Expert Systems with Applications 40 (2013) 551562

Contents lists available at SciVerse ScienceDirect

Expert Systems with Applications

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

Location and allocation decisions for multi-echelon supply chain network

A multi-objective evolutionary approach
B. Latha Shankar a, S. Basavarajappa b, Jason C.H. Chen c,, Rajeshwar S. Kadadevaramath a
a

Dept. of Industrial Engg. & Management, SIT, Tumkur, India

Mechanical Department, UBDT, Davangere, India
c
Graduate School of Business, Gonzaga University, Spokane, WA, USA
b

a r t i c l e

i n f o

Keywords:
Four-echelon supply chain architecture
Evolutionary approach
Non-dominated sorting algorithm
MOHPSO
Swarm intelligence

a b s t r a c t
This paper aims at multi-objective optimization of single-product for four-echelon supply chain architecture consisting of suppliers, production plants, distribution centers (DCs) and customer zones (CZs). The
key design decisions considered are: the number and location of plants in the system, the ow of raw
materials from suppliers to plants, the quantity of products to be shipped from plants to DCs, from
DCs to CZs so as to minimize the combined facility location and shipment costs subject to a requirement
that maximum customer demands be met. To optimize these two objectives simultaneously, four-echelon network model is mathematically represented considering the associated constraints, capacity, production and shipment costs and solved using swarm intelligence based Multi-objective Hybrid Particle
Swarm Optimization (MOHPSO) algorithm. This evolutionary based algorithm incorporates non-dominated sorting algorithm into particle swarm optimization so as to allow this heuristic to optimize two
objective functions simultaneously. This can be used as decision support system for location of facilities,
allocation of demand points and monitoring of material ow for four-echelon supply chain network.
 2012 Elsevier Ltd. All rights reserved.

1. Introduction
Supply chain (SC) is an integrated system of facilities and activities that synchronizes inter-related business functions of material
procurement, material transformation to intermediates and nal
products and distribution of these products to customers. Supply
chain management is a set of approaches utilized to efciently
integrate suppliers, manufacturers, warehouses, and stores, so that
merchandise is produced and distributed at the right quantities, to
the right locations, and at the right time, in order to minimize system-wide costs while satisfying service level requirements
(Simchi-Levi, Kaminsky, & Simchi-Levi, 2000). Above denition reveals that there are many independent entities in a supply chain
each of which tries to maximize its own inherent objective functions in business transactions. This is a complicated problem as
too many factors are involved and needs more than one objective
to be satised simultaneously. Such a problem is called multiobjective optimization problem and has many Pareto solutions.
The nal decision is made taking the total balance over all criteria
into account. This balancing over criteria is called trade-off.
Since today, the success measures for the companies are
thought as lower costs, shorter production time, shorter lead time,
Corresponding author.
E-mail address: chen@jepson.gonzaga.edu (J.C.H. Chen).
0957-4174/$ - see front matter  2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.eswa.2012.07.065

less stock, larger product range, more reliable delivery time, better
customer services, higher quality, and providing the efcient coordination between demand, supply and production, the trade-off
between cost investment and service levels may change over time.
Hence the supply chain performance needs to be evaluated continuously and supply chain managers should make timely and right
decisions (Shen, 2007).
The key issues in supply chain management can broadly be divided into three main categories: (i) supply chain design (ii) supply
chain planning and (iii) supply chain control. In the supply chain
design phase, strategic decisions, such as facility location decisions
and technology selection decisions play major roles. It is very
important to design an efcient supply chain to facilitate the
movements of goods. These strategic decisions lead to costly, time
consuming investment as the facilities located today, are expected
to remain in operation for long time. Environmental changes during the facilitys lifetime can drastically alter the appeal of a particular site, turning todays optimal location into tomorrows
investment blunder. Determining the best locations for new facilities is thus an important strategic challenge (Owen & Daskin,
1998). Once the supply chain conguration is determined, the focus shifts to decisions at the tactical and operational levels, such
as inventory management decisions on raw materials, intermediate products, and end products and distribution decisions within
the supply chain (Chopra & Meindl, 2005).

552

B. Latha Shankar et al. / Expert Systems with Applications 40 (2013) 551562

In traditional supply chain management, the focus of the designs of supply chain network is usually on single objective, minimum cost or maximum prot. But the design, planning and
scheduling projects are usually involving trade-offs among different incompatible goals such as fair prot distribution among all
members, customer service levels, ll rates, safe inventory levels,
volume exibility, etc. (Chen & Lee, 2004). Hence real supply
chains are to be optimized simultaneously considering more than
one objective. Many of the problems that occur in supply chain
optimization are combinatorial in nature and picking a set of optimal solutions in the case of multi-objective formulations requires a
algorithm that can efciently search the entire objective space
using small amounts of computation time. Literature shows that
evolutionary algorithms perform well in this respect and give good
optimal results when applied to many combinatorial problems.
This work proposes the utility of non-dominated sorting particle
swarm optimization algorithm for simultaneous optimization of
two objectives, minimizing total supply chain cost and maximizing
ll rate for a four-echelon supply chain architecture so as to arrive
at an efcient supply chain design and optimal transportation/
shipment plan which can be used as decision support system.

2. Prior related work

Since this work considers location decisions, allocation decisions, multi-objective optimization using particle swarm optimization algorithm, this section deals with prior work related to all
these areas. Supply chain management has vast scope and encompasses the decisions about (1) where to produce, what to produce,
and how much to produce at each site, (2) what quantity of goods
to hold in inventory at each stage of the process, (3) where to locate plants and distribution centers etc. Out of these, location decisions may be the most critical and most difcult of the decisions
needed to realize an efcient supply chain as their effect is long
lasting. Both the cost of whole supply chain and the level of customer service provided by the system are signicantly affected
by the number, size and locations of facilities, as well as by the
allocation decisions as to which customers to be served from which
upstream supplier. Consequently, a signicant amount of research
has been devoted to the development of efcient supply chain design. The early study of location theory begin in 1909 with Alfred
Webers work on positioning a single warehouse so as to minimize
the total distance between it and several customers (Weber, 1929).
After that considerable work in location theory is done by Hakimi
(1964), who work on locating switching centers in a communications network and police stations in a highway system. Later many
researchers work on basic facility location problem formulations
recognized as static and deterministic which take constant, known
quantities as inputs and derive a single solution to be implemented
at one point in time. These fundamental location problems are categorized into median problems (Hakimi, 1964; ReVelle, 1986), covering problems (Daskin, 1995; Church & ReVelle, 1976), center
problems (Daskin, 1995), etc. Later focus is shifted to location-allocation problems which simultaneously locate facilities and dictate
ows between facilities and demands. These problems are reviewed by Scott (1971). Additional variety of problems include
models with multiple commodities, unreliable supply, etc. Warszawski and Peer (1973) and Warszawski (1973) are among the
rst to study the multi-commodity location problem. These models
consider xed location costs and linear transportation costs, and
assume that each warehouse can be assigned at most one commodity. Geoffrion and Graves (1974) consider the capacitated version of the multi-commodity location problem and present a
model to solve the problem of designing a distribution system with
optimal location of the intermediate distribution facilities between

plants and customers. The risks arising from the use of heuristics in
distribution planning are discussed early on by Geoffrion and Vanroy (1979). They present three examples in the area of distribution
planning demonstrating the failure of common sense methods to
come up with the best solution.
In literature, another set of problems considered is called xed
charge facility location problems which consider xed charge associated with locating at each potential facility site. There are two
types of problems capacitated and uncapacitated plant location
problems. Uncapacitated and capacitated plant location models
are extensively dealt in Mirchandani and Francis (1990) and ReVelle, Eiselt, and Daskin (2008) and capacitated plant location models in Sridharan (1995). Also basic facility location problems are
given a new orientation with integrated approach. This is due to
the realization of fact that location decisions taken without considering inventory and shipment costs can lead to sub-optimality.
Hence facility location models are developed considering location
associated costs as well as production, inventory and distribution
costs. Integrated decision making models in particular focus at
coordination of any two of the three important supply chain decisions. Based on the factors considered they are catagorized into (1)
locationrouting (LR) models; (2) inventoryrouting (IR) models;
and (3) locationinventory (Li) models. These problems are extensively reviewed by Shen (2007).
Dynamic facility location models next evolved make an attempt
to capture many of the characteristics of real-world location problems. Ballou (1968) rst use dynamic programming to determine
optimal location and relocation strategy for the planning period.
Wesolowsky (1973) examines another, unconstrained, version of
the single facility location problem over a nite planning horizon
with explicit facility relocation costs. Scott (1971) develop multiple
dynamic facility location-allocation problem. Wesolowsky and
Truscott (1976) present an integer programming model to extend
the analysis of multi-period node location-allocation problems,
allowing facilities to be relocated in response to predicted changes
in demand. Erlenkotter (1981) compares the performance of several heuristic solution approaches on a single problem formulation.
He examines a dynamic, xed charge, capacitated, cost minimization problem with discrete time intervals.
In traditional supply chain management, minimizing costs or
maximizing prot is the primitive objective in most of the supply
chain network design models (Cohen & Lee, 1989; Tsiakis, Shah,
& Pantelides, 2001). But for a supply chain, producing products at
minimum cost is not the only objective, satisfying customers is
also equally important. Later some researchers start incorporating
more than one competing objectives such as improving customer
service and reducing cost in their models. Different methodologies
found in literature for treating multiobjective optimization problems are the weighted-sum method, the e-constraint method and
the goal-programming method, fuzzy method, etc. (Azapagic &
Clift, 1999; Chen & Lee, 2004; Cheng-Liang, Wang, & Wen-Cheng,
2003; Zhou, Cheng, & Hua, 2000).
Sabri and Beamon (2000) develop a model for supply chain
management by combining strategic and operational design and
planning decisions and solve it using an iterative procedure. They
adopt multi-objective decision analysis and optimize simultaneously cost, ll rate and exibility. Nozick and Turnquist (2001)
present an optimization model which minimizes cost and maximizes service. They use a linear function to approximate the safety
stock inventory cost function, which is then embedded in a xedcharge facility location model. For a review of other multi-objective location models, publication by Shen, Coullard, and Daskin
(2003) can be referred. Chen and Lee (2004) propose a model
which simultaneously optimize conicting objectives such as each
participants prot, the average customer service level, and the
average safe inventory level. Guillna, Melea, Bagajewiczb,

B. Latha Shankar et al. / Expert Systems with Applications 40 (2013) 551562

Espuaa, and Puigjanera (2004) formulate a multiobjective stochastic Mixed Integer Linear Programming model for supply chain
design, which is solved by using the standard e-constraint method,
and branch and bound techniques. This formulation takes into account not only Supply Chain prot and customer satisfaction level,
but also uncertainty by means of the concept of nancial risk,
which is dened as the probability of not meeting a certain prot
aspiration level. Shen (2006) addresses prot-maximizing supply
chain design model where in a company can choose whether to
satisfy a customers demand.
Particle swarm optimization (PSO) is one of the evolutionary
computation techniques and is a population-based search algorithm which is initialized with a population of random solutions
called particles. Each particle in PSO is also associated with a velocity. Particles y through the search space with velocities which are
dynamically adjusted according to their historical behaviors. Due
to this the particles have a tendency to y towards the better
and better search area over the course of search process. This technique is originally proposed by Kennedy and Russell Eberhart as a
simulation of social behavior, and it is initially introduced as an
optimization method in 1995 (Kennedy & Eberhart, 1995). PSO
can be easily implemented and it is computationally in expensive,
since its memory and CPU speed requirements are low. The main
difculty in Multi-objective Particle Swarm Optimization (MOPSO)
is to pick suitable global best (gbest) and personal best (pbest) to
move the particles through search space. In general, a good MOPSO
method must obtain solutions with (a) a good convergence, and (b)
a diversity spread along the Pareto-optimal front (Coello, 1999).
Parsopoulos and Vrahatis (2002) are one among to study rst
the performance of the PSO in multi-objective optimization problems and nd Pareto front in weighted aggregation cases. Hu and
Eberhart (2002) present a MOPSO that uses dynamic neighborhood
strategy to obtain the best local guide for each particle in multiobjective optimization problems. Coello and Lechuga (2002) propose a MOPSO where the objective space is divided to hyper cubes.
Fieldsend and Singh (2002) propose a MOPSO which uses a dominated tree for the search of the best local solution. Mostaghim and
Teich (2003) propose a sigma method and turbulence operator to
improve the convergence and diversity for multi-objective optimization. Hu, Eberhart, and Shi (2003) adopt a secondary population
called extended memory to improve their previous dynamic neighborhood PSO approach presented in (2002). Mostaghim and Teich
(2004) propose a new method which divides population of the covering MOPSO into sub-swarms. The sub-swarms try to cover the
gaps between the non-dominated solutions found in the initial
run. In later research in the eld of MOPSO, different approaches
have been introduced to identify the local best solution based on
memorizing all the non-dominated solutions. These approaches
when implemented on standard test problems show that keeping
the particle archive improves signicantly the effectiveness of
the technique (Jrgen & Sanaz, 2006). Mahnam, Yadollahpour, Famil-Dardashti, and Hejazi (2009) develop an inventory model for
an assembly supply chain network where in the performance of
supply chain is assessed by two criteria total cost and ll rate using
a fuzzy expert system. To solve this bi-criteria model, hybridization
of multi-objective particle swarm optimization and simulation
optimization are considered. Results indicate the efciency of proposed approach in performance measurement. Guo, Li, Mileham,
and Owen (2009) apply effectively a variant of PSO to integrate
process planning and scheduling. Recently Guo and Hou (2010)
develop a simulation model to describe the operations of a
third-party logistics provider. They develop an MOPSO algorithm
combining with the simulation model to identify non-dominated
solutions that constitute the trade-off curve between ll rate and
total inventory. Marinakis and Marinaki (2010) propose a new
hybrid algorithmic nature inspired approach based on particle

553

swarm optimization (PSO), Greedy Randomized Adaptive Search

Procedure (GRASP) and Expanding Neighborhood Search (ENS)
Strategy and apply it to the Probabilistic Traveling Salesman
Problem (PTSP). They test the algorithm on numerous benchmark
problems and show satisfactory results.
Tsou, Yang, Chen, and Lee (2011) present a bi-objective model
to represent a xed order system under lost sales. They consider
cycle stock investment and service level to be simultaneously optimized. They utilize a solver based on MOPSO to nd the inventory
management policies and show that combining local search and
MOPSO will hopefully improve the chance of nding better
trade-offs in inventory cost and customer service. Study of Chengming (2011) focus on swarm intelligence where particles adjust
their ying time adaptively through generations. This insures escape from local minima. They call this algorithm as Improved Discrete Particle Swarm optimization and apply it to optimize
capacitated vehicle routing problem. An integrated optimization
approach using an articial neural network and a bidirectional particle swarm is proposed by Navalertporn and Afzulpurkar (2011).
They apply the approach to the real-life ceramic tile pressing process where multiple responses are considered and their experimental results indicate superior performance of the proposed
approach over the conventional ones.
The work of Mukhopadhyay and Banerjee (2012) emphasizes
the stochastic nature of chaotic carrier by embedding the merits
of it in a Swarm Intelligence optimization technique. The outcome
is a modied improved version of particle swarm optimization and
they apply this for global optimization of the unknown parameters
of a laser system derived from MaxwellBlochs Equations. Their
comparative analysis reveals that the proposed version is effectively tackling the problem of premature convergence with enhanced convergence speed. An optimization hybrid swarm
algorithm namely particle-bee algorithm is developed by Lien
and Cheng (2012). They apply their algorithm to the construction
site layout (CSL) design problems and show that the above said
algorithm can be efciently employed to solve CSL problems with
high dimensionality. Mousa, El-Shorbagy, and Abd-El-Wahed
(2012) propose a hybrid multi-objective evolutionary algorithm
combining two heuristic optimization techniques, genetic algorithm (GA) and particle swarm optimization (PSO). They apply hybrid algorithm to various kinds of multi-objective (MO) benchmark
problems to stress the importance of hybridization algorithms in
generating Pareto optimal sets for multi-objective optimization
problems. Prasanna Venkatesan and Kumanan (2012) propose discrete particle swarm algorithm (MODPS) to optimize the supply
chain network with the objectives of minimization of supply chain
cost, minimization of demand fulllment lead time and maximization of volume exibility. In the rst stage the performance of two
global guide selection techniques are evaluated and in the second
stage proposed (MODPS) is compared with non-dominated sorting
genetic algorithm-II. The results indicate that the proposed
approach is effective in producing high-quality Pareto-optimal
solutions. Che (2012) present a modied PSO method formultiechelon unbalanced supply chain planning. The experimental results
infer that the modied method they develop obtain a better quality
solution compared to classical GA and PSO. Kadadevaramath, Chen,
Latha Shankar, and Rameshkumar (2012) has proposed modeling
and optimization of a three echelon supply chain network using
the particle swarm optimization to address the demand uncertainty and constraints posed by the every echelon in the supply
chain design operations.
The above literature survey indicates that very little research
has been carried out to implement MOPSO algorithm in supply
chain network optimization. The purpose of this paper is to formulate and analyze a strategic plant location-allocation model for single product four-echelon supply chain network with respect two

554

B. Latha Shankar et al. / Expert Systems with Applications 40 (2013) 551562

conicting objectives; minimizing total supply chain cost consisting of production, transportation and distribution cost and maximizing ll rate using MOHPSO algorithm. The proposed strategy
should lead to a nal SC design which would represent the desired
compromise among the different objectives from the decision-makers perspective.

3. Modeling and mathematical formulation

This section discusses the objectives of the network optimization model, modeling of the supply chain network (SCN) and the
mathematical programming of the designed model. During supply
chain network design, in one of the phases, managers must decide
on location and capacity allocation for each facility which can be
made jointly. The demand allocation decisions can be altered on
regular basis as costs change and markets evolve (Chopra &
Meindl, 2005). This work considers a general supply chain network
consisting of four different level enterprises. The rst-level enterprise is the CZ from which the products are sold to customers subject to a given lower bound of customer service. The second-level
enterprise is the DC from which the products are transported to
CZ. The third level enterprise is the plant or manufacturer and
fourth level is the supplier. Revenues come from sale of products
and costs arise from facilities, labor, transportation, material and
inventories. In order to overcome difculty in mathematical formulation and convergence, a minimum target for the demand satisfaction, which must be attained in all the scenarios, is
incorporated as a constraint within the existing formulation.
Assumptions

Objective functions
 To minimize total cost of supply chain which includes raw
material making and transportation, plant location, production
and inventory costs in plant, distribution cost from plant to
DC, throughput cost in DC, distribution cost from DC to CZ, etc.
 To maximize the ll rate.
Product availability reects a rms ability to ll a customer
order out of available inventory. A stock-out results if a customer
order arrives when product is not available. Product ll rate (fr)
and cycle service level (CSL) are the ways to measure product
availability. Product ll rate is the fraction of product demand that
is satised from product in inventory. CSL is the fraction of replenishment cycles that end with all the customer demand being met
(Chopra & Meindl, 2005). This work considers a constrained
bi-objective problem formulation for a pull based supply chain
and proposes the utility of using hybrid MOPSO algorithm for the
simultaneous optimization of both cost and product ll rate.
4. Problem description
This work considers four-echelon capacitated plant locationallocation network model as shown in Fig. 1. The notations used
in this network model are listed in Table 1 with their meaning.
The model denes decision variables as shown in Table 2.
The problem is then formulated as the following mixed integer
model:

Objective 1 : Min
Product produced needs three raw materials.
Capacities of suppliers for different raw materials, cost of making each raw material at each supplier, cost of transportation
per raw material per supplierplant pair are known in advance.
Potential location sites for plants and DCs are known.
Production, inventory costs per product and distribution costs
per plantDC pair are known and xed.
Inventory costs per product and cost of transportation per DC
CZ pair are known and xed.
Minimum ll rate is to be maintained.
The overall problem can thus be stated as follows:
Given
 Number of suppliers and their capacities for each raw material.
 Number of potential plants, DCs and their capacities.
 Cost of making of each raw material at each supplier and transportation cost per supplierplant pair.
 Production and inventory costs in each plant.
 Distribution costs per product per plantDC pair.
 Product throughput cost in each DC and transportation cost per
DCCZ pair.
 Number of CZs and their demands.
 Minimum ll rate (fraction of demand satised) to be
maintained.
To determine
 Quantity of each raw material to be shipped from each supplier
to a plant.
 Number and location of plants.
 Flow of products from located plants to DCs.
 Number and location of DCs.
 Allocation of CZs to DCs.

p
n
t
n X
l X
X
X
X
fi yi
fe ye
cchi xhci
e1

i1

i1 h1 c1

n X
t
t X
m
X
X

cie xie
cej xej
i1 e1

Pt

e1

Objective 2 : Max

e1 j1

Pm

Pm

j1 xej

j1 Dj

Subject to

3
n
X
xhci 6 Sch for h 1; . . . :l; c 1; ::p

i1
t
X
xej 6 Dj for j 1; . . . ; m

e1
t
X
xie 6 K i yi for i 1; . . . :n
e1
m
X

xej 6 K e ye for e 1; . . . ; t

6
7

j1
l
t
X
X
xhci 
xie P 0 for i 1; . . . ; n;
h1

e1

c 1; . . . ; p
n
m
X
X
xie 
xej P 0 for e 1; . . . ; t
i1

0:80 6

j1

Pt

e1

8
9

Pm

Pm

j1 xej

j1 Dj

61

yi ; ye 2 f0; 1g for i 1 . . . . . . n; e 1; . . . ; t

10
11

The objective function (1) minimizes the total cost (xed + variable)
of setting up and operating the network and objective function (2)
maximizes ll rate.
The constraint in Eq. (4) species that the total quantity shipped
from a supplier cannot exceed the suppliers capacity. The

B. Latha Shankar et al. / Expert Systems with Applications 40 (2013) 551562

555

Fig. 1. Four-Echelon supply chain network of 3 suppliers, 5 plants, 6 dcs and 7 CZs.

Table 1
Notations and their meaning used in the mathematical formulation.
Notations
used

Meaning

Number of customer zones (markets) or demand points

(j = 1, 2 . . . , m)
Number of warehouse (DC) locations (e = 1, 2, . . . , t)
Number of potential plant locations (i = 1, 2 . . . , n)
Number of suppliers (h = 1, 2, . . ., l)
Number of components (c = 1, 2, . . . , p)
Average demand from market j/period
Potential capacity of plant i
Potential capacity of warehouse e
Supply capacity at supplier h for component c
Annualised xed cost of keeping plant i open/year
Annualised xed cost of keeping warehouse e open/year
Cost of making and shipping a component c from supply source
h to plant i/unit
Cost of producing, stocking and shipping one unit from plant i
to warehouse e/unit
Cost of throughput and shipping one unit from warehouse e to
customer zone j/unit
Cost of making a component c by supplier h/unit
Transportation cost of a component c from supplier h to plant i/
unit
Manufacturing cost at plant i/unit
Inventory cost at plant i/unit
Inventory of component c at plant i
Plant transportation cost from plant i to warehouse e/unit

t
n
l
p
Dj
Ki
Ke
Sch
fi
fe
cchi
cie
cej
CSch
STCchi
ci
ICi
Ici
PTCie

Table 2
Decision variables used in the mathematical formulation.
Notations used

Meaning

yi
ye
xhci
xie
xej

1, if plant i is open, 0 otherwise.

1, if warehouse e is open, 0 otherwise.
Quantity of component c shipped from supplier h to plant i
Quantity shipped from plant i to warehouse e
Quantity shipped from warehouse e to customer zone j

that no warehouse can supply more than its capacity. The

constraint in Eq. (8) states that the quantity shipped out of a plant
cannot exceed the quantity of raw material received. The
constraint in Eq. (9) states that the quantity transported from a
warehouse cannot exceed the quantity received. The constraint
in Eq. (10) states that ll rate can vary from 80% to 100%. The
constraint in Eq. (11) enforces that each plant is either open
(yi = 1) or closed (yi = 0).
The solution has the decision variables which characterize the
network conguration. They include binary variables that represent the existence of plants and warehouses (DCs) of the SC and
the continuous ones that represent the ows of materials from various suppliers to various plants from plants to different warehouses and the allocation of regional demand to these warehouses.

5. Methodology
constraint in Eq. (5) requires that the demand at each regional
market be satised to the maximum extent. The constraint in Eq.
(6) states that no plant can supply more than its capacity. Capacity
is 0 if the plant is closed and Ki if it is open. Similarly Eq. (7) states

5.1. Introduction to particle swarm optimization algorithm

Particle swarm optimization (PSO) algorithm is a promising
new population-based optimization technique developed by

556

B. Latha Shankar et al. / Expert Systems with Applications 40 (2013) 551562

Kennedy and Eberhart (1995). This algorithm models a set of potential problem solutions as a swarm of particles moving about
in a virtual search space. PSO is based on the behavior of communities that have both social and individual communication, similar
to birds searching for food. The bird would nd food through social
cooperation with other birds around it (within its neighborhood).
In PSO algorithm, each individual (particle) represents a solution in a n-dimensional space. Each particle also has knowledge
of its previous best experience and knows the global best experience (solution) found by the entire swarm. Each particle updates
its direction using the following equations:

v ij w  v ij c1  r1  pij  xij c2  r2  pgj  xij

xij xij v ij
where w is the inertia factor inuencing the local and global abilities of the algorithm, vij is the velocity of the particle i in the jth
dimension, c1 and c2 are weights affecting the cognitive and social
factors, respectively. r1 and r2 are uniform random variables between 0 and 1, and two random values are generated independently. pij stands for the best value found by particle i (pbest) and
pg denotes the global best found by the entire swarm (gbest). After
the velocity is updated, the new position i in its jth dimension is calculated. This process is repeated for every dimension and for all the
particles in the swarm. Eventually the swarm as a whole, like a ock
of birds collectively foraging for food, is likely to move close to an
optimum of the tness function. PSO algorithm is becoming very
popular due to its simplicity of implementation and ability to
quickly converge to a reasonably good solution. This is achieved
by adjusting easily few parameters in PSO algorithm to have more
global search ability at the beginning of the run and have more local
search ability near the end of the run.
5.2. Particle swarm model for binary decision
This model proposes that the probability of an individuals
deciding yes or no, true or false or making some other binary decision, is a function of personal and social factors and is given by,

Pxid t 1 f xid t  1; v id t  1; pid ; pgd

to represent a positive random number drawn from a uniform distribution with a predened upper limit so that the two u limits sum
to 4.0. Thus the formula for binary decision is

v id t v id t  1 u1 pid  xid t  1 u2 pgd  xid t  1

If qid < sv id t then xid t 1; else xid t 0
Then qid is a vector of random numbers, drawn from a uniform distribution between 0.0 and 1.0. The constant parameter Vmax is often
set at .0, so that there is always at least a chance of
s(Vmax) 0.0180 that a bit will change state (Kennedy & Eberhart,
2001).
5.3. The particle swarm in continuous numbers
The position of a particle i is assigned the algebraic vector symbol x~i . There can be any number of particles and each vector can be
of any dimension. Change of position of a particle is called v~i the
velocity. Velocity is a vector of numbers that are added to the position co-ordinates in order to move the particle from one time step
to another. As the system is dynamic, position of each individual is
changing. The direction of movement is a function of current position and velocity, the location of individuals previous best success
called pbest and best position found by any member in the swarm
called gbest of the neighborhood (Kennedy & Eberhart, 2001).
5.4. The hybrid PSO
A hybrid swarm optimizer combines both binary and real valued parameters in one search. This hybrid optimizer is used for
such problems which deal with both continuous and binary variables. It simply operates on binary inputs with binary particle
swarm algorithm and treat the continuous variables with real valued particle swarm. Binary PSO algorithm is used to take the location decisions (whether to locate a facility at a given candidate
site), while the allocation decisions are obtained by continuous
PSO algorithm (Kennedy & Eberhart, 2001).
5.5. Multi-objective optimization using MOHPSO

where
 P(xid(t) = 1) is the probability that individual i will choose 1.
 (xid(t)) is the current state of individual i.
 t is the current time step, and t  1 is the previous step.
 vid(t  1) is a measure of the individuals current probability of
deciding 1.
 pid is the best state found so far, for example, it is 1 if the individuals best success occurred when xid was 1 and 0 if it was 0. It
is referred as pbest.
 pgd is the neighborhood best, again 1 if the best success attained
by any member of the neighborhood was when it was in the 1
state and 0 otherwise. It is referred as gbest.
The parameter vid(t), will determine the probability threshold. If
vid(t) is higher, the individual is more likely to choose 1, and lower
values favor the 0 choice. Such a threshold stays in the range
[0.0, 1.0]. The following sigmoid function is used to determine
the probability threshold.

sv id

1
1 expv id

In any situation, whether individual learning (pbest) is stronger or

social-inuence (gbest) is stronger is unknown, both will be
weighed by random numbers, so that sometimes the effect of one
and sometimes the other will be stronger. The symbol u is used

The general Multi-Objective Optimization can be dened as

the process of simultaneously optimizing two or more conicting
objectives subjected to certain constraints. Since multi-objective
optimization problem has several objective functions, the solution method aims at nding several trade-offsolutions rather
than a single solution. The concept of Pareto optimality in multi-objective optimization is the one proposed by Vilfredo Pareto
in 1986.
A point x 2 X is Pareto optimal if "x 2 X and I = {1, 2, . . . , k}
either:
"i 2 I(fi(x) 6 fi(x)) and, there is at least one i e I such that
fi(x) < fi(x).
This denition says that x is Pareto optimal if there exists no
feasible x which would decrease some criteria without causing a
simultaneous increase in at least one other criterion (Coello &
Lechuga, 2002).
Multi-objective optimization algorithm makes use of domination concept to arrive at optimal solutions. In this algorithm, two
solutions are compared on the basis of whether one dominates
the other solution or not. If there are M objective functions then
a solution x is said to dominate the other solution y, if both the following conditions are true:
1. The solution x is not worse than y in all objectives.
2. The solution x is strictly better than y in at-least one objective.

557

B. Latha Shankar et al. / Expert Systems with Applications 40 (2013) 551562

Table 3
Supplier capacities, plant capacities and xed costs.
Vendor capacities

Plants capacities and xed costs

Vendors

Components

V1
V2
V3

C1

C2

C3

36
40
42

62
65
70

50
55
60

Plants

XCapacity

Fixed Cost

P1
P2
P3
P4
P5

18
24
37
22
41

7650
3500
500
4100
2200

Warehouses

Capacity

WH1
WH2
WH3
WH4
WH5
WH6

15
12
14
13
12
16

Table 4
Cost of making and shipping cost matrix per unit for rst stage of SCN.
Vendors

Components

Cost of making the component at vendors place

Supplier transportation costs

Plants
1

V1

C1
C2
C3

300
115
90

10
6
3

13
7
4

8
5
5

11
8
4

15
4
5

V2

C1
C2
C3

320
120
85

17
6
6

14
5
6

12
7
5

12
5
6

15
7
4

V3

C1
C2
C3

290
125
75

13
6
3

12
5
6

14
3
3

11
4
2

9
5
3

Table 5
Manufacturing cost, inventory cost, transportation cost matrix per unit for second stage of SCN.
Plants

P1
P2
P3
P4
P5

Warehouses
WH1

WH 2

WH 3

WH 4

WH 5

WH 6

7
12
8
10
8

12
10
10
12
10

15
11
14
13
11

17
13
15
14
15

18
15
18
13
11

20
17
21
18
12

Manufacturing cost at plants

Inventory cost at plants

3900
2010
1945
1855
1975

50
45
55
48
52

Table 6
Handling cost and transportation cost matrix per unit for third stage of SCN.
Warehouses

WH1
WH2
WH3
WH4
WH5
WH6
Monthly demand Dj

Customer zones

Handling cost at warehouse

CZ1

CZ2

CZ3

CZ4

CZ5

CZ6

CZ7

8
5
9
3
7
5
3

3
8
3
9
6
6
5

9
7
8
2
3
7
4

6
6
6
2
9
8
6

7
3
7
5
4
3
4

3
2
5
4
9
2
5

4
8
4
8
4
9
3

If any of the above conditions is violated, the solution x does not

dominate the solution y. Since multi-objective optimization has
more than one objective to be optimized simultaneously, there
cannot be a single optimum solution which simultaneously optimize all objectives. The resulting outcome is a set of optimal solutions with a varying degree of objective values. This set of solutions
is called non-dominated set. The following two conditions must be
true for a non-dominated set:
1. Any two solutions of non-dominated set must be non-dominated with respect to each other.
2. Any solution not belonging to non-dominated set is dominated
by at least one member of non-dominated set.

55
50
60
54
55
45

Such a non-dominated set is also called as Pareto-optimal set.

Because minimization of total SC cost and maximization of ll
rate cannot be achieved at the same time as they are conicting
in nature, there exists a trade-off between them. This type of system clearly represents a multiobjective optimization situation
which is a procedure looking for a compromise policy, the result,
called a Pareto-optimal or noninferior solution, consists of a number of options. Hence the Pareto set solutions and their corresponding decision variables should be provided, from which the
decision-maker can select a solution to satisfy the industrial need.
Many researchers have suggested many classical multiobjective optimization methods such as weighted sum approach,
e-constraint method, weighted metric method, value function

558

B. Latha Shankar et al. / Expert Systems with Applications 40 (2013) 551562

Table 7
Pareto-front objective function values.
Scenario

Total xed
cost

Total
variable cost

Total SC
cost

Demand
satised

Actual
demand

1
2
3
4
5
6
7

30,050
27,150
29,750
15,600
26,950
20,900
17,950

112,256
114,235
110,879
104,304
90,868
82,390
82,530

142,306
141,385
140,629
119,904
117,818
103,290
100,480

30
29
28
27
26
25
24

30
30
30
30
30
30
30

Table 8
The optimal material ows from suppliers to plants corresponding to Pareto front
points.
Supplier plants

Fig. 2. Feasible solutions where in series 2 represents Pareto-front with six nondominated points showing trade-off between SC cost and ll rate and series 1
represents other inferior points.

Fig. 3. Convergence behavior of HPSO.

method, goal programming method etc. But all these methods convert a multi-objective optimization problem into a single-objective
optimization problem. These methods are subjective and only one
Pareto-optimal solution can be expected in one simulation run.
Since then there has been increasing interest to use evolutionary
algorithms in solving multi-objective optimization because of their
ability to nd multiple optimal solutions in single run and complexity of location-allocation problems.
5.6. Selecting the global best and personal best
In MOHPSO, each particle has to change its position as guided
by two leaders pbest and gbest which must be selected from the
updated set of non-dominated solutions stored in the archive.
The main difculty in MOHPSO is to pick suitable global best
(gbest) and personal best (pbest) to move the particles through
search space to attain a good convergence and diversity along
the Pareto-optimal front.
In this study, method of selection of pbest and gbest used is inspired by NSGA-II. For selecting pbest a method called Prandom is
employed, according to which a single pbest is maintained. Pbest is
replaced if new value < pbest, otherwise, if new value is found to
be mutually non-dominating with Pbest, one of the two is
randomly selected to be the new pbest. ( Eveson, Fieldsend, &
Singh, 2002). To select gbest from archive, this work makes use

Component 1

Component 2

Component 3

V1

V2

V3

V1

V2

V3

V1

V2

V3

Scenario 1
P1
P2
P3
P4
P5

0
5
15
11
0

0
0
0
0
0

8
11
0
0
0

10
10
18
0
0

0
12
0
12
0

0
0
0
0
0

0
0
0
15
0

12
22
12
0
0

0
0
0
0
0

Scenario 2
P1
P2
P3
P4
P5

0
0
10
0
0

1
18
9
0
11

4
0
0
0
0

8
16
0
0
0

0
0
17
0
10

0
0
0
0
0

8
9
0
0
0

0
12
0
0
9

0
0
20
0
0

Scenario 3
P1
P2
P3
P4
P5

7
0
0
8
0

0
0
0
0
0

10
6
0
0
0

13
0
0
10
0

5
8
0
0
0

0
0
0
0
0

5
0
0
7
0

0
0
0
0
0

14
9
0
0
0

Scenario 4
P1
P2
P3
P4
P5

0
0
23
0
0

0
0
0
0
0

0
0
0
15
17

0
0
25
0
20

0
0
0
16
0

0
0
0
0
0

0
0
21
0
0

0
0
0
0
0

0
0
0
18
22

Scenario 5
P1
P2
P3
P4
P5

0
0
0
10
0

11
6
0
0
0

0
0
0
0
0

0
0
0
9
0

0
6
0
0
0

12
0
0
2
0

8
0
0
12
0

3
0
0
0
0

0
7
0
0
0

Scenario 6
P1
P2
P3
P4
P5

0
0
6
0
0

0
7
5
0
0

0
11
0
0
9

0
0
10
0
0

0
0
0
0
12

0
14
0
0
0

0
19
12
0
0

0
0
0
0
7

0
0
0
0
0

Scenario 7
P1
P2
P3
P4
P5

0
0
17
0
0

0
0
0
0
0

9
0
0
0
0

0
0
18
0
0

0
0
0
0
0

10
0
0
0
0

0
0
0
0
0

0
0
20
0
0

11
0
0
0
0

of crowding distance measure and niche count (Deb, 2001). The

non-dominated solutions from the last generations are kept in
the archive. The archive is an external population, in which nondominated solutions are kept after each ight cycle. Such an
archive will allow the entrance of a solution only if: (a) it is nondominated with respect to the contents of the archive or (b) it
dominates any of the solutions within the archive (Coello &
Lechuga, 2002). Finding a relatively large set of Pareto-optimal
trade-off solutions is possible by running the MOHPSO for many
generations.

B. Latha Shankar et al. / Expert Systems with Applications 40 (2013) 551562

Table 9
Optimum location and distribution of products from plants to DCs corresponding to
Pareto front points.
Supply plant

D1

D2

D3

D4

D5

D6

Optimal location
of plants

Scenario 1
P1
P2
P3
P4
P5

0
1
8
0
0

0
3
0
2
0

0
1
0
2
0

2
7
0
2
0

0
0
0
0
0

4
0
2
0
0

1
1
1
1
0

Scenario 2
P1
P2
P3
P4
P5

1
2
2
0
0

0
0
0
0
0

0
0
9
0
0

1
6
0
0
3

0
4
1
0
0

1
2
2
0
4

1
1
1
0
1

Scenario 3
P1
P2
P3
P4
P5

0
0
0
0
0

4
0
0
2
0

1
2
0
2
0

8
0
0
3
0

4
0
0
0
0

0
4
0
0
0

1
1
0
1
0

Scenario 4
P1
P2
P3
P4
P5

0
0
2
2
2

0
0
0
0
0

0
0
0
0
0

0
0
5
4
0

0
0
5
2
1

0
0
2
2
9

0
0
1
1
1

Scenario 5
P1
P2
P3
P4
P5

1
0
0
0
0

3
0
0
0
0

4
0
0
3
0

0
0
0
0
0

3
0
0
4
0

0
6
0
3
0

1
1
0
1
0

Scenario 6
P1
P2
P3
P4
P5

0
0
5
0
0

0
3
4
0
0

0
7
0
0
0

0
0
0
0
6

0
4
0
0
1

0
0
0
0
0

0
1
1
0
1

Scenario 7
P1
P2
P3
P4
P5

3
0
0
0
0

4
0
5
0
0

0
0
0
0
0

1
0
5
0
0

0
0
0
0
0

1
0
6
0
0

1
0
1
0
0

Niche count nci for ith solution is calculated using sharing function sh(dij) as follows:

8


< 1  d a ; if d 6 r
N
X
share
r
share
nci
shdij and shd
:
j1
0;
otherwise

It provides an estimate of extent of crowding near a solution. Here

dij is the distance between the ith and jth solutions. Its value is always greater than or equal to one because sh(dii) = 1. rshare is calculated imagining that the n-dimensional hypersphere of radius r
must be divided among q optima equally. After calculating nc for
each solution in the archive the gbest is selected from the archive
whose nc is the smallest. This ensures diversity being maintained
as one which is comparatively less crowded is selected (Deb,
2001). The MOHPSO algorithm used in this work includes following
steps:
 Generate randomly n particles constituting swarm, initialise
their initial velocities, such that the particle velocity in the kth
dimension is limited by some maximum value, vkmax.
 Set iteration counter j = 0.
 Calculate tness values of each particle, using two objectives.

559

 Initialize pbest and gbest values

 Do

Calculate new velocity using pbest and gbest values.

Update position for each particle using new velocity.
Calculate new tness value.
Update pbest if the new tness value is better than current
one.
Determine gbest if the new tness value is better than current one.
Identify the particles that give non-dominated solutions
with respect to present Gbest and store them in an archive
called nonDomList. The solutions that are not non-dominating with respect to present gbest are deleted so as to
limit the list.
Calculate the crowding distance for each particle in the
Swarm for niching.
Sort all particles in the current non-dominated front in
descending order, and then choose best with least niche
count for next iteration.
j=j+1
} till j <= maximum iteration or till termination condition is
reached.

6. Numerical example
In order to illustrate the capabilities of the proposed model and
algorithm, a hypothetical case study of four echelon supply chain
network connecting 3 suppliers, 5 plants, 6 DCs and 7 CZs as shown
in Fig. 1 is considered. As per the problem, the company has manufacturing plants located in ve different geographical locations
with total capacity of 142 which serve markets in seven different
places with total demand of 30 through six warehouses with total
capacity of 82. In the rst stage of SCN, each vendor supplies three
components which are required for producing one product. Supplier capacities, plant capacities, warehouse capacities and xed
costs per month at each plant are given in Table 3. Supplier cost
of making a component and supplier transportation cost per unit
shipped are given in Table 4. Variable inventory, production and
transportation costs per unit shipped and demand at different
CZs are given in Table 5. Handling cost and transportation cost matrix per unit for third stage of SCN is given in Table 6. This problem
is analyzed with respect to two conicting objectives, minimizing
total cost and maximizing ll rate, with the condition that minimum ll rate to be attained is 80% of actual demand (24). There
are totally 128 decision variables to be optimized and 48 constraints. This problem results in seven Pareto solutions. The nal
decision is made among them taking the total balance over all criteria into account. This is a problem of value judgment of decision
maker. Also these decisions will be revisited every year as demand
and costs change.
6.1. The proposed MOPSO implementation
The proposed MOHPSO based approach was implemented using
C language and the developed software program was executed on a
1.8-GHz Pentium 4 personal computer. The random seeds were
generated by a random number generator incorporated in the program. Initially, several simulation runs have been done with different values of the PSO key parameters such as the maximum
allowable velocity. Other parameters selected are: number of particles n = 20, decrement constant = 0.8, c1 = 1.05, c2 = 0.05. The time

560

B. Latha Shankar et al. / Expert Systems with Applications 40 (2013) 551562

Table 10
Optimum distribution of products from DCs to CZs corresponding to Pareto front points.
Supply DC

CZ1

CZ2

CZ3

CZ4

CZ5

CZ6

CZ7

Optimal location of DCs

Scenario 1
DC1
DC2
DC3
DC4
DC5
DC6
Total quantity supplied
Demand

0
0
1
1
0
1
3
3

0
0
2
2
0
1
5
5

2
1
0
1
0
0
4
4

3
1
0
2
0
0
6
6

0
1
0
1
0
2
4
4

1
0
0
2
0
2
5
5

2
1
0
0
0
0
3
3

1
1
1
1
0
1
30
30

Scenario 2
DC1
DC2
DC3
DC4
DC5
DC6
Total quantity supplied
Demand

1
0
1
1
0
0
3
3

1
0
2
0
1
1
5
5

1
0
1
0
2
0
4
4

1
0
2
2
1
0
6
6

0
0
0
1
0
3
4
4

1
0
0
0
0
4
5
5

0
0
0
2
0
0
2
3

1
0
1
1
1
1
29
30

Scenario 3
DC1
DC2
DC3
DC4
DC5
DC6
Total quantity supplied
Demand

0
1
0
1
1
0
3
3

0
0
0
2
1
1
4
5

0
0
0
3
0
0
3
4

0
3
3
0
0
0
6
6

0
1
0
2
0
1
4
4

0
1
0
3
1
0
5
5

0
0
2
0
0
1
3
3

0
1
1
1
1
1
28
30

Scenario 4
DC1
DC2
DC3
DC4
DC5
DC6
Total quantity supplied
Demand

0
0
0
0
2
0
2
3

0
0
0
4
0
1
5
5

0
0
0
0
0
3
3
4

2
0
0
0
3
1
6
6

2
0
0
0
0
2
4
4

2
0
0
0
2
0
4
5

0
0
0
0
0
3
3
3

1
0
0
1
1
1
27
30

Scenario 5
DC1
DC2
DC3
DC4
DC5
DC6
Total Quantity Supplied
Demand

0
0
1
0
1
0
2
3

0
0
1
0
2
1
4
5

0
0
1
0
3
0
4
4

0
3
0
0
0
3
6
6

0
0
2
0
0
1
3
4

0
0
1
0
0
4
5
5

0
0
1
0
1
0
2
3

0
1
1
0
1
1
26
30

Scenario 6
DC1
DC2
DC3
DC4
DC5
DC6
Total Quantity Supplied
Demand

0
0
1
0
1
0
2
3

0
2
0
0
2
0
4
5

2
1
0
0
0
0
3
4

1
2
2
1
0
0
6
6

0
0
0
3
0
0
3
4

0
1
3
0
0
0
4
5

2
0
1
0
0
0
3
3

1
1
1
1
1
0
25
30

Scenario 7
DC1
DC2
DC3
DC4
DC5
DC6
Total quantity supplied
Demand

0
1
0
0
0
1
2
3

0
3
0
1
0
0
4
5

0
0
0
1
0
2
3
4

0
3
0
2
0
0
5
6

2
0
0
0
0
2
4
4

0
2
0
0
0
2
4
5

0
0
0
2
0
0
2
3

1
1
0
1
0
1
24
30

required to obtain solutions ranges from 363 to 1772 s depending

on PSO key parameters values.

7. Results and discussion

Fig. 2 depicts the obtained optimum feasible solutions. In this
Pareto front is represented by series2 and all other inferior points
by series 1. The convergence behavior of MOHPSO is shown in

B. Latha Shankar et al. / Expert Systems with Applications 40 (2013) 551562

Fig. 3 for total demand of 27 for both best and worst cases. The
solution space has seven non-dominated optimum decision points,
whose corresponding objective values are given in Table 7 and corresponding decision variables representing optimum material
ows from suppliers to plants for seven different scenarios are
given in Table 8, optimum location and distribution of goods from
5 plants to 6 DCs are tabulated in Table 9 and optimum location
and ow of goods from 6 DCs to 7 CZs are tabulated in Table 10.
These points clearly demonstrate tradeoffs in objective functions
total SC cost to be minimized and ll rate to be maximized from
which an appropriate solution can be compromisingly chosen as
per the requirements of the corresponding industrial environment.

8. Conclusion
In this paper, an analytical model is formulated for the location
and allocation of facilities of four-echelon supply chain network for
the optimal facility location and capacity allocation decisions.
Fixed location and variable material cost, production, inventory
and transportation costs are considered while making strategic
decisions. Two objective functions of minimizing total SC cost
and maximizing ll rate are considered. A heuristic based hybrid
MOPSO is used as optimizer. This algorithm is mainly aimed at
characterizing the Pareto Optimal front by computing well-distributed non-dominated solutions. These solutions represent trade-off
solutions which facilitate decision makers to develop management
policies under a changing environment. Further this can be used as
decision support system for strategic supply chain design and
monitoring of material ow. Above explained model and algorithm
further can consider safety stock costs, risk related and reliability
costs. Also the model can optimize more than two conicting
objectives simultaneously.
References
Azapagic, A., & Clift, R. (1999). The application of life cycle assessment to process
optimisation. Computer and Chemical Engineering, 10, 15091526.
Ballou, R. H. (1968). Dynamic warehouse location analysis. Journal of Marketing
Research, 5, 271276.
Che, Z. H. (2012). A particle swarm optimization algorithm for solving unbalanced
supply chain planning problems. Applied Soft Computing, 12, 12791287.
Cheng-Liang, Chen, Wang, B. W., & Wen-Cheng, Lee. (2003). Multi-objective
optimization for a multi-enterprise supply chain network. Industrial and
Engineering Chemistry Research, 42, 18791889.
Chengming, Qi. (2011). Application of improved discrete particle swarm
optimisation logistics distribution routing problem original research. Procedia
Engineering, 15, 3673-3677.
Chen, Cheng-Liang, & Lee, Wen-Cheng (2004). Multi-objective optimization of
multi-echelon supply chain networks, with uncertain product demands and
prices. Computers and Chemical Engineering, 28, 11311144.
Chopra, Sunil, & Meindl, Peter (2005). Supply chain management, strategy, planning
and operation. India: Pearson Prentice Hall.
Church, R. L., & ReVelle, C. S. (1976). Theoretical and computational links between
the p-median location set-covering and the maximal covering location problem.
Geographical Analysis, 8, 406415.
Coello, C. A. C. (1999). A comprehensive survey of evolutionary-based
multiobjective optimization techniques. Knowledge and Information Systems,
1(3), 269308.
Coello, C. A. C., & Lechuga, M. S. (2002). MOPSO: A proposal for multiple objective
particle swarm optimization. Congress on Evolutionary Computation 2002 (Vol. 2,
pp. 10511056). Piscataway, New Jersey: IEEE Service Center.
Cohen, M. A., & Lee, H. L. (1989). Resource deployment analysis of global
manufacturing and distribution networks. Journal of Manufacturing and
Operations Management, 81104.
Daskin, M. S. (1995). Network and Discrete Location: Models Algorithms and
Applications. New York: Wiley.
Deb, K. (2001). Multi-objective Optimization Using Evolutionary Algorithms.
Chichester: John Wiley and Sons.
Erlenkotter, D. (1981). A comparative study of approaches to dynamic location
problems. European Journal of Operational Research, 6, 133143.
Eveson, R.M., Fieldsend, J.E., Singh, S., 2002. Full elite sets for multi-objective
optimisation. In Parmee, I.C. (Ed.), Adaptive computing in design and manufacture
(pp. 343354).

561

Fieldsend, J.E., & Singh, S. (2002). A multi-objective algorithm based upon particle
swarm optimization, an efcient data structure and turbulence. In Proceedings
of the 2002 UK workshop on computational intelligence, Birmingham, UK (pp. 37
44).
Geoffrion, A. M., & Graves, G. W. (1974). Multi commodity distribution system
design by benders decomposition. Management Science, 20, 822844.
Geoffrion, A. M., & Vanroy, T. J. (1979). CautionCommon-sense planning methods
can be hazardous to your corporate health. Sloan Management Review, 20,
3142.
Guillna, G., Melea, F. D., Bagajewiczb, M. J., Espuaa, A., & Puigjanera, L. (2004).
Multiobjective supply chain design under uncertainty. Chemical Engineering
Science, 15351553.
Guo, Shin-Ming, & Hou, Chih Wei (2010). Multi- objective optimization for VMI
operations in the LCD supply chain. Computers and Industrial Engineering, 16,
16.
Guo, Y. W., Li, W. D., Mileham, A. R., & Owen, G. W. (2009). Applications of particle
swarm optimisation in integrated process planning and scheduling. Robotics
and Computer-Integrated Manufacturing, 25, 280288.
Hakimi, S. L. (1964). Optimum locations of switching centers and the absolute
centers and medians of a graph. Operations Research, 12, 450459.
Hu, X., & Eberhart, R. C. (2002). Multiobjective Optimization Using Dynamic
Neighborhood Particle Swarm Optimization. Congress on Evolutionary
Computation, 2, 16771681.
Hu, X., Eberhart, R.C., & Shi, Y. (2003). Particle swarm with extended memory for
multiobjective optimization. In Proceedings of the 2003 IEEE swarm intelligence
symposium, Indianapolis, IN, USA (pp. 193197).
Jrgen, B., & Sanaz, M. (2006). About selecting the personal best in multi-objective
particle swarm optimization. In Proceedings of the 9th international conference on
parallel problem solving from nature PPSN IX, Reykjavik, Iceland (pp. 523532).
Kadadevaramath, R. S., Chen, J. C. H., Latha Shankar, B., & Rameshkumar, K. (2012).
Application of particle swarm intelligence algorithms in supply chain network
architecture optimization. Expert Systems with Applications, 39, 1016010176.
Kennedy, J., & Eberhart, R.C., 1995. Particle swarm optimization. In Proceedings of the
IEEE international conference on neural networks, Perth, Australia (pp. 1942
1948).
Kennedy, J., & Eberhart, R. C. (2001). Swarm intelligence. Morgan Kaufmann
Publishers.
Lien, Li-Chuan, & Cheng, Min-Yuan (2012). A hybrid swarm intelligence based
particle-bee algorithm for construction site layout optimization. Expert Systems
with Applications, 39(10), 96429650.
Mahnam, Mehdi, Yadollahpour, Mohammad Reza, Famil-Dardashti, Vahid, & Hejazi,
Seyed Reza (2009). Supply chain modeling in uncertain environment with biobjective approach. Computers & Industrial Engineering, 56, 15351544.
Marinakis, Yannis, & Marinaki, Magdalene (2010). A hybrid multi-swarm particle
swarm optimization algorithm for the probabilistic traveling salesman
problem. Computers & Operations Research, 37(3), 432442.
Mirchandani, P. B., & Francis, R. L. (Eds.). (1990). Discrete location theory. Wiley.
Mostaghim, S., & Teich, J., 2003. Strategies for nding good local guides in
multiobjective particle swarm optimization (MOPSO). In Proceedings of the 2003
IEEE swarm intelligence symposium, Indianapolis, IN, USA (pp. 2633).
Mostaghim, S., & Teich, J. (2004). Covering pareto-optimal fronts by subswarms in
multiobjective particle swarm optimization. In Proceedings of the IEEE congress
on evolutionary computation 2004, Portland, Oregon, USA (pp. 14041411).
Mousa, A. A., El-Shorbagy, M. A., & Abd-El-Wahed, W. F. (2012). Local search based
hybrid particle swarm optimization algorithm for multi-objective optimization.
Swarm and Evolutionary Computation, 3, 114.
Mukhopadhyay, Sumona, & Banerjee, Santo (2012). Global optimization of an
optical chaotic system by Chaotic Multi Swarm Particle Swarm Optimization.
Expert Systems with Applications, 39(1), 917924.
Navalertporn, T., & Afzulpurkar, N. V. (2011). Optimization of tile manufacturing
process using particle swarm optimization. Swarm and Evolutionary
Computation, 1(2), 97109.
Nozick, L., & Turnquist, M. (2001). Inventory, transportation, service quality and the
location of distribution centers. European Journal of Operations Research, 129,
362371.
Owen, S. H., & Daskin, M. S. (1998). Strategic facility location: A review. European
Journal of Operational Research, 111, 423447.
Parsopoulos, K.E., & Vrahatis, M.N., 2002. Particle swarm optimization method in
multiobjective problems. In Proceedings of the ACM 2002 symposium on applied
computing (pp. 603607).
Prasanna Venkatesan, S., & Kumanan, S. (2012). A multi-objective discrete particle
swarm optimisation algorithm for supply chain network design. International
Journal of Logistics Systems and Management, 11(3), 375406.
ReVelle, C. (1986). The maximum capture or sphere of inuence location problem:
Hotelling revisited on a network. Journal of Regional Science, 26(2), 343358.
ReVelle, C. S., Eiselt, H. A., & Daskin, M. S. (2008). A bibliography for some
fundamental problem categories in discrete location science. European Journal
of Operational Research, 184, 817848.
Sabri, E. H., & Beamon, B. M. (2000). A multi-objective approach to simultaneous
strategic and operational planning in supply chain design. Omega, 28(5),
581598.
Scott, A. J. (1971). Dynamic location-allocation systems: Some basic planning
strategies. Environment and Planning, 7382.
Shen, Z. J. (2006). A prot maximizing supply chain network design model with
demand choice exibility. Operations Research Letters, 34, 673682.

562

B. Latha Shankar et al. / Expert Systems with Applications 40 (2013) 551562

Shen, Z.-J. M. (2007). Integrated supply chain design models: A survey and future
research directions. Journal of Industrial and Management Optimization, 3(1),
127.
Shen, Z. J., Coullard, C. R., & Daskin, M. S. (2003). A Joint location-inventory model.
Transportation Science, 37, 4055.
Simchi-Levi, D., Kaminsky, P., & Simchi-Levi, E. (2000). Designing and managing the
supply chain. New York: Irwin McGraw-Hill.
Sridharan, R. (1995). The capacitated plant location problem. European Journal of
Operational Research, 87, 203213.
Tsiakis, P., Shah, N., & Pantelides, C. C. (2001). Design of multi-echelon supply chain
networks under demand uncertainty. Industrial and Engineering Chemistry
Research, 40, 35853604.
Tsou, Ching-Shih, Yang, Dong-Yuh, Chen, Jyun-Hao, & Lee, Ying-Hao (2011).
Estimating exchange curve for inventory management through evolutionary

multi-objective optimization. African Journal of Business Management, 5(12),

48474852.
Warszawski, A. (1973). Multi-dimensional location problems. Operational Research
Quarterly, 24, 165179.
Warszawski, A., & Peer, S. (1973). Optimizing the location of facilities on a building
site. Operational Research Quarterly, 24, 3544.
Weber, A. (1929). Alfred Webers theory of the location of industries. University of
Chicago.
Wesolowsky, G. O. (1973). Dynamic facility location. Management Science, 19(11),
12411248.
Wesolowsky, G. O., & Truscott, W. G. (1976). The multiperiod location-allocation
problem with relocation of facilities. Management Science, 22(1), 5765.
Zhou, Z., Cheng, S., & Hua, B. (2000). Supply chain optimization of continuous
process industries with sustainability considerations. Computer and Chemical
Engineering, 24, 11511158.