Académique Documents
Professionnel Documents
Culture Documents
1.1 Overview
The problem of Facility Location and Planning deals with the optimum distribution of one or
more facilities, given a set of customers at multiple locations who need to be serviced. Recently
this problem has achieved particular importance due to the development of large, distributed
supply chains that must remain flexible and react quickly to changing customer demands, while
at the same time maintaining low fixed facility costs. While the facility location problem is in
general NP-hard, a number of variants have been proposed, such as: restricted number of
facilities; opening cost associated with one or more facilities; facilities with or without capacity
constraints; and a variety of objective functions that need to be optimized. Some of the factors
that influence the location of facilities are: location of supply centers; potential market demand;
projected facility cost in terms of labor and raw material costs; transportation costs from various
facilities; response time from the facility; inventory maintenance cost; tax structure and
regulatory laws of the state or country in which the facilities are to be located; etc. An extensive
body of literature has studied many aspects of these variants, including classic ones such as the
multiple-source Weber problem and the P-median problem. Correspondingly, many different
approaches have been presented to solve these problems, involving the use of metaheuristics,
simulation and optimization methods. This review will review some of these problem variants
Facility location problems were first shown to be NP-hard by Hansen and Mladenovic (1997),
who tried to minimize the total transportation cost while setting up p facilities in a set of L
locations and for a set of U locations of users. Also known as the P-median problem, the authors
proposed a class of Variable Neighborhood Search (VNS) heuristics and compared these with
another heuristics known as the Tabu search (originally proposed by Fred Glover in 1986). The
VNS algorithm considers a finite set of neighboring locations having k (k=1,..,kmax), so that
k(x) is the set of solutions in the kth neighborhood of x. for multiple neighborhoods, the
objectives usualy are: find out which k is to be used and how many; their search order; and the
The simplest type of these problems is single facility location, some examples of which
are the location of a new warehouse, or a hospital in a city, or a classroom in a school, or a node
within a computer network. The more complex problem corresponds to location of multiple
facilities, in which there must be at least one link with each other between every new facility.
Another category is the Location Allocation problem, originally proposed by Cooper in 1963, in
which the new facilities locations must be optimized to service customers at minimum
transportation cost. On the other hand, the Quadratic Allocation Problem studies the case when
the facility cost is proportional to its distance from other facilities. The Covering Problem
specifies a critical coverage distance, Dc, the value of which determines whether or not a
customer will receive service from a particular facility. Perhaps the most important type of
facility location problem is the median problem, as mentioned above, and which can be either 1-
Median or P-Median. Two other variations are the Uncapacitated Facility Location Problem
(UFLP) and Capacitated Facility Location Problem (CFLP). In UFLP, the sum of fixed and
distributed costs is sought to be minimized for specified sets of facility locations and customer
locations, there being only a fixed cost. In CFLP, the same costs are sought to be minimized
under the same conditions as before, but with the added constraint that facilities also have
Classical facility location models are usually of four types: analytical, which make several
simplifying assumptions; continuous problems, including the Weber type, which represent reality
geometrically and treat locations as continuous; network models consisting of links and nodes;
and discrete models, such as the N-median discrete problem or the uncapacitated problem in
The single facility location problem, also known as the Fermat-Weber problem, is the
simplest, and will be formulated first: it is one of locating a facility X* so as to minimize the
() = (, )
=1
where i is the facility index, wi is the weight assigned to the i-th facility, ( , ) is distance
between facility i and a new one, and = ( , ) is the co-ordinated for the i-th facility.
problem by implementing a Lagrangian brand and bound algorithm. They observed that the
Lagrangian relaxation technique could effectively solve most problems within a few seconds of
CPU time, except for the case where the randomness in the capacity considered was too high.
The multi facility location problem is one of locating more than one facility to serve
multiple customers, so that the facilities are located optimally and the customers are also
assigned optimally (hence it is also known as the location allocation problem). One of the earliest
formulations of the problem was by Elzinga, Hearn and Randolph (1976), as a distance objective
function MiniMax task; for m given facilities, locate n new facilities such that their maximum
1
= { [( )2 + ( )2 ]2 = 1, , ,
1
= 1, , ; [( )2 + ( )2 ]2 1 < < }
where represent weights between new and existing facilities, represent weights between
two new facilities and : ( , ) represent co-ordinates of existing facilities. In addition to the
above formulation, other variations exist, such as: discrete or continuous solving for the area;
distance; stochastic or deterministic parameters; etc. (Eds. Farahani and Hekmatfar, 2009). Not
all of the above variations have been studied in detail; many of the studies reported in literature
relax the NP-hard nature of the problem by using probabilistic assignments. For example, Iyigun
and Ben-Israel (2010) used the well-known iterative Weiszfeld technique for solving a multi-
demands, and the problem of locating K facilities such that 1 < (N being the set of data
points). Customers were each assigned to a facility such that the weighted sum was minimized:
( , )
1 , 2 , ,
=1
where ck represents facility locations and represents the customer cluster at the k-th facility.
The problem is NP-hard in 1 < < , and the authors used the assumption that customers were
1
more likely to be assigned to closer facilities: ()(, ) = (),
1, , where x is
the existence of an exclusion zone, or barrier, where travel is not allowed for example
mountains, water bodies or other natural obstacles, or commercial zoning restrictions in urban
areas. While this problem was first studied by I. N. Katz and L. Cooper in 1981, using circular
barriers, a more general case of polyhedral barriers was investigated by Bischoff, Fleischmann
and Klamroth (2006) as a mixed integer optimization problem. While such barriers can also be
2
(where facilities can be set up and traversed) was defined by the function = (). They also
defined a path to be feasible if it did not intersect inside a barrier zone, and hence reformulated
( , )
=1 =1
= 1, = 1, ,
=1
{0,1}, = 1, , , = 1, ,
, = 1, ,
The authors showed the distances to be Euclidean, and developed a numerical method to solve
reasonably large problem sets within a reasonable amount of computation time. For example
they solved a well-known barrier location problem earlier introduced by Aneja and Parlar (1994),
in which 18 existing location facilites and 12 exclusion zones were considered. The co-ordinates
of existing facilities and extrema of the exlusion zones were both specified, and a problem set of
1-18 new facilities was considered. Figure 1A below shows the best locations and allocation
clusters obtained for the case N=3 (3 new facilities), while the 1B shows a graph of the lowest
objective function values as the number of new facilities being condiered increases (N=1-18):
Figure 1: A multi-facility problem solution set. (A) Location and cluster allocation for 3 new
facilities (B) lowest function values as new facilities considered increases, with and without
For larger facility location problems of 300 facilities and 300 customers, Avella, Boccia, Sforza
and Vasilev (2009) used a maximized Lagrangian relaxation method, followed by a branch and
cut algorithm to determine an upper limit to a CFLP. Considering a set of I facilities, J clients, dj
as the demand of the j-th client and fi as the fixed cost for the i-th new facility, they formulated
With the following constraints: each clients entire demand must be met; the demand to a facility
does not exceed its capacity ( ); and closed facilities do not serve clients.
The authors used their algorithm consisting of a relaxation, core selection and branch-and-cut on
the core set, on up to 2000 facilities and were able to solve the problems within a maximum time
In contrast to CFLPs, UFLPs have perhaps been studied more in literature, and they have
a variety of conditions. A general formulation for a static UFLP with a MiniSum objective
function is as follows:
() = min { + }
where the symbols have their previous meanings (Galvao 2004). Another type of UFLP is that of
dynamic location, which can be extended from the class of static allocation problems by
excluding fixed costs. An early study of an extension to this class of problems was conducted by
(installation, maintenance and transportation costs) problem. A closely related formulation of the
= min { + }
where fik denotes the sum of establishment and maintenance costs, and is the fixed cost incurred
for opening facility i during period k. A variety of algorithms have been suggested for solving
this problem, including dynamic programming, a combination of exact and heuristic phase in-
1. Hansen, P. and Mladenovic, N. (1997) Variable Neighborhood Search for the P-median,
3. Tragantalerngsak, S., Holt, J. and Ronnqvist, M. (2000). An exact method for the two-
5. Iyigun, C. and Ben-Israel, A. (2010). A generalized Weiszfeld method for the multi-
Allocation Problem with Polyhedral Barriers. Computers & Operations Research, 36(5),
1376-1392.
7. Aneja, Y. P. and Parlar, M. (1994). Algorithms for Weber facility location in the presence
8. Avella, P., Boccia, M., Sforza, A. and Vasilev, I. (2009). An effective heuristic for large-