in ship. The solution concerns the containers’
the integer programming, and develop a heuristic algorithm. retrieval sequences in yard and loading slots in ship. Also,

207502 VOLUME 8, 2020


J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

if some containers need to be relocated in yard, its relocation hmax : The largest stowing tier of yard stack, hmax ∈ H .
stack should be defined. (s, h) : The yard slot (s, h) at stack s and tier h, s ∈ S,
Due to the surplus transport capacity in shipping market h ∈ H.
and relatively limited capacity of ship size, the CSSP in (j, i) : The ship slot (j, i) at stack j and tier i, j ∈ J , i ∈ I .
inland shipping emphasizes the higher ship capacity utiliza- SWj : Maximum load weight of ship stack j (Unit: ton),
tion. Thus, the objective of minimizing the number of occu- j ∈ J.
pied stacks in ship is introduced into SPBRP. Keeping the STj : Maximum stacking tiers (or capacity limit) of ship
containers at current port clustered and using as few stacks stack j (Unit: TEU), j ∈ J .
as possible can help increase the ship capacity utilization and 1LG : Maximum longitudinal weight tolerance of ship
decrease re-handles at future ports in liner shipping. More- (Unit: ton).
over, the objective of minimizing re-handles in ship could 1CG : Maximum horizontal weight tolerance of ship
be modified by the constraint of avoiding re-handles, which (Unit: ton).
means the container with further destination port cannot stow 0-1 Binary Variables:
on another one in the same stack of ship. More detailed αksht : αksht = 1, if container k is stowed in yard slot (s, h)
descriptions can be found in Section 4.2. at stage t; otherwise αksht = 0. ∀k ∈ K , s ∈ S, h ∈ H , t ∈ T .
xksht : xksht = 1, if container k is retrieved from yard slot
IV. MATHEMATICAL MODELS (s, h) at stage t; otherwise xksht = 0. ∀k ∈ K , s ∈ S, h ∈ H ,
The integer programming formulation for SPBRP is con- t ∈ T.
structed based on following assumptions: yksht : yksht = 1, if container k is relocated into yard slot
(a) The process of planning is divided into different stages. (s, h) at stage t; otherwise yksht = 0. ∀k ∈ K , s ∈ S, h ∈ H ,
Each stage contains either a retrieval operation or a relocation t ∈ T.
operation. At the last stage, all the containers can be loaded zkjit : zkjit = 1, if container k is loaded aboard into ship
aboard. slot (j, i) at stage t; otherwise zkjit = 0. ∀k ∈ K , j ∈ J , i ∈ I ,
(b) Only the standard 20-ft containers are considered. t ∈ T.
(c) Each container in yard can be relocated at most once βji : βji = 1, if the container in ship slot (j, i) needs
for retrieving the target container. re-handle after all the containers have been loaded aboard;
(d) The stack is considered as the storage unit instead of otherwise βji = 0. ∀j ∈ J , i ∈ I .
bay both in yard and ship.
2) OBJECTIVE FUNCTION
A. SPBRP-I The mathematical model SPBRP-I based on 0-1 integer pro-
A typical inland container ship in the Yangtze River is shown gramming can be defined as follows. In this model, the stud-
in Figure 3 (Li et al. [1]). For the mathematical model con- ied objective includes the total number of relocations in yard
struction, the stacks in ship are numbered one by one from and re-handles in ship.
front to back, and from left to right. In Figure 3, the stacks XXXX XX
are divided into different sets namely front half, back half, min f = yksht + βji (1)
left half, and right half. t∈T k∈K s∈S h∈H j∈J i∈I
XX XX XX
xksht = yksht + zkjit ,
1) NOTATIONS
s∈S h∈H s∈S h∈H j∈J i∈I
Sets:
∀k ∈ K , t∈T (2)
T : Set of stages in the planning process. XXX
K : Set of containers. xksht ≤ 1, ∀t ∈ T (3)
D : Set of container destination ports. k∈K s∈S h∈H
XXX
S : Set of yard stacks. yksht ≤ 1, ∀t ∈ T (4)
H : Set of tiers for the yard stack. k∈K s∈S h∈H
XXX
J : Set of ship stacks, J = JF ∪ JA = JL ∪ JR . zkjit ≤ 1, ∀t ∈ T (5)
JF : Subset of ship stacks in front half of ship. k∈K j∈J i∈I
JA : Subset of ship stacks in back half of ship. XXX XXX
JL : Subset of ship stacks in left half of ship. xksh(t+1) ≤ xksht ,
k∈K s∈S h∈H k∈K s∈S h∈H
JR : Subset of ship stacks in right half of ship.
∀t ∈ T \{tmax } (6)
I : Set of tiers for the ship stack. X X
Parameters: αksht + yksht ≤ 1, ∀s ∈ S, h ∈ H , t ∈ T
t : The stage t, t ∈ T . k∈K k∈K
tmax : The last stage, T = {1, 2, . . . , tmax }. X X X X
(7)
k : The container k, k ∈ K . αks(h+1)t + xksht + yks(h+1)t ≤ αksht ,
wk : The weight of container k (Unit: ton), k ∈ K . k∈K k∈K k∈K k∈K
dk : The destination port of container k, dk ∈ D, k ∈ K . ∀s ∈ S, h ∈ H \{hmax }, t ∈ T (8)

VOLUME 8, 2020 207503


J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

FIGURE 3. The structure of a typical inland container ship.

X X X X
αksht + αks(h+1)t + xksht + yksht

X X X X XXXX
k∈K k∈K k∈K k∈K

wk zkjit − wk zkjit
X k∈K j∈JL i∈I t∈T k∈K j∈JR i∈I t∈T
+ yks(h+1)t ≤ 2,
≤ 1CG (21)
k∈K
∀s ∈ S, h ∈ H \{hmax }, t ∈ T (9) αksht , xksht , yksht ∈ {0, 1}, ∀k ∈ K , s ∈ S,
h ∈ H,
XX XX
yksht + xksh(t+1) ≤ 1, t∈T (22)
k∈K h∈H k∈K h∈H zkjit ∈ {0, 1}, ∀k ∈ K , j ∈ J , i ∈ I , t ∈ T (23)
∀s ∈ S, t ∈ T \{tmax } (10) βji ∈ {0, 1}, ∀j ∈ J , i ∈ I (24)
XX XX
xksht + yksh(t+1) ≤ 1,
k∈K h∈H k∈K h∈H In SPBRP-I, the objective function (1) minimizes total
∀s ∈ S, t ∈ T \{tmax } (11) number of relocations in yard and re-handles in ship. Con-
X straint (2) shows that the container can be either relocated
αkshtmax = 0, ∀s ∈ S, h ∈ H (12)
or loaded aboard after being retrieved at any stage. Con-
k∈K
XXX straint (3) restricts that at most one container can be retrieved
zkjit = 1, ∀k ∈ K (13) in yard at any stage. Constraints (4) and (5) represent at most
j∈J i∈I t∈T one container can be relocated or loaded aboard at any stage.
XX
zkjit ≤ 1, ∀j ∈ J , i ∈ I (14) Constraint (6) restricts all the retrieval operations should be
k∈K t∈T arranged without idle. Constraint (7) defines any yard slot
X 1 −1
X tX can at most stow one container at any stage. Constraint (8)
zkjit1 ≤ zkj(i−1)t , specifies that the container above the bottom tier is stowed on
k∈K k∈K t=1 another one in the yard stack, and the retrieval or relocation
∀j ∈ J , i ∈ I \{1}, t1 ∈ T \{1} (15) operation can only P be handled on Pthe top of yard stack.
Specifically, if α = 0, k∈K αks(h+1)t = 0 and
XXX
wk zkjit ≤ SWj , ∀j ∈ J (16) P k∈K ksht
k∈K i∈I t∈T
y
Pk∈K ks(h+1)t = 0 mean no container hanging
P in the air; if
α α
P
XXX
k∈K ksht = 1 and k∈K ks(h+1)t = 1, k∈K xksht = 0
zkjit ≤ STj , ∀j ∈ J (17)
means the container in yard slot (s, h) cannot be retrieved
k∈K i∈I t∈T P
XX and y
k∈K ks(h+1)t = 0 means the yard slot (s, h + 1)
βji ≤ zkjit , ∀j ∈ J , i ∈ I
P k∈K αksht = 1
(18)
P
cannot be relocated with any container; if
and k∈K αks(h+1)t = 0, k∈K xksht + k∈K yks(h+1)t ≤
P P
k∈K t∈T
P P P P
dk zkjit − dk zkj(i−1)t 1 means the container in yard slot (s, h) can be retrieved
k∈K t∈T k∈K t∈T
βji ≥ , or the yard slot (s, h + 1) can be relocated with anther
kDk container. Similarly, constraint (9) specifies that when the
∀j ∈ J , i ∈ I \{1} (19) upper and lower adjacent yard slots have been occupied,

both slots (s, h) and (s, h + 1) cannot be relocated with any

X X X X XXXX

w z
k kjit − w z
k kjit

container and the container in lower slot (s, h) cannot be
k∈K j∈JF i∈I t∈T k∈K j∈JA i∈I t∈T retrieved. Constraints (10) and (11) avoid the unless oper-

≤ 1LG (20) ations that one container cannot be retrieved and relocated

207504 VOLUME 8, 2020


J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

XX XX
within the same yard stack during two adjacent stages. dk zkjit ≤ dk zkj(i−1)t ,
Constraint (12) ensures all the container in yard can be loaded k∈K t∈T k∈K t∈T
aboard at the last stage. Constraint (13) represents each con- ∀j ∈ J , i ∈ I \{1} (29)
tainer in yard can be loaded into the ship. Constraint (14) γj ∈ {0, 1}, ∀j ∈ J (30)
shows each ship slot can at most stow one container. Con-
straint (15) ensures the container above the bottom tier needs In this mathematical model, f1 denotes the number of relo-
to be stowed on another container in any ship stack. Con- cations in yard, and f2 denotes the number of occupied stacks
straints (16) and (17) represent the capacity and load weight in ship. Constraint (28) defines the relationship between γj
constraints of each ship stack. Constraints (18) and (19) and zkjit . IfPthere are
P containers being stowed in ship stack j at
> 0 and γj = 1; otherwise
P
last stage, z
define
P the Prelationships between zP kjit andPβji . Specifically, k∈K i∈I t∈T kjit
γ
P P P
if 0, β 0; if z = 0 and =0. Constraint (29) ensures
z = = t∈T zkjit = 1 and
kjit ji k∈K i∈I t∈T kjit j
k∈K t∈T k∈K
P P P P
≤ 0, the objec- the container with further destination port cannot be stowed
k∈K t∈T dk zkjit − k∈K t∈T dk zkj(i−1)t
tive P
function (1) will force β 0; if
P P on top of the one with nearer destination port. Constraint (30)
ji = k∈K t∈T zkjit = 1
and k∈K t∈T dk zkjit − k∈K t∈T dk zkj(i−1)t > 0, βji =
P P P defines the value of variable.
1. The ||D|| in constraint (19) shows the number of elements Constraint (29) can ensure the solution of SPBRP-II with
in the set D which means the total number of container no re-handles, then the variable βji could be removed together
destination ports. The longitudinal and horizontal weight with constraints (18) and (19). Based on SPBRP-I and nota-
tolerances of ship are enforced by constraints (20) and (21) tions above, SPBRP is formulated as a new model SPBRP-II:
F = min(f1 , f2 ) : (26) ∼ (27), (2) ∼ (17),
 
for the ship stability (Ambrosino et al. [8]). Constraints (22)
to (24) define the values of variables. (SPBRP-II) .
(28) ∼ (29), (20) ∼ (24), (30)

B. SPBRP-II V. MODEL SOLVING METHOD


SPBRP-I lacks the consideration of keeping containers clus- For solving SPBRP-I and SPBRP-II, the value of last stage
tered in ship for the high capacity utilization. When the tmax should be defined in advance. The value of tmax is
number of containers to be loaded is small, the containers may directly related to the number of variables and constraints
be stowed dispersedly and occupy extra stacks in ship. This which can affect the solving effectiveness of proposed mod-
will bring negative effects on the loading/unloading process els. If tmax is set too small, some containers cannot be loaded
and stowage planning at subsequent ports in liner shipping. aboard. Thus, the models will not be solved as constraints (12)
As mentioned in Section 3, the objective of minimizing the and (13) cannot be satisfied. If tmax is set too large, it may
number of occupied stacks in ship is considered into SPBRP. bring some unless calculations for solving the models.
Based on SPBRP-I, this objective is added into the objective A solution method which updates the value of tmax
function to propose a new model SPBRP-II. It can achieve the iteratively is adopted to solve the models (Parreño-Torres
efficient capacity utilization of the occupied stack and reserve et al. [42]). Specifically, the value of tmax is first set as its
empty stacks as much as possible for the stowage planning at lower bound tlb , i.e., tmax = tlb . If the model can be solved
subsequent ports. correctly, the iterative optimization procedure ends and the
current solution is outputted. If the model cannot be solved,
the value of tmax is updated as tmax = tmax + 1 and the
1) NOTATIONS
iterative optimization procedure continues. The procedure
L : A large positive integer, suppose that L = 1000. repeats until the model can be solved with a solution or
γj : 0-1 binary variable. γj = 1, if there are containers being the tmax has reached its upper bound. Method 1 describes
stowed in ship stack j at last stage; otherwise γj = 0. ∀j ∈ J . the iterative optimization procedure for solving the proposed
models. In this method, the last stage tmax is the main con-
2) OBJECTIVE FUNCTIONS trolling parameter, it will be tuned step-by-step from its lower
The number of occupied stacks in ship is added into the bound to upper bound.
objective function, and the minimization of re-handles in ship Lemma 1: Let N be the number of containers to be loaded
is modified by the constraint of avoiding re-handles. A new in yard. Then, the value of lower bound tlb equals to N ,
mathematical model SPBRP-II can be defined as follows. i.e., tlb = N .
Proof: f1 is defined as the number of relocations in
min F = (f1 , f2 ) (25) yard in equation (26). Based on the assumption (a) and (c),
XXXX all the containers in yard can be loaded aboard at the last
f1 = yksht (26)
stage, and each stage contains either a retrieval operation or a
t∈T k∈K s∈S h∈H
X relocation operation. Thus, the value of last stage tmax equals
f2 = γj (27) to the total number of retrieval operations and relocation
j∈J operations. The retrieval operations correspond to load all
XXX
γj ≤ zkjit ≤ L · γj , ∀j ∈ J (28) the containers in yard aboard, which means each container
k∈K i∈I t∈T contains one retrieval operation. Therefore, the number of

VOLUME 8, 2020 207505


J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

Model 1 Solving Method for SPBRP-I and SPBRP-II retrieval and local search strategies. HA consists of three parts
Require: SPBRP-I, A mathematical model for SPBRP including the yard container retrieving, yard container relo-
SPBRP-II, A new mathematical model for SPBRP cation and ship stowage planning. Firstly, the yard container
tmax , The last stage retrieving defines the containers’ retrieval sequences and the
tlb , Lower bound of the last stage tmax target container at current stage. Then, for the container needs
tub , Upper bound of the last stage tmax to be relocated, the yard container relocation determines its
Container information in yard at current port relocation stack in yard. Finally, for the container that is
Ship information at current port retrieved to load aboard, the ship stowage planning defines
A mathematical programming solver Gurobi its stacking slot in ship.
Ensure: X : Solutin, F: Objective function value of solu-
tion
1: Initial set tmax = tlb A. YARD CONTAINER RETRIEVING
2: {X , F}←Solving SPBRP-I or SPBRP-II by using For the yard container retrieving, the current container to be
Gorubi retrieved namely the target container should be determined.
3: while SPBRP-I or SPBRP-II cannot be solved and The containers’ retrieval sequences are defined by the priority
tmax <= tub do sorting rules. The container with a higher priority could be
4: tmax ← tmax + 1 retrieved earlier. Based the practical operation rules and the
5: {X , F}←Solving SPBRP-I or SPBRP-II by using preference of avoiding re-handles in ship, the yard container
Gorubi retrieving is defined based on priority sorting rules below.
6: if SPBRP-I or SPBRP-II can be solved then (a) The container with further destination port has a higher
7: {X , F}←Solving SPBRP-I or SPBRP-II priority.
8: end while (b) For the containers with same destination ports, the one
9: end if which will cause no relocations or smaller number of reloca-
10: end while tions has a higher priority.
11: return {X , F} (c) If there are more than one container without causing
relocations, the one which stowing closer to the previous
operation stack in yard has a higher priority.
(d) If there are more than one container with smallest
retrieval operations equals to the number P
of containers.
P P P Then, number of relocations, the one which has the container with
we have tmax = f1 + N , where f1 = yksht .
t∈T k∈K s∈S h∈H furthest destination port stowed above gets a higher priority.
Considering yksht is an 0-1 binary variable and yksht ≥ 0(∀k ∈ Following the rules above, if there are containers stowing
K , s ∈ S, h ∈ H , t ∈ T ), f1 ≥ 0. Under the best situation, the upon the target container at current stage, the target container
relocation operations can be avoided in yard, and the value of will be updated as the one stowing on the top of same stack
f1 equals to its theoretical lower bound, i.e., f1 = 0. Therefore, in yard. In this situation, the target container could be either
tmax = f1 + N ≥ N and tlb = N . relocated in yard or loaded aboard corresponding to two
For the model solving, the upper bound of f1 can be simply different retrieval strategies.
set as the number of containers minus the one on top of stacks Retrieval strategy 1 (RS1): the target container is retrieved
in yard based on assumption (c). The number of containers on to load aboard for avoiding the relocation in yard, which
top of stacks in yard is denoted as M , then upper bound of f1 means the procedure of algorithm goes to the ship stowage
equals to N − M . Therefore, tmax = f1 + N ≤ N − M + N planning and continues.
and tub = 2N − M . Retrieval strategy 2 (RS2): the target container is relocated
to anther stack in yard, which means the procedure of algo-
VI. HEURISTIC ALGORITHM rithm goes to the yard container relocation and continues.
The existing algorithms are mostly developed for solving the
CSPP or BRP separately and mainly concerns maritime con-
tainer shipping. They cannot be directly utilized to properly B. YARD CONTAINER RELOCATION
optimize the SPBRP in inland shipping. For solving the CSPP For the yard container relocation, the relocation stack for
or BRP, the solving methods focus on the design of heuris- target container is determined preferentially without causing
tic algorithms considering their NP-hardness. The effective new relocations in yard. The number of containers, which
heuristic algorithm can achieve the high-quality solutions have higher priorities than the target container in stack s,
within very short time. is denoted by ms , ∀s ∈ S. The rules of yard container
Due to the particularities of inland container shipping, relocation for choosing the best relocation stack are presented
the CSSP for it contains some different problems and objec- as follows:
tives. Thus, a heuristic algorithm (HA) is proposed based (a) The empty stack is chosen first. If there are more than
on the previous study in Li et al. [1] and some reloca- one empty stack, the one closer to the previous operation
tion strategies in BRP, which contains some priority sorting, stack in yard should be chosen.

207506 VOLUME 8, 2020


J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

(b) If there are no empty stacks, the stack with ms = 0 is


chosen. Similarly, if there are more than one stack, the one
closer to the previous operation stack is chosen first.
(c) If there are no stacks with ms = 0, the one with the
smallest value of ms is chosen first. Similarly, if there are
more than one stack, the one closer to the previous operation
stack would be chosen.

C. SHIP STOWAGE PLANNING FIGURE 5. Swapping all the containers in two different stacks in ship.
For the ship stowage planning, the constraints of ship stabil-
ity, strength, and capacity should be fulfilled to define the
stowing slots for containers in ship. The ship stowage plan-
ning is achieved by the following stowage rule considering
minimizing the number of occupied stacks in ship. When the
target container is retrieved to load aboard, the stacks in ship
are examined in an ascending order to choose the first one
satisfying the following conditions.
(a) The stack has at least one empty slot;
(b) The load weight constraint of stack could still be
ensured with the target container;
(c) If the stack is not empty, the target container should
have the nearest destination port comparing to these already
in this stack, or the same destination port with the nearest one;
(d) If the condition (c) cannot be satisfied, the first empty
stack in ship is chosen. If there are no empty stacks, the one
first satisfying both conditions (a) and (b) would be chosen.
In some cases, the stowage plan from the rule may not
fulfill the stability constraints of the ship. Two different local
search strategies for modifying the longitudinal and horizon-
tal weight tolerances of ship are presented, i.e., LS1 and LS2
(Li et al. [1]). In LS1, all the containers in one stack are
moved into another empty one in ship in Figure 4. In LS2,
all the containers in two different stacks are swapped in ship
in Figure 5.

FIGURE 6. The flowchart of proposed HA.

The pseudo code of HA is shown in detail in Algorithm 1.


The three parts of HA are implemented one by one. The
yard container retrieving ends until the containers’ retrieval
sequences have been defined. For the current target container,
the yard container relocation ends until its best relocation
FIGURE 4. Moving all the containers in the stack to another empty one in
ship.
yard stack has been determined. The ship stowage planning
ends until the best stowing slot for the container in ship has
Both LS1 and LS2 cannot change the stowing states been found. After all the containers have been loaded aboard,
of containers in one stack, and the stacks are replaceable the ship’s stability will be checked. Otherwise, the two local
to each other in view of the ship’s rectangular structure search strategies are implemented one by one until the ship’s
(See Figure.3). Thus, the containers’ retrieval sequences in stability has been modified within its limit or it cannot be
yard will not be influenced by the strategies. If the stowage further improved.
plan cannot fulfill the ship’s stability constraints, LS1 and
LS2 are implemented one by one until the ship’s stability is VII. COMPUTATIONAL EXPERIMENTS
within its limit. Otherwise, each strategy is implemented until A. EXPERIMENTAL SETUP
the ship’s stability cannot be further improved. The flowchart Considering the inland container shipping along the Yangtze
of proposed HA is shown in Figure 6. River, three typical inland container ships are listed for the

VOLUME 8, 2020 207507


J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

Algorithm 1 HA for SPBRP TABLE 1. Information of container ships on the Yangtze River.
Require: Container information in yard at current port
Ship information at current port
K , Set of containers to be loaded in yard
||S||, Number of yard stacks
||J ||, Number of ship stacks
k, The container k, k ∈ K
dk , The destination port of container k, k ∈ K
qk , The retrieval sequences of containers in yard containers in yard are considered in Table 2. The instances
rk , The relocation plan of container k, k ∈ K are denoted as S1_1, S1_2 and so on. In table 2, S1_1 to
pk , The stowage plan of container k, k ∈ K S1_4 show the four instances including S1_1, S1_2, S1_3,
RS1, Retrieval strategy 1 and S1_4; and {3,4,5,6} shows the number of destination
RS2, Retrieval strategy 2 ports corresponding to these four instances successively.
Ensure: X1 : Solution, F1 : Objective function value of X1
1: for all k ∈ K B. COMPUTATIONAL RESULTS
2: qk ← Yard container retrieving (dk ) The proposed SPBRP-I and SPBRP-II are solved by using
3: end Gurobi with Method 1. Following this model solving
4: for all k ∈ K method, the CPU time limit is first set at 600 s under each
5: Update the container k with smallest qk as the tmax . If the model cannot be optimally solved, it is extended
target container to 3600 s and the first feasible solution is output.
6: if the target container is stowing on the top of yard stack For the two objectives in SPBRP-II, the hierarchical
7: pk ← Ship stowage planning (dk , ||J ||) approach is utilized in Gurobi. It assigns a priority to each
8: end if objective and optimizes for the objectives in a decreasing
9: if the target container isn’t stowing on the top of priority order. The priorities of f1 and f2 are set as 1 and
yard stack 0 separately. At each step, the approach finds the best solution
10: Update the container k stowing on the top of same for the current objective, but only from among those that
stack as the target container would not degrade the solution quality for higher-priority
11: if the strategy RS1 is chosen objectives (https://www.gurobi.com/documentation/ 8.1/ref-
12: rk ← Yard container relocation (dk , ||S||) man/ working_with_multiple_obje.html).
13: end if The algorithms with two different retrieval strategies (i.e.,
14: if the strategy RS2 is chosen RS1 and RS2) are denoted as HA_RS1 and HA_RS2 respec-
15: pk ← Ship stowage planning (dk , ||J ||) tively. The only difference between HA_RS1 and HA_RS2 is
16: end if the retrieval strategy and all the other setting of them keeps the
17: end if same. They are programmed in Python 3.6. All computational
18: end experiments are run on an Intel Core I7-5500U 2.40 GHz
19: for all k ∈ K processor with 4 GB of RAM. The effectiveness of proposed
20: Solution X1 ← {qk , rk , pk } models and algorithms is validated by a considerable of com-
21: end putational experiments. Based on the test instances in Table 2,
22: if the ship’s stability is not within its limit the solution quality, solving quantity, and computational time
23: repeat are concerned for the comparison of different methods.
24: {X1 , F1 }←Update {X1 , F1 } with RS1 and RS2 In view of the management in container terminal, how
25: until X1 cannot be improved to reduce the berthing time of ships is very important for
26: end if improving its operational efficiency. An effective way of
27: return {X1 , F1 } speeding up the loading/unloading operations of ships at the
container terminal is to use the idle time before the arrival of
a ship for sorting the containers in yard in advance. Thus,
experiments according to the Chinese national standard GB/T two different scenarios are concerned for the comparison
19283-2010. In Table 1, the detailed information of each ship of algorithms by considering whether the pre-marshalling
is listed corresponding to the notations in the mathematical operations have been implemented in advance.
models.
For the containers to be loaded in yard, 4 different number 1) SCENARIO 1
of destination ports are set. Considering the general opera- Under this scenario, all the containers are stowed in yard fol-
tional height limitation of the common yard crane, the max- lowing PSCW rule in advance, i.e., the containers were clus-
imum stacking tier in yard is set as 4 to 5. Here, a series tered into different stacks considering their characteristics of
of test instances based on these ships are designed as listed destination port (P), size (S), category (C), and weight (W).
in Table 2. For each type of ship, 3 different layouts of For different types of ships, the results of different models

207508 VOLUME 8, 2020


J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

TABLE 2. Experimental instances of ships on the Yangtze River.

TABLE 3. Results of different models and algorithms for ship S1 under scenario 1.

TABLE 4. Results of different models and algorithms for ship S2 under scenario 1.

and algorithms are shown in Tables 3-5. In Tables 3-5, f1 of re-handles in ship; T0 represents the CPU time (unit: s); and
represents the number of relocations in yard; f2 represents the the ‘‘-’’ represents that a model could not be solved within the
number of occupied stacks in ship; f3 represents the number time limit.

VOLUME 8, 2020 207509


J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

TABLE 5. Results of different models and algorithms for ship S3 under scenario 1.

P
In Tables 3-5, due to the PSCW rule in yard, the con- k∈K wk means the total weight of containers to be loaded
tainers could be retrieved to loaded aboard without any in yard; SW is the maximum load weight of the stack in
relocations. For the models, SPBRP-I and SPBRP-II found ship as listed in Column 6 of Table 1; and d e rounds up the
the best solutions without relocations in yard by setting value inside. The f2 min can be used as the lower bound of f2
tmax = tlb when they were solved within the time limit. because it only concerns the capacity constraint of each stack
The solution of SPBRP-II always had a bigger number of in ship and ignores the re-handles and stability constraints of
occupied stacks (i.e., f2 ) than the solution of SPBRP-I. For the ship.
the number of relocations in yard (i.e., f1 ) and the num- HA_RS1 and HA_RS2 could always find the lower bounds
ber of re-handles in ship (i.e., f3 ), SPBRP-I and SPBRP-II of f1 and f3 in Tables 3-5. In order to further verify
always found the best results when they could be solved the effectiveness of the algorithms, the results of f2 from
within the time limit of 600 s. SPBRP-II always took more HA_RS1 and HA_RS2 were compared with its lower bound
CPU tome since it considered no re-handles into constraints. f2 min in Table 6. In the table, gap represents the gap in
In Tables 4 and 5, SPBRP-I and SPBRP-II could not guar- units of f2 compared with f2 min from HA_RS1 and HA_RS2
antee a feasible solution within 3600 s for some instances (unit: %). In Table 6, the results of f2 from HA_RS1 and
with the rapid growing number of variables and constraints. HA_RS2 were very close to its lower bound. For all the
Overall, although SPBRP-II may spend more computational instances, the average results of HA_RS1 and HA_RS2 had
time than SPBRP-I, it always found a better solution. a small gap with the value of lower bound.
In Tables 3-5, the results of HA_RS1 and HA_RS2 were To make the results statistically more convincing, the
same while the CPU time for some instances may be little dif- Kruskal-Wallis ANOVA method with the significance level
ferent. Comparing with SPBRP-I and SPBRP-II, HA mainly of 0.05 is utilized to analyze the difference of the models and
contains some rules and strategies. Thus, HA_RS1 and algorithms. Firstly, as the models SPBRP-I and SPBRP-II
HA_RS2 took much less CPU time than the proposed mod- cannot find feasible solutions for some instances (e.g., ship
els, and they could solve all instances within an average S3), only the instances which can be solved by all the models
time of 0.16 s. For the objectives of SPBRP, HA_RS1 and and algorithms simultaneously are used for the statistical
HA_RS2 performed better than SPBRP-I and had very small tests. Secondly, the lower bounds of f1 and f3 are both 0,
differences with SPBRP-II. Overall, although SPBRP-II and the results of them from different methods are mostly
could find a better number of occupied stacks in ship than 0. Thus, the test method is mainly applied for f2 to analyze
the algorithms, the differences were quite small and the algo- the difference of the models and algorithms as shown in
rithms significantly reduced the computational time. Figure 7. It is obvious that there are significant differences
Obviously, the lower bound of f1 and f3 are both 0 when from the perspective of f2 among the models and algorithms
considering them separately. The lower bound of f2 could be since the p-values is less than 0.05. For those test instances,
obtained through the following definition: the SPBRP-II, HA_RS1 and HA_RS2 achieve better results
n lX mo of f2 than the SPBRP-I, which is consistent with the result
f2 min = max dN /H e , wk /SW (31)
k∈K analysis from Tables 3-5.
where f2 min is the lower bound of f2 ; N represents the total In general, when the containers in yard following PSCW
number of containers to be loaded; H is the number of tiers rule, SPBRP-II could be utilized to optimize SPBRP for
for the stack of each ship as listed in Column 4 of Table 1; small ship S1. For the bigger ships S2 and S3, HA_RS1 and

207510 VOLUME 8, 2020


J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

TABLE 6. Results of f2 from HA_RS1 and HA_RS2 comparing with f2 min under scenario 1.

the number of relocations in yard (i.e., f1 ). HA_RS1 always


found a better solution for optimizing f1 . Due to the extra
relocations in HA_RS2, its results of f2 and f3 slightly out-
performed the ones from HA_RS1(e.g., S1_2, S1_4, S2_6,
S3_5, S3_7). In general, HA_RS1 and HA_RS2 took much
less CPU time than the proposed models and the differ-
ences between algorithms were quite small. For the objec-
tive optimizations, HA_RS1 performed more balanced than
HA_RS2. HA_RS1 could effectively optimize the different
objectives at the same time. Comparing with SPBRP-I and
SPBRP-II, the algorithms had great advantages in compu-
tational time and solving quantity. For the small ship S1,
SPBRP-II performed better than the algorithms, but the dif-
ferences between it and HA_RS1 were very little. For the
ship S2 and S3, HA_RS1 had a much better performance than
FIGURE 7. Kruskal-Wallis ANOVA significance test results of f2 for the
models and algorithms under scenario 1. SPBRP-II.
In Tables 7-9, HA_RS1 could always find the lower bounds
of f1 . For the ship S1 and S2, HA_RS1 found the lower
HA_RS2 have better performances and always obtain good bounds of f3 . For the ship S3, only few results (e.g., S3_7,
solutions efficiently. S3_10, and S3_12) from HA_RS1 had quite small differences
with the lower bounds. Like the results in Table 6, the result
2) SCENARIO 2 of f2 from HA_RS1 was compared with its lower bound f2 min
Under this scenario, all the containers to be loaded aboard in Table 10. The results of f2 from HA_RS1 were very close
in yard don’t follow the PSCW rule, i.e., the containers to its lower bound. An average result from HA_RS1 had a
are stowed into stacks in yard according to their arriving reasonable gap with the value of lower bound.
sequences without any pre-marshalling operations. For dif- Like the test result in Figure 7, the same Kruskal-Wallis
ferent ships, the results of different models and algorithms ANOVA method is utilized under scenario 2. Figure 8
are shown in Tables 7-9. The definitions in Tables 7-9 have presents the test result of f2 for the models and algorithms.
the same meanings with the ones in Tables 3-5. There are significant differences among the models and
In Tables 7-9, the results from HA_RS1 show that all algorithms since the p-values is less than 0.05. For those
the instances can be optimized with no relocations in yard, test instances, the performances of SPBRP-II, HA_RS1 and
which means the setting of tmax = tlb should be suitable for HA_RS2 are better than the SPBRP-I, which is consistent
solving the mathematical models. Although the containers with the previous result analysis.
in yard didn’t follow PSCW rule, SPBRP-I and SPBRP-II In conclusion, for the computational experiments under
found solutions without relocations in yard. Like the results two typical scenarios, SPBRP-I, SPBRP-II and HA_RS1 can
under Scenario 1, SPBRP-II performed better than SPBRP-I find the solutions without relocations in yard. For the con-
for optimizing the number of occupied stacks and re-handles sideration of high ship capacity utilization, SPBRP-I has
in ship (i.e., f2 and f3 ). the worst results. With the increasing of ship size, SPBRP-I
For HA_RS1 and HA_RS2, the results from them were dif- and SPBRP-II cannot guarantee a solution within the time
ferent owing to the different retrieval strategies, especially for limit. Especially for the ship S3, the models cannot find

VOLUME 8, 2020 207511


J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

TABLE 7. Results of different models and algorithms for ship S1 under scenario 2.

TABLE 8. Results of different models and algorithms for ship S2 under scenario 2.

TABLE 9. Results of different models and algorithms for ship S3 under scenario 2.

a feasible solution for each instance. HA_RS1 outperforms relocations in yard. Thus, SPBRP-II can be utilized for small
the models and HA_RS2 since it always achieves a good ship like S1; HA_RS1 can effectively optimize SPBRP for all
solution for all objectives of SPBRP efficiently with no ships.

207512 VOLUME 8, 2020


J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

TABLE 10. Results of f2 from HA_RS1 comparing with f2 min under be considered into the problem at each port. The container
scenario 2.
ship stowage planning will be linked to the actual stability
formulas like the shear force, bending moment, trim and so
on, which is closer to the reality operation. Furthermore,
some larger ships which can sail through the lower reaches
of the Yangtze River to the sea terminals will be consid-
ered for further verifying the proposed methods, especially
for the algorithm. Some machine learning techniques will
be inserted into the algorithm, such as clustering analysis
(CA), opposition-based learning (OBL), and reinforcement
learning.

REFERENCES
[1] J. Li, Y. Zhang, J. Ma, and S. Ji, ‘‘Multi-port stowage planning for inland
container liner shipping considering weight uncertainties,’’ IEEE Access,
vol. 6, pp. 66468–66480, Nov. 2018.
[2] O. Dubrovsky, G. Levitin, and M. Penn, ‘‘A genetic algorithm with a
compact solution encoding for the container ship stowage problem,’’
J. Heuristics, vol. 8, pp. 585–599, Nov. 2002.
[3] D. Ding and M. C. Chou, ‘‘Stowage planning for container ships:
A heuristic algorithm to reduce the number of shifts,’’ Eur. J. Oper. Res.,
vol. 246, no. 1, pp. 242–249, Oct. 2015.
[4] F. Parreño, D. Pacino, and R. Alvarez-Valdes, ‘‘A GRASP algorithm for
the container stowage slot planning problem,’’ Transp. Res. E, Logistics
Transp. Rev., vol. 94, pp. 141–157, Oct. 2016.
[5] I. D. Wilson and P. A. Roach, ‘‘Principles of combinatorial optimization
applied to container-ship stowage planning,’’ J. Heuristics, vol. 5, no. 4,
pp. 403–418, 1999.
[6] I. D. Wilson and P. A. Roach, ‘‘Container stowage planning: A method-
ology for generating computerized solutions,’’ J. Oper. Res. Soc., vol. 51,
no. 11, pp. 1248–1255, Nov. 2000.
[7] J.-G. Kang and Y.-D. Kim, ‘‘Stowage planning in maritime container
transportation,’’ J. Oper. Res. Soc., vol. 53, no. 4, pp. 415–426, Apr. 2002.
[8] D. Ambrosino, A. Sciomachen, and E. Tanfani, ‘‘Stowing a containership:
FIGURE 8. Kruskal-Wallis ANOVA significance test results of f2 for the The master bay plan problem,’’ Transp. Res. A, Policy Pract., vol. 38, no. 2,
models and algorithms under scenario 2. pp. 81–99, Feb. 2004.
[9] D. Ambrosino, D. Anghinolfi, M. Paolucci, and A. Sciomachen, ‘‘A new
three-step heuristic for the master bay plan problem,’’ Maritime Econ.
Logistics, vol. 11, no. 1, pp. 98–120, Mar. 2009.
VIII. CONCLUSION
[10] A. Sciomachen and E. Tanfani, ‘‘A 3D-BPP approach for optimising
Considering the practical operations in inland container ship- stowage plans and terminal productivity,’’ Eur. J. Oper. Res., vol. 183, no. 3,
ping, this article investigates a new problem called SPBRP. pp. 1433–1446, Dec. 2007.
The optimization methods including mathematical modeling [11] A. Moura, J. Oliveira, and C. Pimentel, ‘‘A mathematical model for the
container stowage and ship routing problem,’’ J. Math. Model. Algorithms
and heuristic algorithm are proposed. Following the prob- Oper. Res., vol. 12, no. 3, pp. 217–231, Sep. 2013.
lem description of SPBRP, the corresponding mathematical [12] Z. Wei-ying, L. Yan, and J. Zhuo-shang, ‘‘Model and algorithm for con-
model SPBRP-I is formulated to minimize the total num- tainer ship stowage planning based on bin-packing problem,’’ J. Mar. Sci.
Appl., vol. 4, no. 3, pp. 30–36, Sep. 2005.
ber of relocations in yard and re-handles in ship. For the [13] D. Pacino, A. Delgado, R. M. Jensen, and T. Bebbington, ‘‘Fast generation
high ship capacity utilization, SPBRP-I is then modified of near-optimal plans for eco-efficient stowage of large container vessels,’’
into a new model SPBRP-II. In view of the problem com- in Proc. Int. Conf. Comput. Logistics (ICCL): Comput. Logistics. Berlin,
Germany: Springer, 2011, pp. 286–301.
plexity, the heuristic algorithm contains different retrieval
[14] D. Ambrosino, M. Paolucci, and A. Sciomachen, ‘‘Experimental evalua-
strategies is proposed. The computational results revealed tion of mixed integer programming models for the multi-port master bay
that SPBRP-II performs better than SPBRP-I; HA_RS1 out- plan problem,’’ Flexible Services Manuf. J., vol. 27, nos. 2–3, pp. 263–284,
performs HA_RS2 especially when the containers in yard Sep. 2015.
[15] D. Ambrosino, M. Paolucci, and A. Sciomachen, ‘‘A MIP heuristic for
don’t follow PSCW rule. Considering the optimization effi- multi port stowage planning,’’ Transp. Res. Procedia, vol. 10, pp. 725–734,
ciency, SPBRP-II is more suitable to find exact solutions Jul. 2015.
for small-sized cases, and HA_RS1 can effectively find high [16] D. Ambrosino, M. Paolucci, and A. Sciomachen, ‘‘Computational eval-
uation of a MIP model for multi-port stowage planning problems,’’ Soft
quality solutions for SPBRP. Comput., vol. 21, no. 7, pp. 1753–1763, Apr. 2017.
In future research, we will expand SPBRP to a multi-port [17] A. Delgado, R. M. Jensen, K. Janstrup, T. H. Rose, and K. H. Andersen,
problem during the inland container liner shipping with some ‘‘A constraint programming model for fast optimal stowage of container
vessel bays,’’ Eur. J. Oper. Res., vol. 220, no. 1, pp. 251–261, Jul. 2012.
larger ships. The ship’s whole route should be investigated to
[18] A. Korach, B. D. Brouer, and R. M. Jensen, ‘‘Matheuristics for slot
make the plans for multiple ports. Thus, the containers which planning of container vessel bays,’’ Eur. J. Oper. Res., vol. 282, no. 3,
have been already loaded aboard in previous ports should pp. 873–885, May 2020.

VOLUME 8, 2020 207513


J. Li et al.: Optimizing the Stowage Planning and Block Relocation Problem in Inland Container Shipping

[19] S. Fazi, ‘‘A decision-support framework for the stowage of maritime con- [41] S. Tanaka and S. Voß, ‘‘An exact algorithm for the block relocation problem
tainers in inland shipping,’’ Transp. Res. E, Logistics Transp. Rev., vol. 131, with a stowage plan,’’ Eur. J. Oper. Res., vol. 279, no. 3, pp. 767–781,
pp. 1–23, Nov. 2019. Dec. 2019.
[20] J. Li, Y. Zhang, S. Ji, L. Zheng, and J. Xu, ‘‘Multi-stage hierarchical [42] C. Parreño-Torres, R. Alvarez-Valdes, and R. Ruiz, ‘‘Integer programming
decomposition approach for stowage planning problem in inland container models for the pre-marshalling problem,’’ Eur. J. Oper. Res., vol. 274, no. 1,
liner shipping,’’ J. Oper. Res. Soc., vol. 71, no. 3, pp. 381–399, Mar. 2020. pp. 142–154, Apr. 2019.
[21] D. Feillet, S. N. Parragh, and F. Tricoire, ‘‘A local-search based heuristic for
the unrestricted block relocation problem,’’ Comput. Oper. Res., vol. 108,
pp. 44–56, Aug. 2019.
[22] C. Zhang, H. Guan, Y. Yuan, W. Chen, and T. Wu, ‘‘Machine learning-
driven algorithms for the container relocation problem,’’ Transp. Res. B, JUN LI received the Ph.D. degree from the School
Methodol., vol. 139, pp. 102–131, Sep. 2020. of Logistics Engineering, Wuhan University of
[23] D. Ku and T. S. Arthanari, ‘‘On the abstraction method for the container Technology, Wuhan, China, in 2019. He is cur-
relocation problem,’’ Comput. Oper. Res., vol. 68, pp. 110–122, Apr. 2016. rently a Teacher with the School of Automobile
[24] S. Tanaka and F. Mizuno, ‘‘An exact algorithm for the unrestricted block
and Traffic Engineering, Wuhan University of Sci-
relocation problem,’’ Comput. Oper. Res., vol. 95, pp. 12–31, Jul. 2018.
[25] M. Caserta, S. Schwarze, and S. Voß, ‘‘A mathematical formulation and
ence and Technology, Wuhan. His current research
complexity considerations for the blocks relocation problem,’’ Eur. J. Oper. interests include the mathematical modeling and
Res., vol. 219, no. 1, pp. 96–104, May 2012. optimization algorithms in operation research.
[26] K. H. Kim and G.-P. Hong, ‘‘A heuristic rule for relocating blocks,’’
Comput. Oper. Res., vol. 33, no. 4, pp. 940–954, Apr. 2006.
[27] M. E. H. Petering and M. I. Hussein, ‘‘A new mixed integer program and
extended look-ahead heuristic algorithm for the block relocation problem,’’
Eur. J. Oper. Res., vol. 231, no. 1, pp. 120–130, Nov. 2013.
[28] C. Expósito-Izquierdo, B. Melián-Batista, and J. M. Moreno-Vega, YU ZHANG received the Ph.D. degree from the
‘‘An exact approach for the blocks relocation problem,’’ Expert Syst. Appl., School of Logistics Engineering, Wuhan Univer-
vol. 42, nos. 17–18, pp. 6408–6422, Oct. 2015. sity of Technology, Wuhan, China, in 2007. He is
[29] E. Zehendner, M. Caserta, D. Feillet, S. Schwarze, and S. Voß, currently a Professor with the Wuhan University of
‘‘An improved mathematical formulation for the blocks relocation prob- Technology. His research interests include supply
lem,’’ Eur. J. Oper. Res., vol. 245, no. 2, pp. 415–422, Sep. 2015. chain management, optimization, and the develop-
[30] V. Galle, C. Barnhart, and P. Jaillet, ‘‘A new binary formulation of the ment of mathematical modeling and algorithms.
restricted container relocation problem based on a binary encoding of
configurations,’’ Eur. J. Oper. Res., vol. 267, no. 2, pp. 467–477, Jun. 2018.
[31] K. E. Y. Quispe, C. N. Lintzmayer, and E. C. Xavier, ‘‘An exact algorithm
for the blocks relocation problem with new lower bounds,’’ Comput. Oper.
Res., vol. 99, pp. 206–217, Nov. 2018.
[32] B. Jin, ‘‘On the integer programming formulation for the relaxed restricted
container relocation problem,’’ Eur. J. Oper. Res., vol. 281, no. 2,
pp. 475–482, Mar. 2020. ZHIXIONG LIU received the Ph.D. degree from
[33] T. Bacci, S. Mattia, and P. Ventura, ‘‘A branch-and-cut algorithm for the the School of Logistics Engineering, Wuhan Uni-
restricted block relocation problem,’’ Eur. J. Oper. Res., vol. 287, no. 2, versity of Technology, Wuhan, China, in 2005.
pp. 452–459, Dec. 2020. He is currently a Professor with the School of
[34] E. Zehendner, D. Feillet, and P. Jaillet, ‘‘An algorithm with performance Machinery and Automation, Wuhan University
guarantee for the online container relocation problem,’’ Eur. J. Oper. Res., of Science and Technology, Wuhan. His research
vol. 259, no. 1, pp. 48–62, May 2017. interests include intelligence algorithms and sim-
[35] C.-J. Ting and K.-C. Wu, ‘‘Optimizing container relocation operations at ulation optimization in logistics systems.
container yards with beam search,’’ Transp. Res. E, Logistics Transp. Rev.,
vol. 103, pp. 17–31, Jul. 2017.
[36] T. Bacci, S. Mattia, and P. Ventura, ‘‘The bounded beam search algo-
rithm for the block relocation problem,’’ Comput. Oper. Res., vol. 103,
pp. 252–264, Mar. 2019.
[37] B. G. Zweers, S. Bhulai, and R. D. van der Mei, ‘‘Optimizing pre- XIAOLEI LIANG (Member, IEEE) received the
processing and relocation moves in the stochastic container relocation
Ph.D. degree from the School of Logistics Engi-
problem,’’ Eur. J. Oper. Res., vol. 283, no. 3, pp. 954–971, Jun. 2020.
neering, Wuhan University of Technology, Wuhan,
[38] F. Tricoire, J. Scagnetti, and A. Beham, ‘‘New insights on the block
relocation problem,’’ Comput. Oper. Res., vol. 89, pp. 127–139, Jan. 2018. China, in 2015. He is currently an Associate Pro-
[39] R. Jovanovic, M. Tuba, and S. Voß, ‘‘An efficient ant colony optimization fessor with the School of Automobile and Traffic
algorithm for the blocks relocation problem,’’ Eur. J. Oper. Res., vol. 274, Engineering, Wuhan University of Science and
no. 1, pp. 78–90, Apr. 2019. Technology, Wuhan. His research includes the
[40] M. Ji, W. Guo, H. Zhu, and Y. Yang, ‘‘Optimization of loading sequence swarm intelligence algorithm and logistics system
and rehandling strategy for multi-quay crane operations in container termi- simulation optimization.
nals,’’ Transp. Res. E, Logistics Transp. Rev., vol. 80, pp. 1–19, Aug. 2015.

207514 VOLUME 8, 2020

You might also like