You are on page 1of 43

Accepted Manuscript

Title: Cooperation and Profit Allocation in Two-echelon


Logistics Joint Distribution Network Optimization

Authors: Yong Wang, Xiaolei Ma, Mingwu Liu, Ke Gong,


Yong Liu, Maozeng Xu, Yinhai Wang

PII: S1568-4946(17)30103-5
DOI: http://dx.doi.org/doi:10.1016/j.asoc.2017.02.025
Reference: ASOC 4074

To appear in: Applied Soft Computing

Received date: 7-5-2016


Revised date: 20-1-2017
Accepted date: 21-2-2017

Please cite this article as: Yong Wang, Xiaolei Ma, Mingwu Liu, Ke Gong, Yong
Liu, Maozeng Xu, Yinhai Wang, Cooperation and Profit Allocation in Two-echelon
Logistics Joint Distribution Network Optimization, Applied Soft Computing Journal
http://dx.doi.org/10.1016/j.asoc.2017.02.025

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
Cooperation and Profit Allocation in Two-echelon Logistics
Joint Distribution Network Optimization

Yong WANGa,b,c, Xiaolei MAd*, Mingwu LIUa, Ke Gonga, Yong LIU a, Maozeng XUa,
Yinhai WANGc,*

(aSchool of Economics and Management, Chongqing Jiaotong University, Chongqing


400074, China)
(bSchool of Management and Economics, University of Electronic Science and
Technology, Chengdu 610054, China)
(cDepartment of Civil and Environmental Engineering, University of Washington,
Seattle, WA 98195-2700, USA)
(dSchool of Transportation Science and Engineering, Beijing Key Laboratory for
Cooperative Vehicle Infrastructure, Systems, and Safety Control, Beihang University,
Beijing 100191, China)
E-mail: yongwx6@gmail.com; xiaolei@buaa.edu.cn*; liumingwu2007@aliyun.com;
gks_cn@163.com; liuevery@gmail.com; xmzzrxhy@cqjtu.edu.cn; yinhai@uw.edu*

*
Corresponding author
Graphical Abstract
Collaborative two-echelon logistics joint distribution network can be organized
through a negotiation process via logistics service providers or participants existing in
the logistics system, which can effectively reduce the crisscross transportation
phenomenon and improve the efficiency of the urban freight transportation system.
This study establishes a linear optimization model to minimize the total cost of
two-echelon logistics joint distribution network. An improved ant colony optimization
algorithm integrated with genetic algorithm is presented to serve customer clustering
units and resolve the model formulation by assigning logistics facilities. A
two-dimensional colony encoding method is adopted to generate the initial ant
colonies. Improved ant colony optimization combines the merits of ant colony
optimization algorithm and genetic algorithm with both global and local search
capabilities. Finally, an improved Shapley value model based on cooperative game
theory and a cooperative mechanism strategy are presented to obtain the optimal
profit allocation scheme and sequential coalitions respectively in two-echelon
logistics joint distribution network. An empirical study in Guiyang City, China,
reveals that the improved ant colony optimization algorithm is superior to the other
three methods in terms of the total cost. The improved Shapley value model and
monotonic path selection strategy are applied to calculate the best sequential coalition
selection strategy. The proposed cooperation and profit allocation approaches provide
an effective paradigm for logistics companies to share benefit, achieve win-win
situations through the horizontal cooperation, and improve the negotiation power for
logistics network optimization.

Highlights:
 A two-echelon logistics joint distribution network optimization model is
developed.
 This model is to minimize the total cost of TELJDN.
 A hybrid algorithm combining ACO and GA operations is proposed.
 A cooperative mechanism strategy for sequential coalitions is studied in
TELJDN.
 An empirical study demonstrates the applicability of the proposed approach.
ABSTRACT
Collaborative two-echelon logistics joint distribution network can be organized through a
negotiation process via logistics service providers or participants existing in the logistics
system, which can effectively reduce the crisscross transportation phenomenon and improve
the efficiency of the urban freight transportation system. This study establishes a linear
optimization model to minimize the total cost of two-echelon logistics joint distribution
network. An improved ant colony optimization algorithm integrated with genetic algorithm is
presented to serve customer clustering units and resolve the model formulation by assigning
logistics facilities. A two-dimensional colony encoding method is adopted to generate the
initial ant colonies. Improved ant colony optimization combines the merits of ant colony
optimization algorithm and genetic algorithm with both global and local search capabilities.
Finally, an improved Shapley value model based on cooperative game theory and a
cooperative mechanism strategy are presented to obtain the optimal profit allocation
scheme and sequential coalitions respectively in two-echelon logistics joint distribution
network. An empirical study in Guiyang City, China, reveals that the improved ant colony
optimization algorithm is superior to the other three methods in terms of the total cost. The
improved Shapley value model and monotonic path selection strategy are applied to calculate
the best sequential coalition selection strategy. The proposed cooperation and profit allocation
approaches provide an effective paradigm for logistics companies to share benefit, achieve
win-win situations through the horizontal cooperation, and improve the negotiation power for
logistics network optimization.

Key words: Two-echelon logistics joint distribution network; Profit allocation; Improved ant
colony optimization; Shapley value model; Monotonic path selection strategy
1 Introduction
Two-echelon logistics joint distribution network (TELJDN) is a crucial component in
multi-echelon logistics distribution systems (Gzara et al. 2014; Wang et al. 2015a;
Mousavi et al. 2015). A two-echelon logistics joint distribution network is usually
composed of the first-echelon logistics facility [e.g., logistics center (LC)], several
second-echelon logistics facilities [e.g., distribution centers (DCs)], and large-scale
customers (Wang et al. 2012; Guo and Li, 2014). A well-designed logistics and
transportation network can effectively improve the efficiency of an entire freight
transportation system (Hajiaghaei-Keshteli, 2011), particularly for the food cold chain
and special drug transportation industries, which can serve customer requirements in a
timely manner. A two-echelon logistics joint distribution network can be established
through cooperation with logistics coordinators [i.e., logistics service providers (LSPs)
or participants existing in the logistics network] to achieve a win-win situation. The
reduced transportation, operational, and infrastructure maintenance costs can gain
more profits for the logistics coordinator and ensure service quality for customers
with low price. Therefore, properly optimizing the logistics distribution network and
designing a reasonable cooperation strategy have become important tasks for logistics
operators.

Two-echelon logistics joint distribution network optimization includes two-echelon


logistics facility (e.g., LC and DCs) location, vehicle routing optimization and
collaborative strategy research procedures among logistics facilities. To alleviate
computational complexity, customer clustering is often applied before vehicle routing
optimization in a large-scale logistics network (Liberatore et al. 2014; Xu et al. 2015;
Wang et al. 2015a). The optimization of a logistics distribution network aims to
satisfy the requirements of customers by designing a reasonable network structure and
allocating logistics facilities in this network (Wang et al. 2014). Therefore, the
selection of collaborative strategies used for multiple logistics facilities is critical in
logistics distribution network optimization procedures; in other words, the logistics
distribution network optimization procedure cannot be achieved if no reasonable
cooperation mechanism exists. In addition, a reasonable network structure study can
simplify network optimization procedure and enhance the negotiation power (Wang et
al. 2015b; Scavarda et al. 2015; Guajardo and Rönnqvist, 2015). Therefore, logistics
network optimization and cooperation mechanism study are closely interrelated,
whereas traditional logistics distribution network optimization neglects such a
cooperative mechanism by assuming that each logistics facility is willing to
collaborate. To take into account the willingness of logistics participants, an
appropriate TELJDN optimization structure should be properly designed with a
reasonable cooperation mechanism.

In the current study, a common hybrid two-echelon logistics joint distribution network
structure is presented, in which goods can be centrally transported between the
first-echelon facility and the second-echelon facilities. To optimize TELJDN, each
customer clustering unit is reasonably assigned to its adjacent logistics facility (LC or
DC) (Guo and Li, 2014; Wang et al. 2015a). A robust solution algorithm can be
developed to optimize the proposed hybrid TELJDN by determining optimal solutions
for the two-echelon logistics joint distribution network, and a profit allocation method
based on cooperative game theory is presented to allocate profits among logistics
entities in the two-echelon logistics joint distribution network. This study aims to
develop such a procedure integrating both the robust solution algorithm and the profit
allocation method for two-echelon logistics joint distribution network optimization.

The remainder of this paper is organized as follows: Relevant studies are initially
reviewed and summarized, and then the two-echelon logistics joint distribution
network structure is further illustrated. The TELJDN is formulated with detailed
definitions and notations. An improved ant colony optimization (IACO) algorithm
combining ant colony optimization (ACO) algorithm and genetic algorithm (GA)
operations is proposed to serve customer clustering units and resolve the model by
assigning logistics facilities. Then, a comparison with a different prevailing approach
is implemented. Shapley value model and monotonic path selection strategy are
utilized to study the profit allocation among multiple logistics entities in TELJDN. An
empirical study in Guiyang City, China, is presented to evaluate the effectiveness of
the proposed approach in three coalition modes. Then, an optimal solution calculation
procedure and a thorough analysis with sequential coalitions and negotiation power
are also presented. Finally, conclusions and future works are discussed at the end of
this paper.

2 Literature Review

Cooperation among multiple logistics facilities in large-scale logistics distribution


networks has become increasingly important and feasible with the advent of big data
era and new technology, such as cloud computing and e-commerce. However, the
metropolitan logistics network usually contains multiple or multi-echelon logistics
facilities with thousands of customers, and the traditional approach may not be
effective to solve such a sophisticated logistics network (Liu et al. 2013a; Wang et al.
2014). Customer clustering can categorize large-scale customers into multiple clusters,
and each customer clustering unit can share certain common features (i.e.,
geographical proximity and demand). Customers within a clustering unit can then
request a service from the same logistics facility, and thus simplify the sophisticated
logistics network.

In recent years, some researchers have proposed various clustering approaches to


reduce the computational complexity and improve the calculation accuracy of
large-scale logistics networks (Dondo and Cerdá, 2007; Nananukul, 2013). Bosona
and Gebresenbet (2011) developed an integrated logistics network by forming clusters
of products through a geographic information system to map the locations of
producers. Özdamar and Demir (2012) presented a multi-level clustering algorithm in
grouping demand points into small clusters to optimize vehicle routing in a large-scale
disaster relief logistics planning. Liu et al. (2013a) proposed a framework that
includes a clustering structure for supply network analysis, and their results showed
that the supply networks with a clustering structure can provide better services on
order lead time and delivery efficiency. Wang et al. (2014) presented a fuzzy-based
customer clustering algorithm with a hierarchical analysis structure to reduce the
complexity of large-scale logistics network optimization. Mesa-Arango and Ukkusuri
(2015) proposed a novel approach to cluster demand lanes based on dependent
sampling historical data and a train of network transformations. The proposed
approach was used to cluster demand lanes in freight logistics networks.

The clustering procedures in the preceding works were utilized in grouping customers
into customer units to further reduce the calculation complexity. Logistics facilities
can be assigned to serve several customer clustering units as a variant of the quadratic
assignment problem (Cordeau et al. 2008; Şeker and Noyan, 2012; Dokeroglu, 2015).
First introduced by Koopmans and Beckmann (1957), quadratic assignment problem
aims to assign n facilities to n locations, such that each facility is assigned to one
exact location. In a large-scale logistics network optimization procedure, each
customer clustering unit should be assigned to a certain facility in the logistics
network for further delivery service (Woo and Saghiri, 2011; Hajiaghaei-Keshteli,
2011).

A group of researchers studied the TELJDN optimization problem, which can be seen
as a basis of research for large-scale logistics networks (Basten and Houtum, 2013;
Cortinhal et al. 2015; Chan, et al. 2016). Hemmelmayr et al. (2012) presented an
adaptive large neighborhood search heuristic to study the two-echelon vehicle routing
problem and the location routing problem. Liu et al. (2013b) established a
two-echelon order allocation model with two objective functions in a logistics service
supply chain, and proposed a model solution method to study the model. Govindan et
al. (2014) presented a multi-objective optimization model and a hybrid intelligent
algorithm to study two-echelon multi-vehicle location routing problem with time
windows for a sustainable supply chain network of perishable food. Wang et al.
(2015a) proposed a hybrid particle swarm optimization-GA to study the two-echelon
logistics distribution region partitioning problem, and the hybrid algorithm improved
both global and local search capability. In addition, the uncertainty of demands and
prices were also studied in the TELJDN optimization problem (Cardoso, et al. 2013;
Gauvin, et al. 2014; Baruah, et al. 2016).

These studies showed that the multi-objective optimization model and the hybrid
intelligent algorithm or heuristic algorithm can be combined to study the TELJDN
optimization problem. Customer clustering units can also enhance the calculation
efficiency, particularly in a metropolitan transportation network. A natural reasoning
is that cooperative mechanisms should be incorporated among logistics entities to
better optimize logistics distribution network. The study of cooperative mechanisms
belongs to the category of cooperative game theory, which can be utilized to analyze
the collaborative behavior and profit allocation among multiple participants (Frisk et
al. 2010; Lozano et al. 2013; Guajardo and Rönnqvist, 2015).

Özener and Ergun (2008) proposed a cost allocation mechanism with well-known
properties including budget balance, stability, and cross monotonicity, based on
cooperative game theory, and presented several cost allocation schemes for
implementable solutions in a collaborative transportation procurement network.
Cruijssen et al. (2010) presented a supplier-initiated outsourcing procedure based on
both operation research and game theoretical insights to study the collaboration and
cost reduction allocation in transportation. Lozano et al. (2013) proposed a linear
optimization model to study cost savings when the transportation requirements are
merged among different potential participants, and studied the problem of allocating
the joint cost savings based on cooperative game theory. Guajardo and Rönnqvist
(2015) established the mixed integer linear programming models to study the
coalition structure in collaborative logistics transportation, and two applications,
including collaborative forest transportation and inventory of spare parts for oil
operations, were tested in the models. Scavarda et al. (2015) developed a
direct/indirect delivery protocol based on collaborative control theory to deal with
changeable conditions among manufacturers, distributors, and retailers by studying an
adaptive direct/indirect delivery through appropriate collaboration.

However, the preceding studies may suffer from the following issues: (1) The
traditional intelligent algorithm and heuristic algorithm cannot be directly applied into
the large-scale logistics distribution network with a considerable number of customers.
(2) TELJDN design procedure rarely considers simultaneously optimizing the
distribution network and the allocation of profits when transportation requirements are
merged or transshipped. (3) The majority of studies focus on profit allocation
mechanism design based on cooperative game theory and collaborative control theory,
but ignore the sequential coalition among participants, while these participants are
usually coordinators in the multi-echelon logistics distribution network.

The present study aims to construct a TELJDN to study the cost savings when the
transportation requirements are merged among two-echelon logistics entities. A profit
allocation approach based on cooperative game theory is proposed according to the
joint cost savings to allocate profits among different logistics entities, and a
cooperative mechanism strategy for sequential coalitions is studied in TELJDN.
Compared with those of previous studies, the main contributions of this study lie in (1)
constructing a linear optimization model to minimize the total cost of the TELJDN, (2)
designing a hybrid algorithm combining ACO and GA to solve the linear optimization
model, (3) utilizing the Shapley value model and the proposed monotonic path
selection strategy to study profit allocation and the sequential coalitions in TELJDN,
and (4) executing a real-world numerical study to demonstrate the applicability of the
proposed approach in three coalition modes. This approach can improve the
practicability of the logistics network optimization, which can also lay a foundation
for further study of multi-echelon logistics network optimization.

3 Problem Statement

Two-echelon logistics joint distribution network can be established through


negotiation between the LC and DCs, and the negotiation procedure can be organized
by players from either the joint distribution network or an LSP (Özener and Ergun,
2008; Nagurney, 2009; Liu et al. 2014). TELJDN can integrate resources and improve
asset allocation. It can also realize resource sharing and reduce the crisscross
transportation phenomenon in complex transportation networks. In our study,
TELJDN contains two-level facilities. The first-level facility is an LC, whereas the
second-level facilities cover several DCs. A large amount of customer
distribution/clustering units also exists in TELJDN. Each distribution unit includes a
group of customers with common features, such as similar temperature-controlled
food cold chain and geographical proximity. The changes of a TELJDN structure
before and after joint distribution are demonstrated in Fig. 1.

Fig. 1(a) illustrates a logistics distribution network without collaborative distribution,


in which a non-optimal condition is exhibited. In this network, the LC serves a large
amount of distribution units, even though some distribution units are adjacent to DCs.
The remaining small amounts of distribution units are served by several DCs. In Fig.
1 (a), long-distance delivery services also exist in the LC because of the time
sequence for logistics facility construction and customer loyalty (Hajiaghaei-Keshteli,
2011; Jouzdani et al. 2013; Ouyang et al. 2015). For instance, an LC is first
established during the process of urban development, and this LC and the customers
maintain long-term cooperation relations. The ever-increasing customer demands
result in the failure of the LC to meet market demands, and several DCs are
subsequently established to serve some new customers. However, some new
customers still choose to accept services from a LC or other DCs because of
long-term good reputation and brand effect, even if these logistics facilities are far
away from them. Therefore, a long-distance and crisscross transportation network
structure is implemented.

A new network structure with cooperation among multiple logistics facilities is


required to optimize this unreasonable transportation network structure. The generated
profits from the new transportation network should be distributed fairly among the
multiple logistics facilities. Therefore, two important issues must be addressed for
TELJDN. The first issue involves the redesigning process of the multiple logistics
facilities transportation network structure. The second issue involves the distribution
of gained profits among multiple logistics facilities in an effective manner. As shown
in Fig. 1(b), a hybrid TELJDN is established for the study (Wang et al. 2015a). The
LC is the first-echelon facility, and the centralized transport goods can be conducted
between the first-echelon facility and the second-echelon facilities. In addition, each
distribution unit is reasonably assigned to its adjacent logistics facility. The goods can
be transported between the LC and DCs through a fleet of semitrailer trucks. Each
facility can also serve the adjacent distribution units through a fleet of vehicles.
TELJDN aims to serve customers timely, reduce the entire logistics system
transportation cost, and improve the operational efficiency of the logistics network.
4 Method

4.1 Model Assumption

This study proposes that each distribution unit contains multiple customers. The
center coordinate of each distribution unit can be expressed as ( xi , yi ) and obtained

Ni Ni
by xi   xij Ni , yi   yij N i , where the coordinate for each customer j in the
j 1 j 1

distribution unit i can be expressed as ( xij , yij ) , and the number of customers is N i .

The distance can be calculated based on the coordinates from logistics facilities and
distribution units. Some necessary assumptions for the TELJDN must be made to
simplify calculation:

Assumption 1: The flow of goods is bidirectional between LC and DCs. The LC and
DCs are served by semitrailer trucks, and the distribution units are served by small
vehicles. Each semitrailer truck and small vehicle need to return to their origins.

Assumption 2: The customer demands are predetermined and relatively stable within
a working period.

Assumption 3: Semitrailer trucks are used in the first-level centralized transport,


whereas small vehicles are used in the direct delivery from the LC and second-level
delivery from the DCs.

Assumption 4: A TELJDN is established through the negotiation organized by the


first-level facility operator in this TELJDN, and the LC and DCs are independent
from one another before the TELJDN is established. Maximum profit and reasonable
profit allocations are the ultimate optimization goal.

Assumption 5: The transportation cost among LC, DCs and general distribution units
is a linear function of delivery quantity and travel distance, and the variable cost of a
DC is directly proportional to the delivery quantity.

4.2 Notations

In a TELJDN, the distribution units can contain an LC, DCs, and general distribution
units. Thus, the total number of distribution units should be n + m + 1. Several related
definitions are used and presented as follows.

Definition 1.
I{i | i  1, 2,3, , m} denotes the set of DCs, and m is the total number of DCs.

J { j | j  1, 2,3, , n} denotes the set of general distribution units in a two-echelon

logistics distribution network, and n is the total number of distribution units.


K{k | k  m  1} denotes the LC in a two-echelon logistics distribution network.

xi , j denotes the delivery quantity from the LC or DC i to the distribution unit j


within a working period.
Li , j denotes the distance from the LC or DC i to the distribution unit j.

di ,m1 denotes the transport quantity from the DC i to the LC m + 1 within a


working period.
d i denotes the delivery capacity of the LC or DC i within a working period.
D denotes the total delivery capacity of the LC and DCs within a working period,
which can be expressed as
D   di , where i  I K.
i

q j denotes the demand quantity of the distribution unit j within a working


period.
Q denotes the total demand quantity of all distribution units within a working
period, which can be expressed as
Q   q j , where j  J .
j

Definition 2. The distribution relation matrix is expressed as


R  {ri , j | ri , j  0,1, i  I K , j  J } , where rij  1 implies that the LC or DC i delivers

goods to the distribution unit j, and rij  0 indicates that LC or DC i does not deliver

goods to the distribution unit j.

4.3 Model Formulation

Several related notations used in the formulation are listed as follows:


Vs : Loading capacity of each semitrailer truck
Vc : Loading capacity of each small vehicle
i : Fixed cost of the DC i
 i : Cost savings provided by the LSP or coordinator from the distribution network for
logistics facility i only when a participant is considered as the candidate in a working
period
 : Variable cost coefficient of the DC
L: Maximum delivery distance
T: Number of working periods
 : Number of trips for a small vehicle within a working period
 ' : Number of trips for a semitrailer truck within a working period
f sr : Freight rate of small vehicle (dollar/pallet·mile), which can be estimated based
on the freight statistics from the local freight agencies.
f tr : Freight rate of semitrailer truck (dollar/pallet·mile), which can be estimated based
on the freight statistics from the local freight agencies.
Fv : Annual maintenance cost of each small vehicle (dollar/year)
Fs : Annual maintenance cost of each semitrailer truck (dollar/year)

Z1 : Total transportation cost includes the transportation cost between LC and DCs,

and the cost from the LC and DCs to the general distribution units within a working
period. In addition, the total transportation cost also considers the returning distance
between the LC or DC and each distribution unit.
Z 2 : Annual maintenance cost from the LC to DCs within a working period
Z 3 : Annual maintenance cost from the LC and DCs to the distribution units within a
working period
Z 4 : Fixed and variable costs of DCs
The model formulation is composed of four sub-formulations, which can be shown as
follows:
Z1 can be expressed as

Z1   (rjJ
m 1, j  xm1, j  Lm1, j  2  f sr  )   [ Lm1,i   (ri , j  xi , j  ftr )   ']
iI jJ
(1)
 [ Lm1,i   (di ,m1  ftr )  ']   (ri , j  xi , j  Li , j  2  f sr  )
iI jJ iI jJ

Z 2 can be expressed as

Z 2    ri , j  xi , j  Vs  T   Fs (2)
iI jJ

Z 3 can be expressed as

Z3   (rm1, j  xm1, j ) (Vc  T )  Fv   (ri , j  xi , j ) (Vc  T )  Fv (3)


jJ iI jJ

Z 4 can be expressed as

Z 4   i   (ri , j  xi , j  i ) (4)
iI iI jJ

where 
iI
i and  (r
iI jJ
i, j  xi , j  i ) represent the fixed cost and variable cost,

respectively.
The objective function for the TELJDN can be expressed as follows:
Min Z  Z1  Z2  Z3  Z4 (5)

Subject to
r
iI K
i, j  1, j  J (6)

ri , j  {0,1}, i  I K , j  J (7)

di   ri , j  xi , j , i  I K (8)
jJ

qj  r
iI K
i, j  xi , j , j  J (9)

D  Q, i  I K , j  J (10)

xi , j  0, i  I K , j  J (11)

Li , j  L , xi , j  0 , i  I K , j  J (12)

Formula (5) aims to minimize the total cost of a TELJDN. Constraint (6) ensures that
a distribution unit can only be served by one LC or one DC. In constraint (7), if the
LC or DC i serves the distribution unit j, then ri , j is set to 1; otherwise, ri , j is set to

0. Constraint (8) guarantees that the delivery capacity is equal to the total delivery
quantity at DC i. Constraint (9) ensures that the demand quantity is equal to the total
delivery quantity from LC or DC to the distribution unit j. Constraint (10) guarantees
that the total delivery quantity is equal to the total demand quantity. Constraint (11)
indicates that the delivery quantity from the LC or DC i to the distribution unit j is
greater than or equal to 0. Constraint (12) assures that the distance from LC or DC i to
the distribution unit j does not exceed the maximum delivery distance. In addition, it
is important to note that the transportation cost depends on the type of cargo, freight
rate, demands, number of trips and number of working periods that can be represented
in eother linear or nonlinear functions of distance, so the model can be calibrated by
converting nonlinear variables into linear variables.

4.4 Proposed IACO Solving Procedure

ACO algorithm is a swarm intelligence stochastic approach. It was initially proposed


by Colorni et al. (1991) and Dorigo (1992). In this scheme, ants randomly choose the
paths in the beginning, and each ant can lay a pheromone trail on the chosen path
between the food source and the nest. The probability of each path can be evaluated
by checking the intensity of the pheromone on the corresponding trail. However,
nearly all of the ants find the shortest path in the end. Therefore, an ant colony can
make continuous communication feedback through pheromones, and ants can
eventually determine the shortest path from the food sources to the nest. GA is also a
stochastic optimization algorithm used to mimic the natural selection process
(Rahmani and MirHassani, 2014; Rostamzadeh, et al. 2015; Chan et al. 2016).
Combinational optimization problems can be solved by applying the combination of
ACO and GA (Hoseini and Shayesteh, 2013; Jiang et al. 2015). In the present study,
we propose that the genetic operations from GA are integrated into the ACO
algorithm procedure as an IACO algorithm.

The IACO algorithm is used to solve the TELJDN. Each distribution unit can be
reasonably assigned into each corresponding LC or DC, which can be completed by
optimizing the objective function for TELJDN in Formula (5) based on the IACO
algorithm. The method can reduce the number of iterations for convergence, and the
probability of obtaining the optimal solution can be increased. The relevant operations
and steps are illustrated as follows.

4.4.1 Relevant Definitions and Notations for the IACO Algorithm


IACO algorithms can enhance the reliability of solving the optimization model. The
relevant notations and definitions for the IACO algorithm (Colorni et al., 1991) are
listed as follows:

 : Size of ant colony


Amax : Maximum number of iterations for the IACO algorithm
AN: Number of colonies used for GA calculation
pc : Crossover probability
pm : Mutation probability
 and  : Pheromone intensity and visibility weighting factors respectively
 : Pheromone decay rate, which has a value in the interval [0, 1]
tabuh (t ) : Tabu list of ant colony h at iteration number t; it is used to prevent any
distribution unit from revisiting
 : Maximum number of iterations used as a stopping criterion
Z h (t ) : Fitness function value of ant colony h at iteration number t
Ph (t ) : Individual best-known solution at iteration number t
Pg (t ) : Global best-known solution at iteration number t
ij (t ) : Visibility parameter at iteration number t; it is generally defined as
1
ij (t )  , where Li , j (t ) denotes the distance between nodes i and j
Li , j (t )
 ij (t ) : Pheromone intensity on arc (i, j) at iteration number t
 ijl (t ) : Amount of pheromone laid by the lth ant on arc (i, j) between iteration
number (t − 1) and t
 ij (t ) : Additional pheromone deposited on arc (i, j) at iteration number t; it can be

expressed as  ij (t )    ijl (t )
l 1
In addition, some formulas used to select and reconstruct the possible attack paths
from the food sources to the nest are shown as follows:
(1) The probability of an ant choosing a path between nodes i and j can be expressed
as
[ ij (t )] [ij (t )]
pij (t )  , (13)

jtabuh ( t )
[ ij (t )] [ij (t )]

Where j  tabul guarantees that node j is a member of a set of nodes neighboring


node i and does not belong to tabuh , that is, the ant has not visited node i.

(2) ACO state update operations. To reconstruct the possible attack path, the amount
of pheromone on arc (i, j) at iteration number t+1 is given as follows:

 ij (t  1)  (1   ) ij (t )   ij (t ) , (14)
n
Where  ij (t )    ijl (t ) , and n is the total number of ants in the colony. The value
l 1

of  ijl (t ) can be obtained with the ant quantity formula presented by Colorni et al.
(1991) as follows:
Q
 Ll if (i, j ) in path visited by l th ant
 ijl (t )   i , j , (15)
0
 otherwise
Where Lli , j is the distance between nodes i and j traversed by the ant l, and Q is a
constant. The pheromone intensity along route (i, j) has an inversely proportional
relationship with Lli , j , as shown in Eq. (15).

In the IACO algorithm, all ants in one colony are randomly distributed on the
distribution unit location, and each ant is recorded in a tabu list to prevent any
distribution unit from revisiting during the search process. The genetic operations,
which include selection, crossover, and mutation procedures, are integrated into ACO
algorithm to update the best-known solutions between ACO and GA (Wang et al.
2015). Finally, the most probable attack paths between the distribution units and the
logistics facilities can be identified by circular searching the paths among multiple
colonies with the highest pheromone intensity along each path.
4.4.2 IACO Algorithm Procedure
A two-dimensional ant colony encoding in a TELJDN is presented in this study. The
first dimension of the colony can be regarded as the location of nests, and it is
expressed as a sequence of natural numbers: 1, 2, . . ., j, . . ., n, where n is the total
number of distribution units. The second dimension can be regarded as the location of
food sources, and it can be expressed as the sequence number of the logistics facility
that is assigned to each ant in the colony. N h , j denotes the logistics facility assigned
to the jth ant (distribution unit) in the hth colony. The two-dimensional ant colony
encoding is shown in Table 1, where Nh, j  1, 2, , i, , m, m  1 . For example,
N 2,3  3 expresses that DC 3 is assigned to the third distribution unit in the second
colony, and N4,5  m  1 indicates that the LC is assigned to the fifth distribution unit
in the fourth colony.

The preceding ant colony encoding method ensures that each distribution unit is
assigned to a certain DC or LC. The initial colonies can be generated through the ant
colony encoding method. Based on the ACO algorithm scheme and ACO state update
operations introduced earlier, the IACO algorithm is detailed as follows:

Step 1. Algorithm initialization. An integer within [1, m  1] is randomly generated


for each N h , j in each colony.  colonies are generated with a random method to
increase the diversity of the initial ant colony. The tabu list tabuh (t ) of ant colony h
at iteration number t can be set as null, and t = 1. The initial pheromone intensity
 ij (1) can be set equal to  ij (1) .

Step 2. The unqualified colonies are found and regenerated based on constraints
(6)–(12). The fitness function value Z h (t ) of each ant colony is calculated with
Formula (5). The visibility parameter value ij (t ) is also calculated.

Step 3. The individual best-known solution Ph (t ) and the global optimal solution
Pg (t ) are found at iteration number t. The corresponding assigned scheme in the ant
colony can also be found.
Step 4. For an ant located at distribution unit j, the assigned logistics facility i is
updated with Formula (16). tabuh (t ) is also updated. The unqualified assignment is
found and regenerated based on constraints (6)–(12).


arg max{[ ij (t )] [ij (t )] } if q  q0 (a)
i jtabu h (t ) , (16)

S otherwise (b)

[ ij (t )] [ij (t )]


 j  tabuh (t )

S  pij (t )   
iI K , jJ
[ ij (t )] [ij (t )] ,


0 otherwise
Where q0 is a user-defined parameter ( 0  q0  1 ), and q is a random number
( 0  q  1).

Step 5. The colony state is updated, the optimal colony is kept, and the individual
best-known value is compared with the global best-known value. The relevant
solutions Ph (t ) and Pg (t ) are updated.

Step 6. AN colonies are selected and used to execute the crossover operations based on
the crossover probability pc and the mutation operations based on the mutation
probability pm . The relevant solutions Ph (t ) and Pg (t ) are updated.

Step 7. When all the distribution units have been assigned to the corresponding
logistics facilities, the pheromone intensity assigned to the most probable path is
recalculated with Formula (14):  ij (t  1)  (1   ) ij (t )   ij (t ) , and t = t + 1. The
relevant solutions Ph (t ) and Pg (t ) are updated, and tabuh (t ) is set as null.

Step 8. The number of iterations t is determined. If it exceeds the maximum number of


iterations Amax or no more improvements can be found over the global optimal solution
within  consecutive iterations, the calculation procedure is terminated or Step 2 is
applied again.
Step 9. The optimal solution (i.e., assigned scheme and fitness function value) is
calculated and selected from all feasible ant colonies. This solution is the final result
for the TELJDN optimization.
In the preceding IACO algorithm, a novel two-dimensional ant colony encoding
method is proposed. Both GA operations and ACO state update operations, which
exist in the IACO algorithm, are reasonably combined. Therefore, the proposed
method provides a robust local and global search capability. The search space is also
extended, and the optimization capability of the proposed approach is enhanced.

4.5 Shapley Value Model and Monotonic Path Selection

Each distribution unit is adequately assigned to its adjacent LC or DC when the


TELJDN optimization structure is achieved. The coordinator is required to estimate
and allocate the profit as a result of the logistics network structure adjustment to each
logistics facility. Shapley value model (Shapley, 1953) is a widely accepted profit and
cost allocation mechanism in cooperative game theory. Cooperative game theory is
used to study the collaborative behavior in a negotiation process within a group of
participants in setting up a contract or collaborative plan of activities, including an
allocation of generated profit and cost savings (Frisk et al. 2010; Guajardo and
Rönnqvist,2015; Martínez-Cánovas et al. 2016).

4.5.1 Shapley Value Model

Shapley value model expresses the profit or cost savings that should be allocated to
participant i as shown in Formula (17), where M is a set of participants, and 2M  1
can be deemed to the number of all subsets of M excluding the null set. The elements
of all subsets are called coalitions. M is the grand coalition, and T is a coalition from
the set M. All T  M , the marginal contribution v(T )  v(T  {i}) can be allocated

to participant i when this participant joins the coalition T. Shapley value model can be
regarded as a completely random procedure for all participants by the average
expected payoff.
( T  1)!( M  T )!
i ( M , v)  
T  M ;iT M!
[v(T )  v(T  {i})] , i  M (17)

Shapley value model is derived based on the four axioms of efficiency, symmetry,
dummy, and additivity. These four axioms exhibit several desirable features, which
include rationality, availability, and relative stability of allocation (Shapley 1953;
Cruijssen et al. 2010; Frisk et al. 2010). Profit allocation procedure can be organized
by either an LSP or participants from the logistics network. In general, a percentage of
profit should be allocated to the organizer (an LSP or participants from the logistics
network). This share of profit is called synergy requirement and is represented as  ,
 [0,1] . The organizer obtains a high prospected profit when  is set as a high
value, whereas the participants are more likely to cooperate by setting  as a low
value. The profit value of coalition T can be calculated as follows:
  
1      C0 (i)  C (T )  if  C (i)  C (T )  0
v(T )    iT  iT
0
(18)
0
 otherwise

In Formula (18), C0 (i) denotes the cost when participant i is in the absence of

cooperation, whereas C (T ) is the total cost of all participants in the coalition T. For

example, if i, j  T , i j  , i j  T , then C (T )  C (i j )  C0 (i)  C0 ( j ) . The

first case occurs in Formula (18); otherwise, v(T ) is equal to zero.

4.5.2 Monotonic Path Selection Strategy


If  is set as the number of sequences in grand coalition M, then M ! different

sequence combinations exist. We can set  as a sequence in these combinations, and


 (i) is the rank of participant i in sequence  . The cost reduction percentages can be

expressed as  (i,  , u) , which can be calculated as follows:

i   , v 
 (i,  , u )    (  )u , M ,  (i)  u (19)
C0 (i )

In Formula (19),  (i,  , u) represents the cost reduction percentage of participant i in

sequence  when the uth participant joins the coalition. Monotonic path selection can
be stated based on the cost reduction percentage. Strictly monotonic path (SMP) can
be explained based on game theory (Cruijssen et al. 2010). If O, K , T  M satisfies

the condition of O  K  M \ T , then v(O  M )  v(O)  v( K  M )  v( K ) . This

condition is called strictly convex. If we suppose that (M, v) belongs to the strictly
convex condition, then sequence    should be an SMP. If some sequences satisfy
the SMP, then the best sequential coalition can be chosen among them. The best
sequential coalition selection process can be stated as follows:
Step 1: The diagonal values are selected from the cost reduction percentage matrix
calculated from Formula (19) in possible SMP sequence coalitions.
Step 2: The participant with the maximal lowest cost reduction percentage is found in
each selected diagonal value from Step 1. If the cost reduction percentages are equal
for all possible sequential coalitions, then another participant with the maximal
second lowest cost reduction percentage is found. The above process is executed until
at least one participant can be found or all participants are searched.
Step 3: The selected sequential coalition is considered as the candidate sequential
coalition strategy used for further profit allocation process. If all participants are
searched, then multiple sequential coalitions meet the requirements. Any sequential
coalition is selected as the candidate monotonic path strategy for profit allocation.

5 Results

5.1 Data Source

A practical study in Guiyang City, China, is utilized to evaluate the applicability of the
proposed methods in TELJDN optimization. Guiyang City, which is a critical
transportation hub, is the capital of Guizhou Province. It is selected as the ideal test
location necessary for this study. The locations of one LC (expressed as LC), four
DCs (expressed as DC1, DC2, DC3, and DC4) and the 85 distribution units
(expressed as C1, C2, . . ., C85) are shown in Fig. 2. The distribution units denoted by
stars are assigned to LC, the distribution units denoted by squares are assigned to DC1,
the distribution units denoted by diamonds are assigned to DC2, the distribution units
denoted by triangles are assigned to DC3, and the distribution units denoted by circles
are assigned to DC4. The original distribution networks are composed of one LC and
four DCs. Logistics service overlap delivery can be found in the original network.
Therefore, further study on distribution network optimization based on the
cooperative transportation service provided by the LC operator is necessary. In
addition, Table A1 (see Appendix) shows the distances measured between the five
logistics facilities (i.e., one LC and four DCs) and the 85 distribution units.
.
The delivery demand of each distribution unit is converted into the standard roll pallet
quantity to simplify the calculation process. The characteristics of one LC and four
DCs contain the number of distribution units. The period demand quantity is
demonstrated in Table 2.

5.2 Improved ACO Parameter Setting and Optimization Results

In this empirical study, parameter settings for IACO can be determined based on
previous studies (Santos et al., 2010; Ting and Chen, 2013; Wang et al. 2016). The
parameters are given as follows:
1.   100 is the number of colonies to increase the diversity of initial ant colony.

2. Amax  1000 is the maximum number of iterations for the IACO algorithm.
3. AN = 40 is the number of colonies used for GA calculation.
4.   1.5 is the pheromone intensity weighting factor, and   1 is the visibility
weighting factor.
5.   0.3 is the decay rate of the pheromone.
6. pc  0.7 is the crossover probability, whereas pm  0.01 is the mutation
probability in genetic operations.
7.   10 is the maximum number of iterations required to obtain the best-known
solution without any improvement.
8. Q = 3 is a constant in Formula (15), a high value of Q results in a rapid
convergence rate.
9. q0  0.5 is the parameter used to assign logistics facility i serving distribution unit
j.
10. m = 4 is the number of DCs, whereas n = 85 is the number of distribution units.
11. Several other parameters used in the model formulation are given as Vs  2000 ,
Vc  200 ,   1.5 , L = 50, T = 52,   10 ,  '  5 , f sr  0.003 , ftr  0.001 ,
Fs  5000 , Fv  1700 , 1  905 , 2  2016 , 3  1267 , 4  1455 , 1  905 ,
2  316 , 3  1067 , 4  1155 , and 5  1896 .
A working period can contain five working days in this study. An LC is also the
participant and organizer in the cooperation process. Thus, 2 4–1+5=20 combinations
of nonempty coalitions that can be served by the LC operator exist. IACO algorithm
is used to assign each distribution unit into each logistics facility by optimizing the
total cost based on the empirical data. The generated profit is then redistributed
among different logistics facilities through the Shapley value model. The optimization
results over a working period from all coalitions are shown in Table 3.
For explanatory purposes, all distribution units assigned to each logistics facility for
grand coalition in TELJDN are listed in Table 4. In the next section, the profit (cost
savings) is obtained from the optimized TELJDN, which is distributed among
logistics facilities based on the Shapley value model.

5.3 Algorithm Comparisons

We implemented the IPSO (Improved Particle Swarm Optimization), EGA (Extended


Genetic Algorithms) and SCTS (Self Controlling Tabu Search) algorithms with the
same dataset for comparison. IPSO constricts the solution space using a specific
approximation approach and determines the maximum flexion portion (Önüt et al.
2008), EGA contains the elitism genetic operator, a specific allotted time and a
slave-mater method (Shima et al., 2006), and SCTS considers a control-theoretic
approach and a flexible diversification strategy to adjust the algorithm parameters
(Fescioglu-Unver and Kokar, 2011). The proposed algorithms along with the three
algorithms are implemented on a laptop with Intel Core i5 Quad CPU and 8 GB RAM.
Each algorithm is executed 50 times, and the best solution is selected as the joint
distribution network optimization results for each approach. The best solution of Önüt
et al., Shima et al., Fescioglu-Unver and Kokar, and the proposed approach is Z =
$46291, Z = $46394, Z = $45987, and Z = $45750, respectively. The three algorithms
are executed 50 times with the optimal cost for convergence. The total cost, number
of iterations and running time are shown in Table 5.

The t-test and p value are conducted and shown in Table 5. The proposed IACO
algorithm outperforms the other three algorithms in terms of total cost and number of
iterations. The total average cost from the proposed algorithm is lower than that of the
HGA, IPSO and SCTS approach by 3.7%, 2.6% and 1.4%, respectively. Meanwhile,
the average number of iterations from the proposed algorithm requires less than 100,
88 and 54 iterations for the HGA, SCTS and IPSO approach, respectively. The
proposed IACO algorithm spends more time completing one iteration than the other
algorithm, causing a longer but acceptable execution time considering its performance
gain. This finding indicates that the possibility of the IACO algorithm to obtain the
optimal solution is higher than that of any of the other three approaches.

Compared with the other three algorithms, IACO has the following merits: (1) IACO
algorithm is a hybrid algorithm combining ACO algorithm and GA operations with
both global and local search capabilities. (2) Implementing GA operations into IACO
algorithm procedures can obtain the best-fit solutions, and thus enhance the capability
of IACO to reach the optimal solutions. As shown in Table 5, the best solution in the
IACO algorithm appears 20 out of 50 times, whereas that in the other three algorithms
only appears three or five times. (3) The IACO algorithm has the inherent global
searching advantage to apply in the large-scale logistics distribution network with
considerable customers. As shown in Table 5, the average number of iterations from
the IACO algorithm is significantly smaller than that from the other three algorithms.
In addition, we utilize the Lingo 12.0 solver to calculate the exact results based on the
objective function for the TELJDN. The data is chosen from RC1_2_1 in Extended
Solomon Datasets and the customer delivery demand is magnified by 20 times as the
input demand (Wang et al. 2015a). The proposed algorithm is randomly executed for
20 times, and the optimal solution can be achieved for 15 times (i.e. equal to the exact
solution).

5.4 Shapley Value Model Application

When the profit is obtained after TELJDN optimization, each distribution unit can be
adequately assigned into each logistics facility. The LC operator as the role of LSP
can allocate the gained profit. As discussed previously, the Shapley value model can
be utilized to achieve the profit allocation process among five logistics facilities. The
synergy requirement value is a key parameter in the negotiation process between the
LC operator and other participants from the coalitions, and the synergy requirement
value is set as   0.2 in this study. All of the possible coalitions, including profit
allocation, are calculated and shown in Table 6.

The cost reduction percentages calculated from Formula (19) and the possible
coalition sequences from Table 6 can be integrated, as shown in Fig. 3. The last line in
Fig. 3 shows that the five candidate participants join the coalitions. They also have
five positive percentage profits, which are compared with the original distribution
network. Line four in Fig. 3 shows that the three coalitions have a positive percentage
profit, except for coalition {LC, DC2}. Line three in Fig. 4 shows that DC2 in
coalition {LC, DC1, DC2}, {LC, DC2, DC3}, and {LC, DC2, DC4} can obtain
−6.7%, −2.4%, and −3.3% cost reduction percentage, respectively. This, the finding
indicates that the DC2 receives a negative cost reduction percentage in these
coalitions. In practice, DC2 does not agree to accept the coalition service. The first
line in Fig. 3 also shows that the grand coalition {LC, DC1, DC2, DC3, DC4} is
reached, and a certain positive percentage of profits are obtained by the five
participants.

5.5 Monotonic Path Selection in Coalitions

Monotonic path selection is necessary to clarify how the different sequential


coalitions affect the profit allocation strategy. SMP selection strategy is introduced in
Section 6.2. The operator from the LC is the coordinator used to enhance the
negotiation power for the five logistics facilities in all coalitions. All sequential
coalitions with cost reduction percentages are shown in Fig. 3. The sequential
coalitions that satisfy the SMP selection strategy are further selected and exhibited in
Table 7.

Table 7 shows that the cost reduction percentages of sequential coalitions {LC, DC3,
DC1, DC4, DC2} and {DC3, LC, DC1, DC4, DC2} are the only two suitable
sequences selected from Fig. 3. These coalitions are strictly monotonically increasing
in each column based on the SMP selection strategy. The next step is to find the best
sequential coalition from these two SMP coalitions as the optimal profit allocation
path selection strategy. The best sequential coalition selection strategy in Section 6.2
is applied. Thus, the maximum lowest cost reduction percentage in the diagonal
values from sequential coalition 1  {LC, DC3, DC1, DC4, DC2} is 4.8% when the

participant LC is initially invited to join, whereas that in


 2  {DC3, LC, DC1, DC4, DC2} is 4.85% when the participant DC3 is first invited
to join and then LC is invited to join. If the five participants are all rational, the
sequential coalition  2  {DC3, LC, DC1, DC4, DC2} likely forms a final grand

coalition based on the best sequential coalition selection strategy. The best
cooperation strategy is described as follows. DC3 is initially invited to join the
coalition. Then, LC enters the coalition to form {DC3, LC}. DC1 then joins the
coalition. Afterward, DC4 joins the coalition. Finally, DC2 joins the coalition, and the
grand coalition is established among these facilities. In this case study, LC is both a
required participant and a coordinator in the cooperation process. Thus, LC does not
need to satisfy the SMP strategy in the cooperation process to achieve the final grand
coalition. Therefore, the candidate coalitions for each individual logistics facility,
which include {LC}, {DC1}, {DC2}, {DC3}, and {DC4}, do not need to be
considered in the cooperation process. The sequential coalitions, except for the LC
that satisfies SMP condition, can be selected, as shown in Table 8.

All possible sequential coalitions, except for the LC for grand coalition based on SMP,
are shown in Table 8. The best sequential coalition selection strategy in Section 6.2 is
applied, and the best sequential coalition should be   {DC4, DC3, DC1, DC2} . The

best cooperation strategy can be described as follows. The DC4 initially cooperates
with the LC for the network optimization invited by the operator from the LC, and the
DC4 obtains 17.2% cost reduction. Then, DC3 joins the coalition, and DC4 and DC3
obtain 17.6% and 14.5% cost reduction respectively. Afterward, DC1 joins the
coalition, and DC4, DC3, and DC1 can accept 19.8%, 23.2%, and 18.7% cost
reduction, respectively. Finally, DC2 enters the coalition, and DC4, DC3, DC1, and
DC2 can obtain 28.5%, 30%, 20.1%, and 7.5% cost reduction, respectively, and then
the grand coalition is established.
6 Analysis and Implication

6.1 Analysis

When the grand coalition is established, the operator from the LC receives 20% of the
total savings (1562 USD in one working period) because the synergy requirement
value is set as   0.2 . In this study, LC is both a participant and a coordinator in the
cooperation process. Therefore, the 20% cost savings should be added into the profit
allocated to the LC in the cooperation process, and LC should be considered as the
first participant in the cooperation process. After the 20% cost savings is added into
the profit allocation process and LC first joins the coalition, the new sequential
coalitions that satisfy the SMP selection strategy can be calculated and selected, as
shown in Table 9. The profit allocation procedure for the new sequential coalitions
based on SMP selection strategy is displayed in Fig. 4. The two best sequential
coalitions are stated and analyzed as follows.

All possible sequential coalitions starting from the LC for grand coalition based on
SMP are shown in Table 9 and Fig. 4. The best sequential coalition can be selected
based on the optimal sequential coalition selection strategy in Section 6.2, namely,
  {LC, DC3, DC1, DC4, DC2} . This finding indicates that when the LC is first
invited to join the coalition, 6% cost savings exists. Then, a coalition is formed
between LC and DC3, and 6.7% and 20.5% cost reduction are obtained for them
respectively. Afterward, DC1 enters the coalition, and then DC4 joins the coalition.
Finally, the grand coalition is established after DC2 joins the coalition. A reasonable
profit allocation scheme is completed among LC, DC1, DC2, DC3, and DC4 through
the Shapley value model.

An interesting phenomenon can be observed from the final grand coalition. The LC
receives the highest profit and a low cost reduction percentage because of numerous
customer requirements and high non-optimization logistics network cost. The profit
and cost reduction percentage for DC2 is the lowest. The probable reason is that the
poor transportation condition and the outdated traffic facilities from DC2 contribute
the least in the current TELJDN structure. The lowest cost savings provided by the LC
operator compared with that of other logistics facilities brings a certain degree of
influence for DC2 to obtain a low profit and cost reduction percentage. The synergy
requirement  can be adjusted in response to the change in the negotiation process,
particularly because the participant is the coordinator. This scenario is similar to our
study case. The low synergy requirement improves the negotiation capacity for
rational candidates. The coordinator can explore different synergy requirements for
different participants by considering their current transportation conditions and traffic
facility completeness in the actual environment.

6.2 Implication

Designing proper cooperation strategies among the logistics distribution network


eventually generates more benefit among logistics participants for win-win situations,
and facilitates the quality of service for customers. TELJDN optimization is a critical
step in the study of the multi-echelon logistics distribution network optimization and
multi-echelon logistics facility cooperation problem. In particular, it is critical for
industries that require infrastructures in serving high real-timeliness merchandise,
such as food cold chain, logistics DCs, and special drug DCs.
6 Conclusions

This paper presents a novel approach to optimize TELJDN and allocate the gained
profit among logistics facilities. A linear programming model and IACO algorithm are
presented to optimize the nonempty coalition logistics network. Then, a profit
allocation method is presented to allocate the gained profits among logistics facilities
from one coalition. The coordinator from the LC is the existing participant in the
TELJDN; it is utilized to organize and complete the aforementioned procedures,
thereby enhancing the robustness and reliability of designing a multi-echelon and
large-scale logistics distribution network.

A two-echelon logistics joint distribution model formulation is initially established to


optimize the total cost of nonempty coalition TELJDN. Moreover, an IACO algorithm,
which includes reasonable loop iteration operations and appropriate exchange optimal
solutions between ACO and GA, is proposed to solve the model. For comparison
purposes, a GA for multiple task assignments from Shima et al. (2006) and a PSO for
multi-level warehouse layout design from Önüt et al. (2008) are implemented. The
results indicate that compared with the other two algorithms, the proposed IACO
algorithm can capture well the optimal cost and the number of iterations. Its capability
to handle a complex logistics distribution network optimization issue is also
demonstrated. Then, a profit allocation approach based on the Shapley value model is
presented among the logistics facilities from nonempty coalitions. Finally, the
monotonic path selection strategy is proposed and used to obtain the optimal
sequential coalitions. The proposed method is successfully applied in a numerical case
study in Guiyang City, China. Monotonic path selection strategy is applied in three
coalition modes: the candidate LC that first joined the sequential coalitions; sequential
coalitions, except for LC; and the participant LC that first joined the sequential
coalition. The three sequential coalitions provide certain significance in reality. The
synergy requirement value can also be adjusted to enhance the negotiation power for
the participants of different conditions in a nonempty coalition logistics distribution
system.

To improve the usability and depth of this study, the following aspects should be
enhanced with further research: (1) The heterogeneity of the synergy requirement
values among multi-echelon logistics entities should be considered to strengthen the
robustness of the logistics distribution network. (2) The mathematical programming
models and profit allocation methods for different network structures in logistics
distribution network system are desired. (3) The exact method used to study a
large-scale logistics network for further cooperation should be a potential research
direction. (4) The process of choosing the LSP or the existing participants in the
logistics network should be analyzed in future study. (5) Integrating VRP into the
current study is another interesting research area. (6) The stochastic demand and
nonlinearity of transportation cost should be taken into account.

Acknowledgments The authors thank the Transportation Planning Department in


Guiyang City for providing valuable location and data for the empirical case study.
This research is supported by the National Natural Science Foundation of China
(Project No. 71402011, 71471024, 51408019, 71432003, 51329801), the China
Postdoctoral Science Foundation (Project No. 2016M600735),
the Natural Science Foundation of Chongqing of China (No. cstc2015jcyjA30012,
cstc2016jcyjA0023).
References

Baruah, P., Chinnam, R. B., Korostelev, A., Dalkiran, E. Optimal soft-order revisions
under demand and supply uncertainty and upstream information. International Journal of
Production Economics, 2016, 182, 14-25.
Basten, R. J. I., Houtum, G. J. V. Near-optimal heuristics to set base stock levels in
a two-echelon distribution network. International Journal of Production Economics, 2013,
143(2): 546-552.
Bosona, T. G., Gebresenbet, G. Cluster building and logistics network integration of local food
supply chain. Biosystems Engineering, 2011, 108(4): 293-302.
Cardoso, S. R., Barbosa-Póvoa, A. P. F. D., Relvas, S. Design and planning of supply chains with
integration of reverse logistics activities under demand uncertainty. European Journal of
Operational Research, 2013, 226(3): 436-451.
Chan, F. R. S., Jha, A., Tiwari, M. K. Bi-objective optimization of three echelon supply chain
involving truck selection and loading using NSGA-II with heuristics algorithm. Applied Soft
Computing, 2016(38): 978-987.
Colorni, A., Dorigo, M., Maniezzo, V. Distributed optimization by ant colonies, in: Proceedings of
the European Conference on Artificial Life, Paris, France, 1991, 134-142.
Cordeau, J. F., Laporte, G., Pasin, F. An iterated local search heuristic for the logistics network
design problem with single assignment. International Journal of Production Economics,
2008, 113(2): 626-640.
Cortinhal, M. J., Lopes, M. J., Melo, M. T. Dynamic design and re-design of multi-echelon,
multi-product logistics networks with outsourcing opportunities: A computational study.
Computers & Industrial Engineering, 2015, 90, 118-131.
Cruijssen, F., Borm, P., Fleuren, H., Hamers, H. Supplier-initiated outsourcing: A methodology to
exploit synergy in transportation. European Journal of Operational Research, 2010, 207(2):
763-774.
Dokeroglu, T. Hybrid teaching-learning-based optimization algorithms for the quadratic
assignment problem. Computers & Industrial Engineering, 2015, 85, 86-101.
Dondo, R., Cerdá, J. (2007). A cluster-based optimization approach for the multi-depot
heterogeneous fleet vehicle routing problem with time windows. European Journal of
Operational Research, 2007, 176(3): 1478-1507.
Dorigo, M. Optimization, learning and natural algorithms, PhD thesis, Polotecnico di Milano,
Italie, 1992.
Fescioglu-Unver, N., Kokar, M. M. Self controlling tabu search algorithm for the Quadratic
Assignment Problem. Computers & Industrial Engineering, 2011, 60(2): 310-319.
Frisk, M., Göthe-Lundgren, M., Jörnsten, K. and Rönnqvist, M. Cost allocation in collaborative
forest transportation. European Journal of Operational Research, 2010, 205(2): 448-458.
Gauvin, C., Desaulniers, G., Gendreau, M. A branch-cut-and-price algorithm for the vehicle
routing problem with stochastic demands. Computers & Operations Research, 2014, 50,
141-153.
Govindan, K., Jafarian, A., Khodaverdi, R., Devika, K. Two-echelon multiple-vehicle location–
routing problem with time windows for optimization of sustainable supply chain network of
perishable food. International Journal of Production Economics, 2014, 152, 9-28.
Guajardo, M., Rönnqvist, M. Operations research models for coalition structure in
collaborative logistics. European Journal of Operational Research, 2015, 240(1): 147-159.
Guajardo, M., Rönnqvist, M. Operations research models for coalition structure
in collaborative logistics. European Journal of Operational Research, 2015, 240(1): 147-159.
Guo, C., Li, X. P. A multi-echelon inventory system with supplier selection and order allocation
under stochastic demand. International Journal of Production Economics, 2014, 151, 37-47.
Gzara, F., Nematollahi, E., Dasci, A. Linear location-inventory models for service
parts logistics network design. Computers & Industrial Engineering, 2014, 69, 53-63.
Hajiaghaei-Keshteli, M. The allocation of customers to potential distribution centers in supply
chain networks: GA and AIA approaches. Applied Soft Computing, 2011, 11(2): 2069-2078.
Hemmelmayr, V. C., Cordeau, J. F., Crainic, T. G. An adaptive large neighborhood search heuristic
for Two-Echelon Vehicle Routing Problems arising in city logistics. Computers & Operations
Research, 2012, 39(12): 3215-3228.
Hoseini, P., Shayesteh, M. G. Efficient contrast enhancement of images using
hybrid ant colony optimisation, genetic algorithm, and simulated annealing. Digital Signal
Processing, 2013, 23(3): 879-893.
Jiang, Z. H., Zhou, M. C., Tong, J. H., Jiang, H. Y., Yang, Y. F., Wang, A., You, Z. H. Comparing
an ant colony algorithm with a genetic algorithm for replugging tour planning of seedling
transplanter. Computers and Electronics in Agriculture, 2015(113): 225-233.
Jouzdani, J., Sadjadi, S. J. Fathian, M. Dynamic dairy facility location and supply chain planning
under traffic congestion and demand uncertainty: A case study of Tehran, Applied
Mathematical Modelling, 2013, 37(18-19): 8467-8483.
Koopmans, T. C., Beckmann, M. Assignment problems and the location of economic activities.
Econometrica, 1957, 25(1): 53-76.
Liberatore, F., Ortuño, M. T., Tirado, G., Vitoriano, B., Scaparra, M. P. A hierarchical compromise
model for the joint optimization of recovery operations and distribution of emergency goods
in Humanitarian Logistics. Computers & Operations Research, 2014, 42, 3-13.
Liu, Q., Zhang, C. Y., Zhu, K. R., Rao, Y. Q. Novel multi-objective resource allocation and
activity scheduling for fourth party logistics. Computers & Operations Research, 2014, 44:
42-51.
Liu, S. Y., Li, C., Feng, Y. P., Rong, G. Clustering structure and logistics: A new framework for
supply network analysis. Chemical Engineering Research and Design, 2013a, 91(8):
1383-1389.
Liu, W. H., Liu, C. L., Ge, M. Y. An order allocation model for the two-echelon logistics service
supply chain based on cumulative prospect theory. Journal of Purchasing & Supply
Management, 2013b, 19(1): 39-48.
Lozano, S., Moreno, P., Adenso-Díaz, B. and Algaba, E. Cooperative game theory approach to
allocating benefits of horizontal cooperation. European Journal of Operational Research,
2013, 229(2), 444-452.
Martínez-Cánovas, G., Val, E. D., Botti, V., Hernández, P., Rebollo, M. A formal model based
on Game Theory for the analysis of cooperation in distributed service discovery. Information
Sciences, 2016(326): 59-70.
Mesa-Arango, R., Ukkusuri, S. V. Demand clustering in freight logistics networks. Transportation
Research Part E: Logistics and Transportation Review, 2015, 81, 36-51.
Mousavi, S. M., Alikar, N., Niaki, S. T. A., Bahreininejad, A. Optimizing a location
allocation-inventory problem in a two-echelon supply chain network: A modified fruit fly
optimization algorithm. Computers & Industrial Engineering, 2015, 87, 543-560.
Nagurney, A. A system-optimization perspective for supply chain network integration: The
horizontal merge case. Transportation Research Part E: Logistics and Transportation Review,
2009, 45(1), 1-15.
Nananukul, N. Clustering model and algorithm for production inventory and distribution problem.
Applied Mathematical Modelling, 2013, 37(24): 9846-9857.
Ouyang, Y. F., Wang, Z. D., Yang, H. Facility location design under continuous traffic equilibrium.
Transportation Research Part B: Methodological, 2015, 81(1): 18-33.
Önüt, S., Tuzkaya, U. R., Doğaç, B. A particle swarm optimization algorithm for the
multiple-level warehouse layout design problem. Computers & Industrial Engineering, 2008,
54(4): 783-799.
Özdamar, L., Demir, O. A hierarchical clustering and routing procedure for large scale disaster
relief logistics planning. Transportation Research Part E: Logistics and Transportation
Review, 2012, 48(3): 591-602.
Özener, O. Ö., Ergun, Ö. Allocating Costs in a Collaborative Transportation Procurement Network.
Transportation Science, 2008, 42(2): 146-165.
Rahmani, A., MirHassani, S. A. A hybrid Firefly-Genetic Algorithm for the capacitated facility
location problem. Information Science, 2014(283): 70-78.
Rostamzadeh, R., Sabaghi, M., Sofian, S., Ismail, Z. Hybrid GA for material routing optimization
in supply chain. Applied Soft Computing, 2015(26): 107-122.
Santos, L., Coutinho-Rodrigues, J., Current, J. R. An improved ant colony optimization
based algorithm for the capacitated arc routing problem. Transportation Research Part B:
Methodological, 2010, 44(2): 246-266.
Scavarda, M., Seok, H., Puranik, A. S., Nof, S. Y. Adaptive direct/indirect delivery decision
protocol by collaborative negotiation among manufacturers, distributors, and retailers.
International Journal of Production Economics, 2015, 167, 232-245.
Shapley, L. S. A value for n-person Games. Annals of Mathematics Studies, 1953, 28, 307-317.
Shima T, Rasmussen S J, Sparks A G, Passino K M. Multiple task assignments for cooperating
uninhabited aerial vehicles using genetic algorithms. Computers & Operations Research,
2006, 33(11): 3252-3269.
Şeker, M., Noyan, N. Stochastic optimization models for the airport gate assignment problem.
Transportation Research Part E: Logistics and Transportation Review, 2012, 48(2): 438-459.
Ting, C. J., Chen, C. H. A multiple ant colony optimization algorithm for the capacitated location
routing problem. International Journal of Production Economics, 2013, 141(1): 34-44.
Wang, P., Lin, H. T., Wang, T. S. An improved ant colony system algorithm for solving the IP
traceback problem. Information Science, 2016, 326: 172-187.
Wang, Y., Ma, X. L., Wang, Y. H., Mao, H. J., Zhang, Y. Location optimization of multiple
distribution centers under fuzzy environment. Journal of Zhejiang University Science A, 2012,
13(10): 782-798.
Wang, Y. Ma, X. L., Lao, Y. T. and Wang, Y. H. A Fuzzy-based Customer Clustering Approach
with Hierarchical Structure for Logistics Network Optimization. Expert Systems with
Applications, 2014, 41(2), 521-534.
Wang, Y., Ma, X. L., Xu, M. Z., Liu, Y., Wang, Y. H. Two-echelon logistics distribution region
partitioning problem based on a hybrid particle swarm optimization-genetic algorithm.
Expert Systems with Applications, 2015a, 42(12): 5019-5031.
Wang, Y., Ma, X. L., Xu, M. Z., Wang, L. K., Wang, Y. H., Liu, Y. A Methodology to Exploit
Profit Allocation in Logistics Joint Distribution Network Optimization. Mathematical
Problems in Engineering, 2015b, Article ID 827021, 15 pages.
Woo, H. S., Saghiri, S. Order assignment considering buyer, third-party logistics provider, and
suppliers. International Journal of Production Economics, 2011, 130(2): 144-152.
Xu, X. F., Zhang, W., Li, N., Xu, H. L. A bi-level programming model of resource matching for
collaborative logistics network in supply uncertainty environment. Journal of Franklin
Institute, 2015, 352(9): 3873-3884.
Logistics Center
Distribution Center
Distribution Unit
First-level Centralized Transport
Direct Delivery from LC
Second-level Delivery
from DC
(a) Before joint distribution (b) After joint distribution

Fig. 1. TELJDN diagram


Fig. 2. Spatial distribution of the LC, DCs, and distribution units
DC3 DC4

17.6%
DC2 DC3 DC4

26.3%
14.5%
29.4%
LC

3.98%
6.5%
DC2 DC4

30.1%
LC

4.84%
DC4

17.2%
-3.3%

DC4
19.6%
LC

4.5%
LC

3.6%
DC1 DC3 DC4

19.8%

Fig. 3. Cost reduction percentages for all possible coalitions


23.2%
DC2 DC3

33.7%

DC3
20.3%
DC3
20.5%
DC1 DC2 DC3 DC4

28.5% 18.7%
-2.4%

LC
4.85%
LC

30% 5.15%
LC

3.8%

LC
7.5% 4.8%
DC1 DC4 13.5%
DC1 DC2 DC4

20.1% 20.2%

DC2
0.0%
14.3%
LC

6.1% 3.48%

LC
0.0%

DC2
3.4%

LC
4.76%
19.4%
LC

4.83%

DC1 DC3
20.8%

DC1
17.4%

DC1
12.7%
12.1%

LC
5.7%

DC1 DC2 DC3


20.98%

LC
4.86%
2.0%

DC1 DC2
-6.7%
18.2%
27.6%

LC
4.79%

LC
4.7%
(a) (b)
Fig. 4. Selected possible sequential coalitions for grand coalition based on SMP:
sequential coalition {LC, DC3, DC1, DC2, DC4} from (a) and sequential coalition
{LC, DC3, DC1, DC4, DC2} from (b).
Table 1 Two-dimensional ant colony encoding table
Distribution unit 1 2 j n
Logistics facility
N h ,1 N h ,2 N h, j N h,n
number

Table 2. Characteristics of LC and four DCs


Logistics Facility LC DC1 DC2 DC3 DC4

Number of DUs 49 10 11 8 7
Periodic demand
86900 16900 19500 10800 11300
(roll pallets)

Table 3 Comparison between initial and optimized network over a working period
Total cost for
Total cost for
S Demand optimized
initial network
network
{ LC} 86900 31463 29567
{ D1} 16900 5683 4778
{ D2} 19500 7481 7165
{ D3} 10800 4209 3142
{ D4} 11300 4726 3571
{ LC,D1} 103800 37146 33686
{ LC,D2} 106400 38944 38971
{ LC,D3} 97700 35672 32690
{ LC,D4} 98200 36189 33410
{ LC,D1,D2} 123300 44627 40881
{ LC,D1,D3} 114600 41355 37492
{ LC,D1,D4} 115100 41872 38184
{ LC,D2,D3} 117200 43153 39977
{ LC,D2,D4} 117700 43670 40511
{ LC,D3,D4} 109000 40398 37031
{ LC,D1,D2,D3} 134100 48836 44366
{ LC,D1,D2,D4} 134600 49353 44553
{ LC,D1,D3,D4} 125900 46081 40338
{ LC,D2,D3,D4} 128500 47879 42267
{ LC,D1,D2,D3,D4} 145400 53562 45750
Table 4 Distribution units assignment based on grand coalition in TELJDN
No. of Distribution Units
C31, C32, C41, C42, C43, C47, C48, C49, C53, C54, C55, C56, C61, C63,
LC
C64, C65, C66, C67, C68, C69, C70
DC1 C1, C2, C3, C4, C7, C8, C9, C13, C14, C15, C20, C21, C22, C23, C28, C30
C5, C6, C10, C11, C12, C16, C17, C18, C19, C24, C25, C26, C33, C34, C35,
DC2
C36, C37, C38
C71, C72, C73, C74, C75, C76, C77, C78, C79, C80, C81, C82, C83, C84,
DC3
C85
C27, C29, C39, C40, C44, C45, C46, C50, C51, C52, C57, C58, C59, C60,
DC4
C62
Table 5 Algorithms Optimization Results Comparison
Total Cost(dollar) Number of iterations Running Time (Seconds)
Sequence
IACO IPSO HGA SCTS IACO IPSO HGA SCTS IACO IPSO HGA SCTS
1 46721 47658 48765 47125 315 458 463 392 13.62 8.26 8.34 9.63
2 45750 48594 48693 46981 386 423 494 453 15.73 8.03 8.82 10.79
3 46832 48762 48752 47059 327 436 505 416 13.95 8.18 8.95 10.36
4 45750 48257 49136 47652 369 395 462 469 15.1 7.15 8.51 10.88
5 45750 46683 48681 46873 372 467 497 478 15.46 8.31 8.8 10.97
6 45825 46591 47528 46362 395 473 453 495 15.88 8.72 8.46 11.32
7 45750 48730 46947 47131 403 445 512 463 16.15 8.21 9.01 10.6
8 46395 47943 47646 47350 356 451 465 458 14.92 8.37 8.55 9.94
9 45750 48768 46872 46519 394 392 528 464 15.81 7.26 9.34 10.85
10 46641 46591 48537 46438 368 416 481 487 15.03 7.98 8.63 11.13
11 46731 47632 47538 46882 320 487 491 463 14.1 8.62 8.79 10.69
12 45750 46589 47231 46173 375 456 529 495 15.46 8.41 9.37 11.45
13 45898 46775 46998 47025 399 432 493 433 15.96 8.13 8.78 10.93
14 45973 46291 47035 46128 387 410 475 504 15.53 8.12 8.83 11.65
15 45750 47351 47392 46851 415 465 499 453 16.27 8.04 9.11 10.88
16 46527 48139 47633 47035 379 397 523 396 15.18 7.52 8.94 9.8
17 45750 47217 48697 46625 412 404 434 453 15.91 8.47 8.1 10.97
18 46296 47369 48752 46967 359 451 471 468 13.9 8.2 8.75 10.56
19 45750 46683 48693 46136 391 433 451 515 15.78 8.85 8.9 11.3
20 46872 47762 48276 47150 336 409 492 421 14.16 7.93 8.63 10.51
21 45750 46775 47934 46981 405 426 443 525 15.84 8.19 8.34 11.73
22 46431 46591 48657 46732 341 447 545 478 14.6 8.52 9.56 10.85
23 46252 47364 46893 47820 373 461 498 460 14.92 8.87 8.84 10.39
24 45750 47535 47680 46173 422 413 512 483 16.31 8.41 9.11 11.12
25 45750 46553 48165 46362 410 468 527 479 16.02 8.95 8.7 11.27
26 45869 47894 46958 47217 397 474 486 461 15.79 8.53 9.06 10.56
27 46176 48165 47852 46851 389 388 501 475 15.66 7.31 8.83 11.02
28 46318 46793 48794 45987 348 420 469 513 14.81 8.6 8.61 11.38
29 45750 47687 47146 47328 396 492 477 468 15.87 8.75 8.59 10.7
30 45828 46582 47035 46539 403 456 480 479 15.93 8.82 8.92 10.95
31 45750 46291 47113 46425 415 433 492 503 16.13 8.73 8.87 11.28
32 46352 47373 47682 46931 396 465 483 517 15.96 8.8 9.02 11.41
33 45894 47116 47538 46579 427 481 434 462 16.35 8.52 8.16 10.34
34 45750 46541 47625 46237 405 468 444 471 16.18 8.91 8.4 11.03
35 46031 46291 46982 45987 398 467 482 526 15.99 8.87 9.16 11.66
36 45973 47199 48276 46136 382 475 511 497 15.63 8.59 9.23 11.32
37 45750 47687 47880 46895 413 419 482 468 15.87 8.58 8.88 10.51
38 46168 46694 47172 46328 393 436 477 489 15.75 8.76 8.81 10.93
39 46342 46591 47693 46237 386 451 438 480 14.62 8.53 8.29 10.77
40 45750 47528 46394 46025 422 419 536 548 16.23 8.44 9.25 11.86
41 46318 47213 47869 47125 376 423 492 428 15.07 8.29 8.71 10.52
42 45825 47562 48136 47018 410 412 476 409 16.21 8.33 8.94 10.05
43 45750 46385 47593 46207 392 471 439 487 15.83 8.52 8.33 10.79
44 46196 46368 46825 45987 381 445 513 551 15.6 8.42 9.28 11.88
45 46078 46291 46791 46662 373 439 534 479 15.15 8.87 9.37 11.03
46 45983 48139 48679 47313 377 396 471 455 15.11 7.57 8.86 10.91
47 45750 47351 49032 46579 407 427 431 470 16.28 8.49 8.22 10.85
48 46395 47568 47238 46362 369 416 466 463 15.13 8.38 8.72 10.47
49 45750 46372 46815 47150 411 478 511 432 16.01 8.65 9.41 10.91
50 46289 46827 47193 46237 390 403 473 493 15.72 7.91 8.68 11.23
Average 46049 47234 47749 46697 385 439 485 473 15.49 8.36 8.82 10.9
t-test -11.16 -16.18 -8.66 -9.91 -17.0 -16.89 67.78 62.93 48.75
p-value 4.6E-15 2.7E-21 1.9E-11 2.7E-13 3.4E-22 4.5E-22 4.3E-50 1.6E-48 3.5E-43
Table 6 Profit allocation in two-echelon logistics joint distribution network optimization

T  iS
C0 (i) C(T) v(T)  (T , v)
{ LC} 31463 29567 1517 (1517; · ; · ; · ; · )
{ DC1} 5683 4778 724 ( · ;724; · ; · ; ·)
{ DC2} 7481 7165 253 ( · ; · ;253; · ; ·)
{ DC3} 4209 3142 854 ( · ; · ; · ;854; ·)
{ DC4} 4726 3571 924 ( · ; · ; · ; · ;924)
{LC,DC1} 37146 33686 2768 (1781;987; · ; · ; ·)
{LC,DC2} 38944 38971 0 (0; · ; 0 ; · ; ·)
{LC,DC3} 35672 32690 2386 (1525; · ; · ;861; · )
{LC,DC4} 36189 33410 2223 (1408; · ; · ; · ;815)
{LC,DC1,DC2} 44627 40881 2997 (1478;1569;-50; · ; ·)
{LC,DC1,DC3} 41355 37492 3090 (1530;685; · ; 875; · )
{LC,DC1,DC4} 41872 38184 2950 (1497;813; · ; · ;640)
{LC,DC2,DC3} 43153 39977 2541 (1197; · ;-75;1419; · )
{LC,DC2,DC4} 43670 40511 2527 (1130; · ;-25; · ;1422)
{LC,DC3,DC4} 40398 37031 2694 (1253; · ; · ;609;832)
{LC,DC1,DC2,DC3} 48836 44366 3576 (1506;1034;153;883; ·)
{LC,DC1,DC2,DC4} 49353 44553 3840 (1519;1104;260; · ;957)
{LC,DC1,DC3,DC4} 46081 40338 4594 (1619;1061; · ;976;938)
{LC,DC2,DC3,DC4} 47879 42267 4490 (1522; · ;486;1238;1244)
{LC,DC1,DC2,DC3,DC4} 53562 45750 6250 (1930;1144;563;1266;1347)
Table 7 Possible sequential coalitions for grand coalition based on SMP
1  {LC, DC3, DC1, DC4, DC2}
Participant i LC DC3 DC1 DC4 DC2
 (i,  ,1) 4.8% - - - -
 (i,  , 2) 4.85% 20.5% - - -
 (i,  ,3) 4.86% 20.8% 12.1% - -
 (i,  , 4) 5.15% 23.2% 18.7% 19.8% -
 (i,  ,5) 6.1% 30% 20.1% 28.5% 7.5%
 2  {DC3, LC, DC1, DC4, DC2}
Participant i DC3 LC DC1 DC4 DC2
 (i,  ,1) 20.3% - - - -
 (i,  , 2) 20.5% 4.85% - - -
 (i,  ,3) 20.8% 4.86% 12.1% - -
 (i,  , 4) 23.2% 5.15% 18.7% 19.8% -
 (i,  ,5) 30% 6.1% 20.1% 28.5% 7.5%
Table 8 Possible sequential coalitions except LC for grand coalition based on SMP
1  {DC3, DC1, DC2, D4}  2  {DC3, DC1, DC4, DC2}
Participant i DC3 DC1 DC2 DC4 Participant i DC3 DC1 DC4 DC2
 (i,  ,1) 20.5% - - -  (i,  ,1) 20.5% - - -
 (i,  , 2) 20.8% 12.1% - -  (i,  , 2) 20.8% 12.1% - -
 (i,  ,3) 20.9% 18.2% 2.0% -  (i,  ,3) 23.2% 18.7% 19.8% -
 (i,  , 4) 30% 20.1% 7.5% 28.5%  (i,  , 4) 30% 20.1% 28.5% 7.5%
 3  {DC4, DC3, DC1, DC2}  4  {DC4, D3, D2, D1}
Participant i DC4 DC3 DC1 DC2 Participant i DC4 DC3 DC2 DC1
 (i,  ,1) 17.2% - - -  (i,  ,1) 17.2% - - -
 (i,  , 2) 17.6% 14.5% - -  (i,  , 2) 17.6% 14.5% - -
 (i,  ,3) 19.8% 23.2% 18.7% -  (i,  ,3) 26.3% 29.4% 6.5% -
 (i,  , 4) 28.5% 30% 20.1% 7.5%  (i,  , 4) 28.5% 30% 7.5% 20.1%
Table 9 New possible sequential coalitions for grand coalition based on SMP
1  {LC, DC3, DC1, DC2, DC4}
Participant i LC DC3 DC1 DC2 DC4
 (i,  ,1) 6.0% - - - -
 (i,  , 2) 6.7% 20.5% - - -
 (i,  ,3) 7.3% 20.8% 12.1% - -
 (i,  , 4) 7.6% 20.9% 18.2% 2.0% -
 (i,  ,5) 11.1% 30% 20.1% 7.5% 28.5%
 2  {LC, DC3, DC1, DC4, DC2}
Participant i LC DC3 DC1 DC4 DC2
 (i,  ,1) 6.0% - - - -
 (i,  , 2) 6.7% 20.5% - - -
 (i,  ,3) 7.3% 20.8% 12.1% - -
 (i,  , 4) 8.8% 23.2% 18.7% 19.8% -
 (i,  ,5) 11.1% 30% 20.1% 28.5% 7.5%
Appendix

Table A1 The distances (miles) between and logistics facilities and first twenty
distribution units
C1 C2 C3 C4 C5 C6 C7 C8 C9 C10
DC1 23.9 17.1 14.6 18.3 36.9 48.6 13.2 13.1 12.8 40.5
DC2 51.8 53.1 49.1 43.4 20.1 21.3 54.1 54.9 38.6 23.3
DC3 75.8 74.5 74.2 68.1 75.4 82.2 73.1 71.9 96.9 108.5
DC4 69.1 57.9 61.6 64.3 73.9 81.1 54.4 51.3 98.6 112.2
LC 51.2 51.9 48.4 43.9 49.9 56.9 47.1 48.1 62.5 84.1
C11 C12 C13 C14 C15 C16 C17 C18 C19 C20
DC1 58.9 78.4 14.5 11.7 11.3 28.5 42.6 55.2 81.9 17.6
DC2 25.6 28.6 73.9 64.7 55.7 33.3 15.6 13.1 19.3 68.3
DC3 106.7 117.1 105.3 98.7 94.7 101 107.1 112.3 117.4 91.2
DC4 116.8 119.6 122.1 74.9 90.5 112.3 112.5 117.6 121.7 64.3
LC 80.1 92.2 70.6 63.1 53.4 57.7 69.1 83.4 93.5 53.3
C21 C22 C23 C24 C25 C26 C27 C28 C29 C30
DC1 8.3 12.8 8.2 36.8 84.9 87.5 35.5 38.8 44.1 28.1
DC2 58.4 50.5 43.6 16.8 12.8 17.8 78.2 66 73.5 70.2
DC3 85.6 82.8 87.1 86.3 109.9 111.6 95.5 83.6 92.9 86.7
DC4 67.3 65.9 72.1 80.9 111.2 115.1 52.3 56.2 46.1 59.4
LC 51.1 49.4 49.3 49.2 70.5 73.7 61.9 45.9 57.8 49.8
C31 C32 C33 C34 C35 C36 C37 C38 C39 C40
DC1 39.1 43.1 48.8 53.3 84.9 67.5 76.7 83.5 52 56.7
DC2 58.6 36.5 27 20.5 11.4 10.3 11.1 8.2 75.2 71.6
DC3 71.5 71.9 79.3 82.5 107.2 65.8 68.4 79.1 82.3 78.4
DC4 83.7 92.8 97.9 105.4 115.5 107.2 110.4 115.7 42.3 69.1
LC 37 32.9 39.8 44.3 71.9 50.2 44.1 56.9 62.7 46.6
C41 C42 C43 C44 C45 C46 C47 C48 C49 C50
DC1 39.5 49.9 63.4 83.2 77.5 64.4 58.7 48.5 61.6 89.1
DC2 40.6 44.3 35.4 89 82.7 75 66.9 61.5 39.7 98.2
DC3 62.8 66.8 72.1 87.7 86.6 82.2 61.3 69.9 63.4 81.7
DC4 85.6 87.9 95.4 40.4 24.7 37.4 68.2 68.6 86 30.8
LC 25 24.3 28.8 80.8 69.6 56.5 19.1 25.7 21 73.9
C51 C52 C53 C54 C55 C56 C57 C58 C59 C60
DC1 79.6 81.9 48.4 61.3 63.7 70.5 83 73.1 77.9 80.7
DC2 95.8 99.7 83.5 78.4 47.3 41.3 120.3 112.4 105.1 101.9
DC3 83.8 72.3 69.8 58.7 69.2 79.5 68.2 62.8 53.4 52.5
DC4 33 27.9 48.9 55.5 86.1 88.4 26.7 19.8 17.5 19.2
LC 71.3 66.7 49.5 43.9 16.8 22.4 72.2 66 55.1 50.7
C61 C62 C63 C64 C65 C66 C67 C68 C69 C70
DC1 72.2 78.3 65.2 67.5 70.9 65.6 66.9 76 84.8 82.5
DC2 77 85.5 68.3 71.3 72.8 66 54.6 71.2 69.3 63.6
DC3 45.9 50.4 51.4 31.7 35.6 45.4 52.3 38.8 43.8 58.5
DC4 47.4 33.9 53.5 45.9 58.3 68.5 79.3 65.3 76.4 91.2
LC 21.9 25.3 16.4 17 13.7 15.2 12.3 26.2 18.8 29.1
C71 C72 C73 C74 C75 C76 C77 C78 C79 C80
DC1 88.3 90.1 84.9 82.1 75.9 82.3 87.6 92.8 99.3 103.6
DC2 98.2 100.8 93.7 89.4 80.8 85.4 91.4 95.8 103.9 105.6
DC3 39.1 37.5 32.3 31.7 26.5 23.1 20.6 10.4 19.7 17.6
DC4 29.5 36.4 35.8 36.1 44.1 50.6 55.3 63.5 40.6 44.3
LC 39 45 35.6 31.9 20.6 30.3 33.9 38 48.6 51.6
C81 C82 C83 C84 C85
DC1 104.6 103.9 100.2 100.6 99.4
DC2 107.9 103.3 102.1 105.7 98.4
DC3 20.6 16.7 12.3 18.2 12.9
DC4 66.1 69.3 69.4 77.3 71.2
LC 60.1 55.7 53.7 54.1 45.4
Note: DCi=The ith distribution center; LC=logistics center; Ci=The ith distribution unit

You might also like