You are on page 1of 14

Map-Route: A GIS-Based Decision Support System for Intra-City Vehicle Routing with Time

Windows
Author(s): G. Ioannou, M. N. Kritikos and G. P. Prastacos
Source: The Journal of the Operational Research Society, Vol. 53, No. 8 (Aug., 2002), pp. 842-854
Published by: Palgrave Macmillan Journals on behalf of the Operational Research Society
Stable URL: http://www.jstor.org/stable/822912
Accessed: 18-05-2015 03:33 UTC
REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/822912?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

Palgrave Macmillan Journals and Operational Research Society are collaborating with JSTOR to digitize, preserve and extend
access to The Journal of the Operational Research Society.

http://www.jstor.org

This content downloaded from 202.43.95.117 on Mon, 18 May 2015 03:33:35 UTC
All use subject to JSTOR Terms and Conditions

Journal of the Operational

Research

Society

(2002) 53, 842-854

? 2002 Operational Research Society Ltd. All rights reserved. 0160-5682/02

$15.00

www.palgrave-journals.com/jors

Map-Route: a GIS-based decision support system


for intra-city vehicle routing with time windows
G Ioannou*, MIN Kritikos and GP Prastacos
Athens Universityof Economics and Business, Athens, Greece
This paperpresentsa Decision SupportSystem (DSS) that enables dispatchers-schedulersto approachintra-cityvehicle
routing problemswith time windows interactively,using appropriatecomputationalmethods and exploiting a custom
knowledge base that contains informationabout trafficand spatialdata. The DSS, named Map-Route,generatesroutes
that satisfy time and vehicle capacityconstraints.Its computationalengine is based on an effective heuristicmethod for
solving the underlying optimization problem, while its implementation is developed using MapInfo, a popular
GeographicalInformationSystem (GIS) platform. Map-Route provides very efficient solutions, is particularlyuserfriendly,and can reachanswersfor a wide varietyof 'what if' scenarioswith potentiallysignificantcost implications.We
have implementedMap-Routein an actualindustrialenvironmentand we reporton the experiencegained fromthis reallife application.
Journal of the OperationalResearch Society (2002) 53, 842-854. doi:l0. 1057/palgravejors.2601375
Keywords: vehicle routing;distribution/logistics;decision supportsystems; heuristics;GIS

The majorityof the systems to-date have been of help to


enterprises;however for most of them a number of drawThe vehicle routingproblemwith time windows (VRPTW)
backs have been reported:(a) they are quite expensive, thus
arises in a variety of pick-up and delivery applicationsand
not preferredsolutions for Small and Medium Enterprises
can be described as the design of optimal delivery/
(SMEs)-the majority of users; (b) they are based on
collection routes from one or several depots to a number proprietarysoftware as opposed to popular GIS platforms
of customers,within a pre-specifiedtime window, at miniwith standardizeduser-interfaces,effective personnel trainmum cost. Many papersin the literaturehave addressedthe
ing, and guaranteedmaintenanceand system upgrades;and
VRPTW problem, and substantialresearcheffort has been
(c) they need to incorporatemore realisticassumptions,and
devoted in developing efficient algorithms for solving a
to improve solutions graphically.As a result, distribution,
variety of VRPTW problems.We refer to Bodin,l Laporte2 pick-up and delivery SMEs still need suitable tools for
and Gendreauet al3 for surveys of the VR literatureand
supportingcomplex decisions related to route planning, in
appropriatepointersto relevantresearchefforts.
order to provide high level service to their customers and
Since the mid-1980s, significantworkhas been performed optimize their resources.
in developing computerized routing software systems.4
The objectiveof this paperis threefold:first,to propose a
Examples are: Geo-route,5Fleet-Manager,6micro-ALTO,7 frameworkfor addressingVRPTWfor intra-citynetworksin
Greentrip Toolkit,8 MACS-VRPTW,9 Dynamic Route
a user-friendlyand effective mannervia efficient solutions
Guidance,'0 and DRIVE." Apart from general and/or
of the underlyingoptimizationproblem. Second, to develop
dynamicVRPTWsoftware,therehave been severalindustry a prototypeDSS based on: (i) a popularGIS platform;and
specific approachessuch as the ones summarizedin Camp(ii) an optimizationmethod coupled with a knowledgebase.
bell and Langevin12for roadwaysnow and ice control, and
And third,to demonstratethe applicabilityof the approach
Road-net,Truck stops, and Micro Vehicle Plan, in the soft
through the results obtained from the implementationof
drink industry.13Commercialsoftware is also availablefor
the DSS to an actual industrialenvironment.The proposed
various applications(see eg, http://www.geocities.com and
DSS can assist logistics operationsin a numberof ways, eg:
http://www.paragon-software.co.uk).
(a) enhance daily operationaltasks of dispatchers-schedulers; (b) provideflexibilityin solving VRPTWby generating
alternativesolutions and reformulatingproblem conditions
*Correspondence.G Ioannou, ManagementSciences Laboratory,Graduate Programin Decision Sciences, Departmentof ManagementScience and
(eg, editingthe underlyingtransportationnetwork,addingor
Technology,Athens Universityof Economics and Business, 8th Floor, 47A
removing customers, defining alternative scenarios etc.),
EvelpidonStreet and 33 LejkadosStreet,Athens 113-62, Greece.
while keeping a good 'eye' on the geographicalreality of
E-mail: ioannougaueb.gr

Introduction

This content downloaded from 202.43.95.117 on Mon, 18 May 2015 03:33:35 UTC
All use subject to JSTOR Terms and Conditions

Gloannou
etal-Map-Route:
aGIS*based
decision
support
system843

the problem;and (c) offer interactivity,ie, allowing users to


employvisual techniquesto formulate-reformulate
problems
and derive solutions that can be easily implemented.
The remainderof the paperis organizedas follows: first,
we presentthe architectureof our Decision SupportSystem
(DSS) and its constituent elements. Then we provide the
overallframeworkof Map-Routeand identifythe interaction
between all of its components.The reporton an industrial
application follows, and the conclusions of our work are
finally presented.
Map-Route architecture
Map-Routeis specifically designed for vehicle fleet routing
for deliveries within a compact large city street network,
ratherthan general VRPTW It consists of four basic components, which are presentedin detail below.
Databases of Map-Route
Map-Route'sspatial databaseincludes a digitized map with
all relevantlocations (depot and customers)and underlying
network (streets, roads, intersections, etc.). The customer
databaseincludes, for each customer,the node identification
number,the demand,and the time window restrictionsand
service time. The nodes databaseincludes, for each node,
the identificationnumberand coordinates.Finally,the street
database includes, for each street segment, name, length,
and address ranges for both sides. The data files of MapRoute can be changeableor permanent.The formerrelateto
the propertiesof the underlyingtransportationnetworkand
the coordinatesof the depot locationnode. The lattermay be
modifiedby the schedulerthroughour DSS by, eg, inputting
a new scenario via tables, entering new customers into a
given problem,removing customersfrom a scenario,inputting a new scenariousing the map underconsideration,and
enteringdatarelatedto the problemusing the browsertable.
Computationalengine of Map-Route
The computationalengine of Map-Route can include any
heuristicor mathematicalprogrammingmethod for solving
the VRPTW The selected method is very importantsince it
determinesthe applicabilityof the solution scheme in reallife situations.The key factorfor the appropriateselection is
the efficiency of the methodand its abilityto providein very
short times high-qualitysolutions. In our approach,we use
IMPACT,14 the basic steps of which are:

The algorithm terminates by providing the number of


routes (equal to active vehicles), the customers that are
assigned to each vehicle, the sequence in which customers
are visited, and the total time-distance-cost of the solution.
IMPACTis very efficient and providesresultscomparableto
meta-heuristicsat a fractionof the computationaleffort.For
a detailed description of IMPACTand the computational
tests that supportits effectiveness, the readeris referredto
Ioannou et al,14 while Table 1 provides a comparison of
IMPACTwith severalheuristics(I-1, PARIS,HE) and metaheuristic (GRASP, Tabu search-TABU-A and RTS, and
genetic algorithms-GENEROUS-20) methods to illustrate
its efficiency ('*' indicates that IMPACToutperformsother
methodsfor the classical datasets RI, Cl, RC1, R2, C2, and
RC2 of Solomon15).

User-interfaceof Map-Route

Algorithm IMPACT
Step 0:

the earliest and latest service times (time


window) for each customer.
Step 1: Select a 'seed' customerto starta route, finding
the farthestcustomerfrom the depot. If there is
no non-routedfeasible customerto starta route,
go to Step 6.
Step 2: Find the feasible non-routed customer u that
minimizes a composite criterion Impact(u),
which includes functions of the relationship
between the arrivaltime to customer u and the
lower bound on the service time of u, the impact
of customer'su insertion on non-routedcustomers, and the impactof customer'su insertionon
customersalreadyroutedwithin the route under
construction.The searchprocedureis as follows:
Step 2a: Examineall possible feasible insertionsof customer u into the currentroute. For each feasible
insertion, calculate the criterion function
Impact(u). Select the insertion location that
results in minimumImpact(u)for this customer.
Step 2b: Repeat Step 2a for all feasible non-routed
customers.
Step 2c: Select customeru with minimumImpact(u).
Step 3.
Insertthe selected customeru, to the best insertion location on the currentroute (see Steps 2a
and 2c). Update the route and set u as a routed
customer.
Step 4: If there are non-routedcustomersthat are feasible for insertioninto the currentroute, returnto
Step 2; otherwiseproceed to Step 5.
Step 5.
If all customershave been scheduled,terminate.
Otherwise,go to Step 1 initiatingnew route.

Initialization.Read the numberof customers,the


vehicle capacity,the inter-customerand depotcustomerdistances (or times, routing costs) and

Map-Route'suser-interfacehas been developed using the


Map-Basic programminglanguage and is based on pulldown menus to provide functionality related to solution
methods, problem initialization(eg, defining the speed of

This content downloaded from 202.43.95.117 on Mon, 18 May 2015 03:33:35 UTC
All use subject to JSTOR Terms and Conditions

844 Journal
oftheOperational
Research
Society
Vol.53,No.8

Table 1 Comparisonbetween literatureheuristicsand new heuristicon averagenumberof routes


Data set
R1
R2
C1

C2
RC1
RC2

I-I

PARIS

HE

GRASP

TABU-A

RTS

GENEROUS-20

--*

*
*

*
*

*
*

_
*

vehicles), data input, formationof local networks,design of


accurate vehicle routes, etc. Furthermore,it allows the
display of spatialmaps allowing the user to zoom on a part
of the map and evaluatethe suggested solution or generate
alternative solutions using logical inference. The user
manualsof MapInfo16 provideall the necessaryfunctionality
information.
Knowledge base of Map-Route
Deriving solutions that follow actual road networks and
complying to trafficpatternsthat favour main city arteries,
large streets and roads with light traffic and open structure
is a very difficult task, especially when optimizationapproaches are employed. The latter are very sensitive to the
route segments that characterizethe underlyingroad-street
networkand cannot handle logical attributessuch as those
described above. The solutions they produce comprise
segments that may not be feasible in actual route planning,
or may not be preferableto drivers. The above problems
may be alleviated either by direct user interaction, ie,
changes in the route structurethat are performedmanually
by experienced users, or by appropriateknowledge bases
that capturethe logic in which such changes are made. The
first approachrequireshigh level of user involvementin the
solution process and is very time consuming for large
problems, since the user has to examine all parts of the
network and make the necessary changes. The latter
approachrequiresa strong set-up phase in which roads are
coupled with specific prioritizationattributesand routes are
examined using a knowledge base containing all these
attributesor additionalinformationconcerningpreferences.
The structure of the knowledge base we propose is
simple: Rules and attributesare assigned to road segments
and inference logic is designed in order to transforman
optimisation solution into a feasible-preferable solution
with minor cost implications. The rules can have, eg, the
following forms relevant to street characteristics,timerelatedtrafficand date peculiarities,respectively:
'IF MULTILINEx IS {main,regular,narrow) streetthen
label = {PREFERRED,NONE, LOW)'
'IF MULTILINEx IS traffic loaded at time t then
label = LOW'

'IF MULTILINEx IS non-preferable on date then


label = LOW'

Labels characterizethe priority for using a particular


route segment (ie, segments with priority LOW will be
used only when necessary to guaranteethe connectivity of
a sub-network).The rules are exhaustivefor all appropriate
road segments and time- and date-relatedinformation,and
areimplementedin conjunctionwith the Map-Basicroutines
using the experience of drivers and planners-schedulers.
The priorities associated with road segments are directly
used when forming a transportationnetwork over which
vehicles are to be routed.17
Solution framework
The proposedsolution frameworkfollows a typical four-step
process: (a) solve an approximationof the VRTPW using
Euclidean distances; (b) break the region down into subnetworks,each correspondingto a vehicle route,andgenerate
a travel path over each sub-networkusing shortest paths
between customers,while preservingthe orderof customers
in routes;(c) modifythe solutionvia the knowledgebase rules
to betterapproachthe actualvehicle paths;and (d) perform
manual modifications, if necessary. The solution process
iteratesamongthese steps until a 'good' solutionis obtained.
Subsequentlythe user acceptsthe solution or modifies some
of its attributesin orderto providethe finalset of routes.The
usercan also modifyproblemparametersandreapplythe four
steps if the system proposed solution is not satisfactory.
Visualizationcan play a criticalrole in this process, and the
GIS platformis very helpful in this direction.
Figure 1 providesan overview of Map-Route'sinfrastructureand organization.MapInfois the centreof the approach.
Databases include digitized maps and data concerning
customers and depot. The knowledge base is represented
as a databasethat is externalbut connectedto MapInfovia
an appropriateAPI. The user can interactwith the spatial
data and also provide adjustnents graphicallyto the solutions provided by the computationalengine. In Figure 1,
lines connect sequentiallyevoked components,while arrows
provide the directionof each sequence.
Map-Routephases
As mentionedbefore, Map-Routeinvolves four interacting
phases. During the first phase, the VRPTW is solved for a
networkwhere the customers and the depot are connected

This content downloaded from 202.43.95.117 on Mon, 18 May 2015 03:33:35 UTC
All use subject to JSTOR Terms and Conditions

a GISbaseddecision
etal-Map-Route:
Gloannou
system 845
support

Algorithms
(Computational
Engine)

7\

g zeMp10

-4

t APNO

Figure 1

/ Pick-u
D~~~~~~~~~~~~~epot

Deie

The architecture of Map-Route.

with straightlines. This enablesthe solutionof problemswith


a large numberof customers,in short computationaltimes.
StandardMapInfo tools estimate the Euclidean distances
between customersand between customersand the depot.
During Phase 2, for every route of Phase 1, a connected
sub-networkis constructedby selecting areas(roadsections)
adjacentto the Euclideanroutes. Sub-networksare adjusted
throughMapInfotools that use proximitycriteriafor actual
route segment inclusion. Furthermore,distances are reevaluated based on the priorityrules of the knowledge base. On
the new networks, accurate routes can be determined
accordingto the following steps:
Step 1: Renumberthe nodes of the selected sub-network
(assigning '1' to the depot)
Step 2: Run the Floyd's Shortest Path algorithm'8
between customerlocations (stops) and depot, in
the same order as in the solution of Phase 1
Step 1: Displaythe finalaccuraterouteon the originalmap
The above steps can be repeated for each sub-network,
leading to accurate routes that incorporate actual road
segments. Note that in Phase 2, the distance between
customer locations is increased, since the multi-lines of
Phase 2 replace the straightlines of Phase 1, and violations
of customers'time windows may occur.This problemcan be
addressedby: (a) selecting sub-networksmore adjacentto
'specific' trips;and (b) tighteningtime windows of specific
customersand iteratingthe whole process of the two phases.
Both approaches have been examined, and the results
showed that multiple iterationswith tighter time windows
are preferable.19
It is importantto note that the two-step approach of
determiningEuclidean routes and transformingthem into
actual street segment-routesmay not be necessary or even

efficient for general VRPTW, especially in the case of large


inter-city routing with significant obstacles and constraints,
where severe problems may arise. Nevertheless, for a
compact intra-city network such as the one we are handling
via Map-Route, the approach can smoothly work.
In Phase 3 the knowledge base rules are evoked and the
solution is transformed to approach better the road segments
employed by the vehicles. This is accomplished by feeding
the solution to the knowledge base via the MapInfo interfaces. When Phase 3 is completed, the actual road network
is determined based on proximity criteria and preferences
residing within the rule constructs.
Finally, Phase 4 is a pure user-driven phase. The user
interacts with MapInfo via the solution of Phase 3 and
provides final adjustments necessary to derive the schedule
of each vehicle. Figure 2 illustrates the four-phase approach
inherent in the Map-Route logic. Note that this is a discrete
and time-including representation of the Map-Route architecture of Figure 1. The elements of each Phase are grouped
in shaded boxes, while the sequence of the approach is
depicted through the directional arcs that connect databases,
applications, results and user adjustments.
Before we proceed to the implementation aspects of MapRoute, we should mention that the optimal solution to intracity routing problems that we consider in this paper might
include more than one daily trip per vehicle. However, the
common policy in all distribution companies we have
interacted with was to load only once the vehicles at the
warehouse and perform all remaining activities the rest of
the day. Furthermore, reaching near-truck load per vehicle
was a key performance indicator. Thus, we did not proceed
in exploiting this potential cost-saving application and
constrained our DSS into single daily loading and single
routes per day per vehicle. Nevertheless, an extension to

This content downloaded from 202.43.95.117 on Mon, 18 May 2015 03:33:35 UTC
All use subject to JSTOR Terms and Conditions

Vol.53,No.8
oftheOperational
Research
846 Journal
Society
: Eg: ; iS)LE
; yi i :: i y:: EL:E
g- yiiLE;iLLi
::

Problem

0..::1X000
.

i...L .........
E:;;

0040-i00ll
QEEi 24iDEdQj
i ERiED
diR[iTiE
iS i -iEEi dEEii

7 ...

X10.

S};000144XXa

l--0000-000000000000000000-;0
QiSEi
i-iEES
E
EA d-253!

f. i MA.INFOf.Sf00tEfff

:j.:fLiff4Jj

t0000St000tt--0000000000tti-00f000000000ftPUser40:EW
d
s tm e n t

ju
0000StS1:X0-S-;::.:00000000A10X,fSjt400

ES000041001;S4.......S:XfS

40E4W

Et..iiE.

tEEttt:;i,;.ST
i.itSN4X!.;;;;;.
tidjf;;
. .
i i;;i;i;L. LLiLa;;;itiz
e d ,..ap;LiLL0;st.L
g.f;

_;,- -

CT:
---IMPi:::XX-XXA
-- j ::i:;e;XX-

::-

TE

R -iE

i~~~~~~~~~~~~~~~~~~~~~~~~~~~

iS;-.9iE- ..
i;
9:D
glD:
E
i-.-iSTi.9:

..

XXESS

rHeuristica us:::104i000 40:00 0d 1 0X0: ' ;X0t:;---40000tV;f;X-F100000000:00f00410


04000t
i F l
SC111

.Eif.

y
tlltt00-0

St00tttl-000iTt000

aTf
S ho

est7.0

d . t.

iLLL,iiESLLSS

iAf,|;ltlt;-fk0-0

-,S,,jttiSl0000
......Results...
f-0.-tk--00-0.000-.00
0--0.

_~~~~~~~~~~~~~~~~~~P
MAPINFOhm
t

.S

. .. .

..

Maps

0~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Foy'sSortest
netw

SolHurition
.giiii

Digitized

basesut

tE

2 Th
::~~~~~~~~~~~~~~Fgr

Map;$-Route: solutio

Map-Route is possible through reduced time available per


day, a factorthat could be interactivelymodified within the
DSS to producevarious solutions.
Map-Routeimplementation
Map-Routehas been implementedon a Pentium PC. The
core of DSS has been written in Map-Basic, and the
algorithmsfor the VRPTW in Fortran,appropriatelyintegrated into MapInfo. The knowledge base uses a Lisp
inference engine and is also integratedwith MapInfo. The
solution providedby Map-Routeis depicted on a real city
map, and generatesinformativeoutputfor the vehicle driver,
while enabling the evaluation of alternativeroutes. It is
importantto restatethe significanceof user involvementin
the solution procedure. No matter how, extensive and
complete the knowledge base is, or effective and comprehensive the optimizationsolution is, the final set of routes is
eitheracceptedor modifiedby the scheduler-planner,whose
experience and flexibility in dynamic daily adjustmentsof
the problemparametersis irreplaceable.
Industrial application
Background
The company for which Map-Route was developed is a
wholesalerand logistics service providerthat suppliesmultiple packaged goods and beverages to a large number of

flgow.tm

local small supermarketsand other small retail outlets


throughoutthe CentralAthens area, in Athens, Greece on
a daily basis. The company operates its own small-vehicle
fleet from a central warehouse located at Pireus (noted as
PIRAIVS at the map of MapInfo provided later in this
section) Street, a main street connecting Pireus to Omonia
Square in the centre of Athens. The company owns 26
delivery vehicles, which are operated by certified drivers.
The overall fleet size though, necessary for satisfying all
customerswas approximately35 vehicles, before the implementationof our DSS; thus, the companyemployedvehicles
owned by individuals on a need-basis, a fact the created
additionalcosts and resultedin severeproblemswith respect
to quality of customer service and adherence to order
fulfilmentgoals.
The numberof customersvaries from435 to 680, depending on the day of the week and the periodof the year (higher
numberof demandpoints duringthe summerseason, when
additionalpoints of sale are open to service the large tourist
populationthat visits Athens). The customersare dispersed
throughoutcentralAthens, and are located at main arteries
of the city (eg, Panepistimiouor StadiouStreet)as well as at
small streets near the archaeologicalsites and particularly
vibrantor densely populatedcity neighbourhoods.Figure 3
illustratesthe distributionof the customerset on the map of
Athens (circles), and the depot (square).
Since a just-in-timeapproachis promotedas the competitive advantageof the company,the replenishmentof goods
follows daily orders from each customer;these ordersmay

This content downloaded from 202.43.95.117 on Mon, 18 May 2015 03:33:35 UTC
All use subject to JSTOR Terms and Conditions

a GIS-based
decision
support
system 847
etal-Map-Route:
Gloannou

Figure 3

Customer distribution and depot location on the map of Athens.

be zero for some Stock Keeping Units (SKU) in a day.


Nevertheless,the overalldemandis relativelyconstant,apart
from peak seasons, and especiallyduringthe summerwhere
beverage consumption increases significantly. The daily
iterativeoperationsstartwith customerorders,whicharefinalized every evening, and can be satisfied by the inventory
held at the warehouse. Inventoryavailabilityis guaranteed
by the largesafetystockheldforeachSKU.The customersare
geographicallydispersedwithina distanceradiusthatallows
for demandto be satisfiedthroughdaily deliveries,as shown
in Figure 3. In addition,the time intervalduringwhich the
deliveryhas to takeplace (timewindow)is also known(fixed
for each customeraccordingto a contract).
The deliveryprocessis performedas follows. Productsare
loaded on vehicles at warehousedocks up to (or sometimes
below, accordingto customerrequests)capacityandthey are
transportedto the customers' locations. At each location,
quantitiesthat equal customer demand for each SKU are
unloaded, and paper work (shipping documents, bills and
invoices) is filled and exchanged;this takes approximately
5 min. Then, vehicles travelto subsequentcustomerswhere
the process is repeated, until all deliveries have been
performedand returnto the depot for the following daily
cycle. It is important to note that before Map-Route's
implementation,the sequence in which a vehicle visited
customerswas not determinedwhen loading at the depot;
driversresponsiblefor a particulararea-customerset were

makingsequencingdecisions.This had a significanteffect of


the complianceto time windows,and affectedcost, customer
satisfactionand qualityof service.
Map-Routeset-up
To generatethe problemwithin MapInfo,we have started
with appropriatemaps of the Central Athens area, and
created 5137 node-objects for the 8231 road segments of
the underlyingmap thatmodel approximately3000 different
streets and covering almost 500 km of road network,using
the configurationtools of MapInfo. This initializationis
required for any subsequent task. Figure 4 provides a
zoomed view aroundthe depot of the road networkmodelled in MapInfofor the application.To input the customer
locationcoordinatesand the dataconcerningtime windows,
we used appropriatefiles for data entry into MapInfo.
Furthermore,we have created special MapInfo screens to
be availableto the plannersfor adjustments,deletions and
additionsof customersand time windows. Figure5 provides
a sample screenthatwas constructedto give multiplepoints
of access to the user (both graphicaland in tabularform).
The road segments (black lines in Figure 3) were characterized as 'preferred', 'unacceptable'or 'non-labelled',
and this informationwas included in the knowledge base.
Filling up the knowledgebase with rules and prioritieswas
the most dauntingand time-consumingtask of the system

This content downloaded from 202.43.95.117 on Mon, 18 May 2015 03:33:35 UTC
All use subject to JSTOR Terms and Conditions

848 Journal
oftheOperational
Vol.
Research
53,No.8
Society

Figure 4 The zoomed map aroundthe depot.

YA~

~~
1844'

189
1841

~~~~~~
~~~~~~~~~~~~~~~~~~~~1642
2g1

Lo

23.73225

37.O70081

23.733177

37.972683

23.732472

23,730Q8M ,

0 00. 00

'23:50

0.0

35

37.973121

0 00XJ0

2 5

3.

0 '010

23

~~Figre-5Dat - ma ipuaion------screen. -----------

This content downloaded from 202.43.95.117 on Mon, 18 May 2015 03:33:35 UTC
All use subject to JSTOR Terms and Conditions

Gloannou
etal-Map-Route:
a GIS*based
decision
support
system 849

set-up process. We used data from the Greek Ministry of


Transportationand Communications concerning traffic
patterns and time-info for various dates of the year and
times of the day. Furthermore,we interviewedthe driversof
the company for routing preferences and considered their
answersfor labellingroad segments.Driverswere requested
to providethe most commonlyused streetsand estimatesof
travellingtimes at these segments duringpeak and off-peak
hours.Finally,for each 'unacceptable'road segment,we run
a special MapInfo procedure to derive via proximity
measures a 'preferable'correspondingroad segment, and
includedit in the knowledgebase. It is importantto note that
the experienceof driversconflictedsome times with official
data;howeverthe company'smanagementinsistedon adhering to drivers preferences, as more reliable information
concerningactual routing paths. Apart from the initial setup of the knowledge base, we have providedscreens within
MapInfo,which allow users to adjustthe labels accordingto
new realities,as the system life cycle evolves, or on a daily
basis, in line with expectationsconcerningcongestion, road
blocking (strikes and marches in the centre of Athens is
commonplace),etc.
The computationalengine of Map-Route, ie, IMPACT,
was integrated in the MapInfo menu. For the particular
instances in the industrialcase, IMPACTrequiredless than
30 s to terminate(for the largerexamples of more than 600
customers). We have also incorporatedFloyd's algorithm
within the DSS. Floyd'simplementationuses dynamictables
in order to provide fast the optimal solutions; for the
particularinstances in the industrial case, the algorithm
took less than 40s to terminate,even in cases where the
sub-networkincludeda largenumberof node-objectsdue to
alternativeroute segments induced by the knowledge base.
The DSS in operation
At the startof a shift, customerdemandis alreadyinto the
system and Phase 1 of Map-Routeis initiated.The result is
the sequence of customersvisited by each vehicle based on
Euclidean distances, which are automaticallycalculatedby
MapInfo. Euclideanroutes appearon the screen with lines
connecting customers and depot; such a screen from the
actualapplicationis providedin Figure6. Note that the user
can make adjustmentsto time windows, demandand customer attributes (existence, location, etc.) before running
IMPACT,if necessary, through appropriateselections in
the MapInfo menu (that invoke the previously discussed
screens).
Subsequently,the user proceeds to the second phase of
Map-Routeto determinethe actualroadpathof each vehicle
using the shortest paths on the real road network. The
procedure is repeated for each vehicle and is as follows:
An initialsub-networkis formedthroughthe knowledgebase
rules of proximity;this sub-network,which includesvarious
road segments, is expandedor adjustedby the user that can

include additionalsegmentsor removesome segmentsbased


on experience and daily data. Given the sub-network,a
routine incorporatedin MapInfo produces the necessary
shortestpathmatrix.Figure7 presentsa samplesub-network
associatedwith one routeof the Euclideansolution of Phase
1 presentedin Figure6 (includesall routesegmentsdepicted
by the thick lines).
Given the distancematrix,the next step is to applyFloyd's
algorithmto determinethe actualvehicle pathsby invokinga
resident MapInforoutine for calculatingshortestpaths and
displaying the results on the MapInfo interface. Figure 8
providesthe actualroadpathfor the sub-networkof Figure7,
and Figure 9 a zoomed view. The procedureis repeatedfor
each initialEuclideanrouteof Phase 1. This loop constitutes
the most time consumingpart of the application,since it is
directly linked to the numberof vehicles employed. If this
numberremainsat the presentlevel (ie, orderof 30 vehicles),
then it is possible to complete the routing procedureand
deriveschedulesfor each vehicle in less than 1h. This time is
acceptable,and allows the companyto smoothlyemploy the
DSS. However,futureplans include the additionof several
more SKUs and customers-locations,a fact that would add
furtherdelay to the application. Thus, we were asked to
automate a combined Phase 1-2 of Map-Route. This
providedfull solutions(actualroadnetworksfor all vehicles)
thatcould be furtherexaminedand improveda posteriori by
the users, if necessary.The automatedprocedureallowedthe
completion of the daily tasks in less than 0.5 h. However,
planners who desired their direct interventionand drivers
who felt thattheirflexibilitywas compromiseddid not deem
the total automation appropriate.Thus, the operational
version of Map-Routeruns with individualroute construction and adjustments,and performs the routing procedure
sequentiallyfor each vehicle.
The resultsof Map-Routeareprovidedto the drivers,who
are requested to follow the prescribed routes (customer
sequences and road segments) in their daily delivery schedule. Appropriateforms that include the sequence of customers to be visited and the expected time of arrivalat each
customer's location are employed to check the delivery
schedule; customers are required to sign the form when
paperworkis exchanged.
Results and discussion of the application
Map-Routewas deployed to two PCs of the company, located at the office of the warehouse. The DSS was standalone, ie, it was not connected to any other Information
Systems (eg, the warehouseand inventorymanagementsystems). Two planners,who were involved in all stages of the
finalization-deploymentof the DSS, were responsible for
dataentryand runningMap-Routeon a daily basis. The system was tested using demanddata from previoustime periods and coveredall peak seasons and severaltrafficloading
scenarios. The key result that impressed the company's

This content downloaded from 202.43.95.117 on Mon, 18 May 2015 03:33:35 UTC
All use subject to JSTOR Terms and Conditions

oftheOperational
Vol.53,No,8
850 Journal
Research
Society

Figure 6

Sample Euclidean routes.

Figure 7 A sub-networkfor a route.

This content downloaded from 202.43.95.117 on Mon, 18 May 2015 03:33:35 UTC
All use subject to JSTOR Terms and Conditions

a GIS-based
decision
support
system 851
Gloannou
etal-Map-Route:
_.!.

.~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~--

- --.:::

---

Figure 8 The accurateroute.

Figure 9 A zoomed view of the accurateroute.

This content downloaded from 202.43.95.117 on Mon, 18 May 2015 03:33:35 UTC
All use subject to JSTOR Terms and Conditions

Vol.
53,No.8
Society
oftheOperational
Research
852 Journal

managementwas the possibility of serving all customers Most of these add-onsarerelatedto reportsthathelp reviews
and managerialcontrol/decision making. All requestswere
even during the peak seasons under heavy traffic loading
optimizing
implemented using standardMapInfo tools and for each
fact
that
conditions. This was attributedto the
the
than
better
Map-Routecan now generatevariousreportsthat can
indeed
route,
was
stage
the routes at the planning
second
The
by the user interactively.Figure 10 providessuch
drivers.
the
by
be
accessed
followed
intuitive schedules
includes the components of all routes with
company's
that
the
a
report
with
was
in-line
that
aspect of Map-Route
times at customerlocations, waiting times
the
arrival
and
to
respect
user-interface
the
of
expectationswas the simplicity
and demand,as well as all the names of
begins,
were
service
until
who
the
users,
'power'thatthe DSS left in the handsof
thatvehicles visit them. Thereare
sequence
in
the
streets
logistics
the
of
the
key decision-makersin the daily operations
can generate, based on
MapInfo
that
reports
such
who
several
the
drivers,
from
came
plan. The thirdpositive reaction
alter the problem
solution,
the
modify
can
the
user
the
which
to
follow
week
one
of
were asked for a trial period
or return
2
of
and
Map-Route,
1
Phases
and
rerun
to
parameters
all
able
were
They
schedules produced by Map-Route.
produced
segments
road
of
set
the
reconfigure
2
and
to
Phase
customers
the
served
finishtheirdeliveryrouteson-time and
within the contractualtime windows. Thus, even the drivers by the knowledge base in orderto derive bettersolutions.
A finalpoint that shouldbe noted relatesto the distances'bought-in'the new application.
times
produced at the various phases of Map-Route. The
some
were
there
above,
views
Apart from the positive
between deliverypoints would generallybe differdistances
full-time
from
apart
that
drivers
some
negativecommentsby
comparing solutions between Phase 1 (straight
ent
when
to
used
that
vehicles
their
own
employment also owned
2 (accurateroutes), or Phase 3 (user adjustPhase
lines),
the
Nevertheless,
seasons.
peak
during
'rent'to the company
differencesmay cause violations of time winThese
ments).
overcompany
to
the
offered
DSS
obvious savings that the
which can be overcome by a combination
constraints,
dow
summary
a
2
Table
provides
came their negative reactions.
(a) select anothersub-networkadjarules:
the
following
of
details.
of Map-Routeimplementation
Phase
1; (b) move earlierthe starttime
of
route
the
cent
to
that
2
in
Table
presented
the
results
from
It is evident
the length of the time window
reduce
(c)
of
scheduling;
and
to
Map-Route,
is
deploy
minimal investment required
in the first phase; and
windows
time
dummy
through
the
for
Furthermore,
SMEs.
for
even
the cost is affordable
ones. Thus, the selected
into
smaller
the
problem
split
(d)
route
in
effective
resulted
particularapplication,Map-Route
in the feasibility of the
role
essential
an
plays
the
sub-network
of
fleet
the
of
existing
use
the
planning by allowing
solution.
The
season.
quality
even
during
peak
vehicles),
company(26
of the solutioncan be inferredby the significantreductionof
both violated time windows and lost sales; note that these
Conclusions
two percentagesare differentdue to the acceptanceof some
off-time window deliveries by several customers. Finally, In this paperwe developed a DSS for the VRPTW for intrauser training on MapInfo and the Map-Routecomponents city fleet planning.Map-Route,is based on the popularGIS
platformMapInfoinstead of proprietarysoftware,and is at
was straightforwardand was completed during the system
the right cost for SMEs. The computational engine is
the
from
were
involved
users
two
the
development (since
IMPACT,a heuristic that provides high quality results in
initial developmentstages). Unfortunately,we did not have
short computationaltimes, coupled with Floyd's shortest
access to commercial software in order to compare our
path algorithmand a knowledge base to allow for efficient
results.
transformationof the Euclideansolutions into effective trips
After the full-scale deployment of Map-Route, several
on the actual street network.The DSS is developed with a
for
were
implementation.
functionalities
requested
additional

Table 2

Summaryof DSS implementationresults

Status before Map-Routeimplementation

Parameter
Requirednumberof vehicles
Optimizedroutes
Violated time windows
Lost sales
IMPACTrunningtime (>600 customers)
Final route construction(time)
Cost of two MapInfouser licenses
Cost of digitized maps (city of Athens)
Softwaremaintenancecost (yearly)
Cost of solution algorithms
Training

Status after Map-Routeimplementation

35
No
20%
- 10%
Ad hoc (Driver)
-

This content downloaded from 202.43.95.117 on Mon, 18 May 2015 03:33:35 UTC
All use subject to JSTOR Terms and Conditions

26
Yes
<5%
<2%
<30 s
Near optimal (< 1 h)
- C 6000
- C 5000
12% of licences
C 0 (freeware)
- :C1000

decision
a GIS-based
Gloannou
etal-Map-Route:
support
system 853

E ?f*Am

21sb it4l ~dow grow$$YRP LOCA-ROUTES

~ ~~~IFROMLEFT

RREER6AI.IE

IERRA
ODSoO
IFRATEOR

ITOLEFT IF6RMR GHT1 TORGHT ILENGTH IL CROE91I SCODESI

II

~~~
~ ~~~
~~~~~17'

IF-ROFANTvIJ1
Z...

1200'i 1EY]6)MOLPIDON

1,

EYM0LPIDVN

IE80FANTVN
IERIFANTVN
IEROFANTVN

FigReA0

3,
3'

2
12'

10;
24

2:

6:
4

2
2'

10

10:

17

lo:

11,

apRotereor

user-centricphilosophy,bringingvehicle routingalgorithms
into the handsof planners-schedulers,whose vast experience
and 'common sense' can be the determinantsuccess factors
for the GIS-basedDSS implementationin realenvironments.
The experiences from the deploymentof Map-Routeto an
actual case in the Greek marketwere presentedto demonstratethe phases of the methodologyinherentin the DSS and
reveal several open issues that need to be handled on an
exception basis by the users and/or the knowledge base.
Throughthis case, the flexibility and ease of adaptationof
the DSS were also illustrated.
Via the four-phaseapproachofferedby Map-Route,a user
can easily find a scheduleas well as alternativescheduleson
intra-citytransportationnetworksfor VRPTW.The use of
visualizationalong with the availabilityof GIS can help users
in making improved decisions when solving real world
routing problems, which are everyday reality in logistic
operations,and become even more criticaldue to the expansion of third-partylogistics. Thus, developingand deploying
effective DSSs is a key prerequisite for the successful
operationof logistics groups. Furtherextensions of MapRoute include: (a) integratingmodern meta-heuristicsto
furtherimprovethe qualityof the final solutions;(b) enhancing the approachto handle inter-citynetworkswith additional constraints and route complications; and (c)
integratingthe DSS with warehousemanagementsystems
(eg, MANTIS)or EnterpriseResourcePlanningSystems(eg,

21

4~
12,
i2
6

51' AlIROITO AlIO


12A1100100
AIIOOIOO
76: 61100100

A11100100

38,A112O010_:41165100
38'A11O00_:A 4116I00
6501100100
11664..~!,1
58 ;RAl100100
7 . Al100100

L ELTA

11664
11664

1043
10435

I11164:
11864,

118660

11864

11864

11854:
I1104.

11864:

AlIIORIOT

11654:

11664'

41162100O
4 l1165100

REA

11664,

,
routes. 0 A 000 114:5 145
on3the actua

SAP or Oracle Apps) for seamless information technology


applications to distribution problems.
Acknowledgements-Theauthorswould like to thank the two anonymous
refereesfor their constructivecommentsand pointersto archivalliterature
that helped improvethe contentand the presentationof the paper.

References
1 Bodin L (1990). Twentyyearsof routingand scheduling.Opns
Res 38: 571-579.
2 LaporteG (1992). The vehicle routingproblem:an overviewof
exact and approximatealgorithms.EurJ Opl Res 59: 345-358.
3 GendreauM, LaporteG and Potvin J (1997). Vehicle routing:
modem heuristics. In: Aarts E and LenstraJK (eds). Local
Search in CombinatorialOptimisation.John Wiley & Sons:
Chichester,UK.
4 Keen PB (1998). Spatialdecision supportsystems for vehicle
routing.Dec SupportSyst 22: 65-71.
5 LapalmeG, RousseauJM, ChapleauS, CormierM, CossetteP
and Roy S (1992). Geo-route:a geographicinformationsystem
applications.CommunACM35: 80-88.
for transportation
6 Basnet C, Foulds L and IgbariaM (1996). Fleet-Manager:a
decision supportsystem for vehicle routmicrocomputer-based
ing. Dec SupportSyst 16: 195-207.
7 PotvinJY, LapalmeG and RousseauJM (1994). A microcomputer assistant for the development of vehicle routing and
schedulingheuristics.Dec SupportSyst 12: 41-56.
8 Concialini A and Hasle G (1997). The Greentriptoolkitsustainable transportationvia intelligent routing. http://
www.cs.strath.ac.uk/research/GreenTrip/ISATA97.doc

This content downloaded from 202.43.95.117 on Mon, 18 May 2015 03:33:35 UTC
All use subject to JSTOR Terms and Conditions

oftheOperational
Vol.53,No.8
Research
Society
854 Journal
9 Gambardella LC, Taillard E and Agazzi G (1999). MACSVRPTW: a multiple ant colony system for vehicle routing
problems with time windows. In: Come D, Dorigo M and
Glover F (eds). New Ideas in Optimization. McGraw-Hill:
London, UK, pp 63-76.
10 Wahle J, Annen 0, Schuster C, Neubert L and Schreckendberg
M (2001). A dynamic route guidance system based on real
traffic data. Eur J Opl Res 131-132: 74-80.
11 Savelsberg M and Sol M (1998). DRIVE: dynamic routing of
independent vehicles. Opns Res 46: 474-490.
12 Campbell JF and Langevin A (2000). Arc routing applications
for roadway snow and ice control. In: Dror M (ed). Arc Routing:
Theory, Solutions and Applications. Kluwer: Amsterdam.
13 Golden BL and Wasil EA (1987). Computerized vehicle routing
in the soft drink industry. Opns Res 35: 6-17.
14 Ioannou G, Kritikos M and Prastacos G (2001). A greedy lookahead heuristic for the vehicle routing problem with time
windows. J Opl Res Soc 52: 523-537.

15 Solomon MM (1987). Algorithms for the vehicle routing and


scheduling problems with time window constraints. Opns Res
35: 254-265.
16 MapInfo (2000). MapInfo Professional: Users Guide. MapInfo
Corporation: Troy, NY.
17 Taylor MA (1990). Knowledge-based systems for transport
network analysis: a fifth generation perspective on transport
network problems. Transport Res A 24: 3-14.
18 Cormen TH, Leiserson CE and Rivest RL (1992). Introduction
to Algorithms. MIT Press: Cambridge, MA.
19 Kritikos MN (1997). Mathematical models and algorithmic
approaches for transportation problems using Geographical
Information Systems. PhD dissertation, Athens University of
Economics and Business.

Received September2001;
accepted January 2002 after one revision

This content downloaded from 202.43.95.117 on Mon, 18 May 2015 03:33:35 UTC
All use subject to JSTOR Terms and Conditions