Available online at www.sciencedirect.

com

Computers & Industrial Engineering 54 (2008) 783–799

www.elsevier.com/locate/caie

A particle swarm optimization algorithm for the

multiple-level warehouse layout design problem
a,*
Semih Önüt , Umut R. Tuzkaya a, Bilgehan Doğaç b

a
Department of Industrial Engineering, Mechanical Faculty, Yildiz Technical University, 34349 Istanbul, Turkey
b
Lojitek Logistics Technologies and Consultancy Company, Istanbul, Turkey

Received 25 January 2007; received in revised form 24 September 2007; accepted 26 October 2007
Available online 4 November 2007

Abstract

Warehouse operation and management is one of the essential parts of manufacturing and service operations. The ware-
house layout problem is a key to warehouse operations. Generally, warehouse layout design models attempt to optimize
different objectives such as the orientation of storage racks, the allocation of space among competing uses, the number of
cranes, the overall configuration of the facility, etc. The warehousing strategies can be classified as distribution-type, pro-
duction-type and contract-type warehouse strategies. In this study, a distribution-type warehouse considered that various
type products are collected from different suppliers for storing in the warehouse for a determined period and for delivery to
different customers. The aim of the study is to design a multiple-level warehouse shelf configuration which minimizes the
annual carrying costs. The turnover rates of the products are classified and they are considered while putting/picking them
to/from shelves regarding the distances between the shelves and docks. Since proposed mathematical model was shown to
be NP-hard, a particle swarm optimization algorithm (PSO) as a novel heuristic was developed for determining the optimal
layout.
 2007 Elsevier Ltd. All rights reserved.

Keywords: Warehouse layout; Order-picking; Class-based storage; Particle swarm optimization

1. Introduction

A supply chain can be considered as a network of individual entities that are collectively responsible for
procurement, manufacturing, warehousing and transportation activities. Performance of any entity in the net-
work depends on the performance of others. The efficiency and effectiveness in any supply network in turn is
highly determined by the operation of the nodes in such a network, like the warehouses (Jayaraman & Ross,
2003). Warehouses provide an important connection among suppliers, manufacturers, distributors and cus-
tomers in the supply network. Warehousing involves all movement of goods within warehouses or distribution
centers consisting of receiving, storage, order picking, accumulation, sorting and shipping (Van den Berg,
1999). Planning, design and control of warehousing systems are complex issues. It includes a large number

*
Corresponding author. Tel.: +90 212 259 70 70x2823; fax: +90 212 258 59 28.
E-mail address: onut@yildiz.edu.tr (S. Önüt).

0360-8352/$ - see front matter  2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cie.2007.10.012
784 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799

of interrelated decisions from a functional description through a technical specification, to equipment selection
and layout determination. Warehouse operation and management is also one of the essential parts of a supply
network. The warehouse layout design problem is a key component of warehousing operations.
Essentially warehousing strategies are classified as distribution-type, production-type and contract-type
warehouses (Van den Berg & Zijm, 1999). Distribution-type warehouses collect different type products from
different suppliers for delivery to different customers. Production-type warehouse is used for storage of the
products having different kinds of characteristics in a production facility. Contract-type warehouse executes
warehousing activities for different customers. In this study, a distribution-type warehouse is considered as
the suitable warehousing strategy.
Warehousing processes may be divided into different phases. These are mainly receiving, storage, order
picking and shipping phases (Rouwenhorst et al., 2000; Van den Berg & Zijm, 1999). In the receiving phase
as the first process, goods are delivered generally by trucks, unloaded at the receiving area, checked and pre-
pared for transportation to the storage area. The goods are transported to a storage location in the storage
area as the storage phase. A storage area can be divided into the following areas: reserve storage area and
forward area. In the reserve area goods are stored till they are required for shipment to the customer. In
the forward area goods are stored and prepared for the retrieving operations carried out by the order pickers.
As the third phase, order picking refers to the retrieval of goods from their storage locations whenever a good
is requested. These operations executed by the order pickers. At the shipping area orders are checked, sorted,
packed and loaded in trucks as the last phase. A cross-docking process can also be used as a warehousing pro-
cess. Cross-docking refers to the process in which goods, cartons or pallet loads are taken directly from the
receiving trucks to the shipping trucks. It provides rapid consolidation and processing (Chen, Guo, Lim, &
Rodrigues, 2006; Heragu, Du, Mantel, & Schuur, 2005).
There are a number of studies related with the design and integration of the different warehousing pro-
cesses. Also different design and integration models have been developed in the literature including mathemat-
ical models and heuristic procedures. A genetic algorithm with a special crossover operator was proposed by
Poulos, Rigatos, Tzafestas, and Koukos (2001), who find the Pareto-optimal solutions in the warehouse
replenishment problem. Lee and Elsayed (2005) considered the storage sizing problem for generic warehouses
under a dedicated storage policy. Since the problem was formulated as a nonlinear optimization model, an
iterative search procedure was developed to solve the model optimally. There are also some studies classified
as the single- or multiple-level warehouse layout design models mentioned below. Lai, Xue, and Zhang (2002)
considered a paper reel layout problem formulating by integer programming, which is shown to be NP-hard.
Following a natural decomposition of the problem, they proposed a two-stage heuristic procedure including
an optimal method and a simulated annealing heuristic. Zhang, Xue, and Lai (2002) investigated a multiple-
level warehouse layout problem considering multiple storage areas in different levels of a warehouse. An inte-
ger programming model was proposed and due to the similarity of the Lai et al. (2002)’s NP-hard problem, a
class of genetic algorithm-based heuristic was developed. Zhang and Lai (2006) considered a multiple-level
warehouse layout problem again and proposed a class of new heuristics by combining a genetic algorithm
and path linking strategy to solve the problem. Heragu et al. (2005) presented a mathematical model and a
heuristic algorithm that jointly determine product allocation to the functional areas in the warehouse.
As mentioned before, order picking refers to the retrieval of goods from their storage locations on the basis
of customer orders and one of the most important activities in a warehouse. This process is also the most
laborious and time-consuming process of all warehousing processes. Hence modeling a warehouse with the
order picking locations and considering all order pickers in the same model is very useful to determine good
order picker routes and to reduce travel times simultaneously. Additionally batching several orders in a
single order picking route can be taken into account to reduce travel times. Various methods have been eval-
uated for order picking processes in the literature (Chen & Wu, 2005; De Koster, Van der Poort, & Wolters,
1999; Hsu, Chen, & Chen, 2005). De Koster and Van der Poort (1998) studied the problem of finding efficient
order picking routes for both conventional and modern warehouses. They used extended polynomial
algorithm to find order picking routes with a minimal length and then compared the performance of this
new algorithm with the well-known S-shape heuristic solution. In another work (Chew & Tang, 1999) a travel
time model with general item location assignment in a warehouse system was developed to analyze order
batching and storage allocation strategies in an order picking system. Roodbergen and De Koster (2001a)
 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799 785

presented a dynamic programming-based routing algorithm to determine the shortest order picking tours for a
warehouse where aisle changing is possible at the front, the rear and in the middle of the warehouse. The same
authors also described a number of extended heuristics to evaluate order picking routes in a warehouse with
two or more cross aisles and they presented a new routing heuristic, called the combined heuristic, based on
dynamic programming again. To analyze the performance of the heuristics, a branch-and-bound algorithm
that generates shortest order picking routes was used (Roodbergen & De Koster, 2001b). Kim, Graves, Her-
agu, and Onge (2002) solved an industrial warehouse order picking problem using an intelligent agent-based
model. In the problem, goods are stored at multiple locations and the pick location of goods can be selected
dynamically. Same authors presented a hybrid intelligent agent-based scheduling and control system architec-
ture for the same problem mentioned above in another survey. They also developed a mathematical model and
a genetic algorithm-based heuristic for the resource assignment problem (Kim, Heragu, Graves, & Onge,
2003). Jewkes, Lee, and Vickson (2004) considered a multiple-picker order picking line which stores nk types
of products in n bins, each with k shelves and determined optimal policy for the problem of product location,
picker home base location and allocating products to each picker for minimizing expected order cycle time
using a dynamic programming algorithm. Another survey (Hsieh & Tsai, 2006) presented the effects on the
order picking system performance for factors, such as quantity and layout type of cross aisles in a warehouse
system, storage assignment policy, picking route, average picking density inside an aisle, order combination
type, etc. using a simulation and analysis tool.
The performance and efficiency of an order picking system depend on the demand pattern of the items, the
layout of the warehouse, storage process, batching method and routing method (Petersen, 1999). Storage pro-
cess can be performed by different storage policies. The most used and preferred policies can be given as ran-
domized storage policy, dedicated storage policy and class-based storage policy. The randomized storage
policy is performed by the allocation of the storage location based on the available space at the time of the
storage job. Storage decision is left to the operator in another word. A dedicated storage policy determines
a particular predetermined location for each product to be stored. A class-based storage policy is a common
and shared policy between randomized and dedicated policies. It divides goods into classes based on some cri-
teria and each class is assigned a block of storage locations. This policy can be called as ABC zoning (Haus-
man, Schwarz, & Graves, 1976; Lai et al., 2002; Larson, March, & Kusiak, 1997; Rosenblatt & Eynan, 1989;
Rouwenhorst et al., 2000).
As seen above, the literature on warehouse design and operations is highly extensive. Much of the research
in this area has been motivated by the need to improve the efficiency of order picking operation and to esti-
mate travel distance or time. For these reasons order picking process, storage strategy, batching method and
routing method have received considerable attention in the warehousing literature. However warehouse layout
design, especially in multiple-level, has received less attention. There are also a limited number of studies con-
cerned with the determination of the number of door in the literature. Additionally a multiple-level warehouse
layout design problem regarding the receiving and shipping processes, order picking process, storage strategy
and optimal number of door has not taken into account by any author. This is the most powerful motivation
to consider this problem. Furthermore, a large number of mathematical models and heuristic procedures have
been proposed for warehouse layout design problems and different warehousing processes.
Warehouse design problem can also be considered as facility layout problem. Especially, the problem type
that assigns the facilities to locations by considering capacity limitations, priorities of facilities, etc. is difficult
to solve. There is no algorithm that solves this kind of problems in polynomial time even small ones (Kusiak &
Heragu, 1987). Therefore these problems, like assigning the goods to shelves in warehouses, belong to NP-
hard problems class. The constituted model in this study has similarities with Zhang and Lai (2006) in terms
of the difficulty of solving multi-level warehouse problem and NP-hardness. For detailed information Garey
and Jhonson (1979) can be seen.
In this paper, we present a mathematical model for determining a multi-level warehouse layout. We also
focus on locating the items in the warehouse (storage policy), picking process from the pick location (order
picking) and specifying the optimal number of dock. The employed storage policy is the class-based storage
strategy. It is widely used in practice because it often leads to a substantial reduction in order pick travel dis-
tance. Essentially a distribution type warehouse considered that various type products are collected from dif-
ferent suppliers for storing in the warehouse for a determined period and for delivery to different customers.
786 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799

The turnover rates of the products are classified and they are also considered while putting/picking them to/
from shelves regarding the distances between the shelves and docks. Since proposed mathematical model was
shown to be NP-hard because of the above mentioned reasons and dealt with nonlinear components in the
objective function and constraints, a particle swarm optimization (PSO) algorithm as a novel heuristic was
developed for determining the optimal layout. The rest of the paper arranged as follows: Section 2 formulates
the addressed multiple-level warehouse layout problem. In Section 3, PSO algorithm is introduced by giving
the original PSO algorithm and improvements related with dealing the constrained problems. Section 4 pre-
sents the proposed PSO algorithm for tackling the problem. The computational results and some scenario
analysis are given in Sections 5 and 6, respectively. Finally, the paper is concluded in Section 7.

2. The formulation of multiple-level warehouse layout design model

In this study, a warehouse layout problem that tries to minimize material handling cost was considered.
Bassan, Roll, and Rosenblatt (1980) were examined the same problem by offering a comparison method
for two alternative shelf arrangement for a rectangular warehouse. They were considered homogeneous items
that have the equal probabilities for picking and putting away in a two-dimensional warehouse. Here, we
extended their study for heterogeneous items stocked multiple-level warehouse.
The yearly throughput of the warehouse is classified into three groups including A, B and C, according to
their turnovers. Considering the turnover rates and stocking periods of the items, probabilities of (picking or
putting) the orders belonging in A, B and C classes are PA, PB and PC, respectively. The main reason of this
classification is to locate the items in the warehouse basing on the closeness to the dock. The total capacity and
yearly throughput of the warehouse, total storage spaces for each item class and the lengths of the aisles and
shelves are entered to the model as the pre-determined parameters. The notation of the parameters and vari-
ables are shown in Table 1. Also, Fig. 1 is given to visualize some of the dimensions given in Table 1.
By using given details, a mathematical model providing the optimal number of storage spaces along a shelf
and the optimum number of shelves is constituted. In other words, the three dimensions of the warehouse
namely, length (u), width (v) and height (h) are obtained by the model solution. The objective function of
the model is constituted from the average travel distances in three dimensions and the unit material handling
cost. Before giving the objective function and constraint formulations, some points should be clarified. The
dock of the warehouse is located at the center of the horizontal wall. In the other word, the distance between
the dock and the left vertical wall is u/2. Also, the probability of carrying an item to right or left side of the
dock is equal. Therefore, the average travel distance in the horizontal axis is given in Eq. (1).

Table 1
Nomenclatures
w Width of the double shelf
L Length of a storage space (pallet)
m Number of the total storage spaces along a shelf
mi Number of the storage spaces allocated to class i items
h Number of the storage levels in the height directions
n Number of the double shelves
K Total warehouse capacity in the storage spaces
a Width of an aisle
u Length of the whole warehouse
v Width of the whole warehouse
d Yearly throughput of the warehouse, in storage units
Pi Probability of an order belonging to class i items
Ni Total number of the storage spaces of i type items
Ch Material handling cost of moving an item in unit length
Tv Average travel distance in vertical axis
Tu Average travel distance in horizontal axis
Th Average travel distance in height axis
 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799 787

mL
V

A a

w a

Fig. 1. A cross-section of the warehouse denoting the parameters.

u
Tu ¼ ð1Þ
4
The average vertical travel distances depend on the probability of the order that belonging the A, B and C
classes. In such a case, the formulation of the average travel distance in the vertical axis can be given in
the following equation:
     
ma L mb L mc L
Tv ¼ a þ Pa þ P b ma L þ þ P c ma L þ mb L þ ð2Þ
2 2 2
After some simplifications, Eq. (2) can be expressed as follows:
    
m Pa Pc NB
T v ¼ a þ L ðN A þ N B Þ 1  þ ðN B þ N C Þ  ð3Þ
K 2 2 2

The average travel distance in the last axis height can be given in the following equation:
h
Th ¼ ð4Þ
2
Multiplying the total travel distances in all dimensions by the unit material handling cost and by the yearly
throughput of the warehouse, the total cost minimizing objective function will constitute as shown in Eq. (5)
     
m Pa Pc NB u h
C ¼ 4dC h a þ L ðN A þ N B Þ 1  þ ðN B þ N C Þ  þ þ ð5Þ
K 2 2 2 4 2

The part of the Eq. (5), which is given in the parenthesis, represents a simplified version of the summation of
the average travel distances in all dimensions. The carrier starts the carrying process of an item at the dock. It
carries the item to the shelf and return to the dock. It means that the distance is passed two times. At the end
of the storage time, the same item is picked back to the dock and this means two more passing the same dis-
tance. This situation is represented by adding a leading coefficient 4 in Eq. (5).
The constraints of the model are given in Eq. (6). The first constraint of the model satisfies the yearly
demand and the others provide designing a warehouse in three dimensions. The details of the calculation
of the total moving distance for picking and putting away the items through a year can be found in Bassan
et al. (1980), but for only one type product and a two-dimensional layout.
788 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799

K  2mnh ¼ 0
mP1
nP1 ð6Þ
hP1
u ¼ nðw þ aÞ

Here, the decision variables are m, n and h. n shows the required number of the shelves in the warehouse
(u = n(w + a)). m shows the number of the storage spaces in a shelf (v = 2a + mL). h shows the number of
storage levels in the height directions. As a result, the model tries to determine the optimal number, length
and height of the shelves. The assignments of the items to the shelves are realized by considering the order
probabilities of the item classes and the quantities of the separated parts of the shelves to the item class.

3. Particle swarm optimization

Particle swarm optimization (PSO) is a stochastic optimization technique and also a population based
search algorithm first proposed by Eberhart and Kennedy (1995) and Kennedy and Eberhart (1995), inspired
by social behavior of bird flocking or fish schooling. PSO is a meta-heuristic approach used for solving hard
global optimization problems. Commonly used meta-heuristics can be briefly summarized as ant colony opti-
mization, genetic algorithms, artificial neural networks, evolutionary algorithms and simulated annealing
(Eussuf, Lansey, & Pasha, 2006). PSO is one of the modern meta-heuristic algorithms under the evolutionary
algorithms. Evolutionary algorithms, like genetic algorithm, and evolutionary programming strategies are
search algorithms based on the simulated evolutionary process of natural selection, variation and genetics.
PSO has been defined as an evolutionary computation algorithm and has typical features of both genetic algo-
rithms and evolution programming strategies. Evolutionary algorithms can also provide a near global solution
(Arumugam & Rao, 2005; Kennedy & Eberhart, 1995). PSO combines local search and global search and
ensures high efficiency. It has a more global searching ability at the beginning of the run and a local search
near the end of the run. Therefore, while solving problems with more local optima, there are more possibilities
for the PSO to explore local optima at the end of the run (Naka, Genji, Yura, & Fukuyama, 2003; Shi & Eber-
hart, 1999). PSO has some common characteristics with genetic algorithm, such as starting with a group of
randomly generated population, having fitness values to evaluate the population, updating the population
and searching for the optimum with random techniques. But PSO does not have genetic operators. Particles
update themselves and they also have own memory (Haq, Sivakumar, Saravanan, & Karthikeyan, 2006). PSO
algorithm has been applied to a wide range of engineering problems in the literature. Recently some attentions
have been focused on hybrid applications and some comparisons with the different heuristics, especially in
electrical, mechanical and industrial engineering. Huang and Mohan (2005) proposed a simple hybrid bound-
ary condition that could be used to obtain a robust and consistent PSO performance for high dimensional
optimization problems regardless of where the global optimum is located in the search space. In another study,
a solution model for the unit commitment problem was obtained using fuzzy logic to address uncertainties in
the problem (Victoire & Jeyakumar, 2006). In order to schedule the generating units based on the fuzzy logic
decisions, hybrid tabu search, PSO and sequential quadratic programming was used. Jeyakumar, Jayabarathi,
and Raghunathan (2006) described an adaptation of the PSO algorithm to solve various types of economic
dispatch problems in power systems and solved these problems using both the PSO method and the classical
evolutionary programming approach. The following examples are related with the manufacturing systems spe-
cifically. Arumugam and Rao (2005) presented several novel approaches of PSO algorithm with new particle
speed equations and inertia weights to solve the optimal control problems of a class of hybrid manufacturing
systems. Xia and Wu (2005) proposed a hierarchical solution approach to solve multi objective flexible job-
shop scheduling problem by using PSO algorithm and simulated annealing algorithm. Another work focused
on the different scheduling mechanisms which are designed to generate optimum scheduling. These nontradi-
tional optimization procedures, such as genetic algorithm, simulated annealing, memetic algorithm and PSO
algorithm were then implemented for solving the scheduling optimization problem of flexible manufacturing
systems and results were compared by Jerald, Asokan, Prabaharan, and Sravanan (2005). An activity network
 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799 789

based multi-objective partner selection model in a supply chain was developed by Zhao, Yu, and Chen (2005).
They proposed a hybrid heuristic algorithm based on PSO and simulated annealing to solve this multi-objec-
tive problem. Although the PSO algorithm has been applied to the different industrial areas and the different
engineering problems, only a few applications are known in the supply chain management, especially in the
layout design, network design or warehouse design in a supply network. There is only one study encountered
in the literature recently. In this work, Paul, Asokan, and Prabhakar (2006) proposed the PSO algorithm to
derive better solutions for unequal-area facility layouts having inner walls and passages. Ant colony optimi-
zation and PSO are the most popular swarm inspired methods in the computational intelligence area. While
ant colony optimization inspires by the behaviors of ants, PSO originates as a simulation of simplified social
system which graphically simulates the composition of a flock of bird or a school of fish.

3.1. Original PSO algorithm

In the implementation of the PSO, the population is referred to as a swarm and each individual as a par-
ticle. It is initialized with a random particles group and then searches the solution space for optima by updat-
ing generations. The general PSO algorithm is represented step by step in Fig. 2.
In PSO, each particle included by social structure keeps in mind its best position and uses this as a factor
affecting its speed. A particle gains speed toward its individual best position considering with how far away
from that point. It also shows the same behavior for the global best position. In other words, while it is scan-
ning the surface, it is affected by the global best position and adjusts its own speed. In the situation of that it is
far from the global best position, there will be a higher change in its speed and direction. Individuals (particles)
of a swarm show inclination to change their movements by using the information below

• Position of the ith particle in kth iteration xki (k = 0, . . . itermax and i = 1, . . ., ,N).
• Speed of the particle i in iteration k V ki ,
• Best position of the particle i (local best) (Pbesti),
• Best position of the particle group (global best) (gbest).

Initialization (for k = 0)

For i = 1 to N
k
Assign particles randomly in solution space ( xi )
k
Generate initial solutions S( xi )
k
Assign Pbesti = initial solutions S( xi )
Assign Gbesti = the obtained best solution among all particles
k
Generate initial velocities randomly ( Vi )
k +1
Add velocities to the corresponding particles ( xi )

Improve the solution (for k = 1 to itermax)

Determine the inertia weight ( wk )
For i = 1 to N
k
Update velocities ( Vi )
k +1
Modify the current positions ( xi )
Update the Pbesti
Update the Gbesti

Finalize the algorithm (k = itermax)

Assign the Gbesti = Ubest and stop.

Fig. 2. Algorithmic schema for general PSO algorithm.

790 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799

Each individual’s speed changes according to the formula in Eq. (7);

V kþ1
i ¼ X ðw  V ki þ C 1  ðRndÞ  ðpBesti  xki Þ þ C 2  ðRndÞ  ðgBesti  xki ÞÞ ð7Þ

V ki ith individual’s speed on kth iteration

xki ith individual’s position on kth iteration
w inertia function
Ci inertia factor
Rnd random number
Pbesti individual’s best position
Gbest global best position
X constriction factor

Inertia value of the equation changes on the each iteration. This change is based on the logic of
decreasing from the value determined to minimum value according to inertia function. The objective is
to converge the created speed by diminishing on the further iterations; hence more similar results can
be obtained.
Inertia function is obtained as follows:
 
wmax  wmin
w ¼ wmax  k ð8Þ
itermax

wmax first inertia force

wmin minimum inertia force
itermax maximum iteration number

The values of Ci inertia factor and wmax and wmin inertia forces are investigated by Shi and Eberhart (1998a,
1998b). It is found that these values should not be changed from a problem to another. They fixed the values
of these parameters as; Ci = 2, wmax = 0.9 and wmin = 0.4. Hence, in this study we also used these fixed values.
Positions of the particles change by speeds as shown in Eq. (9)
xkþ1
i ¼ xki þ vxkþ1
i ð9Þ
Same procedure is reiterated for each dimension.
As it can be seen above, the advantages of the PSO are easiness to implement and having few parameters to
adjust. However, there are some difficulties related with applying PSO on constricted models even it has been
successfully applied in many areas, such as function optimization, artificial neural network training, fuzzy sys-
tem control, and other areas (Eberhart & Shi, 1998).

3.2. Difficulties of using PSO on constricted models and improvements

PSO determines minimum or maximum value of any function easily between specific bounds. However, it
has some difficulties when it is necessary to ensure some constriction equations. According to PSO, the differ-
ence between a constricted model and a function is based on a high contradiction probability of any chosen
point and constriction equation.
Another effect of this problem is no points ensuring the constrictions on the first iteration, hence no global
best can be found. If a global best can be found on the first iteration, then algorithm can reach to better results
around this point. But if global and local bests can not be determined on the first iteration, then PSO is
useless.
Considering all this cases, keeping sensibility low on the first iteration and redounding step by step of the
procedure will grow up the efficiency of PSO. On the first iteration, the best integer will be chosen, then other
points will be investigated on decimal degree and this process will reach to the desirable sensibility about the
last iterations. Therefore the probability of finding the best points on the first iteration arises and no need to
redound particle number too much so no need to extend solving time.
 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799 791

30

35
Constriction = 15 + H

20 Constriction = 15

Constriction
value
15
Solution Space
10
Constriction = 15 - H
5

0
20 40 60 80 100 120 140 160 180 200
Iteration
Fig. 3. Funnel effect.

When solution set is small, enlarging the solution set to cover some of the neighbor points can be a useful
method to use. However, this wideness must shrink while the iteration number grows up, hence deviation from
the constrictions can be diminished.
Stretching the constrictions including the points on a specific approximation to the solution set is useful to
overcome this problem. But this elasticity should not be permanent; it should disappear to the maximum iter-
ation so that we can reach to the real points ensuring the constrictions. This effect is shown in Fig. 3. On the
each iteration, original constriction value is stretched by a certain quantity. This elasticity changes according
to the direction of the constriction too. Equation constrictions are positively and negatively stretched to reach
an interval. If it is bigger than constriction, then the value is pulled to higher value, else the value is pulled to a
lower value.
+n  H value, shown on the graphic, is calculated as in Eq. (10)
H ¼ Constriction  M ep  expðIter=M iter  0:1Þ ð10Þ
where in the each iteration
Mep maximum flexion portion
Miter maximum iteration number

Hence the constriction will be stretched by Mep on the first iteration.

To adapt standard PSO to this technique, some changes should be done into the criteria ensuring best point
choice. According to this, if any point is the best, then it will continue on the solution set. However, if the
point remains out of the solution set, even it is still the best one, the first point caught in the solution set will
be chosen instead of that point. This will prevent missing a better point in the solution set.
The pseudo code of the developed PSO algorithm embedding the required improvements to deal with the
shortcomings of the standard PSO algorithm is detailed below.

Initialization
Solve the objective function for the worst value (by using general PSO algorithm)
Assign Gbesti = The worst solution of general PSO algorithm

(for k = 0)
For i = 1 to N
Assign particles randomly in solution space
If the sensitivity is increasing step by step, round the solution to integer value
Generate initial solution Sðxki Þ
Generate initial speed randomly ðV ki Þ
792 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799

If the sensitivity is increasing step by step, round the speed to integer value
AssignPbesti ¼ Sðxki Þ
Keep Gbesti = The worst solution of general PSO algorithm

Improve the solution (k = 1 to itermax)

Determine the inertia weight
h   i
wmax wmin
wk ¼ wmax  itermax
k

For i = 1 to N
Update speed ðV ki Þ as
V kþ1
i ¼ X ðw  V ki þ C 1  ðRndÞ  ðPbesti  xki Þ þ C 2  ðRndÞ  ðGbesti  xki ÞÞ
Modify the current positions ðxki Þ as
½xkþ1
i ¼ xki þ vkþ1
i 
kþ1
If xi is still in the boundaries
Accept the current position
Else
Reduce or increase to the boundaries
Endif
Calculate the objective function value Sðxkþ1 i Þ
If the funnel effect is activated;
Repeat for each constraint
H = Constraint value * Mep * exp (-e no/Miter * 0,1)
if the constraint is an equality (=) constraints
Separate into (<=) and (>=) constraints
Endif
Add H value to the right hand side values of (<) and (<=) constraints
Subtract H value from the right hand side values of (>) and (>=) constraints
Else
Consider the current right hand side values for the constraints
Endif
Control whether the constraints are ensured
If Gbesti position ensures constraints then
Assign tempBoolean = True
Else
Assign tempBoolean = False
End if
If the constraints are ensured
IfSðxkþ1 i Þ > Pbesti
Assign Pbesti ¼ Sðxkþ1 i Þ
Else
Assign Pbesti ¼ Sðxki Þ
Endif
Else
Assign Pbesti ¼ Sðxki Þ
Endif
If the Pbesti > = Gbesti OR if tempBoolean = False
Gbesti = Pbesti

Finalize the algorithm (k=itermax)

Assign the Gbesti = Ubest and stop.
 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799 793

In order to solve an optimization problem with this improved PSO algorithm, solution space should be con-
stricted and the numbers of particles, iterations, max/min inertia forces and factors have to be defined at the
beginning. In addition, considering the funnel effect activation, maximum flexion portion should be
determined.

4. Proposed PSO algorithm for multiple-level warehouse layout design problem

The developed warehouse design model is applied to a distribution-type warehouse, which is considered on
the construction phase, and the solution is obtained by using improved PSO algorithm. The warehouse is
planned to serve for six product groups, i.e., personal cleaning products, house and kitchen cleaning products,
food products, chemical raw materials, electronics, and ceramic objects of which include 10, 14, 6, 2, 11 and 7
item types, respectively. Items belonging to different product groups are stored for different periods in the
shelves of the warehouse.
A classification phase that separates the storage products belonging to the different groups is realized before
applying the model to warehouse design problem. The aim is to determine which product should be put in which
class and the required size for each class, i.e., A, B, or C. Here, the closeness to the warehouse door decreases
from class A products to class B and C products. Three criteria, namely, turnover rate (circulation speed of prod-
ucts on shelves), fragility, and weights are chosen to rate the products by using the analytic hierarchy process
(AHP) methodology. The AHP is a multi-criteria decision-making method that uses hierarchic or network
structures to represent a decision-making problem and develops priorities for the alternatives based on the deci-
sion maker’s judgments throughout the system (Saaty, 1980). The reasons for including the AHP in our study
are especially the qualitative data and weighting requirement of the criteria because the qualitative factors are
often complicated and sometimes conflicting. The calculations of the product ratings are realized by Expert
Choice Package Program. First, the criteria are compared with each other and the priorities for turnover rate,
fragility and weights are obtained as 0.669, 0.088 and 0.243, respectively. Then, the products are rated for each
criterion by verbal judgments such as, very high, high, normal, low and very low. At the end of this evaluation
process, the priorities for all items of each product groups are determined. Considering with this data, the items
of product groups are allocated to the A, B, and C classes as 14, 24 and 12 items respectively. And the ordering
probabilities of each class are determined as 0.6, 0.3 and 0.1 mainly considering the turnovers. The throughput
of the warehouse is 120,000 palletized products in a year and the whole capacity of the warehouse is 6000 pallets.
The total storage capacity of the warehouse is divided to the classes into the sizes of 3000 (NA), 2000 (NB), and
1000 (NC) pallets, considering the ordering probabilities and storage periods.
The second data set is related with the dimensions in the storage area. A back to back shelve warehouse
design is considered and the width of 2.2 m, length of 0.9 m and height of 1.0 m for a storage space in a double
shelf is determined. The width of an aisle between shelves is 2.0 m and also the width of a dock is 4 m which is
determined considering the aisles’ width. The width of a dock can be unconsidered in the situation of using one
dock. But if the number of docks increases, it should be added to the formulation as a parameter. The only one
cost factor, material handling cost, is calculated by taking the cost of workers, forklift usage, fuel consump-
tion, and depreciations rates into account and 1.13 * 103 $/m is obtained.

5. Computational results and discussions

Above-mentioned data are inserted to the model formulation as the parameters. Then, the model is simpli-
fied to make it convenient to solve with the PSO algorithm. The PSO algorithm, which uses funnel affect to be
able to embedding the equality constraints to the objective function, is coded by using Visual Basic package
program. Sixty particles are used in each iteration and the maximum number of iterations is determined as
200. Fig. 4 shows the model inputs and obtained results screen of the program. The solution process is ended
in 22 s and at 62nd iteration. Due to the integer values are tried to be obtained, the decimal sensitivity of the
algorithm is constant and zero.
The variables X1, X2 and X3 shown in Fig. 2 denote the m, n and h variables of the model, respectively. The
obtained result is a warehouse which includes 25 shelves, 8 storage spaces in each shelf, and height of 15 stor-
794 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799

Fig. 4. Model inputs and obtained results screen.

Fig. 5. Improving of the model variables in the iterations.

age spaces. The changes in the variable values and objective function value, while iterations are increasing, are
shown in Figs. 5 and 6.
The objective function value can also be denoted as the total distance of picking and putting away of the
palletized products for one year as 13,362,000.00 m and the cost of $15,099.06. This version of the model does
not consider any physical constraint, but it can provided by putting some limitations to the dimensions and by
increasing the dock numbers for getting more realistic results or solutions for specific situations.
 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799 795

Fig. 6. Improving of the objective function value in the iterations.

6. Scenario analysis

There are some different situations that have minor or major affects on the results of the multi-level ware-
house design model solution. Firstly, all three dimensions of the travel distances in the warehouse are given in
equal weights in the cost minimization objective function. However, the height dimension may be handled dif-
ferently as the speed along this dimension is unlikely to be close to the speed along the other two dimensions.
Moreover as the maximum height to be reached increases, the investment cost of the storage and retrieval
equipment needed will likely increase at a nonlinear rate. Due to the speed differentiations, traveling time will
increase exponentially while the height of the shelves increasing. By using the technical data of a few material
handling equipments, it is decided to change the average travel distance Th, which may also be thought as tra-
vel time, to a nonlinear function (Eq. (11))
phffi
Th ¼ e 2 ð11Þ
Another alteration in the model is to assume that there is not only one dock in the horizontal wall of the ware-
house. In order to avoid the possibility of waiting goods in the dock queue and to increase the service quality
and throughput capacity, having multiple docks on the horizontal wall for entrance and exit may be a neces-
sity. To embed this situation to the model, it should be assume that the warehouse has r docks with a width of
2a. The lk is the distance between left wall of the warehouse and the middle of the kth dock (Eq. (12))
kðu þ 2aÞ
lk ¼ a ð12Þ
rþ1
The possibility of moving to left or right from the kth dock for picking/putting a good from A, B or C classes
is luk and ul
u
k
, respectively. And the average travel distances to left and right directions are l2k and ul
2
k
, respec-
tively. Thus, the average traveling distance (tk) to left or right directions can be expressed as follows.
lk lk ðu  lk Þ ðu  lk Þ
tk ¼ þ ð13Þ
u 2 u 2
When lk and the probabilities of picking and putting a good from different classes is embedded to the model
formulation and the required calculations are completed, the obtained average distance traveled in the hori-
zontal axis of the warehouse for all the docks will be as shown in Eq. (14)
796 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799

aða þ uÞ ðu þ 2aÞ X
2 r
Tu ¼  þ 2
k2 ð14Þ
2u urðr þ 1Þ k¼1
After these improvements in the model, the objective function that represent the total annual handling cost
(Cver1) for the shelf configuration will be derived as shown in Eq. (15)
"     #
aða þ uÞ ðu þ 2aÞ X
2 r
m Pa Pc NB 2 h
C ver1 ¼ 4dC h a þ L ðN A þ N B Þ 1  þ ðN B þ N C Þ   þ 2
k þ ð15Þ
K 2 2 2 u urðr þ 1Þ k¼1 2

Also, Eq. (15) is differentiated by applying the changes on height dimension that is given in Eq. (11). The last
version of the objective function (Cver2) has obtained as shown in Eq. (16)
"     #
m Pa Pc NB aða þ uÞ ðu þ 2aÞ X
2 r pffih
C ver2 ¼ 4dC h a þ L ðN A þ N B Þ 1  þ ðN B þ N C Þ   þ 2
k2 þ e 2 ð16Þ
K 2 2 2 u urðr þ 1Þ k¼1
This improved nonlinear objective functions with previously given nonlinear capacity constraints are solved
by using PSO algorithm for one to nine docks. As a result, 18 different scenarios are obtained for multiple-
level warehouse design problem (Fig. 7).
The results, which are given in Table 2 show that, while the dock numbers are increasing, the annual
handling cost increases independently from Th function. The main reason of this situation is that the good
arrivals to any dock have the equal probabilities. For example, a good can arrive to ninth dock to be
placed in the first double-shelf which has the longest horizontal path. Due to the above mentioned pos-
sibilities, the total handling cost increases. Also, the change on the obtained values of the variables, which
represents the dimensions of the warehouse, is an expected situation while the numbers of docks increase.
Another main point is the constriction of the height dimension by using a coefficient or a function that
does not permit to high level increase which is infeasible for a warehouse layout. As it is seen, the height
dimension h is decreasing from 15 or 20 to 10 for all dock numbers. Due to the exponential increase of h
function, there is no necessity to put an upper bound for the height dimension. It already prevents to
exceed 10 shelves for the height dimension.
There is one more main point related with the solution time of the PSO algorithm. All the scenarios are also
solved by LINGO Release 9.0 which uses branch and bound algorithm (B and B) for pure integer nonlinear
problems. LINGO is able to find only local optimum results with longer solution times in this problem. When
the solution performance of PSO algorithm is compared with B and B, it is superior for most of the situations
in terms of the solution times and number of iterations to reach to the best solution especially for scenarios
that uses exponential height dimensions (Table 3). Furthermore, for all situations, PSO finds less than or equal
cost objective values. As a result, even it is hard to compare an algorithm that gives optimal result and another
algorithm that gives approximate result, it is proven that PSO is superior in terms of obtained objective values
and solution times.

Multiple-level
Warehouse Design
Problem

h h
T =
h 2 T =e 2
h

Nine solutions for one to nine docks Nine solutions for one to nine docks
(Scenarios 1 to 9) (Scenarios 10 to 18)

Fig. 7. Scenarios applied to the main problem.

 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799 797

Table 2
The obtained results for 18 scenarios
Number of docks
1 2 3 4 5 6 7 8 9
h
Th ¼ 2 Objective function 15099.06 15733.01 16049.98 16233.62 16335.58 16408.4 16525.44 16578.27 16981.86
X1 (m) 25 25 25 25 25 25 25 25 20
X2 (n) 8 8 8 6 6 6 8 8 10
X3 (h) 15 15 15 20 20 20 15 15 15
pffih
Th ¼ e 2 Objective function 18323.07 19082.14 19461.68 19689.4 19841.21 19949.65 20030.98 20094.24 20144.84
X1 (m) 30 30 30 30 30 30 30 30 30
X2 (n) 10 10 10 10 10 10 10 10 10
X3 (h) 10 10 10 10 10 10 10 10 10

Table 3
Comparison of the solution times of PSO algorithm versus branch and bound algorithm
Number of docks Linear height dimension Exponential height dimension
Iteration number Solution time (s) Iteration number Solution time (s)
PSO B and B PSO B and B % Change PSO B and B PSO B and B % Change
1 62 6152 23 21 0.05 27 1514 22 22 0.05
2 39 9683 22 34 0.35 119 489 22 9 1.44
3 67 8327 22 27 0.19 103 5396 22 48 0.54
4 147 9934 22 34 0.35 127 9469 22 104 0.79
5 150 6507 22 21 0.05 89 10273 22 126 0.83
6 106 3565 22 10 1.2 121 7872 22 77 0.71
7 98 14467 20 54 0.59 61 5280 22 112 0.82
8 107 1786 22 7 2.14 103 4563 22 97 0.77
9 144 9406 22 30 0.27 93 5723 22 87 0.75

However, if the problem should be solved by continuous variables in the situation of having possibility to
use decimal data, the improved PSO algorithm has some shortcomings related with the solution time. In that
condition, the solution space is getting larger and the required particle numbers and iteration numbers are
increasing. For example, solving the problem by using continuous variables for two docks and exponential
height dimension, the required particle numbers increase from 60 to 100 and required iteration numbers from
119 to 343. As a result of these changes, solution time increases from 22 to 127 s and the deviation in the con-
straints is occurred as 0.0099.

7. Conclusions

This study aims to model the problem of designing a multiple-level warehouse considering the handling
costs in three dimensions. One of the contribution of our proposed model is to enhance the two-dimen-
sional warehouse design (Bassan et al., 1980) to the multiple-level warehouse design considering a class-
based storage strategy which includes three palletized product types, namely, A, B and C. But, the main
difficulty of solving this kind of designing problems is to struggle with nonlinearity in the variables and
the constraints for finding an optimal solution. To overcome this difficulty, we used a novel algorithm
PSO which is able to find near optimal results in a short time. To adapt the standard PSO algorithm
to the constrained problems, which can be applied to only unconstrained problems, it is modified by fun-
nel effect and penalty methods.
Then, the dock numbers on the same wall of the warehouse is increased one by one from one to nine
docks to see the changes on the total handling cost and the change on the configuration of the warehouse
in three dimensions. It is seen that the total handling cost increases by the increase of the dock numbers
even the configuration is changed or not. This is another contribution to multiple-warehouse layout design
798 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799

problem. However, some future directions can be added to the existing study. For example, there is a
trade off between total material handling cost and vehicle waiting cost for delivering the goods at the dock
while the dock number increases. Therefore, to determine the optimal dock number of this kind of ware-
houses, both the material handling and vehicle waiting costs should be considered together. Vehicle wait-
ing costs can be calculated by simulation techniques and embedded to the warehouse design problem in a
future work.

References

Arumugam, M. S., & Rao, M. V. C. (2005). On the optimal control of single-stage hybrid manufacturing systems via novel and different
variants of particle swarm optimization algorithm. Discrete Dynamics in Nature and Society, 3, 257–279.
Bassan, Y., Roll, Y., & Rosenblatt, M. J. (1980). Internal layout design of a warehouse. AIIE Transactions, 12(4), 317–322.
Chen, M. C., & Wu, H. P. (2005). An association-based clustering approach to order batching considering customer demand patterns.
Omega, 33, 333–343.
Chen, P., Guo, Y., Lim, A., & Rodrigues, B. (2006). Multiple crossdocks with inventory and time windows. Computers and Operations
Research, 33, 43–63.
Chew, E. P., & Tang, L. C. (1999). Travel time analysis for general item location assignment in a rectangular warehouse. European Journal
of Operational Research, 112, 582–597.
De Koster, M. B. M., Van der Poort, E. S., & Wolters, M. (1999). Efficient order batching methods in warehouses. International Journal of
Production Research, 37(7), 1479–1504.
De Koster, R., & Van der Poort, E. S. (1998). Routing order pickers in a warehouse: a comparison between optimal and heuristic solution.
IIE Transactions, 30, 469–480.
Eberhart, R. C., Kennedy, J. (1995). A new optimizer using particle swarm theory. In Proceedings of 6th Symposium Micro Machine and
Human Science, Nagoya, pp. 39–43.
Eberhart, R. C., Shi, Y. (1998). Comparison between genetic algorithms and particle swarm optimization. In Proceedings of 7th ICEC, pp.
611–616.
Eussuf, M., Lansey, K., & Pasha, F. (2006). Shuffled frog-leaping algorithm: A memetic meta-heuristic for discrete optimization.
Engineering Optimization, 38(2), 129–154.
Garey, M. R., & Jhonson, D. S. (1979). Computers and intractability: a guide to the theory of NP-completeness. Newyork: W.H. Freeman.
Haq, A. N., Sivakumar, K., Saravanan, R., & Karthikeyan, K. (2006). Particle swarm optimization algorithm for optimal machining
allocation of clutch assembly. International Journal of Advanced Manufacturing Technology, 27, 865–869.
Hausman, W. H., Schwarz, L. B., & Graves, S. C. (1976). Optimal storage assignment in automatic warehousing systems. Management
Science, 22(6), 629–638.
Heragu, S. S., Du, L., Mantel, R. J., & Schuur, P. C. (2005). Mathematical model for warehouse design and product allocation.
International Journal of Production Research, 43(2), 327–338.
Hsieh, L. F., & Tsai, L. (2006). The optimum design a warehouse system on order picking efficiency. International Journal of
Manufacturing Technology, 28, 626–637.
Hsu, C. M., Chen, K. Y., & Chen, M. C. (2005). Batching orders in warehouses by minimizing travel distance with genetic algorithms.
Computers in Industry, 56, 169–178.
Huang, T., & Mohan, A. S. (2005). A hybrid boundary condition for robust particle swarm optimization. IEEE Antennas and Wireless
Propagation Letters, 4, 112–117.
Jayaraman, V., & Ross, A. (2003). A simulated annealing methodology to distribution network design and management. European Journal
of Operational Research, 144, 629–645.
Jerald, J., Asokan, P., Prabaharan, G., & Sravanan, R. (2005). Scheduling optimization of flexible manufacturing systems using particle
swarm optimization algorithm. International Journal of Advanced Manufacturing Technology, 25, 964–971.
Jewkes, E., Lee, C., & Vickson, R. (2004). Product location, allocation and server home base location for an order picking line with
multiple servers. Computers and Operations Research, 31, 623–636.
Jeyakumar, D. N., Jayabarathi, T., & Raghunathan, T. (2006). Particle swarm optimization for various types of economic dispatch
problems. Electrical Power and Energy Systems, 28, 36–42.
Kennedy, J., & Eberhart, R. C. (1995). Particle swarm optimization. Proceedings of IEEE International Conference on Neural Networks, 4,
1942–1948.
Kim, B., Graves, R. J., Heragu, S. S., & Onge, A. St. (2002). Intelligent agent modeling of an industrial warehousing problem. IIE
Transactions, 34, 601–612.
Kim, B., Heragu, S. S., Graves, R. J., & Onge, A. St. (2003). A hybrid scheduling and control system architecture for warehouse
management. IIE Transactions on Robotics and Automation, 19(6), 991–1001.
Kusiak, A., & Heragu, S. S. (1987). The facility layout problem. European Journal of Operation Research, 29, 229–251.
Lai, K. K., Xue, J., & Zhang, G. (2002). Layout design for a paper reel warehouse: A two-stage heuristic approach. International Journal
of Production Economics, 75, 231–243.
Larson, T. N., March, H., & Kusiak, A. (1997). A heuristic approach to warehouse layout with class-based storage. IIE Transactions, 29,
337–348.
 S. Önüt et al. / Computers & Industrial Engineering 54 (2008) 783–799 799

Lee, M. K., & Elsayed, E. A. (2005). Optimization of warehouse storage capacity under a dedicated storage policy. International Journal of
Production Research, 43(9), 1785–1805.
Naka, S., Genji, T., Yura, T., & Fukuyama, Y. (2003). A hybrid particle swarm optimization for distribution state estimation. IEEE
Transactions on Power Systems, 18(1), 60–68.
Paul, R. C., Asokan, P., & Prabhakar, V. I. (2006). A solution to the facility layout problem having passages and inner structure walls
using particle swarm optimization. International Journal of Advanced Manufacturing Technology, 29, 766–771.
Petersen, C. G. (1999). The impact of routing and storage policies on warehouse efficiency. International Journal of Operations and
Production Management, 19(10), 1053–1064.
Poulos, P. N., Rigatos, G. G., Tzafestas, S. G., & Koukos, A. K. (2001). A Pareto-optimal genetic algorithm for warehouse multi-objective
optimization. Engineering Applications of Artificial Intelligence, 14, 737–749.
Roodbergen, K. J., & De Koster, R. (2001a). Routing order pickers in a warehouse with a middle aisle. European Journal of Operational
Research, 133, 32–43.
Roodbergen, K. J., & De Koster, R. (2001b). Routing methods for warehouses with multiple cross aisles. International Journal of
Production Research, 39(9), 1865–1883.
Rosenblatt, M. J., & Eynan, A. (1989). Deriving the optimal boundaries for class-based automatic storage/retrieval systems. Management
Science, 35(12), 1519–1524.
Rouwenhorst, B., Reuter, B., Stockrahm, V., Van Houtum, G. J., Mantel, R. J., & Zijm, W. H. M. (2000). Warehouse design and control:
Framework and literature review. European Journal of Operational Research, 122, 515–533.
Saaty, T. L. (1980). The analytic hierarchy process. Pittsburgh, PA: RWS Publications.
Shi, Y., & Eberhart, R. C. (1998a). A modified particle swarm optimizer. In Proceedings of the IEEE international conference on
evolutionary computation (pp. 69–73). Piscataway, NJ: IEEE Press.
Shi, Y., & Eberhart, R. C. (1998b). Parameter selection in particle swarm optimization. In Evolutionary Programming, VII: Proc. EP98
(pp. 591–600). New York: Springer-Verlag.
Shi, Y., & Eberhart, R. C. (1999). Empirical study of particle swarm optimization. In Proceedings of the congress on evolutionary
computation (pp. 1945–1950). New Jersey: IEEE Service Center.
Van den Berg, J. P. (1999). A literature survey on planning and control of warehouse systems. IIE Transactions, 31, 751–762.
Van den Berg, J. P., & Zijm, W. H. M. (1999). Models for warehouse management: Classification and examples. International Journal of
Production Economics, 59, 519–528.
Victoire, T. A. A., & Jeyakumar, A. E. (2006). A tabu search based hybrid optimization approach for a fuzzy modeled unit commitment
problem. Electric Power Systems Research, 76, 413–425.
Xia, W., & Wu, Z. (2005). An effective hybrid optimization approach for multi-objective flexible job-shop scheduling problems. Computers
and Industrial Engineering, 48, 409–425.
Zhang, G. Q., & Lai, K. K. (2006). Combining path relinking and genetic algorithms for the multiple-level warehouse layout problem.
European Journal of Operational Research, 169(2), 413–425.
Zhang, G. Q., Xue, J., & Lai, K. K. (2002). A class of genetic algorithms for multiple-level warehouse layout problems. International
Journal of Production Research, 40(3), 731–744.
Zhao, F. Q., Zhang Q. Y., Yu, D. M., Chen, X. H. Yang, Y. H. (2005). A hybrid algorithm based on PSO and simulated annealing and its
applications for partner selection in virtual enterprises. Advances in Intelligent Computing, PT 1, Proceedings Lecture Notes in
Computer, Science 3644, pp. 380–389.