Anda di halaman 1dari 15

Sociedad de Estadistica e Investigacidn Operativa

Top (1998) Vol. 6, No. 2, pp. 261-275

The Many-to-Many Location-Routing Problem


G. Nagy
The Universityof Greenwich, U.K.
S. Salhi
School of Mathematics and Statistics
The Universityof Birmingham, U.K.

Abstract
In this paper the many to many location routing problem is introduced, and its
relationship to various problems in distribution management is emphasised. Useful
mathematical formulations which can be easily extended to cater for other related
problems are produced. Techniques for tackling this complex distribution problem
are also outlined.
K e y Words: Location, Routing, Heuristics.
A M S subject classification: 90B80, 90C10

1 Introduction

In this paper the many-to-many location-routing problem (MMLRP) is ad-


dressed. The paper is divided into four main parts: problem description,
connection with other problems, mathematical formulation and a solution
framework. In the first part we define the M M L R P and describe the issues
concerned. Then we proceed to point out the connections of the M M L R P
to several distribution problems encountered in practice. To aid our under-
standing of the problem structure and to illustrate its connections to the
above problems, we present a mathematical programming formulation for
the M M L R P . The final part is an outline of our solution methodology.

2 Problem Description

We are given a list of n customer locations with their respective distances.


For each ordered pair of customers (i, j) we know the a m o u n t qij which
customer i wishes to send to customer j . This a m o u n t refers to a differ-
ent commodity for each pair. We assume t h a t all customer locations are

Received: December 1997; Accepted: October 1998


262 G. Nagy and S. Salhi

feasible sites for establishing a terminal. For these locations, the costs of
establishing and maintaining a terminal is also given. The drop time of
customers (the time required to service their needs) is known. Finally, we
have a fleet of available vehicles with known cost parameters. These belong
to two classes: inter-hub and local delivery/collection vehicles. Both classes
are homogeneous.
We wish to design a system, consisting of terminals and vehicle routes,
so that from each customer the goods it wishes to send are taken away,
and the goods it wishes to receive are supplied to it, by some vehicle to
and from some terminal. It is also necessary to have a fleet of vehicles to
operate between the terminals, and we may suppose t h a t they are cheaper
per unit a m o u n t transported. These vehicles are assumed to link terminals
directly to one another and no tours are allowed. We assume t h a t all
terminals are connected to all the others. On the other hand, access level
vehicles start from some terminal, serve a number of customers and then
return to the same terminal. Serving customers here may mean both a
delivery and a pickup, which may or may not be made at the same time. A
customer has to be visited by at least one vehicle but it is possible t h a t two
vehicles will serve the needs of the same customer (one for pickups and one
for deliveries). The possibility of vehicles returning to their terminals and
then resuming their journey is forbidden. Furthermore, sorting of goods
cannot be performed on vehicles.
Let us now look at what we mean by "designing the system". We have
to find the number and location of terminals, and the number and routing
of vehicles. The result is a two-level network as shown in Figure 1. The two
levels are referred to as hub level and access level respectively. A locational
decision has to be m a d e at the hub level and routing decisions have to be
made at the access level.
Our aim is to design the above system, with the objective of cost minimi-
sation, such t h a t the requirements of the customers and the maximum ca-
pacity and maximum distance (time) constraints on the access level vehicles
are met. The costs involved can be divided into two categories: terminal
costs and transportation costs. Only fixed costs are considered in the first
category. Transportation costs may be further divided into linehaul (hub-
to-hub) and local delivery/collection (hub-to-customers) costs. We assume
t h a t they are both linear functions of route lengths.
Furthermore we note t h a t since the decisions referred to above relate
The Many-to-Many Location-Routing Problem 263

9 terminals 9 customers _ hub level routes -~ access level routes

F i g u r e 1: An illustration of the m a n y - t o - m a n y location-routing problem

to different time horizons, we need to specify the overall time interval over
which we aim to minimise costs. As the effects of a Iocational decision last
longer than those relating to routing or fleet composition, it is appropriate
to specify the time interval to relate to this decision. This implies that
during the period over which we wish to minimise costs both the vehicle
routes and the composition of the vehicle fleet may change.
Note that the problem outlined above is a representative of a d u s t e r of
similar problems. A number of assumptions were made about the problem.
These are in line with the assumptions generally made in the literature,
see e.g. Campbell (1994), pp.33-35., but we note that these assumptions
may need to be modified according to the particular real-life problem being
tackled.
Possible applications for the M M L R P may occur in the freight transport
or in the postal industries. As the primary aim of all location-routing
problems is the long-term problem of location, the M M L R P can be used in
the location of terminals. The algorithms developed may be used to locate:
264 G. Nagy and S. Salhi

9 freightliner terminals for road transport,

9 sorting yards for rail transport,

9 container terminals for combined road/rail transport, or

9 mail sorting offices for postal companies.

The advantages of using an M M L R P framework rather than traditional


hub location models in the above problems are the same as those of us-
ing location routing to solve depot location problems, as pointed out by
Balakrishnan, Ward and Wong (1987), and Salhi and Rand (1989). The
M M L R P also entails designing the routes in the above-mentioned systems
and the methods we will propose can be used to solve routing problems.
Similarly, if a problem arising in practice can be classified as a sub-problem
of the MMLRP, than it may be possible to use the relevant components of
the proposed methodologies to solve that problem.

3 Connections with Existing Problems and Methods

In this section we shall point out the connections between the problem out-
lined above and various problems of distribution management. Although
we have adopted the name "the many-to-many location-routing problem",
it will be made clear that it can be viewed as an extension to some other
distribution problems. The inter-connection between the various problems
is illustrated in Figure 2.

3.1 Distribution management

As there are several origins and destinations, and there is a possible de-
mand for goods to be transported from any location to any other, we are
dealing with a many-to-many distribution problem. This necessitates the
case of different commodities, which is satisfied, as each supply-destination
pair determines a different commodity. The structure of local collection,
linehaul, and local delivery vehicles is also present.
If we ignore the underlying routing problem, then our problem reduces
to the hub location problem, see Campbell (1994). While this is in many
The Many-to-Many Location-Routing Problem 265

Terminal Location WarehouSeproblemLOCation [


Problem

Many to Many Location


T
Location-Routing
Problem I
Routing Problem

t
Many to Many [ Vehicle Routing
Transportation Problem ] Problem

Figure 2: The interrelation of various distribution problems. Arrows point


from a problem to its sub-problems.

cases solved as a planar problem, the definition in section 2 clearly puts our
problem to the class of discrete location problems. Thus, the hub location
problem can be solved as a simple extension to the discrete warehouse
location problem, see Mirchandani and Francis (1990).
If we consider the location of all the terminals as fixed then the M M L R P
reduces to the many-to-many transportation problem, also known as the
freight transport problem, see Daganzo (1996). This can be modelled and
solved as an extension to the vehicle routing problem, see Laporte (1992).

3.2 Location theory

The problem at hand involves determining the number and location of


terminals. However we may - after suitable modifications - use Iocational
methods which are not specific to the problem of locating terminals.
We will rely on an updated version of a known neighbourhood search
heuristic to solve both the depot and the hub location problems, see Kuehn
and Hamburger (1963).
The assumptions made in the previous section classify our problem as a
discrete uncapacitated location problem. All customer locations are consid-
ered to be feasible d e p o t / h u b locations, although it is possible to restrict
ourselves to just a subset of all customer locations.
266 G. Nagy and S. Salhi

3.3 Vehicle routing

Fixing the location of the terminals and disregarding inter-terminal trans-


portation costs, the M M L R P can be transformed into a multi-depot vehicle
routing problem. The supply and demand of customer i are the sums Y~j qji
and Y~j qij respectively. As all customers can simultaneously be delivery
and pickup points, and as sorting is not performed on the vehicles, we are
dealing with a simultaneous pickup and delivery problem, see Min (1989).
This can be solved using suitable extensions to multi-depot routing meth-
ods.
By fixing the supply of all customers to zero and setting up a fictitious
super-origin the problem further reduces to the basic multi-depot vehicle
routing problem, see Salhi and Sari (1997).

3.4 Location-routing

The problem we consider unifies the terminal location problem (in its g l o b a l
structure) with the pickup-and-delivery problem (in its local structure) in
the framework of location-routing. Although the problem could be solved
without using this field of research - by solving the two sub-problems sep-
arately - it is clear from the literature that the use of location-routing
methods improves the solution quality, see Balakrishnan, Ward and Wong
(1987), Salhi and Rand (1989), and Salhi and Fraser (1996).

4 Mathematical Formulation

4.1 Introduction

This being a new problem, no ILP formulations exist for it in the literature
as yet. Before showing how to solve this problem, we have considered it
desirable to develop an ILP formulation to aid our understanding of the
problem structure. Although such formulations may also be used as part of
the solution procedure, for the ILP may be solved by some exact method,
we shall not a t t e m p t to do this. The reason for having to rely on a different
approach is that ILPs of large size may not be solved efficiently, and our
ILP formulation contains a large number of variables even for a relatively
small number of customers.
The Many-to-Many Location-Routing Problem 267

The ILP presented below is an extension of problem (LRP1) of Laporte


(1989). While his formulation was a pure integer program, we use a 3mixed
ILP 1formulation as the load carried on each arc has to be explicitly in-
cluded in the capaOty constraints. Another problem is that in the M M L R P
vehicles are allowed to serve the demands and supplies of customers on sep-
arate occasions. Thus, we decided to form a set of dummy customers, for
each customer i who is served twice the delivery call remains denoted by i,
but the pickup instance is denoted by n + i, where n is the total number
of customers. If i is served only once, then n + i is a d u m m y customer
who must not be part of any vehicle tour. The distance matrix is extended
accordingly.

4.2 Notation

7t the set of customers, 7 / = {1, 2, ..., n}


ff the duplicate set of customers, f f = {n + 1, n + 2 , . . . , 2n}
/~ the set of vehicles,/g = { 1 , 2 , . . . , m }
fi the fixed cost of maintaining a terminal at customer location i
dij the distance between customers i and j. To account for d u m m y
customers, we define di(j+n) -~ d(i+n)j ~- d(i+n)(j+n) = dij for i,j E 7/
variable routing cost of local delivery/collection vehicles
variable routing cost of inter-terminal vehicles
qij the quantity of goods sent from customer i to customer j
drop time measured in units of distance
D the m a x i m u m distance which access level vehicles may cover in a tour
Q the maximum capacity of access level vehicles
wij t h e load on interhub arc ij
tijk the load on arc ij of vehicle route k

(~ if location i is used as a hub


Yi = otherwise

{~ ifhubsiandjareconnected
zij = otherwise

{10 ifarcijispartofroutek
~Cijk ~ otherwise
268 G. Nagy and S. Salhi

10 if the pickup and delivery of i are served separately


vi= if the pickup and delivery of i are served simultaneously

4.3 Formulation

Min Z fiyi + Z Z Y~ v~dijxok + Y~ Y~ t3dijzo


iEH kEIC i E H U J jEHU,.q" iEH j E H

subject to
(Z Z (dij + ~)xijh) - ~ < D (k 9 If.) (4.1)
i E "Hu,.'T j E 7../u,.7"

i,j 9 7 / U J (4.2)
tijk < Q k9
zo <_yi (i, j 9 7/) (4.3)
zij < yj (i, j 9 7/) (4.4)
zij > Yi "at-Yj - - 1 (i, j 9 7/) (4.5)
Z Z xOk --> 1 (j 9 7/) (4.6)
kEICiE~tUff
(j 9 7/) (4.7)
kEICiET-luJ kEKiE't.luJ
Z Z Xijk =~)J-n (j 9 j ) (4.8)
kEICiET-IU,.7"
Z Z Z ;gijk~
k EK iESuT j E ( ' H u f f ) - (807")

> 1 - (y~ yh) + (Y~ ,,g-.) - ITI


2 < ISUTI ~2n(4.9 )
scT/,Tc
hE8 gET

Z xijk.= Z xjik i, j e T l u J (4.1o)


kEIC
j E'HU,3" jE'~uJ

(k e/C) (4.11)
i E'H o,.q" j E'H O ,.7"

vy < l - y j (j e ~) (4.12)
(i,j e 7/) (4.13)
gE?.lhE'H
i, j E 7 / U f l (4.14)
kEK
gET.lhEH
The Many-to-Many Location-Routing Problem 269

Z Z
kEK iE~luff iE~l iE~
=
"-- Z Z tjik + Z w j i - ( Z q j i ) ( 1 - v j ) (j E 7/) (4.15)
kE~iE~uff iE~ iE~
Z ~ tijk-l-Zwij=
kEKiE~Uff
=ZZ tjik + Z wji - ( E qji)vj-n (j e 3.) (4.16)
kE~iE~uff iE~t iE74
Yi E {0, 1} (i e 7/) (4.17)
zij E {0, 1} (i,j E 7/) (4.18)
i, j E 7 / U 3 . (4.19)
Xijk E {0, 1} k E K~
v, E {0, 1} (i E 7/) (4.20)
Wij )_ 0 (i,j r 7/) (4.21)
i,j E 7/U 3" (4.22)
t~jk > 0 k E/C

4.4 Discussion

In the above formulation, the objective function consists of terminal fixed


costs and variable routing costs for vehicles at both levels.
Constraints (4.1.) and (4.2.) are the maximum distance and maximum
capacity constraints respectively. Constraints (4.3.) and (4.4) specify that
inter-hub links may exist between hubs only if both hubs are open; con-
straints (4.5.) ensure that all hubs are connected. Constraints (4.6.) to
(4.8.) stipulate that every customer belongs to one and on|y one route,
except terminals which may belong to more than one route. Constraints
(4.9.) ensure that every customer is on a route connected to the set of ter-
minals. Constraints (4.10.) stipu[ate that every customer is entered and left
by the same vehicle. Constraints (4.11.) guarantee that a vehicle can de-
part only once from a customer. Constraints (4.12.) specify that terminals
have no duplicates. Constraints (4.13.) and (4.14.) stipulate that flow is
only present between customers and hubs who are connected. Constraints
(4.15.) and (4.16.) are flow conservation equations. Finally, constraints
(4.17.) to (4.22.) represent integrality and non-negativity conditions.
We note that the above formulation contains 4n2m + n 2 + 2n zero-one
270 G. Nagy and S. Salhi

and 4n2m + n 2 continuous variables, 4n2m + n 2 q- 4rim + 3n equations and


22n + + 2 0 n 2 m + 6n 2 + 8urn + 9n + 2m - 1 inequalities. If we set m = n and
replace equations by inequalities, then we find that for problems involving
20 customers there are nearly 17 thousand variables and about 17 million
inequalities. While we do not claim that our formulation is the simplest
possible, this observation nevertheless shows the complexity of the task we
face if we elect to solve the M M L R P using exact methods.

4.5 Extensions and reductions

E x t e n s i o n s . The formulation presented in subsection (4.3) can be mod-


ified to cater for additional specifications on the problem. Capacity con-
straints may be set for terminals by specifying an upper bound for ~je7-t (wij
+wji) for each feasible location i. Depending on the problem specifications,
the maximum capacity constraint for inter-hub vehicles is expressed as an
upper bound either on (wij + wji) or on wij for all pairs of locations i, j. If
the access level vehicle fleet is heterogeneous, then D and Q in inequalities
(4.1.) and (4.2.) should be replaced by Dk and Qk respectively: these lat-
ter expressions specify the maximum distance and capacity levels for each
vehicle k. The different running costs of the vehicles can be accounted for
by replacing c~ with c~k in the objective function. We note that some other
assumptions, such as time windows or hub level routing, may be much more
difficult to incorporate.
R e d u c t i o n s . In order to stress the relationship of the M M L R P to other
distribution problems, we shall show below how the ILP formulation pre-
sented in subsection (4.3) can be reduced into ILPs for a number of known
problems in distribution management. These reduced versions will of course
not be the simplest ones available for these problems. We note that for
all problems where customers do not send goods, expressions of the form
~ i e u qij are replaced by q, the demand of customer j.
The basic location-routing problem can be formulated by adding the set
of equations zij = 0 (i, j E 7t) and deleting constraints (4.3.), (4.4), (4.5.)
and (4.13.). Two alternative formulations for this problem can be found in
Nagy (1996, p.56), and Nagy and Salhi (1996b).
We may find a formulation for the problem of locating terminals by
adding the following sets of equations: vi = 0 (i E 7/) and xijk = xjik (i,j E
7 / U if, k E /~). The warehouse location problem can be formulated from
The Many-to-Many Location-Routing Problem 271

the above by adding constraints zij -- 0 (i, j E 7/) and deleting inequalities
(4.3.), (4.4.), (4.5) and (4.13.).
The m a n y 4 o - m a n y transportation problem can be formulated by adding
the set of equations Yi = 1 for all terminals i and the set yj : 0 for other
customers. This can be reduced to a formulation for the vehicle routing
problem with pickups and deliveries by adding equations zij = 0 (i, j E 7-0
and deleting constraits (4.3.), (4.4.), (4.5) and (4.12.). An alternative for-
mulation for this problem can be found in Nagy (1996, p.57). By also
setting vi = 0 for all i we arrive at a formulation for the basic vehicle
routing problem.

5 Solution Framework

5.1 Introduction

Location-routing problems, such as MMLRP, are usually solved in three


stages: location, routing and their inter-relation. There are many cases,
when the third stage is omitted or ignored. However, we know from the
literature that such an approach (called sequential methods) leads to subop-
timal solutions. The most frequent methodology which does take into ac-
count the interrelation between location and routing is the case of iterative
methods, where the first two stages are solved in an iterative framework,
with information flowing from one stage to another, and viceversa. The
main drawbacks of iterative methods is that they treat the first two stages
as if they were on the same footing. However, it can be observed that lo-
cation routing problems are mainly problems of location, with the routing
component taken into consideration. The above observation suggests an
appropriate solution method: an algorithm, where location is the master
problem and routing is a subproblem. This method is referred to as nested
method, because the routing stage is embedded into the location stage, see
Nagy and Salhi (1996a). In the remainder of this section, we describe our
respective methodologies for the above three stages.

5.2 T e r m i n a l location

Noting the close relationship between the terminal and the basic location
problems, as asserted by Kara and Tansel (1996), we use a suitable modified
272 G. Nagy and S. Salhi

version of the a d d / d r o p / s h i f t heuristic of Kuehn and Hamburger (1963)


as our 'master program', see Nagy and Salhi (1996a). The new terminal
configuration is obtained by either a removal of one terminal, an addition
of one terminal, or a simultaneous addition and removal. This heuristic is
further enhanced by the use of an efficient implementation of tabu search.
In this approach the inter-hub routes are derived without additional effort
as they are uniquely determined by the set of terminals.

5.3 A c c e s s level routing

At this level, we employ a many-to-many extension of the integrated pick


up and delivery heuristic of Nagy and Salhi (1998). This algorithm is based
on the multi-depot vehicle routing heuristic of Salhi and Sari (1997) which
uses the concept of borderline customers to create flexibility in generat-
ing routes. The solutions are further enhanced by improvement heuristics
and reduction tests. These improvement methods are then extended and
adapted to combat infeasibilities arising from loading constraints. The
idea of accepting infeasible solutions with penalties attached and the Use
of strategic oscillation are introduced into this approach.

5.4 The interrelation of routing and location

In the neighbourhood search heuristic of the locational stage we need to use


information from the routing stage in two situations. When implementing
an actual move, we always rely on the full results of the underlying ac-
cess level routing (pickup and delivery) algorithm. However, due to a large
number of neighbourhoods which need to be investigated during the search
procedure, a quicker method of getting routing information is called for "
when investigating possible moves. One possibility is to create a particular
(geographical) area, or region, within which the routing solution is likely to
change and calculate changes in routing costs only within that region. This
approach is called REGIONAL, see Nagy and Salhi (1996a). Another ap-
proach is to extend this by approximating routing costs within such regions
using suitable route length estimation formulae, referred to as ESTIMATE,
see Nagy and Salhi (1996b). The latter approach has been found to be
considerably faster with a negligible loss in solution quality.
The Many-to-Many Location-Routing Problem 273

Sequential method Proposed method


Total cost 3439 3226
Radial cost 3818 4391
Number of terminals 10 8
Number of vehicles 38 34
Number of iterations 124 39
Total computing time (mins) 5.27 123
CPU time per iteration (secs) 2.6 189

Table 1: Summarising the results for the MMLRP

5.5 I l l u s t r a t i v e results

In order to illustrate the usefulness of the proposed method we constructed


an M M L R P instance with 249 customers. Fixed terminal costs and the
inter-terminal routing costs were determined so as to provide about 10 ter-
minals. In order to show the superiority of this method to the sequential
approach of solving the locational and the routing parts separately, the
combined M M L R P method was initiated with the outcome of the sequen-
tial one. The algorithm was stopped after two hours of running time. the
results are tabulated in Table 1. The above results indicate that it is - just
like for the location-routing problem - better to use combined methods than
the sequential approach when locating terminals with less-than-truckload
transportation taking place at access level. However, further testing with
other examples and different versions would be desirable, including a con-
sistency and robustness analysis, see Salhi and Nagy (1998). Though the
above proposed method is relatively slow, it is possible to modify it so as
to attain a running speed comparable to that of the sequential one.

6 Conclusion and Suggestions

In this work, we have investigated the many-to-many location-routing prob-


lem. An integer linear programming formulation for this problem was given.
An example problem instance was solved for illustration.
We believe that the many-to-many location-routing problem and its
sub-problems are both interesting and challenging areas of research, due to
274 G. Nagy and S. Salhi

their practical and theoretical importance, and we hope that further results
will support and reinforce our view.

Acknowledgements

We would like to thank Dr Kathrin Klamroth from the University of Kaisers-


lautern for her helpful suggestions. We are also grateful to the referees for
their comments and suggestions. Dr Nagy is grateful to the School of
Mathematics and Statistics of the University of Birmingham and to the
C o m m i t t e e of Vice Chancellors and Principals for financial support.

References

Balakrishnan A., J.E.Ward and R.T.Wong (1987): Integrated facility location and
vehicle routing models: recent work and future prospects, American Journal
of Mathematical and Management Sciences, Vol.7.,pp.35-61.
Campbell J.F. (1994): A survey of network hub location, Studies in Locational
Analysis, Issue 6, pp.31-49.
Daganzo C.F. (1996): Logistics Systems Analysis, Springer, Berlin.
Eilon S., C.D.T.Watson-Gandy and N.Christofides (1971): Distribution Manage-
ment: Mathematical Modelling and Practical Analysis, Griffin, London.
Kara B.Y. and B.C.Tansel (1996): On the equivalence of the allocation part of
the hub location and the multimedian location problems, paper presented at
the E WGLA '9 meeting, Birmingham.
Kuehn A.A. and M.J.Hamburger (1963): A heuristic program for locating ware-
houses, Management Science, Vol.9., pp.643-666.
Laporte G. (1989): A survey of algorithms for location-routing problems, lnvesti-
gacion Operativa, Vol.1, pp.93-123.
Laporte G. (1992): The vehicle routing problem: an overview of exact and ap-
proximate algorithms, it European Journal of Operational Research, Vol. 59,
pp.345-358.
Min H-K (1989): The multiple vehicle routing problem with simultaneous delivery
and pick-up points, Transportation Research A, Vol.23A, pp.377-386.
Mirchandani P.B. and R.L.Francis (1990): Discrete Location Theory, Wiley, New
York.
The Many-to-Many Location-Routin 9 Problem 275

Nagy G. (1996): Heuristic Methods for the Many-to-Many Location-Routing Prob-


lem, PhD thesis, University of Birmingham.
Nagy G. and S.Saihi (1996a): Nested heuristic methods for the location-routeing
problem, Journal of the Operational Research Society, Vol.47., pp.l166-1174.
Nagy G. and S.Salhi (1996b): A nested location-routing heuristic using route
length estimation, Studies in Locational Analysis, Issue 10, pp.109-127.
Nagy G. and S.Salhi (1996e): The multi-depot vehicle routing problem with pick-
ups and deliveries, paper presented at IFORS'96, Vancouver.
Salhi S. and M.Fraser (1996): An integrated heuristic approach for the combined
location-vehicle fleet mix problems, Studies in Locational Analysis, Vol. 8,
pp.3-22.
Salhi S. and G.Nagy (1998): Consistency and robustness in location-routing, Stud-
ies in Locatmnal Analysis, Issue 13. (in press).
Salhi S. and G.K.Rand (1989): The effect of ignoring routes when locating depots,
European Journal of Operational Research, Vol.39., pp.150-156.
Salhi S. and M.Sari (1997): Models for the multi-depot vehicle fleet mix problem,
European Journal of Operational Research, Vol. 103., pp.95-112.

Anda mungkin juga menyukai