Sunteți pe pagina 1din 5

2010 Fourth Asia International Conference on Mathematical/Analytical Modelling and Computer Simulation

Hub Covering Location Problem under Capacity Constraints


Reza Ghodsi, Assistant
Professor

Mehrdad Mohammadi, MSc


Student

Industrial Engineering Dept.


University of Tehran
Tehran, Iran
ghodsi@ut.ac.ir

Industrial Engineering Dept.


University of Tehran
Tehran, Iran
mehrdadmohamadi@ut.ac.ir

Hamideh Rostami, BSc Student


Industrial Engineering Dept.
University of Tabriz
Tabriz, Iran
neshani_13@yahoo.com

interconnected as shown in Fig. 1, where the squares and


circles represent hub and non-hub nodes, respectively.

Abstract The hub location problem appears in a variety of


applications including airline systems, cargo delivery systems,
and telecommunication network design. When we analyze hub
location applications separately, we observe that each area has
its own characteristics. In this paper, we study the single
allocation hub covering problem under capacity constraints
(or CSAHCLP - Capacitate Single Allocation Hub Covering
Location Problem) over complete hub networks and propose a
mixed-integer programming formulation to this end. The aim
of our model is to find the location of hubs and allocate nonhub nodes to the located hub nodes so much that the travel
cost between any hub-node pair is within a given cost bound
and hubs are considered under capacity constraint. Unlike [1]
we prepare new formulation with covering radius. In general
this paper attempts to propose a new mixed-integer
programming formulation and adapt the imperialist
competitive algorithm to solve the hub covering location
problem. Also unlike previous studies, we adapt new solution
algorithm (Imperialist competitive algorithm) for solving our
problem that has not used yet.

Figure1. Hub-and-spoke network designs.

As for how non-hub nodes are allocated to the located


hub nodes, two fundamental types of hub networks are
defined in the single and multiple allocations. In single
allocation hub networks, each non-hub node is allocated to
exactly one hub; in multiple allocation networks, a non-hub
node can be allocated to more than one hub. O'Kelly [2] at
first introduced the hub location problem. Also in [3] he
presented the first recognized mathematical formulation for
a hub location problem by studying airline passenger
networks. His formulation is referred to as the single
allocation p-hub median problem. The objective of his
model was to minimize the total transportation cost of
demand flow. In order to reflect the economies of scale in
hub-to-hub connections, [3] introduced a constant discount
factor, [0, 1], for using inter-hub connections. OKelly
[3] introduced a data set based on the airline passenger
interactions between 25 US cities in 1970 evaluated by the
Civil Aeronautics Board (CAB). The rest of the literature on
the hub location problem primarily focused on the
linearization of the quadratic model proposed in [3], for
example, as in [4], [1], [5], and [6]. These studies introduce
different mathematical formulations and solution procedures
for the minimization of the total transportation cost.
Campbell [7] introduced different hub location problems (phub center, hub covering) to the literature and considered
different objective functions. In particular, the hub covering
problem minimizes the total cost of establishing hub
networks, so that the cost (or travel time) between any
origin-destination pair is within a given bound. Campbell [7]
provided quadratic as well as linear formulations for both
single and multiple allocation variants of the problem. The
first attempt to provide computational results for the single
allocation hub covering problem, however, was from Kara
and Tansel [8]. They also suggested various linear

Keywords:
Hub location; Hub covering problem; Network
design; Capacity constraint.

I.

INTRODUCTION

Hub-and-spoke systems are prevalent in many areas of


everyday life from passenger travel through an airlines
network of airports, to postal delivery, communication, and
public transportation networks. These applications have led
to more researches and researcher on the hub location
problem, which finds the best design of a hub-and-spoke
network by locating hubs and assigning non-hub nodes to
the hub nodes. The hub location problem includes the
selection of the location of hub facilities and the allocation
of the demand nodes to these located hub facilities. The hub
nodes have two roles: they collect demand from the non-hub
nodes and transport the demand to the other hubs in the
network, or they distribute demand that received from other
hub nodes in the network, to the non-hub nodes that they
serve. In this paper the hub nodes are assumed to be fully

978-0-7695-4062-7/10 $26.00 2010 IEEE


DOI 10.1109/AMS.2010.132

204

formulations and proved the NP-hardness of the hub


covering problem. Ernst, Jiang, and Krishnamoorthy [9]
proposed a better mathematical formulation for the hub
covering problem using the radius idea. For the single
allocation uncapacitated hub location problem Labbe and
Yaman [10] derived a family of valid inequalities that
generalizes the facet-defining inequalities and that can be
separated in polynomial time.
Ernst and Krishnamoorthy [11] presented two new
formulations for the capacitated single allocation hub
location problem. Their formulations are a modified version
of the previous mixed integer formulations developed for the
p-hub median problem. Similar to other hub location studies,
we use a constant cost discount factor [0, 1] to represent
the economies of scale in hub-to-hub connections and also
we use a constant cost discount factor [0, 1] to represent
the economies of scale in hub-to-node or node-to-hub
connections (so much that >), and we do not allow direct
connections between the non-hub nodes. Unlike Ernst,
Jiang, and Krishnamoorthy [9], we consider the sum of cost
of all flow between each pair nodes and fixed cost of
opening new hub for objective function. Also in our problem
we consider the capacity constraint for each hub. Each hub
has a radius that its use for covering the non-hubs. Non-hub
nodes can allocate to one hub when they be in the coverage
area of that hub.
Unlike Ernst, Jiang, and Krishnamoorthy [9], we
consider the hub covering version of the problem. We do not
locate a fixed number of hub arcs and we force the hub arc
network to be connected. Also we do not impose any
structure on the hub network. Our integer programming
formulation involves O (n4) decision variables and O (n4)
constraints. In order to solve realistically sized instances, we
adapt a heuristic for this problem. In contrast to other hub
location problems, constructing feasible solutions for the
hub covering problem, especially with tight radius and
capacity bounds, is a challenge. Many studies in the
literature apply different heuristic approaches to hub
location problems, for example, the tabu search heuristics
proposed by Klincewicz [12] and Skorin-Kapov [13], the
simulated annealing heuristic by Ernst and Krishnamoorthy
[1], and the Lagrangean relaxation based heuristic by Pirkul
and Schilling [14] for p-hub median problems.
Additional contributions include a shortest path based
heuristic by Ebery, Krishnamoorthy, Ernst, Boland [15], a
genetic algorithm by Cunha and Silva [16], a hybrid
heuristic by Chen [17], and a dual- ascent heuristic by
Cnovas, Garcia, Marin [18] for hub location problems with
fixed costs. Lastly, proposals for p-hub center problems
include a tabu search based heuristic by Pamuk and Sepil
[19] and a greedy heuristic by Ernst, Hamacher, Jiang,
Krishnamoorthy, Woeginger [20] The reader should note
that, for hub location problems, nearest allocation strategy
(assigning a non-hub node to its nearest hub) does not
necessarily give optimum solutions for the hub location
problem. In this paper, we relax the complete hub network
assumption in the hub covering problem. To be able to
handle real sized problems, we propose a tabu search based
heuristic for our problem. As will be apparent with the

computational studies, the heuristic behaves quite


effectively. To the best of our knowledge, both the
formulation and the heuristic are new for the hub covering
problem. We propose an integer programming formulation
in the second section of this paper. In the third section, we
present and explain our heuristic algorithm. The fourth
section is dedicated to the computational analysis. The last
section is devoted to concluding remarks.
II.

MATEMATICAL FORMULATION

This formulation has been formulated before. But in this


paper we prepare a new and complete formulation. In this
paper we assume that there is a given node set N with n
nodes where some of them can be located such as hubs. The
mathematical model locates hubs, constructs the hub
network, and allocates the remaining nodes in set N to these
hubs, such that any hub allocates only those nodes that it can
cover. Also just a limited number of nodes can be allocated
to specific hub, because all hubs are under capacity
constraints.
The objective of our mathematical model is to minimize
the total cost of flow between any origin-destination pair
and the total cost of establishing hubs. The parameters of the
model are as fallows. The parameter Wij is the flow between
nodes i and j, Cij is the transportation cost of a unit of flow
between i and j, Fk is the fixed cost of opening a hub at node
k. and rk is the maximum collection/distribution cost
between hub k and nodes that are allocated to hub k. Chubk
is the capacity of hub k. And
, is cost discount
is the same
factor for between tow hubs and also
but it is for between non-hub nodes and hubs. That most
likely be higher than and it is expected to be a number
close to 1. We define the decision variables of the model as
follows:
Xik = 1 If node i is allocated to hub at node k; 0 otherwise.
Is the total amount of flow of commodity i (i.e.,
traffic emanating from node i) that is routed between hubs k
and l.
Let
be the total amount of flow originating at
node i. Also M is a big number.
More specifically, the allocation decisions are taken
care by the classical
variables. Consistent with the
for some k nodes, it means
literature, if the variable
that node k is a hub node. The objective function of our
mathematical model is to minimize the total cost of flow
between any origin-destination pair and the total cost of
establishing hubs. With the previously defined parameters
and decision variables, the objective function is expressed as
follows:

In the objective function, in the first term, we sum the


total cost of flow between non-hub nodes and hub nodes; in
the second term we sum total cost of flow between each tow
hubs and in the third term we calculate the total cost of

205

Each bit or bit string expresses a node in the network, where


its value explains the number of the hub or the node is
allocated to. Further, when the value of the country is equal
to the number of the node, the node is considered a hub. For
example we can represent one country as follow:

establishing hubs. After objective function we present our


model constraints as follows:

Country = [1 7 3 7 3 3 7 1 9 9]

In this country, 1, 3, 7, and 9 are hubs. Nodes 2 and 4


allocated to hub 7. Also nodes 5 and 6 allocated to 3 and
node 10 allocated to 9 and 8 allocated to 1.
B.

Initial countries
The initialization process begins with creating the initial
countries. The size of the countries is predefined. For the
hub covering problem with single allocation under capacity
constraints, the number of hubs in the network is undefined.
After that hubs located in the network, the remaining spokes
should be allocated to a located hubs. For the initial
population of countries, spokes are allocated to their nearest
hub, based on cost (radius) values. Also all hubs should
cover some nodes that not exceeded from their capacity.
Design of each network that satisfies all constraints is
challenging.

Constraints (2) and (10) ensure that each node is


assigned to exactly one hub. Eq. (3) is the flow balance
equation (divergence equation) for commodity i at node k
where the demand and supply at the node is determined by
the allocations . Constraints (4) ensure that the allowed
nominal capacity of the hub is not exceeded by preventing
cargo from entering. Constraint (5), makes sure that node i
can only be allocated to k, if cost
between i and k is at
of k. constraint (6) states that a node
most the radius
cannot be allocated to another node unless that node is a hub

C. Function evaluation
The evaluation function is an operation to evaluate how
good the network configuration of each individual is,
making the comparison between different solutions possible.
The evaluation function consists of calculating the value of
the objective function of the network represented by each
country.
D.

node. Constraints (7), (8) and (9) ensure that


can be
higher than zero only when k and l are hubs. Finally the

of best country (those countries have minimum cost =


)
) as colonies. After
as emperor. We consider reminders (
that we allocate reminded countries to emperor according
their authority. To perform this, we compute the normalized
cost for each emperor as follow:

constraint (11) ensure that the variable


be higher than
zero, because it is the amount of flow and need to be more
than zero.
III.

Imperialism
For producing the initial imperialist, we consider some

ADAPTED IMPERIALIST COMPETITIVE


ALGORITHM

In which,
is the cost of nth emperor, and
is the
normalized cost for nth emperor. If we have the normalized
cost, we can compute the proportional power for each
emperor for allocating the colonies to emperor as fallow:

It is hard to solve most of the NP-complete problems to


optimality for realistically sized instances. For the problem
at hand, even finding a feasible solution is challenging. With
this motivation, we decided to develop a heuristic algorithm
for our problem. In this section, according to AtashpasGargari E. and Lucas C. [21], the adapted Imperialist
Competitive Algorithm for the hub location problem is
presented and described.

Finally the numbers of colonies that each emperor has


are:

A. Representation of the solution


According to Kelly Takano and Makoto Arai [22], the
solution of the hub location problem should represent the
network by showing the location of the hubs and the
allocation of the remaining nodes to a valid hub. In this
algorithm each solution is one country. The procedure uses
integer numbers to represent a given network. The length of
the vector is equal to the number of nodes in the network.

In which,
is the number of colonies that nth
emperor has. After allocating the colonies to emperors the
imperialism competition will be started. Each emperor tries
to develop its colonies. In next section we present three
methods for developing or assimilating colonies.

206

E. Assimilation methods
1) Near building policy
In this approach we assimilate colonies with emperors.
For this work we determine the hub of emperor and use the
half of them for locating hubs in some colonies. Those
colonies which select to assimilation are limited. We choose
some colonies randomly. After hub selection for colonies we
allocate remained nodes to located hubs. Finally colonies
will be assimilated. In this process, if one colony will find
better cost then the position of emperor and best assimilated
colony will be changed. For example if the array matrix of
emperor is Emp= [1 2 7 6 6 7 2 2 9 10] therefore the hub
location is [1 2 6 7 9 10]. Next we select half of hub location
matrix randomly. For example: [6 9 10]. Then for each
colony that has chosen we locate nodes 6, 9, and 10 as hubs.
Finally we allocate remained nodes to this hub set.
2) Association policy
In this method each emperor combines its feature with
colony. This approach is similar to cross over in genetic
algorithm. We can say our solution algorithm is hybrid of
Imperialist competitive algorithm and genetic algorithm.
Because of this contribution our solution algorithm solves
the problem in lower time. After each association the
country will passed from filtering step in program to
guarantee that the country will have a valid structure at the
end of the process. For example if emperor is Emp= [1 2 7 6
6 7 2 2 9 10] and the colony is Col=[2 2 3 4 7 3 7 8 8 2].
Then we chose the association point randomly. For example
if association point is 6 then we combine [1 2 7 6 6 7] with
[7 8 8 2]. Therefore the colony will be changed to [1 2 7 6 6
7 7 8 8 2].
3) Revolution policy
In our algorithm mutation consists of creating the new
solution. It likes the population production as mentioned
above. This step is for releasing from local search in solution
space. In some imperialism some of colonies will be
revolted. In our algorithm we have revolution rate that is
closed to 0.1. For example if the imperialism has 20
populations, then just two of their colonies will be revolted.
F. Cost of each imperialism
To compute the cost or fitness function for any
imperialism we act as follow:

algorithm after predefined iterations. If in each imperialism


just emperor remained then that imperialism will be
eliminated and allocate to strongest imperialism.
H. Pseudo code for Imperialist Competitive Algorithm
In this section we present the pseudo code for our
solution algorithm in 7 steps.
1. Create some random point on solution space and cause the
initial imperialism.
2. Assimilate the colonies to their emperor.
3. If in each imperialism one colony had lower cost than
emperor then change the position of emperor and that
colony.
4. Compute the total cost of imperialism (Consider the cost of
emperor and its colonies).
5. Chose the weakest colony from weakest imperialism and
allocate that to strongest imperialism.
6. Eliminate the weakest imperialism.
7. If only one imperialism is remained, stop or go to 2.
Figure2. Pseudo code for Imperialist Competitive Algorithm

IV.

Previous data in the literature is not suitable for our


problem. Hence, we produce random parameters so that they
create feasible solutions. For these data we compare both of
the exact and heuristic solutions. For the exact solution we
use LINGO software and for the heuristic solution we use
our proposed genetic algorithm. For comparison we consider
five amounts of nodes which are 10, 11, 12, 15, and 20. For
each of these, all parameters are as follows:
Table I. Computational comparison of the IP model and the heuristic
algorithm with n = 10, 11, 12, 15, 20
Number
of
Nodes n

10
11
12
15
20

0.75
0.75
0.75
0.75
0.75

In which, is a positive number between 0 and 1. Also


is the total cost of nth imperialism. If is low then the
cost of nth imperialism close to nth emperor. But if is big
(lower than 1), then the effect of colonies cost is higher.
Usually = 0.3.
G. Imperialist competition
In this algorithm we have imperialist competition
between each two emperors. It means that the strongest
emperor occupy the weakest colony in weakest imperialism.
During this competition each emperor who cannot develop
their colonies will be eliminated and will be occupied with
stronger imperialism. This process will be continued that
just one imperialism remaining. Or we can stop the

COMPUTATIONAL RESULTS

Parameter

Computational Result

The
OFV
with
LINGO

The OFV
with
heuristic

Number
of hubs

Gap
(%)

0.95
0.95
0.95
0.95
0.95

1434.8
1878.9
2537.45
3850.65
7442.3

1434.8
1878.9
2562.3
3850.65
7469.6

4
5
5
8
13

0
0
0.96
0
0.36

In problems with 10, 11, and 15 nodes our algorithm


does not have any gap. But with 12 and 20 nodes, we have
0.96, and 0.36 percent gaps, respectively. In larger size, the
LINGO takes a lot of time (seven hours or more) to solve
the problem. Thus, for large size problems it is better to use
the heuristic method. Also, our algorithm results in few gaps
for large problems. The average gap for our test problems is

207

just 0.264%. Therefore, it is reasonable to use the heuristic


method.
V.

CONCLUSION

In this paper, we studied the single allocation hub


covering location problem under capacity constraint. Unlike
[1] we provided a new formulation with covering radius and
unlike [9], we considered the sum of costs of all flows
between each pair of nodes and the fixed costs of opening
new hubs in the objective function. Also in our problem we
considered the capacity constraint for each hub. In contrast
to other hub location problems, constructing feasible
solutions for the hub covering problem, especially with tight
radius and capacity bounds, is a challenge. To the best of the
authors knowledge, our solution and coding is unique in the
literature proposed for the hub covering problem. In order to
solve real size problem instances, we adopted the imperialist
competitive algorithm for the first time. The imperialist
competitive algorithm is proposed originally for continuous
problems but we adapted this algorithm for discrete
parameters. The algorithm has very short CPU time and
small acceptable gaps for large size problems. The heuristic
algorithm was tested using random data. We compared the
performance of our heuristic with the LINGO software and
the results were promising. The average gap for our test
problems is just 0.264% and it is reasonable to use the
heuristic method to solve real size problems.
REFERENCES
[1]

[2]
[3]

[4]
[5]

[6]

[7]

[8]

[9]

[10]

Ernst, A.T., Krishnamoorthy, Efficient algorithms for the


uncapacitated single allocation p-hub median problem.
Location Science 4 (3), 139154. 1996.
OKelly, M.E. The location of interacting hub facilities.
Transportation Science 20 (2), 92105. 1987.
OKelly, M.E. A quadratic integer program for the location
of
interacting hub facilities. European Journal of
Operational Research 32, 393404. 1995.
Campbell, J.F. Hub location and the p-hub median
problem. Operations Research 44 (6), 113. 1996.
OKelly, M.E., Skorin-Kapov, D., Skorin-Kapov, J. Lower
bounds for the hub location problem. Management Science
41 (4), 713721. 2005.
Skorin-Kapov, D., Skorin-Kapov, J., OKelly, M. Tight
linear programming relaxations of incapacitated p-hub
median problems. European Journal of Operational
Research 94, 582593. 1996.
Campbell, J.F. Integer programming formulations of
discrete hub location problems. European Journal of
Operational Research 72, 387405. 1994.
Kara, B.Y., Tansel, B.C. The single-assignment hub
covering problem: Models and linearizations. Journal of the
Operational Research Society 54, 5964. 2003.
Ernst, A.T., Jiang, H., Krishnamoorthy, M. Reformulations
and computational results for incapacitated single and
multiple allocation hub covering problems. Unpublished
Report, CSIRO Mathematical and Information Sciences,
Australia. 2005.
Labbe, M., Yaman, H. Projecting flow variables for hub
location problems. Networks 44 (2), 8493. 2004.

[11] Ernst, A.T., Krishnamoorthy, M. Solution algorithms for the

[12]

[13]

[14]

[15]

[16]

[17]
[18]

[19]

[20]

[21]

[22]

208

capacitated single allocation hub location problem. Annals


of Operations Research 86, 141159. 1999.
Klincewicz, J.G. Avoiding local optima in the p-hub
location problem using tabu search and GRASP. Annals of
Operations Research 40, 283302. 1992.
Skorin-Kapov, D., Skorin-Kapov, J. On tabu search for the
location of interacting hub facilities. European Journal of
Operational Research 73, 502509. 1994.
Pirkul, H., Schilling, D.A. An efficient procedure for
designing single allocation hub and spoke systems.
Management Science 44 (12), 235242. 1998.
Ebery, J., Krishnamoorthy, M., Ernst, A., Boland, N. The
capacitated multiple allocation hub location problems:
Formulations and algorithms. European Journal of
Operational Research 120, 614631. 2000.
Cunha, C.B., Silva, M.R.. A genetic algorithm for the
problem of configuring a hub-and- spoke network for a LTL
trucking company in Brazil. European Journal of
Operational Research 179, 747758. 2007.
Chen, J.F. A hybrid heuristic for the uncapacitated single
allocation hub location problem. Omega 35, 211220. 2007.
Cnovas, L., Garca, S., Marin, A. Solving the
uncapacitated multiple allocation hub location problem by
means of a dual-ascent technique. European Journal of
Operational Research 179, 9901007. 2007.
Pamuk, F.S., Sepil, C. A solution to the hub center problem
via a single-relocation algorithm with tabu search. IIE
Transactions 33 (5), 399411. 2001.
Ernst, A., Hamacher, H., Jiang H., Krishnamoorthy, M.,
Woeginger, G. Uncapacitated single and multiple allocation
p-hub center problems. Unpublished Report, CSIRO
Mathematical and Information Sciences, Australia. 2002.
Atashpas-Gargari E. and Lucas C., Colonial competitive
algorithm. 2007 E Congress on Evolutionary Computation
(CEC 2007).
Kelly Takano and Makoto Arai (2009). A genetic algorithm
for the hub-and-spoke problem applied to containerized cargo
transport. J Mar Sci Technol 14:256274. 2009.

S-ar putea să vă placă și