This article was downloaded by: [181.114.126.

147]
On: 01 June 2014, At: 15:03
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK
International Journal of Production Research
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/tprs20
Multi-commodity warehouse location and distribution
planning with inventory consideration
Ronald G. Askin
a
, Ilaria Baffo
b
& Mingjun Xia
a
a
School of Computing, Informatics, and Decision System Engineering, Arizona State
University, Tempe, Arizona, United States of America.
b
Institute of Industrial Research and Automation, National Research Council of Italy, Rome,
Italy.
Published online: 10 Jun 2013.
To cite this article: Ronald G. Askin, Ilaria Baffo & Mingjun Xia (2014) Multi-commodity warehouse location and
distribution planning with inventory consideration, International Journal of Production Research, 52:7, 1897-1910, DOI:
10.1080/00207543.2013.787171
To link to this article: http://dx.doi.org/10.1080/00207543.2013.787171
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the Content) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Multi-commodity warehouse location and distribution planning with inventory consideration
Ronald G. Askin
a
, Ilaria Baffo
b
and Mingjun Xia
a
*
a
School of Computing, Informatics, and Decision System Engineering, Arizona State University, Tempe, Arizona, United States of
America;
b
Institute of Industrial Research and Automation, National Research Council of Italy, Rome, Italy
(Received 21 February 2013; nal version received 13 March 2013)
We consider the problem of designing a distribution network for a logistics provider that acquires products from multiple
facilities and then delivers those products to many retail outlets. Potential locations for consolidation facilities that
combine shipments for cost reduction and service improvements are considered. The problem is formulated with direct
shipment and consolidation opportunities. A novel mathematical model is derived to solve a complex facility location
problem determining: (i) the location and capacity level of warehouses to open; (ii) the distribution route from each
production facility to each retailer outlet; and (iii) the quantity of products stocked at each warehouse and retailer. A
genetic algorithm and a specic problem heuristic are designed, tested and compared on several realistic scenarios.
Keywords: facility location; distribution; inventory; genetic algorithm; consolidation
1. Introduction
Logistics systems are designed to provide ore to door transport. The system transforms materials into products via a
production network and then delivers products to customers through a distribution network. Strategically located distri-
bution centres or warehouses are commonly used to store items and meet delivery goals during distribution. Modelling
their logistics network and solving a facility location problem (FLP) is a key element in strategic planning for a wide
range of private and public rms. The FLP has its roots in the pioneering work of Weber (1909) who considered the
Fermat-Weber problem of locating a single facility to minimise the total travel distance between the site and a set of
customers. High costs associated with property acquisition and facility construction make facility location or relocation
projects long-term investments, and many other contributing factors such as actual road network and congestion,
customer response time demands and dynamic customer bases complicate site selection and facility design.
Strategic decisions, such as location and capacity of warehouses, are generally planned separately from tactical
issues such as the mode of transport and service level, and operational issues such as order sizes and lead times. How-
ever, this separate treatment may generate a suboptimal solution. As shown in the literature review in the next section,
the available research on integrated models is limited. Relevant issues include: multiple-commodity ows, random
demand, capacity considerations, lead time and safety stock; and few authors have considered all these parts together.
However, this combined problem is quite general in practice, multiple distribution options exist, inventory cost is a
major factor, retailers often face random demands for many commodities, and safety stock is used in the presence of
uncertain demand to guarantee a service level. The need to integrate these issues drives the motivation for this paper.
We consider the selection of warehouse locations and capacity levels from a predetermined nite set of options. We
also consider the subsequent choice of distribution paths from multiple product suppliers to retailers in a three-echelon
(facility, warehouse and retailer) supply chain system. Our objective is to optimise the whole system and minimise the
total cost which includes xed location cost, inventory cost and transportation cost. We consider two shipment methods
for products to each retailer: direct shipment from facility to retailer and indirect shipment from facility to warehouse
and then from warehouse to retailer. We also group multiple products into sets based on environmental or other factors
and allow consolidation in transportation. With respect to inventory, we include both safety stock and regular inventory
and consider the trade-off between inventory and transportation costs when delivery time requirements must be met or
replaced by safety stock. Another major contribution of our work is in capturing nonlinear aspects of shipping cost and
directly addressing inventory requirement to meet a service objective as a function of the selected multi-commodity
distribution network architecture.
*Corresponding author. Email: Mingjun.Xia@asu.edu
International Journal of Production Research, 2014
Vol. 52, No. 7, 18971910, http://dx.doi.org/10.1080/00207543.2013.787171
2013 Taylor & Francis
D
o
w
n
l
o
a
d
e
d

b
y

[
1
8
1
.
1
1
4
.
1
2
6
.
1
4
7
]

a
t

1
5
:
0
3

0
1

J
u
n
e

2
0
1
4

The remainder of this paper is organised as follows. Section 2 reviews related research work. Section 3 presents the
proposed model for the facility location problem with multiple plants, multiple commodities, safety stock consideration
and direct shipment option. Section 4 presents two heuristics and examines their computation feasibility, solution quality,
and robustness. A conclusion is provided in Section 5.
2. Literature review
Our research builds upon the standard xed charge capacitated facility location problem (CFLP) with single sourcing.
Numerous research has appeared on this topic. Cornejols, Sridharan, and Thizy (1991), Sridharan (1995), Owen and
Daskin (1998), Revelle, Eiselt, and Daskin (2008) and Melo, Nickel, and Saldanha-da-Gama (2009) present summaries
of FLP/CFLP. More details about general characterises in FLP/CFLP can be found in these papers.
Traditional CFLP only considers xed location cost and transportation cost. Daskin, Coullard, and Shen (2002) and
Ozsen, Coullard, and Daskin (2008) are among the rst authors who consider inventory control in CFLP. When
considering inventory control, only cycle inventory is necessary for constant/known demand. However, in the case of
stochastic/unknown demand, safety stock plays an important role to overcome the unexpected demand. Sourirajan,
Ozsen, and Uzsoy (2007, 2009) included the inventory, lead time, and service level in their papers. Nozick and
Turnquist (2001) considered the trade-off among facility costs, inventory costs, transportation costs and customer
responsiveness.
Among the available research, the multiple-commodity case has received limited attention. There may be two rea-
sons for this: a multiple-commodity problem can be translated to a single-commodity problem based on an independent
assumption; the complexity of a multiple-commodity problem.
The complexity of CFLP has also limited much of the facility location literature to simplied static and deterministic
models. The rst paper, published by Ballou (1968), recognised the limited application of static and deterministic loca-
tion models. Dynamic programming was then used to determine the best schedule for opening a subset of these sites as
an optimal location and relocation strategy for the planning period. Qi and Shen (2007) introduced an integrated
supply chain design model that considers unreliable supply. Santoso et al. (2004) and Shen and Qi (2007) outlined a
model for the stochastic supply chain design problem.
Table 1 summarises the major characteristics of our referenced papers in this research. The table classies papers by
the number of products, the type of demand, whether Inventory (I), Transportation (T), Safety Stock (SS), Stock Out
(SO) is considered, and also the main solution methods used in each paper.
3. Consolidation facility location and demand allocation model (CFLDAM)
We consider the selection of warehouse locations and sizes from a predetermined nite set of options and the subse-
quent choice of distribution paths from multiple product suppliers to retailers in a three-echelon supply chain (facility,
warehouse and retailer) system. Manufacturing facilities already exist and each provides one specic product. Ware-
houses can be located at potential locations with alternative sizes and need hold both cycle inventory and safety stock.
They may also effectively serve as cross-docking points. Retailers locations are also known in advance and the demand
rate for each product at each retailer is assumed to have a known distribution per time (assumed later to be normally
distributed for simplicity of presentation). Each retailer can order from a warehouse or directly from manufacturers but
chooses a single route for each product. In practice, this decision is based on cost and delivery lead time. In addition to
regular cycle inventory, retailers hold safety stock if the lead-time of replenishing one order is above a specic threshold
value (for example, one day).
Products are divided into sets based on environmental or other factors. Consolidation is allowed for shipping prod-
ucts in the same product set, but products from different product sets must be shipped separately. The holding cost rate
for products is the same for products in the same set. For instance, in a food chain, certain products may require refrig-
erated trucks. In other environments security or handling considerations may dictate compatibility of products.
Indices:
I Set of products.
S Set of product sets.
K Set of warehouses.
J Set of possible warehouse sizes small, medium and large.
R Set of retailers.
n Set of subscripts, n =1, , 5, where 1 means from a facility to a retailer, 2 means from a facility to a
warehouse, 3 means from a warehouse to a retailer, 4 means at a warehouse, 5 means at a retailer.
1898 R.G. Askin et al.
D
o
w
n
l
o
a
d
e
d

b
y

[
1
8
1
.
1
1
4
.
1
2
6
.
1
4
7
]

a
t

1
5
:
0
3

0
1

J
u
n
e

2
0
1
4

Table 1. Literature review summary.
Product Demand I T SS SO Main solution method
Bidhandi and Yusuff (2011) Multiple Stationary X X X Sample average approximation
Deterministic Benders decomposition
Dasci and Verter (2001) Single Stationary X An extension of handling headways
Stochastic
Daskin, Coullard, and Shen (2002) Single Stationary X X X X Lagrangian relaxation
Stochastic
Eskigun et al. (2005) Single Stationary X Heuristic algorithm based on LP-
relaxation Deterministic
Gebennini, Gamberini, and Manzini
(2009)
Single Dynamic X X X X Recursive heuristic
Stochastic
Hinojosa, Puerto, and Fernndez (2000) Multiple Dynamic X Lagrangian relaxation
Deterministic
Jawahar, Gunasekaran, and Balaji
(2012)
Single Stationary X X X Heuristic algorithm based on
LP-relaxation Deterministic
Lee and Park (2010) Single Stationary X Heuristic algorithm based on
LP-relaxation Deterministic
Lu and Bostelb (2007) Single Stationary X Heuristic algorithm based on
LP-relaxation Deterministic
Manzini and Gebennini (2008) Single Stationary X X X Mixed integer linear programming
solver Multiple Deterministic
Miranda and Garrido (2004) Single Stationary X X X Lagrangian relaxation
Stochastic
Miranda and Garrido (2009) Single Stationary X X X X Heuristic algorithm
Stochastic
Melo, Nickel, and Saldanha-da-Gama
(2005)
Multiple Dynamic X X Mixed integer linear programming
solver Deterministic
Novaes et al. (2009) Single Stationary X Voronoi diagrams
Stochastic
zceylan and Paksoy (2013) Single Stationary X Mixed integer linear programming
solver Deterministic
Ozsen, Coullard, and Daskin (2008) Single Stationary X X X Lagrangian relaxation
Stochastic
Pourakbar, Farahani, and Asgari (2007) Single Stationary X X Genetic algorithm
Deterministic
Qi and Shen (2007) Single Stationary X X Lagrangian relaxation
Stochastic
Rudi, Kapur, and Pyke (2001) Single Stationary X X Mathematical analysis
Stochastic
Santoso et al. (2004) Multiple Stationary X X Sample average approximation
Deterministic Benders decomposition
Shen and Qi (2007) Single Stationary X X X Lagrangian relaxation
Stochastic
Shu, Teo, and Shen (2005) Single Stationary X X X Column generation algorithm
Stochastic
Silva and de la Figuera (2007) Single Stationary X X X Heuristic algorithm
Stochastic
Snyder and Daskin (2005) Single Stationary X Lagrangian relaxation
Deterministic
Sourirajan, Ozsen, and Uzsoy (2007) Single Stationary X X Lagrangian relaxation
Stochastic
Sourirajan, Ozsen, and Uzsoy (2009) Single Stationary X X Genetic algorithm
Stochastic
Tagaras and Vlachos (2001) Single Dynamic X X Mathematical analysis
Deterministic Simulation
Yao et al. (2010) Multiple Stationary
Stochastic
X X X Heuristic algorithm
Yigit et al. (2006) Single Dynamic
Deterministic
X Simulated annealing
Notes: I, Inventory; T, Transportation; SS, Safety stock; SO, Stock out.
International Journal of Production Research 1899
D
o
w
n
l
o
a
d
e
d

b
y

[
1
8
1
.
1
1
4
.
1
2
6
.
1
4
7
]

a
t

1
5
:
0
3

0
1

J
u
n
e

2
0
1
4

Parameters:
f
kj
Fixed cost of opening one warehouse at location k with size j.
U
kj
Capacity of one warehouse at location k with size j.
lt
nij
Lead time from point i to j(n = 1, 2, 3).
C
nij
Capacity of one truck used for shipping from point i to j(n = 1, 2, 3).
A
nij
Setup cost of each order from point i to j(n = 1, 2, 3).
a
nij
Fixed transportation cost per trip for using one truck from point i to j(n = 1, 2, 3).
bl
nij
Variable transportation cost from point i to j(n = 1, 2, 3).
h
nij
Holding cost of product i at point j per time based on average inventory level (n = 4, 5).
h
s
r
Holding cost of product set S at retailer r.
t
nij
1 if the lead time from point i to j is greater than threshold value (one day), 0 otherwise (n = 1, 3).
D
ir
Demand mean of product i at retailer r per time.
r
2
ir
Demand variance of product i at retailer r.
z
a
Inverse of the Standard Normal for a probability of .
M A very big number.
Decision variables:
w
kj
1 if opening one warehouse at location k at size j, 0 otherwise.
x
ir
1 if retailer r orders product i from facility i directly, 0 otherwise.
x
ikr
1 if retailer r orders product i from warehouse k, 0 otherwise.
Q
nij
Quantity of one order of product i from the facility to j(n = 1, 2).
Q
3ikr
Quantity of one order of product i from warehouse k to retailer r.
Q
s
3kr
Quantity of one order of one product set s from warehouse k to retailer r.
The objective is to minimise the total cost including: xed warehouse location costs (depreciation), regular inventory
cost, safety stock cost, order cost and transportation cost. When calculating safety stock at a warehouse, risk-pooling is
applied for each product. At a warehouse, the total safety stock for one product is shown in Equation (1) and is deter-
mined as the desired condence multiplier times the standard deviation of cumulative product demand served by that
warehouse.
SS
i
z
a

r
r
2
ir
lt
2ik
x
ikr
_
(1)
When calculating the transportation cost, we assume a xed cost of using a truck per shipment along with a variable
cost related to number of units and shipping distance. Thus, cost for each order is equal to: xed cost number of
trucks + variable cost distance quantity of one order. Assuming full truck load order size, the cost is shown in
Equation (2).
Transport Cost=Shipment a
Q
C
_ _
bl Q (2)
An economic order quantity model is used to determine the initial optimal order quantity. And since the existence of
order cost, we can show that it is near optimal to use multiple full-truck loads at one time instead of sending one full-
truck load several times. Let Q
0
be the optimal economic order quantity assuming xed truck costs are linearised. Due
to the relative insensitivity of actual cost to quantity and the economics of full truckload shipments, the actual order
quantity used is selected from the oor or ceiling function of Q
0
as either Q dQ
0
=Ce C or Q
0
bQ
0
=Cc C. Bounds
on maximal loss from considering only full truck loads are derived in the Appendix. The model is easily extended to
allow for multiple capacity truck options in the case that the natural order size is signicantly different than the capacity
of the normal truck for such shipments. The modeller could include options for 20, 40 and 53 containers for instance
or even smaller delivery trucks for local deliveries from warehouses to retailers. Indeed, structurally, options such as
mail packages could even be considered.
1900 R.G. Askin et al.
D
o
w
n
l
o
a
d
e
d

b
y

[
1
8
1
.
1
1
4
.
1
2
6
.
1
4
7
]

a
t

1
5
:
0
3

0
1

J
u
n
e

2
0
1
4

The proposed model assuming shipment quantities are multiples of truck loads, and the formulation is then:
Minimise

j
f
kj
w
kj

i
Q
2ik
h
4ik
2

i
z
a
h
4ik

r
r
2
ir
lt
2ik
x
ikr
_
_ _

s
Q
s
3kr
h
s
r

i
Q
1ir
h
5ir
2

i
z
a
h
5ir

k
r
ir
t
3kr
x
ikr

lt
3kr
_
r
ir
t
1ir
x
ir

lt
1ir
_
_ _ _ _

i
A
1ir
a
1ir
Q
1ir
C
1ir
_ _
bl
1ir
Q
1ir
_ _
D
ir
x
ir
Q
1ir

i
A
2ik
a
2ik
Q
2ik
C
2ik
_ _
bl
2ik
Q
2ik
_ _
r
D
ir
x
ikr
Q
2ik

s
A
3kr
a
3kr
Q
s
3kr
C
3kr
_ _
bl
3kr
Q
s
3kr
_ _
i2S
D
ir
x
ikr
Q
s
3kr
_ _
(3)
Subject to:
Q
2ik

j
Mw
kj
8i; k (4)
Q
s
3kr

j
Mw
kj
8s; k; r (5)
Q
s
3kr

i2S
Q
3ikr
8s; k; r (6)

j
w
kj
1 8k (7)

k
x
ikr
x
ir
1 8i; r (8)

i
Q
2ik
2

r
r
2
ir
lt
2ik
x
ikr
_
0:8

j
U
kj
w
kj
8k (9)
Q
1ir
; Q
2ik
; Q
s
3kr
0 8i; s; k; r (10)
x
ir
; x
ikr
2 f0; 1g 8i; k; r (11)
The objective function (3) has six terms the xed warehouse location costs, the average inventory costs at ware-
houses, the average inventory costs at retailers, order cost, the transportation cost from facilities to retailers directly, the
transportation costs from facilities to warehouses and the transportation costs from warehouses to retailers when needed.
Safety stock at a site is based on desired percentiles of its replenishment lead time demand. When delivery lead time
(distance) is below an acceptable threshold, safety stock is not needed. The order quantities are computed by rst nd-
ing the optimal continuous economic order quantity and then costing out the options of that quantity against the rounded
up and down full truck load alternatives. Based on the choice of x
ir
, x
ikr
we nd the candidate continuous optimal order
quantity value using the typical EOQ model as shown in Equations (12) to (14). As described above, these values are
then rounded to nd the appropriate Q value for use in the model.
Q
0
1ir

2A
1ir
D
ir
x
ir
h
5ir
_
(12)
International Journal of Production Research 1901
D
o
w
n
l
o
a
d
e
d

b
y

[
1
8
1
.
1
1
4
.
1
2
6
.
1
4
7
]

a
t

1
5
:
0
3

0
1

J
u
n
e

2
0
1
4

Q
0
2ik

2A
2ik
(

r
D
ir
x
ikr
)
h
4ik

(13)
Q
s0
3kr

2A
3kr
(

i2S
D
ir
x
ikr
)
h
s
r

; Q
0
3ikr
Q
s0
3kr
D
ir

i2S
D
ir
_ _
(14)
The cost model is adaptable. For instance, suppose orders are for multiple truck loads. The model above assumes all
loads are shipped at once. However, if truck loads are spaced in time by the ratio of truck capacity to demand, then the
inventory terms in (3) would be replaced by expressions of the form Ch / 2 (a similar change is used in Equation (9)).
Constraint sets (4) and (5) require that shipping quantities for one plant to one warehouse or from one warehouse to
one retailer can be greater than 0 only when we open this warehouse. Constraint set (6) sets the total shipping quantity
of one product set equal to the summation of all the shipping quantities of products in this product set. Constraint set
(7) limits opening at most one warehouse at one potential warehouse location. Constraint set (8) requires only one
supplier for each retailer-product combination; the retailer can order directly from the plant or order from one ware-
house. Constraint set (9) assumes random access and guarantees the average inventory level at each warehouse should
be less than the effective capacity of this warehouse. Average inventory includes cycle stock plus safety stock. Effective
capacity is nominally set at 80% of total space (but is easily adjusted). Average inventory is equal to average cycle
inventory plus a safety stock based on replenishment lead time and total product volume. Constraints set (10) and (11)
are non-negative and binary constraints.
4. The model application
Deriving optimal or near-optimal solutions to location problems has fed the growth of the eld of location analysis over
the past three decades (Jamarillo, Bhadury, and Batta 2002). The large number of integer variables makes it computa-
tionally difcult to solve. For this reason a genetic algorithm approach is applied.
4.1 The genetic algorithm overview
The concept of a genetic algorithm (GA) was developed by Holland (1975) and then described by Goldberg (1989).
GAs are stochastic research techniques based on the mechanism of natural selection with the search for better organ-
isms (problem solutions) over time as the organisms evolve by random changes in their genes. Changes that enhance
survival are more likely to be passed down to future generations via reproduction (Konak, Coit, and Smith 2006).
GA starts by creating an initial Population of individuals wherein each individual is dened by a chromosome
consisting of a sequence of genes. An associated tness function (objective function) measures the survival probabil-
ity (solution quality) of a chromosome as a function of its gene sequence. A new generation in GA is built using
operators generally known as Crossover and Mutation operators. In Crossover, two chromosomes, called parents, are
combined together to form two new chromosomes, called offspring. Mutation operators randomly change the value
of genes. A set number of individuals are retained in each generation with the likelihood of survival being based on
the tness function. In the metaheuristic adaptation of evolution, the best individual found is retained as the
incumbent solution.
4.2 GA for warehouses location problem
In recent years genetic algorithms (GAs) have been used to solve several optimisation problems, but applications
of GAs to location models have been relatively few. Hosage and Goodchild (1986) and Chaudhry, He, and
Chaudhry (2003) present an application of GA for the p-median problem. Gen and Syarif (2005) propose a span-
ning tree-based GA to solve a location facility problem considering multi products and multi periods. Finally,
Jamarillo, Bhadury, and Batta (2002) and Zhou, Min, and Gen (2003) propose GA application to two simple
models for location-allocation problems. We present a GA application for a complex model of location consider-
ing several factories with single product production, several potential sites for opening warehouses, multiple cus-
tomers having different and continuous demand for each product and the choice of direct or indirect shipment
from factories and retailers. In the next section we present the GA heuristic in detail and then report the outcome
of empirical tests.
1902 R.G. Askin et al.
D
o
w
n
l
o
a
d
e
d

b
y

[
1
8
1
.
1
1
4
.
1
2
6
.
1
4
7
]

a
t

1
5
:
0
3

0
1

J
u
n
e

2
0
1
4

4.3 GA for the proposed model
4.3.1 Chromosome representation
The chromosome representing the problem solution is composed for each product by three sub-strings representing
respectively: (i) the direct shipment of products from a single factory to networks retailers (i to r), (ii) the link among a
single factory and the possible sites for opening warehouses (i to k), and (iii) the shipment of products from warehouses
to retailers (k to r). (The i to k link is unnecessary since it may be inferred by the other two, but we include it for ease
of description. During implementation, its presence did not impact performance). For each product-retailer there are
K + 1 possible tours for shipment. However, only one route must be chosen. A retailer can be supplied by one of the
warehouses or directly by the plant as showed by Figure 1.
Figure 2 shows a feasible solution where Product 1 (i1) is directly shipped from plant to retailer 1 and through ware-
house k1 for retailer 2. Retailers 1 and 2 are both supplied from warehouse k1 for Product 2. In this scenario warehouse
k2 is not opened and the corresponding chromosome is:
The genes are represented by binary values shown in Table 2. The gene in position 0 that assumes value equal to 1
means that retailer r1 is supplied directly by plant i1 for product 1. The 1 in position 5 indicates warehouse 1 is the
inter model shipment point for product 1 to retailer 2. The gene equal to 1 in position 10 means that there is a link
between plant i2 and warehouse k1. The gene in position 13 means that retailer r2 is supplied by warehouse k1 for
product 2. With respect to the model presented in Section 2, this string represents only the values of variables called
x. The values of variables called w are calculated with a simple method that cheques the value of the chromosomes
genes and decides to open a warehouse when a warehouse supplies at least 1 retailer. With respect to the chromosome
presented in Table 2, the corresponding string for the warehouses opened is shown is Table 3. The value 1 in position 0
means that warehouse k1 will be opened, the value 0 in position 1 means that warehouse k2 will be not opened.
i k r
x
ir
x
ikr
Figure 1. Shipment directions.
i1 k1 r1
i2 k2 r2
Figure 2. A feasible solution.
Table 2. Chromosome representation.
Product 1 (P1) Product 2 (P2)
From i to r From i to k From k to r From i to r From i to k From k to r
1 0 1 0 0 1 0 0 0 0 1 0 1 1 0 0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
International Journal of Production Research 1903
D
o
w
n
l
o
a
d
e
d

b
y

[
1
8
1
.
1
1
4
.
1
2
6
.
1
4
7
]

a
t

1
5
:
0
3

0
1

J
u
n
e

2
0
1
4

4.3.2 Constraint feasibility and tness function evaluation
For each chromosome we need to control the feasibility with respect to the following constraints:

Single sourcing constraint: for each product a retailer has to be supplied and has to be supplied by the plant or
by just one warehouse.

Warehouses capacity limit.

The total ow entering a warehouse has to be equal to the ow exiting for each product.

If required, the limit about the service level (delivery time).

For the feasible individuals the next step is to calculate the tness function ( ff ) value corresponding to the objective
function of the model presented in Section 2. When the individual chromosome is not feasible with respect to ware-
houses capacity or service level, then the value of ff is assumed to be equal to a big integer called M. When the individ-
ual is not feasible with respect to other constraints then two methods called ToBuilt_1() and ToBuilt_2() are applied in
order to build a feasible solution from a infeasible solution. The rst method operates xing the rst sub-string of the
chromosome and building the rest of it. The second xes the last sub-string of the chromosome and builds the remain-
ing parts respecting the above-mentioned constraints.
4.3.3 Operators
In the general crossover, given a pair of parent strings, an arbitrary cut-off point is picked. The only difference of
our crossover operator with respect to the normal one is the choice of the cut-off point. In order to obtain feasi-
ble offspring, we chose the cut-off randomly from a predened set of possibilities. These points correspond to the
end of substring relative to a product. With respect to Table 1, the unique possible cut-off is between positions 7
and 8.
Mutation is an operation at the genes level. We adopt the typical manner to use this operator. With a pre-dened
probability a gene changes its value, from 1 to 0 or from 0 to 1. We only dene three kinds of mutations depending on
which substring of chromosome is changing. After this operation, the ToBuilt_1 and ToBuilt_2 () methods are applied
as necessary to re-establish the solutions feasibility.
4.3.4 Evolution mechanism
We adopt the mechanism known in the literature as Elitism. For every generation 10% of population represents the
Elit of the set and it is composed of the best solutions found during the evolution process. The remaining 90% of
population changes on basis of pre-determined percentages for applying Crossover and Mutation operators. The evolu-
tion stops when reaching the number of iterations declared or when we do not improve the best solution for a specied
large number of iterations.
4.4 Proposed greedy construction heuristic (GCH)
Merging concepts from opportunity cost and steepest approach, we develop a greedy heuristic for comparison to GA.
The GCH heuristic builds the solution step by step using a cascade method. Iteration makes a decision for a product
retailer pair and includes the decision taken in the previous iterations. The procedure is as follows:
Step 1: Set t = 0.
Step 2: To build a table with I R rows and K + 1 columns and evaluate the feasibility of the solution with respect
to warehouses capacity. If the constraint is satised then calculate the objective function for each product-
mode-retailer combination OF
ikr
(t) comparing the K + 1 possibilities of shipment (directly by plant or by K
warehouses). Otherwise put the OF
ikr
(t) equal to a big integer called M.
Table 3. Locations variables values.
k1 k2
Value 1 0
Position 0 1
1904 R.G. Askin et al.
D
o
w
n
l
o
a
d
e
d

b
y

[
1
8
1
.
1
1
4
.
1
2
6
.
1
4
7
]

a
t

1
5
:
0
3

0
1

J
u
n
e

2
0
1
4

Step 3: Comparing the value of OF
ikr
(t) for each row to select the minimum and the second smallest for each row
respectively called:
Min
ir
= min
k
{OF
ikr
(t)}, SecMin
ir
= min
k
{OF
ikr
(t)/Min
ir
}.
Step 4: Calculate
ir
as the difference between Min
ir
and SecMin
ir
(potential regret).
Step 5: Select the max
ir
{
ir
} and in correspondence to the column, x the solution for the relative product/retailer
couple. Set t = t + 1.
Step 6: Repeat the steps 25 for I R iterations.
Step 7: End.
As an alternative to Step 4, selection may be based solely on Min
ir
.
Table 4 shows an iteration of the heuristic described above. With 3 warehouses, two products, and four retailers, at
the rst iteration, a direct shipment for the couple plant i1 and retailer r1 is xed. The rest of the solution is built
though a cascade method. The selected product-mode-retailer combination is xed and removed from the table. All
affected values are then updated for the next iteration. Thus, the complexity is of O(I
2
R
2
K) consolidation policy and
objective function evaluations.
4.5 Empirical evaluation
4.5.1 Parameters and scenarios denition
Some products can be consolidated for shipment. In our model, we grouped those multiple products into two different
product sets. Consolidation is allowed for shipping products in the same product set, but products from different product
sets must be shipped separately. Holding costs for products are different for different product sets and different holding
places.
Manufacturing facilities and retailers are chosen as major cities in the United States. Potential warehouses can be
located at the locations of retailers. Each warehouse has three possible sizes small, medium and large. The distances
among cities are supplied by Daskin (1995). The xed cost of each warehouse is calculated on the basis of home value
in the respective cities, which is also supplied by Daskin (1995), and capacity of the warehouse which is set according
to potential service amount.
Demands of products at each retailer are normally distributed. The mean is proportional to the population around
that retailer. The variance of demand is calculated using coefcient of variation times mean demand. We set our service
level to be approximately 98% (Z value of 2).
Using trucks to distribute products, lead time between two cities depends on the distance and speed of a truck (500
miles/day). Each truck has specied capacity. Shipping cost of one order is computed as the xed cost of using trucks
plus variable costs which depends on distance and shipping quantity.
We compare eight scenarios dening the set of plants, possible sites for opening warehouses, customers locations
and kind of function used to dene the batch size. Table 5 illustrates the eight scenarios.
Using the chromosome representation described in the Section 4.3.1, the length of chromosome is dened as: Chro-
mosomes length = (number of plants number of customers) + (number of plants number of possible sites for ware-
houses) + (number of possible sites for warehouses number of customers). The length of chromosome joined with the
number of iterations required to reach a feasible solution determines the complexity of the algorithm and consequently
the computation time (CPU). Each element of the chromosome is called a gene so length is dened as number of genes.
Table 4. Heuristic algorithm example.
OF
Direct
shipment
Warehouse 1
opened
Warehouse 2
opened
Warehouse 3
opened Min SecMin Delta
i
1
r
1
680.386 2.327.768 78.905.221 372.698.028 680.386 2.327.768 1.647.382
i
1
r
2
20.002.904 21.318.949 20.992.233 181.298.158 20.002.904 20.992.233 989.329
i
1
r
3
168.036.279 169.608.843 181.057.583 169.554.096 168.036.279 169.554.096 1.517.817
i
1
r
4
82.576.957 84.028.851 92.402.173 199.285.905 82.576.957 84.028.851 1.451.894
i
2
r
1
74.843.490 75.753.105 84.512.660 98.935.406 74.483.490 75.753.105 1.269.615
i
2
r
2
21.002.242 29.974.560 21.741.500 29.248.200 21.002.242 21.741.500 739.258
i
2
r
3
8.939.292 84.280.026 73.877.307 10.039.649 8.939.292 10.039.649 1.100.357
i
2
r
4
36.870.220 80.561.996 67.847.045 36.866.944 36.866.994 36.870.220 3.276
Max 1.647.382
International Journal of Production Research 1905
D
o
w
n
l
o
a
d
e
d

b
y

[
1
8
1
.
1
1
4
.
1
2
6
.
1
4
7
]

a
t

1
5
:
0
3

0
1

J
u
n
e

2
0
1
4

4.5.2 Results
For each scenario above, we test and compare the performance of the genetic and the two versions of the heuristic algo-
rithms. For the GA, we set the runs parameters as follows. The population size is equal to 100 individuals for the rst
six scenarios and equal to 10 for the last two scenarios. The number of iterations has been xed equal to 50,000. The
GA stops when it fails to improve the solution for 10,000 continuous iterations. Table 6 gives the results obtained for
the scenarios presented in the previous section. Heuristics introduced here are applied in Microsoft Visio Studio C++,
and all the computational time is obtained on a Intel(R) Core(TM)2 T5550 at 1.83 GHz using Windows 7.
The heuristic proved computationally efcient and provided the best solution in all but one case. The delta form of
the heuristic (making the selection based on difference between the best and second best options) outperformed the
min form in two cases and the min form performed best in one case. As expected, both forms of the heuristic per-
formed at least as well as direct shipments in all cases and better in ve of eight cases for the delta version and six of
eight for the min version. The genetic algorithm found the unique best feasible solution in the last case and tied for best
in two additional cases where no warehouses were opened. However the genetic algorithm required signicantly longer
computation time.
In addition to the two heuristics discussed earlier, the cost for all direct-shipping is also calculated. The
performances in terms of results are comparable, but the heuristic seems better in terms of CPU time. We also
tested larger scenarios for examples containing a set of 88 customers, two or ve plants, and 190 possible sites
were opening warehouses as suggested by Daskin. In these cases the GA gives solutions in a reasonable time
depending by the choice of population size. We do not report these results due to difculty evaluating the accuracy
of these solutions.
5. Conclusion and future research
This paper presents a new model to guide the design of a distribution network for shipping multiple products, each orig-
inating from its unique production plant, to retailers. Shipments may be direct or use intermediate warehouses for ship-
ment consolidation and/or inventory pooling prior to nal delivery to retail demand points. Facility costs, inventory
costs and shipping costs are considered. The model is exible and may consider factors such as multiple types of deliv-
ery trucks for each segment, full or less than full truck shipments and different service requirements. A bound is derived
on the maximum cost penalty that could be incurred from restricting all deliveries to full truck loads.
The output of the model indicates: (i) the location and capacity level of warehouses to open; (ii) the distribution
route from each production facility to each retailer; (iii) the quantity of products stocked at each warehouse and retailer;
and (iv) the average cost per time for the specied conguration and operating policy. Two versions of a greedy con-
struction heuristic and a genetic algorithm are developed to solve the model. The solution approaches are tested on a set
of problems ranging from two production plants and 10 customer regions to ve production facilities and 49 customer
regions. The construction heuristics are shown to provide computationally efcient approaches to obtain good solutions
as compared to a direct shipment strategy. Given a set of possible warehouse locations and standard cost data for ship-
ping alternatives and storage, the user can evaluate one or multiple scenarios and generate a system design by applying
the heuristic. The genetic algorithm also provides good, feasible solutions but requires greater computational effort to
produce comparable results.
In this work we assume a continuous demand distributed, experimentation further assumed a Gaussian distribution.
Future work could explore discrete demand. Another possible improvement can be the consideration of an additional
Table 5. Design of test problems.
Scenario
Number of
Plants
Number of Locations for
warehouses
Number of
Customers
Function to dene the
batch size
Length of Chromosome
(genes)
1 2 10 10 Floor 140
2 2 10 10 Ceiling 140
3 5 10 10 Floor 600
4 5 10 10 Ceiling 600
5 2 10 49 Floor 608
6 2 10 49 Ceiling 608
7 5 10 49 Floor 2745
8 5 10 49 Ceiling 2745
1906 R.G. Askin et al.
D
o
w
n
l
o
a
d
e
d

b
y

[
1
8
1
.
1
1
4
.
1
2
6
.
1
4
7
]

a
t

1
5
:
0
3

0
1

J
u
n
e

2
0
1
4

T
a
b
l
e
6
.
C
o
m
p
u
t
a
t
i
o
n
a
l
r
e
s
u
l
t
s
.
G
e
n
e
t
i
c
H
e
u
r
i
s
t
i
c
(
M
a
x
{
D
e
l
t
a
=
S
e
c
M
i
n

M
i
n
}
)
H
e
u
r
i
s
t
i
c
(
M
i
n
{
M
i
n
}
)
A
l
l
d
i
r
e
c
t
-
s
h
i
p
p
i
n
g
C
P
U
t
i
m
e
(
s
)
N
u
m
b
e
r
o
f
i
t
e
r
a
t
i
o
n
s
O
b
j
e
c
t
v
a
l
u
e
(
$
)
N
u
m
b
e
r
o
f
o
p
e
n
w
a
r
e
h
o
u
s
e
s
C
P
U
t
i
m
e
(
s
)
N
u
m
b
e
r
o
f
i
t
e
r
a
t
i
o
n
s
O
b
j
e
c
t
v
a
l
u
e
(
$
)
N
u
m
b
e
r
o
f
o
p
e
n
w
a
r
e
h
o
u
s
e
s
C
P
U
T
i
m
e
(
s
)
N
u
m
b
e
r
o
f
i
t
e
r
a
t
i
o
n
s
O
b
j
e
c
t
v
a
l
u
e
(
$
)
N
u
m
b
e
r
o
f
o
p
e
n
w
a
r
e
h
o
u
s
e
s
O
b
j
e
c
t
v
a
l
u
e
(
$
)
1
9
8
4
2
2
3
.
0
5
6
7
E
+
8
4
1
2
0
3
.
0
3
0
3
E
+
8
3
1
2
0
3
.
0
3
0
3
E
+
8
3
4
.
2
1
1
3
E
+
8
2
1
2
3
3
2
9
3
.
0
5
5
1
E
+
8
4
1
2
0
3
.
0
3
0
2
E
+
8
3
1
2
0
3
.
0
3
0
3
E
+
8
3
4
.
2
1
1
2
E
+
8
3
1
4
3
5
9
2
3
8
1
.
1
8
0
7
E
+
9
4
3
5
0
9
.
8
2
7
2
E
+
8
6
3
5
0
9
.
8
4
6
2
E
+
8
7
1
.
1
8
0
0
E
+
9
4
1
2
0
3
1
3
9
9
1
.
1
7
9
7
E
+
9
4
3
5
0
9
.
7
9
1
4
E
+
8
7
2
5
0
9
.
8
4
6
2
E
+
8
7
1
.
1
8
0
1
E
+
9
5
8
0
6
5
2
5
.
0
6
0
2
E
+
7
0
1
0
9
8
5
.
0
6
0
2
E
+
7
0
1
0
9
8
5
.
0
6
0
2
E
+
7
0
5
.
0
6
0
2
E
+
7
6
3
9
5
8
9
5
.
0
6
1
8
E
+
7
0
1
0
9
8
5
.
0
6
1
8
E
+
7
0
9
9
8
5
.
0
6
1
8
E
+
7
0
5
.
0
6
1
8
E
+
7
7
1
7
,
7
7
1
2
3
,
7
6
0
1
.
7
4
3
0
E
+
8
1
4
9
2
4
5
1
.
5
4
5
4
E
+
8
0
2
5
2
4
5
1
.
5
4
5
2
E
+
8
1
1
.
5
4
5
4
E
+
8
8
2
0
,
4
8
6
2
5
,
9
0
0
1
.
5
1
5
2
E
+
8
1
4
9
2
4
5
1
.
5
4
5
2
E
+
8
1
2
4
2
4
5
1
.
5
4
5
2
E
+
8
1
1
.
5
4
5
4
E
+
8
International Journal of Production Research 1907
D
o
w
n
l
o
a
d
e
d

b
y

[
1
8
1
.
1
1
4
.
1
2
6
.
1
4
7
]

a
t

1
5
:
0
3

0
1

J
u
n
e

2
0
1
4

level of consolidation that would allow for early consolidation from multiple plants for long shipments to demand
regions which are then divided into delivery orders at local warehouses within demand regions. This can be modelled
by allowing transhipments between storage facilities. With regards to model solution, other chromosome denitions may
be considered for the genetic algorithm. The use of integers instead of binary values could improve the running time as
a result of chromosomes length reduction. Sensitivity to shipping policies could able be investigated.
References
Ballou, R. H. 1968. Dynamic Warehouse Location Analysis. Journal of Marketing Research 5 (3): 271276.
Bidhandi, H. M., and R. M. Yusuff. 2011. Integrated Supply Chain Planning under Uncertainty Using an Improved Stochastic
Approach. Applied Mathematical Modelling 35 (6): 26182630.
Chaudhry, S. S., S. S. He, and P. E. Chaudhry. 2003. Solving a Class of Facility Location Problems Using Genetic Algorithms.
Expert System 20 (2): 8691.
Cornuejols, G., R. Sridharan, and J. M. Thizy. 1991. A Comparison of Heuristics and Relaxations for the Capacitated Plant Location
Problem. European Journal of Operational Research 50 (3): 280297.
Dasci, A., and V. Verter. 2001. A Continuous Model for Productiondistribution System Design. European Journal of Operational
Research 129 (2): 287298.
Daskin, M. S. 1995. Network and Discrete Location. New York: John Wiley.
Daskin, M. S., C. R. Coullard, and Z.-J. M. Shen. 2002. An Inventory-Location Model: Formulation, Solution Algorithm and Com-
putational Results. Annals of Operations Research 110 (1): 83106.
Eskigun, E., R. Uzsoy, P. V. Preckel, G. Beaujon, S. Krishnan, and J. D. Tew. 2005. Outbound Supply Chain Network Design with
Mode Selection, Lead times and Capacitated Vehicle Distribution Centers. European Journal of Operational Research 165
(1): 182206.
Gebennini, E., R. Gamberini, and R. Manzini. 2009. An Integrated Production Distribution Model for the Dynamic Location and
Allocation Problem with Safety Stock Optimization. International Journal of Production Economics 122 (1): 286304.
Gen, M., and A. Syarif. 2005. Hybrid Genetic Algorithm for Multi-Period Production/Distribution Planning. Computers & Indus-
trial Engineering 48 (4): 799809.
Goldberg, D. E. 1989. Genetic Algorithms in Search, Optimization and Machine Learning. Boston: Addison-wesley Longman.
Hinojosa, Y., J. Puerto, and F. R. Fernndez. 2000. A Multiperiod Two-echelon Multicommodity Capacitated Plant Location Prob-
lem. European Journal of Operational Research 123 (2): 271291.
Holland, J. H. 1975. Adaptation in Natural and Articial Systems. Cambridge: The MIT.
Hosage, C. M., and M. F. Goodchild. 1986. Discrete Space Location-allocation Solutions from Genetic Algorithms. Annals of Oper-
ations Research 6 (2): 3546.
Jamarillo, J. H., J. Bhadury, and R. Batta. 2002. On the Use of Genetic Algorithm to Solve Location Problems. Computers & Oper-
ations Research 29 (6): 761779.
Jawahar, N., A. Gunasekaran, and N. Balaji. 2012. A Simulated Annealing Algorithm to the Multiperiod Fixed Charge Distribution
Problem Associated with Backorder and Inventory. International Journal of Production Research 50 (9): 25332554.
Konak, A., D. Coit, and A. E. Smith. 2006. Multi-objective Optimization Using Genetic Algorithms: A Tutorial. Reliability Engi-
neering and System Safety 91 (9): 9921007.
Lee, J. I. Moon, and J. Park. 2010. Multi-level Supply Chain Network Design with Routing. International Journal of Production
Research 48 (13): 39573976.
Lu, Z., and N. Bostelb. 2007. A Facility Location Model for Logistics Systems including Reverse Flows: The Case of Remanufac-
turing Activities. Computers & Operations Research 34 (2): 299323.
Manzini, R., and E. Gebennini. 2008. Optimization Models for the Dynamic Facility Location and Allocation Problem. Interna-
tional Journal of Production Research 46 (8): 20612086.
Melo, M. T., S. Nickel, and F. Saldanha-da-Gama. 2005. Dynamic Multi-commodity Capacitated Facility Location: A Mathematical
Modeling Framework for Strategic Supply Chain Planning. Computers & Operations Research 33 (1): 181208.
Melo, M. T., S. Nickel, and F. Saldanha-da-Gama. 2009. Facility Location and Supply Chain Management A Review. European
Journal of Operational Research 196 (2): 401412.
Miranda, P. A., and R. A. Garrido. 2004. Incorporating Inventory Control Decisions into a Strategic Distribution Network Design
Model with Stochastic Demand. Transportation Research Part E 40 (3): 183207.
Miranda, P. A., and R. A. Garrido. 2009. Inventory Service-Level Optimization within Distribution Network Design Problem. Inter-
national Journal of Production Economics 122 (1): 276285.
Novaes, A. G. N., J. E. S. Cursi, A. C. L. Silva, and J. C. Souza. 2009. Solving Continuous Locationdistricting Problems with
Voronoi Diagrams. Computers & Operations Research 36 (1): 4059.
Nozick, L. K., and M. A. Turnquist. 2001. Inventory, Transportation, Service Quality and the Location of Distribution Centers.
European Journal of Operational Research 129 (2): 362371.
1908 R.G. Askin et al.
D
o
w
n
l
o
a
d
e
d

b
y

[
1
8
1
.
1
1
4
.
1
2
6
.
1
4
7
]

a
t

1
5
:
0
3

0
1

J
u
n
e

2
0
1
4

Owen, S. H., and M. S. Daskin. 1998. Strategic Facility Location: A Review. European Journal of Operational Research 111 (3):
423447.
zceylan, E., and T. Paksoy. 2013. A Mixed Integer Programming Model for a Closed-loop Supply-chain Network. International
Journal of Production Research 51 (3): 718734.
Ozsen, L., C. R. Coullard, and M. S. Daskin. 2008. Capacitated Warehouse Location Model with Risk Pooling. Naval Research
Logistics 55 (4): 295312.
Pourakbar, M., R. Z. Farahani, and N. Asgari. 2007. A Joint Economic Lot-size Model for an Integrated Supply Network Using
Genetic Algorithm. Applied Mathematics and Computation 189 (1): 583596.
Qi, L., and Z.-J. M. Shen. 2007. A Supply Chain Design Model with Unreliable Supply. Naval Research Logistics 54 (8):
829844.
Revelle, C. S., H. A. Eiselt, and M. S. Daskin. 2008. A Bibliography for Some Fundamental Problem Categories in Discrete Loca-
tion Science. European Journal of Operational Research 184 (3): 817848.
Rudi, N., S. Kapur, and D. F. Pyke. 2001. A Two-location Inventory Model with Transshipment and Local Decision Making. Man-
agement Science 47 (12): 16681680.
Santoso, T., S. Ahmed, M. Goetschalckx, and A. Shapiro. 2004. A Stochastic Programming Approach for Supply Chain Network
Design under Uncertainty. European Journal of Operational Research 167 (1): 96115.
Shen, Z.-J. M., and L. Qi. 2007. Incorporating Inventory and Routing Costs in Strategic Location Models. European Journal of
Operational Research 179 (2): 372389.
Shu, J., C. P. Teo, and Z.-J. M. Shen. 2005. Stochastic Transportation-inventory Network Design Problem. Operational Research
53 (1): 4860.
Silva, F. J. F., and D. S. de la Figuera. 2007. A Capacitated Facility Location Problem with Constrained Backlogging Probabilities.
International Journal of Production Research 45 (21): 51175134.
Snyder, L. V., and M. S. Daskin. 2005. Reliability Models for Facility Location: The Expected Failure Cost Case. Transportation
Science 39 (3): 400416.
Sourirajan, L., L. Ozsen, and R. Uzsoy. 2007. A Single Product Network Design Model with Lead Time and Safety Stock Consider-
ations. IIE Transactions 39 (5): 411424.
Sourirajan, L., L. Ozsen, and R. Uzsoy. 2009. A Genetic Algorithm for a Single Product Network Design Model with Lead Time
and Safety Stock Considerations. European Journal of Operational Research 197 (2): 599608.
Sridharan, R. 1995. The Capacitated Plant Location Problem. European Journal of Operational Research 87 (2): 203213.
Tagaras, G., and D. Vlachos. 2001. A Periodic Review Inventory System with Emergency Replenishments. Management Science
47 (3): 415429.
Weber, A. 1909. Theory of the Location of Industries. Chicago: The University of Chicago Press.
Yao, Z., L. H. Lee, W. Jaruphongsa, V. Tan, and C. F. Hui. 2010. Multi-source Facility Location-allocation and Inventory Problem.
European Journal of Operational Research 207 (2): 750762.
Yigit, W., M. Emin Aydin, and O. Turkbey. 2006. Solving Large-Scale Uncapacitated Facility Location Problems with Evolutionary
Simulated Annealing. International Journal of Production Research 44 (22): 47734791.
Zhou, G., H. Min, and M. Gen. 2003. A Genetic Algorithm Approach to the Bi-Criteria Allocation of Customers to Warehouses.
International Journal of Production Economics 86 (1): 3545.
Appendix
The proposed model assumes use of full-truck loads in transportation. However, the number of full-trucks in each order may be
greater than 1. In practice, these loads may be staggered but we assume all are shipped jointly in our current inventory calculations.
The model could be readily adjusted for other shipping scenarios as discussed earlier in the paper. Here we develop an upper bound
on the loss from use of full truck load shipments.
For each suppliercustomer pair, we want to minimise the total cost for this pair given a policy. Using the same parameters as in
the paper:
Minimise:
TC A a
Q
C
_ _
bl Q
_ _

D
Q
h
Q
2
(A:1)
Let the optimal quantity Q mC, where m may not be an integer. Total Cost function (A.1) becomes
TC(m) (A a dme bl m C)
D
m C
h
m C
2
(A:2)
International Journal of Production Research 1909
D
o
w
n
l
o
a
d
e
d

b
y

[
1
8
1
.
1
1
4
.
1
2
6
.
1
4
7
]

a
t

1
5
:
0
3

0
1

J
u
n
e

2
0
1
4

We want to know the maximum loss between this Q and the better of the oor and ceiling function multiples of C. Dene:
m
1
bmc; m
2
dme;
d
minfTC(m
1
) TC(m) ; TC(m
2
) TC(m)g; if m[1
TC(m
2
) TC(m); if 0\m\1
0; if m 1
_
_
_
TC(m
1
) TC(m)
AD
C
1
m
1

1
m
_ _

aD
C
dm
1
e
m
1

dme
m
_ _

hC
2
(m
1
m)
Note that m 1 m
1
m
TC(m
1
) TC(m)
AD
C
1
m 1

1
m
_ _

aD
C
1
dme
m
_ _

hC
2
(m m)
TC(m
1
) TC(m)
AD
Cm(m 1)
TC(m
2
) TC(m)
AD
C
1
m
2

1
m
_ _

aD
C
dm
2
e
m
2

dme
m
_ _

hC
2
(m
2
m)
Note that m m
2
m 1
TC(m
2
) TC(m)
AD
C
1
m 1

1
m
_ _

aD
C
1
dme
m
_ _

hC
2
(m 1 m)
TC(m
2
) TC(m)
hC
2
d
min
AD
Cm(m 1)
;
hC
2
_ _
; if m > 1
hC
2
; if 0\m\1
0; if m 1
_

_
.
In conclusion, d
hC
2
.
1910 R.G. Askin et al.
D
o
w
n
l
o
a
d
e
d

b
y

[
1
8
1
.
1
1
4
.
1
2
6
.
1
4
7
]

a
t

1
5
:
0
3

0
1

J
u
n
e

2
0
1
4