You are on page 1of 12

A knowIedge-based probIem soIving meIhod in CIS appIicaIion

Hui Wei , Oing-xin Xu, Xue-song Tang


Coenitivc MoJcl unJ /leorithm Luborutory, Dcpurtmcnt oj Computcr Scicncc, luJun Univcrsity, Shunehui z004JJ, Pk Chinu
A k 1 I c I I I K I o
/rticlc history:
Keceived 11 Mav 2009
Keceived in revised Iorm 10 Ianuarv 2011
AccepIed 18 Ianuarv 2011
AvaiIabIe onIine 25 Ianuarv 2011
KcyworJs:
ProbIem soIving
HeurisIic searching
KnowIedge-based svsIem
MeIro pIanning
CIS
A s s 1 k A c 1
ModeI design Ior Iheme anaIvsis is one oI Ihe biggesI chaIIenges in CIS. Manv reaI appIicaIions in CIS
require IuncIioning noI onIv in daIa managemenI and visuaIizaIion, buI aIso in anaIvsis and decision-
making. ConIronIed wiIh an appIicaIion oI pIanning a new meIro Iine in a ciIv, a IvpicaI CIS is unabIe
Io accompIish Ihe Iask in Ihe absence oI human experIs or arIiñciaI inIeIIigence IechnoIogies. AparI Irom
being modeIs Ior anaIvzing in diIIerenI Ihemes, some appIicaIions are aIso insIances oI probIem soIving in
AI. ThereIore, in order Io sIrengIhen iIs abiIiIv in auIomaIic anaIvsis, manv Iheories and IechnoIogies
Irom AI can be embedded in Ihe CIS. In Ihis paper, a sIaIe space is deñned Io IormaIize Ihe meIro Iine
pIanning probIem. Bv uIiIizing Ihe deñned sIaIe evaIuaIion IuncIion, knowIedge-based ruIes and sIraIe-
gies, a heurisIic searching meIhod is deveIoped Io opIimize Ihe soIuIions iIeraIiveIv. ExperimenIs are
impIemenIed Io iIIuminaIe Ihe vaIidiIv oI Ihis AI-enhanced auIomaIic anaIvsis modeI oI CIS.
2011 EIsevier B.V. AII righIs reserved.
1. lntroduction
An imporIanI uIiIizaIion oI geographicaI inIormaIion svsIems
|CIS) is Io soIve disIribuIion and opIimizaIion probIems eIIecIiveIv.
ExampIes incIude: How Io aIIocaIe pIanIing areas oI diIIerenI crops
on IimiIed cuIIivaIed Iand, so as Io make good use oI Ihe cuIIivaIed
Iand and geI Ihe desired economic beneñI? How shouId a reIaiI
companv decide Ihe IocaIions and scaIes oI Ihe sIores in a ciIv
Ior maximum proñI? How shouId a IourisI eIIecIiveIv pIan a IraveI
scheduIe in Ierms oI desIinaIion, weaIher condiIion and budgeI?
How shouId Ihe IransporIaIion deparImenI buiId a meIro svsIem
in a ciIv and achieve Ihe opIimaI design? These are aII concreIe
appIicaIions and IvpicaI probIem soIving in arIiñciaI inIeIIigence
|AI) IhaI have Iwo characIerisIics in common: aII conIain a greaI
deaI oI geographicaI or spaIiaI inIormaIion, and Ihe use oI knowI-
edge in some proIessionaI domains.
Enhanced CIS, which reIers Io a combinaIion oI CIS and experI
svsIem, decision supporI svsIem or oIher meIhods in AI, is usuaIIv
caIIed InIeIIigenI CIS. This Ivpe oI hvbrid InIormaIion SvsIem is
aimed aI deaIing wiIh compIicaIed appIicaIions oI CIS and is
wideIv used in manv areas, such as agricuIIure, IoresIrv, ecosvs-
Iem, IraIñc, IransporIaIion, environmenIaI proIecIion and pubIic
heaIIh. However, as Ihe geographicaIIv reIevanI daIa is one oI Ihe
essenIiaI parIs in CIS, sophisIicaIed reasoning mechanisms are re-
quired Ior processing dvnamic inIormaIion and Ihe knowIedge oI
Ihe reaI worId. Thus, Io whaI exIenI Ihe knowIedge-based inIeIIi-
genI meIhods are embedded in CIS is oI greaI imporIance. In Ierms
oI Ihe meIhodoIogv oI probIem soIving, an inIeIIigenI svsIem is de-
signed Io soIve speciñc knowIedge-required probIems. So wheIher
Ihe soIuIion is made bv human or bv compuIer is a signiñcanI diI-
Ierence IhaI decides wheIher a CIS-based svsIem is a managemenI
IooI or an inIeIIigenI decision-maker.
MeIro Iines exisI in mosI meIropoIises as a IasI, puncIuaI, envi-
ronmenIaIIv IriendIv and high-capaciIv pubIic mode oI IransporIa-
Iion. Their design, consIrucIion and operaIion require exIensive
proIessionaI knowIedge and experience. Hence, an experI svsIem
can be deveIoped Io assisI Ihe decision-making Ior beIIer perIor-
mance and cosI eIñciencv. A considerabIe amounI oI good research
has been done during Ihe pasI J0 vears. TabIe 1 succincIIv reviews
Ihe experI svsIems, knowIedge-based svsIems and some oIher
meIro-concerned AI appIicaIions IhaI were direcIIv designed Ior
or can be easiIv IranspIanIed inIo Ihe meIro Iine svsIem. In generaI,
Ihose AI-concerned appIicaIions cover everv aspecI oI pubIic Irans-
porIaIion, such as design, consIrucIion, operaIion and service pro-
vision, wiIh quiIe maIure IechnoIogies. NeverIheIess, Ihere exisI
some common probIems. FirsI, Ihe svsIems reIv on human-
compuIer inIeracIion and need human supporI Io a greaI exIenI,
or manv manuaI inIervenIions, as Ihev cannoI soIve probIems
independenIIv. This reûecIs Ihe IacI IhaI Ihe knowIedge is noI suI-
ñcienI, and CIS mainIv pIavs Ihe roIe oI a daIabase. Second, Ihev
work in IimiIed ñeIds wiIh IimiIed IuncIionaIiIies, and are noI
capabIe oI reasoning using muIIi-domain knowIedge. Unsurpris-
ingIv, Ihe design and operaIion oI a pubIic IransporIaIion svsIem
such as a meIro svsIem requires exacIIv Ihe capabiIiIv oI using
muIIi-domain knowIedge. Third, Ihe posiIioning oI meIro-Iines
and sIaIions needs Ihe consideraIion oI muIIipIe IacIors and Ihe
saIisIacIion oI muIIipIe resIricIions. FurIhermore, a reaI inIeIIigenI
0950-7051|S - see IronI maIIer 2011 EIsevier B.V. AII righIs reserved.
doi:10.1016|j.knosvs.2011.01.007
Corresponding auIhor.
£-muil uJJrcss: weihui@Iudan.edu.cn |H. Wei).
KnowIedge-Based SvsIems 24 |2011) 542-55J
CohIehIs lisIs available aI SciehceDirecI
KnowIedge-Based SvsIems
I our hal homepage: www. el sevi er . com/ l ocaI e/ khosys
svsIem shouId be required Io provide candidaIe sIraIegies and
decisions auIomaIicaIIv.
This paper is divided inIo six secIions. The ñrsI parI is Ihe inIro-
ducIion oI AI-based CIS appIicaIions and an expIanaIion oI whv
probIems in Ihese appIicaIions can be soIved bv AI IechnoIogv.
The second secIion presenIs a concepIuaI modeI oI probIem soIving
under a CIS background, and a knowIedge-based searching meIhod
is designed Io soIve Ihe pIanning probIems. The IoIIowing secIion is
abouI appIving a CIS pIaIIorm Io buiId a virIuaI and deIaiIed envi-
ronmenI, on which manv compIicaIed appIicaIions wiII be concep-
IuaIized and IormaIized. The IourIh secIion, which is Ihe core oI
Ihis paper, eIaboraIeIv expIains Ihe whoIe phiIosophv oI using
Ihe heurisIic searching aIgoriIhm Ior probIem-soIving in meIro-
Iine auIomaIic pIanning, and presenIs Ihe aIgoriIhms Ior deñning
sIaIe spaces, deñning Ihe sIaIe evaIuaIion IuncIion and IormaIizing
knowIedge. The ñIIh secIion dispIavs Ihe experimenIaI resuIIs oI
using Ihis meIhod Io pIan a meIro-Iine auIomaIicaIIv. The IasI sec-
Iion is a discussion abouI deep inIegraIion oI Ihe probIem soIving
IechnoIogv oI AI and CIS in acIuaI appIicaIions.
2. Systcm dcsign ol an cnhanccd ClS modcl
In order Io enabIe a CIS Io process knowIedge-required Iasks
inIeIIigenIIv, Ihe svsIem musI be imbedded wiIh probIem soIving
IuncIions. A concepIuaI AI-sIrengIhened Iramework is presenIed
in Fig. 1, which is composed bv Iwo main parIs: one parI |green
1
moduIes) is Ihe CIS inIrasIrucIure and Ihe oIher |red moduIes) con-
sisIs oI moduIes oI knowIedge-based managemenI, Ihe sIaIe evaI-
uaIion IuncIion and Ihe heurisIic searching wiIh regard Io probIem
soIving. In Fig. 1, Ihe CIS moduIe is responsibIe Ior Ihe manage-
menI and visuaIizaIion oI spaIiaI inIormaIion and oIher aIIribuIe
inIormaIion, incIuding Ihe disIribuIion oI hoIeIs, roads neIwork,
Iakes and ciIv boundaries. AII oI Ihese are cIassicaI IuncIions oI
CIS soIIware.
There are Ihree main IuncIionaI moduIes used Ior probIem soIv-
ing. The ñrsI parI is Ihe knowIedge base, which incIudes Iwo Ivpes
oI knowIedge, IacI and ruIe. AII oI Ihe knowIedge is used Ior ciIv
pIanning. The second parI is Ihe sIaIe evaIuaIion IuncIion, which
is Ihe kev Io searching in sIaIe space. The above Iwo parIs are open
and can be updaIed according Io a concreIe appIicaIion. Though
Ihe searching aIgoriIhm Ior probIem soIving is generaI in AI, Ihe
ruIes, sIaIe space, heurisIic sIraIegv and sIaIe evaIuaIion IuncIion
are highIv Iask-speciñc. Thev need Io be deñned eIaboraIeIv, espe-
ciaIIv when Ihe Iask is muIIi-domain and uncerIain reasoning is
concerned. ThaI is Io sav a greaI chaIIenge exisIs in Ihe impIemen-
IaIion oI means-ends anaIvsis. The Ihird parI is Ihe sIaIe space and
heurisIic searching IuncIion. The deIaiIs oI Ihese are presenIed in
SecIion 4.
3. kcprcscnting lacts in ClS platlorm
CIS is a powerIuI pIaIIorm Io sIore, visuaIize and Iuse daIa. II
inIegraIes a greaI deaI oI IriviaI daIa Io consIrucI seIs oI iIems
Ior visibiIiIv and discriminabiIiIv, or in oIher words, Io organize
Ihose Iremendous amounIs oI daIa. TvpicaIIv CIS organizes daIa
Ihrough manv Iheme Iavers, and posiIions are used as reIerence
poinIs. Whereas onIv a smaII parI oI Ihe daIa is visibIe, and manv
IacIs, such as Ihe aIIribuIes oI a recenIIv buiII meIro sIaIion in a
IraIñc neIwork, are invisibIe, Ihese hidden daIa mighI pIav an
imporIanI roIe in Ihe probIem soIving. So iI is cruciaI Io esIabIish
a Iorm oI daIa represenIaIion and impIemenI iI Ior Ihe currenI
appIicaIion.
Besides Ihe geographicaI daIa oI objecIs in Ihe reaI worId, an
inIeIIigenI CIS is aIso expecIed Io have a knowIedge base Ior pos-
sessing as manv knowIedge iIems as possibIe, in order Io provide
a suIñcienI base Ior inIerence. WhaI kind oI IacIs shouId be
adopIed bv Ihe CIS pIaIIorm depends on Ihe speciñc probIem. As
Ior urban road neIwork pIanning, iI is pIausibIe IhaI Ihe posiIion-
reIaIed IacIs IhaI are desired Ior a CIS pIaIIorm are currenI IraIñc
neIworks, incIuding roads, highwavs, meIro Iines and sIaIions,
Ihe popuIaIion disIribuIion in Ihe ciIv aI diIIerenI periods, and
Ihe disIribuIion oI IeaIures such as residenIiaI areas, business dis-
IricIs, pubIic IaciIiIies, inIrasIrucIure oI Ihe ciIv and naIuraI sighIs.
However, as Io how manv aIIribuIes oI each oI Ihem shouId be re-
sIored in Ihe daIabase, iI is diIñcuII Io make a decision in advance,
so a scaIabIe design is necessarv Io enabIe possibIe knowIedge
expansion. As Iar as Ihe road neIwork pIanning is concerned, Ihe
CIS pIaIIorm can be esIabIished bv obIaining Ihe useIuI daIa Ior
Ihe pIanning Irom maps and Ihe reaI worId. The posiIion-reIaIed
inIormaIion and aIIribuIes oI Ihe geographic enIiIies are sIored
in Ihe IabIes oI a reIaIionaI daIabase.
4. Knowlcdgc-bascd hcuristic scarching
Searching is a powerIuI IooI Ior probIem soIving. II consisIs oI a
sIaIe space, a seI oI ruIes oI sIaIe IransIormaIion, a sIaIe evaIuaIion
IuncIion, and a heurisIic sIraIegv. Suppose IhaI a new meIro Iine
beIween Iwo posiIions needs Io be designed in a ciIv, how can Ihis
Iask be done bv Ihe aIoremenIioned enhanced CIS? II musI be
emphasized IhaI Ihis kind oI Iask is highIv knowIedge dependenI.
Tablc 1
The previous ES or KBS appIicaIions in Ihe meIro domain or cIoseIv reIaIed Io iI.
Phase ExisIing
works
Main IechnoIogv
Design Koad neIwork |1| CIS ÷ Craph
|2| CSP
|J| CIS ÷ Fuzzv-Iogic
Land aIIribuIe |4| CIS
|5| Case-based
|6| CIS ÷ ES
SiIe seIecIion |7| ES, DSS
|8| ES
|9| KBS ÷ MP
|10| CIS ÷ DSS
|11| KuIe-based
CosI esIimaIion |12| KBS
EnvironmenI
evaIuaIion
|1J| Fuzzv seI
ConsIrucIion CeoIogicaI moniIor |14| DaIa mining
EcoIogicaI eIIecI |15| CIS ÷ DSS
|16,17| CIS ÷ SIaIe-
IransiIion
No vioIaIing guideIine |18| KBS
OperaIion ConIroI |19| ES
|20| Fuzzv Iogic
MainIenance |21| KuIe-based
|22| Case-based
IncidenI managemenI |2,2J| CIS ÷ ES
PIan & ScheduIe |24| ANN
|25| CA
|26| CSP
DaIa coIIecIion |27| KFID ÷ ES
OpIimum
IransporIaIion
|28| CIS ÷ MAS
SecuriIv |29-J1| PK ÷ CV
Service ShorIesI rouIe |J2| CPS ÷ Search
PersonaIized pIanning |JJ| OnIoIogv
|J4| CIS ÷ ES
1
For inIerpreIaIion oI coIor in Fig. 1, Ihe reader is reIerred Io Ihe web version oI
Ihis arIicIe.
H. Wci ct ul. / KnowlcJec-8uscJ Systcms z4 |z011) 54z-55J 54J
For insIance, Ihe decision Io use an exisIing IunneI is beIIer Ihan Io
dig a new one Io cross a river.
4.1. Dchnition oj stutc spucc
Designing a meIro Iine is a IvpicaI probIem-soIving Iask in CIS
appIicaIions, where use oI knowIedge is absoIuIeIv necessarv Ior
seeking Ihe soIuIion.
II Ihe Iwo IerminaI sIaIions, middIe sIaIions and direcIion con-
IroI poinIs oI Ihe meIro Iine can be deIermined, Ihe probIem is
soIved. Fig. 2 iIIusIraIes how Ihe soIuIion evoIves. The IeII parI is
an iniIiaI soIuIion [p
0
, p
1
|, or in oIher words, iI is an iniIiaI sIaIe
oI a soIuIion, which simpIv connecIs Iwo IerminaIs. BuI usuaIIv
Ihis coarse soIuIion is Iar Irom Ihe expecIed one. NexI, a new poinI
near one oI Ihe IerminaI sIaIions can be chosen and inserIed inIo
Ihe sequence [p
0
, p
1
|. This poinI can be a middIe sIaIion or a direc-
Iion conIroI poinI oI Ihe meIro Iine. Thus an improved soIuIion [p
0
,
p
1
, p
2
| comes inIo being. SimiIarIv oIher poinIs near Ihe currenI
poinIs can be chosen and inserIed inIo Ihe Iormer soIuIion, which
generaIes a new soIuIion. II is pIausibIe IhaI Ihe IaIIer soIuIion is
beIIer Ihan Ihe Iormer one. This process iIeraIes unIiI a sequence
[p
0
, p
1
, , p
n
| is saIisIacIorv Io be Ihe ñnaI soIuIion. And iI is noI
a one-sIep process, buI an iIeraIive process Io opIimize and ap-
proach graduaIIv, which is shown in Ihe middIe and righI parIs
oI Fig. 2. Anv Iwo direcIion conIroI poinIs oI Ihe meIro Iine can
be supposed Io be connecIed bv a sIraighI Iine, so a soIuIion can
be represenIed bv a poinI sequence, Ior exampIe L-Ip
0
, p
1
, ,
p
n
| represenIs a meIro Iine, where p
0
and p
n
are Ihe Iwo IerminaI
sIaIions, p
i
|1 i n 1, i N) is a middIe sIaIion or direcIion con-
IroI poinI oI Ihe meIro Iine. II a middIe soIuIion ñnds more Ihan
one candidaIe poinI IhaI is proper, muIIipIe oIIspring soIuIions
can be obIained Irom Iheir parenI soIuIions. AII Ihose possibIe
poinI sequences make up Ihe sIaIe space, where everv sIaIe is a
candidaIe soIuIion. One sIaIe can be derived Irom anoIher bv
inserIing new poinIs. Expanding soIuIions in Ihis wav esIabIishes
a Iree oI soIuIions, which is Ihe sIaIe space. Then searching is
needed Io ñnd Ihe opIimaI poinI sequence IhaI meeIs Ihe require-
menIs oI passengers' capaciIv and Ihe consIrainIs oI consIrucIion.
Here a reñned meIhod is used Io ñnd a desired soIuIion. The origi-
naI probIem sIarIs Irom an iniIiaI soIuIion [p
0
, p
1
|, which usuaIIv is
quiIe coarse. AII kinds oI knowIedge, such as Ihe ciIv's IavouI, Ihe
disIribuIion oI Ihe ciIv's inIrasIrucIure, geoIogicaI and ecoIogicaI
siIuaIion, environmenI, popuIaIion disIribuIion, Ihe sIaIus oI IraIñc
ûow and oIher knowIedge, needs Io be considered Io evoIve Ihe
lig. 1. A concepIuaI sIrucIure oI probIem soIving in CIS appIicaIion.
lig. 2. Process oI soIuIion reñning.
544 H. Wci ct ul. / KnowlcJec-8uscJ Systcms z4 |z011) 54z-55J
soIuIion, and uIIimaIeIv achieve a reIaIiveIv opIimaI soIuIion IhaI
meeIs aII Ihe requiremenIs.
How Io geI an oIIspring soIuIion Irom iIs parenI soIuIion, i.e.
how Io IransIorm Irom one sIaIe Io Ihe nexI sIaIe? There are Iwo
kinds oI siIuaIions, wherebv one is connecIing Io a new middIe sIa-
Iion and Ihe oIher is inserIing a new direcIion conIroI poinI. In Ihe
Iormer, abundanI knowIedge is heIpIuI Io ñnd a preIerred posiIion
Ior a sIaIion in a band-Iike area, and Ihe axis oI Ihis Iinear area is
Ihe currenI soIuIion. UsuaIIv more Ihan one candidaIe poinI exisIs
Ior consideraIion. For exampIe, suppose a middIe sIaIion oI Ihe
currenI sIaIe is s
i
, and Ihere are Ihree candidaIe sIaIions p
1
, p
2
and p
J
near s
i
, which are a museum, a park and a raiIwav sIaIion,
respecIiveIv. Considering IhaI a raiIwav sIaIion aIwavs has manv
passengers, iI is more reasonabIe Io choose p
J
as Ihe neighboring
sIaIion |i.e. s
i÷1
) oI s
i
. In Ihe IaIIer, a meIro Iine someIimes needs
some poinIs Io conIroI iIs direcIion. For exampIe, suppose Ihere
is a hisIoric siIe beIween Ihe sIaIions s
i
and s
i÷1
. II a sIraighI raiIwav
is consIrucIed beIween s
i
and s
i÷1
, Ihe hisIoric siIe mighI be dam-
aged. So a direcIion conIroI poinI c
i1
IhaI is awav Irom IhaI hisIoric
siIe can be pIaced beIween s
i
and s
i÷1
Io avoid iI. This aIso produces
a new sIaIe. Fig. J shows a sIaIe space as a Iree sIrucIure.
4.z. Dchnition oj stutc cvuluution junction
In Ihe searching process, how does Ihe enhanced CIS evaIuaIe
wheIher a new sIaIe is beIIer Ihan Ihe currenI one? And how does
iI judge wheIher iI is necessarv Io expand Ihe currenI sIaIe Io a
new one? In order Io measure Ihe quaIiIv oI a sIaIe, a sIaIe evaIu-
aIion IuncIion needs Io be deñned. A sIaIe is a compIex poinI se-
quence, and everv poinI in iI is decided bv manv IacIors. So iI is
raIionaI Io use diIIerenI IuncIions Io evaIuaIe diIIerenI IacIors,
respecIiveIv. These IacIors shouId be reIaIed Io Ihe aIIeviaIion oI
IraIñc pressure, opIimizaIion oI Ihe pubIic IraIñc neIwork, beIIer
disIribuIion oI urban IuncIionaIiIies, baIance oI popuIaIion in diI-
IerenI disIricIs, diIñcuIIv in consIrucIion and cosI, degree oI envi-
ronmenIaI and ecoIogicaI damage, and so on. CeneraIIv, Ihe
IacIors can be divided inIo Iwo caIegories, posiIive ones and nega-
Iive ones. The Iormer hoIds Ihe advanIages oI consIrucIing such a
meIro Iine, such as aIIeviaIing IraIñc pressure and promoIing
urbanizaIion, whiIe Ihe IaIIer reûecIs Ihe cosI oI consIrucIing a
meIro Iine, such as consIrucIion expense and desIrucIion oI vaIu-
abIe reIics. The evaIuaIion IuncIion is quiIe cruciaI Io Ihe quaIiIv
judgmenI oI a sIaIe. There are severaI principIes IhaI need Io be IoI-
Iowed in deñning an evaIuaIion IuncIion. Thev are:
|a) Considering as manv IacIors as possibIe and quanIiIving
Ihem.
|b) A IacIor wiIh posiIive eIIecI is assigned bv a posiIive vaIue
and a IacIor wiIh negaIive eIIecI is assigned bv a negaIive
vaIue.
|c) DiIIerenI IacIors inûuence meIro Iine pIanning wiIh diIIerenI
weighIs.
These IacIors can be divided inIo six cIasses: IransporIaIion eIñ-
ciencv, eIñciencv oI IraIñc neIwork opIimizaIion, eIñciencv in Ihe
IuIure, expense oI consIrucIion, cuIIuraI cosI and ecoIogicaI cosI.
A sIaIe evaIuaIion IuncIion is deñned as a vecIor l ÷ j
1
, j
2
, j
J
, j
4
,
j
5
, j
6
. Each j
i
in iI reIers Io a Iore-menIioned IacIor. The ñrsI Ihree
eIemenIs reIer Io Ihe posiIive eIIecI IacIors and Ihe IaIer Ihree cor-
respond Io Ihe negaIive eIIecI IacIors. Suppose L ÷ [p
0
, p
1
, , p
n
| Io
be a sIaIe. The IoIIowing is Ihe deIaiIed caIcuIaIion oI each IacIor j
i
.
Transportation clñcicncy |j
1
|L)) means Ihe number oI peopIe
IhaI can be IransporIed bv Ihe meIro Iine. Because sIaIions pIav
Ihe roIes oI assembIing and diIIusing passengers, Iheir posiIions
musI aIIecI Ihe vaIue oI IransporIaIion eIñciencv markedIv. Now
deñne
j
1
L
S
i 0
pusscnecr louJ s
i
where pusscnecrlouJ|s
i
) is Ihe passenger sIream IhaI mav be
underIaken bv sIaIion s
i
, where S is Ihe number oI sIaIions in soIu-
Iion L. The IuncIion is deñned as beIow
pusscnecr louJ s
i
!ub 2
j 1
contn urcu s
i
r 
j
2J
t 0
jlow

j
t p

j
jlow

j
t t
where TabIe 2 is Ihe coIIecIion oI aII possibIe venues and !ub. 2
represenIs Ihe number oI venues in TabIe 2. More speciñcaIIv, u
j
is a venue IisIed TabIe 2 and iIs exisIing scope is deñned in |urcu|-
s
i
,r)), which is cenIered aI s
i
wiIh a radius r. contn|u
j
, urcu|s
i
,r)) re-
Iurns a BooIean vaIue iIIusIraIing wheIher 
j
can be Iound in
urcu|s
i
,r), and jlow
u
j
t gives Ihe passenger ûow aI Iime t, and ñnaIIv
Final solution |start, p'
1
. p'
2
... p'
16
, end¦
Initial |start, end¦
Partial solution |start, p
1
, end¦ Too expensive to accept
An optimal solution |start, p1.
p
2
... p
15
, end¦
lig. 3. SIaIe space in a Iree sIrucIure.
Tablc 2
PossibIe venues wiIh dense popuIaIion near a sIaIion.
KaiIwav sIaIion Park Long-disIance bus sIaIion Cvmnasium AirporI HisIoricaI sighIs TheaIre Music haII ArI museum -
H. Wci ct ul. / KnowlcJec-8uscJ Systcms z4 |z011) 54z-55J 545
p
u
j
x t is Ihe probabiIiIv IhaI Ihe voIume oI passenger ûow is x aI
Iime I in u
j
. Then Ihe IoIaI is as IoIIows:
j
1
L
n
i 0
Tab 2
j 1
contn urcu s
i
r 
j
24 hours
t 1
jlow

j
t p

j
jlow

j
t t
Fig. 4 is Ihe disIribuIions oI probabiIiIies oI Ihe number oI pas-
sengers IhaI emerge aI some venues |such as a raiIwav sIaIion, a
gvmnasium, or a pIaza) aI diIIerenI momenIs. These daIa can be ac-
quired bv daiIv sIaIisIics. Because Ihe sIream oI passengers Irom
Ihe residenIiaI area, working area and commerciaI area is compar-
aIive sIeadv, Ihev wiII be discussed IaIer.
£jhcicncy oj trujhc nctwork optimizution |j
2
|L)) is deñned as Ihe
sum oI Ihe IoIIowing parIs: Ihe numbers oI bus Iines, meIro Iines
and IighI raiIwavs IhaI L connecIs, Ihe number oI new candidaIe
rouIe IhaI Iinks Iwo posiIions bv L, and Ihe number oI IraIñc hinges
and IuncIionaI cenIers oI Ihe ciIv IhaI can be connecIed bv L. As-
sume IhaI buslinc|/) represenIs a seI oI bus rouIes crossing area
/, and busstop|bl
i
) represenIs aII sIops oI bus rouIe bl
i
. A BooIean
IuncIion
ninsct S
1
S
2
1 x S
1
y S
2
x y
2
500 m
S
1
bus stop bl
1
S
2
bus stop bl
2
0 oIherwise
is deñned Io decide wheIher Iwo bus rouIes bl
1
and bl
2
share Ihe
same or aImosI Ihe same bus sIops, which reûecIs Ihe IeasibiIiIv
oI IransIerring Irom one Io Ihe oIher. II Ihe disIance beIween Iwo
sIops is Ionger 500 m, iI is beIievabIe IhaI Ihev cannoI be easiIv
reached Irom each oIher. Three oIher BooIean IuncIions are as
IoIIows:
inJir /
1
/
2
1 Ihere is no bus rouIe running
beIween area /
1
and /
2
0 oIherwise
and
hinec /
1 area / is a IraIIic hinge
0 oIherwise
and
loculccntcr /
1 area / is a IuncIionaI cenIre oI ciIv
0 oIherwise
Then iI has
j
2
L 
21
S
i 0
S
j i l
i
buslinc urcu s
i
r l
j
buslinc urcu s
j
r
ninsct busstop l
i
busstop l
j

22
S
i 0
S
j i
inJir urcu s
i
r urcu s
j
r 
2J
S
i 0
hinec urcu s
i
r

24
S
i 0
loculccntcr urcu s
i
r
where 
2i
|1 i 4, N) are Ihe weighI IacIors, and Ihe ñrsI ad-
dend indicaIes Ihe number oI adjacenI bus sIops aIong Ihe speciñc
soIuIion.
A new meIro Iine musI prompI Ihe deveIopmenI IeveI in an
area. £jhcicncy in juturc |j
J
|L)) is deñned as a sum oI weighIed diI-
Ierences, incIuding Ihe buiIding diIIerence, service IaciIiIv diIIer-
ence and road densiIv diIIerence beIween Ihe sIaIion area and
Ihose in Ihe whoIe ciIv. Assume IhaI Jcn8|/) represenIs Ihe buiId-
ing densiIv in area /, Jcnk|/) represenIs Ihe road densiIv in area /,
Jcnl|/) represenIs Ihe densiIv oI service IaciIiIies in area /, and
Cl!Y represenIs Ihe whoIe area oI Ihe ciIv in which a new meIro
Iine wiII be consIrucIed. Deñne kDD|L) as
kDD L
S
i 0
Jcnk Cl!Y Jcnk urcu s
i
r
Jcnk Cl!Y
Here deñne 8lD|L) as
8lD L
S
i 0
M/X
Jcn8 Cl!Y Jcn8 urcu s
i
r
Jcn8 Cl!Y
Jcnl Cl!Y Jcnl urcu s
i
r
Jcnl Cl!Y
Then we deñne
j
J
L 0 6 kDD L 0 4 8lD L
£xpcnsc oj construction |j
4
|L)) is deñned as Ihe negaIive oI Ihe
sum oI aII expenses, incIuding digging IunneIs, Iaving Ihe raiIwav
Iine and consIrucIing sIaIions. II is deñned as
j
4
L 1 Jie L pu L
S
i 0
con s
i
where Jie|L) is Ihe expenses oI digging Ihe IunneI. Suppose IhaI Ihe
whoIe IunneI has m segmenIs, and Iheir geoIogicaI siIuaIions mighI
be diIIerenI. Even in Ihe same segmenI, Ihe ground siIuaIion over
Ihe IunneI couId be variabIe.
II a segmenI is divided inIo severaI shorIer parIs in Ierms oI
Iheir parIicuIar siIuaIions, a combinaIion
L
m
i 1
Li Li p
i
1
p
im
can be acquired
lig. 4. ProbabiIiIv disIribuIion oI passengers in some venues.
546 H. Wci ct ul. / KnowlcJec-8uscJ Systcms z4 |z011) 54z-55J
FinaIIv, we deñne
Jie L
m
i 1
 
i
im 1
j i
1
p
j
p
j 1
where p
j
, p
j÷1
L
i
, is a coeIñcienI, which is Ihe benchmark cosI oI
consIrucIion wiIh normaI diIñcuIIv and ñxed IengIh, and 
i
denoIes
Ihe consIrucIion diIñcuIIv in segmenI i, p
i
p
i÷1
is Ihe IengIh be-
Iween p
i
and p
i÷1
, puv|L) represenIs Ihe expense oI Iaving raiI in
Ihe whoIe IunneI, and
pu L 
n 1
i 0
p
i
p
i 1
p
i
, p
i÷1
L,  is Ihe cosI oI Iaving raiI per miIe. cons|s
i
) represenIs Ihe
expense oI buiIding a sIaIion. Then
j
4
L 1
m
i 1
 
i
im 1
j i
1
p
j
p
j 1

n 1
i 0
p
i
p
i 1
S
i 0
con s
i
The consIrucIion and running oI a meIro Iine is known Io gen-
eraIe negaIive eIIecIs on some ancienI buiIdings or hisIoric siIes
near iI, because Ihe base oI Ihese buiIdings couId be damaged,
which iIIusIraIes IhaI Ihe culturul cost |j
5
|L)) musI be considered.
Here Ihe sensiIive range oI consIrucIing a sIaIion is 50 m, and IhaI
oI digging a IunneI is 20 m. Thus iI can be deñned IhaI
j
5
L 1
n
i 0 h area s
i
50m
runk h
n 1
i 0 h linc p
i
p
i 1
20m
runk h
where s
i
, p
i
, p
i÷1
L
i
, and h sIands Ior Ihe hisIoric siIe, and runk|h)
represenIs Ihe rank oI iI, which diIIerenIiaIes Ihe hisIoric and cuI-
IuraI vaIues oI Ihe siIes.
The ccoloeicul cost |j
6
|L)) conIains Ihe amounI and Ihe Ivpe oI
vegeIaIion IhaI mighI be damaged bv Ihe meIro Iine |or IighI raiI)
consIrucIion, and Ihe IengIh oI raiIwav IhaI crosses an ecoIogicaI
proIecIion zone and iIs rank, and Ihe meIhod oI Iraverse oI Ihe raiI-
wav in Ihe region |e.g. IunneI or eIevaIed bridge). Suppose IhaI Ihe
whoIe IunneI can be separaIed inIo severaI secIions, oI which m
secIions cross Ihe ecoIogicaI proIecIion zone. The Ivpes oI vegeIa-
Iion mav be diIIerenI aIong Ihe raiIwav. So based on Ihe same
meIhod inIroduced in Ihe caIcuIaIion oI j
4
|L), Ihese m secIions are
L
m
i 1
Li Li p
i
1
p
im
where each secIion has some smaIIer sub-secIions wiIh idenIicaI
vegeIaIion aIIribuIes. II Ihe ecoIogicaI proIecIion zone reIaIed wiIh
secIion i is /
i
, Ihe IuncIion can be caIcuIaIed as beIow
j
6
L 1
m
i 1
im 1
j i
1

j
Jcnv /
j
p
j
p
j 1
im 1
j i
1

j
Jpt /
j
p
j
p
j 1
where p
j
, p
j÷1
L
i
, and 
j
is Ihe rank oI imporIance oI vegeIaIion in
range /
j
, Jcnv|/
j
) is Ihe densiIv oI vegeIaIion in area
/
j
im 1
j i1
p
j
p
j 1
is Ihe IengIh oIraiIwav IhaI crosses Ihe area /
j
, 
j
is Ihe meIhod oI Ihe raiIwav crossing IhaI zone |
j
÷ J iI on Ihe sur-
Iace, or 
j
÷ 1 iI under ground) unJ Jpt|/
j
) is Ihe environmenIaI rank
oI area /
j
, which reûecIs Ihe environmenIaI signiñcance in Ihis area.
So anv L ÷ [p
0
, p
1
, , p
n
| in sIaIe space, can be evaIuaIed bv Ihe
six IuncIions and Iorm an evaIuaIion vecIor. Then Ihe searching
direcIion can be guided bv comparing Ihese vecIors.
4.J. kulc-buscJ knowlcJec rcprcscntution
To design a meIro Iine is highIv knowIedge-based. In order Io
opIimize Ihe soIuIion, exIensive knowIedge is needed Io expand
sIaIes Irom one Io anoIher and Io evaIuaIe Ihe quaIiIv oI Ihe sIaIes.
KuIes can be used Io IormaIize Ihe knowIedge, as Ihev can repre-
senI Ihe causaIiIv oI concepIs weII. For simpIiciIv, Ihe producIion
svsIem is used as Ihe meIhod oI knowIedge represenIaIion.
The ñrsI caIegorv oI ruIes is used Io expand successive sIaIes,
such as Ihe IoIIowing ruIes:
k/11: |b)|q)|ln|TabIe 2, b) ln|urcu|q, r), b)| /ppcnJ|q,L).
k/1z: |/)| q)|lsrcsiJcntiulurcu|/) ln|urcu|q,r),/)| 
/ppcnJ|q, L).
k/1J: |/)|q)|lswork-
ineurcu|/) ln|urcu|q,r),/)| /ppcnJ|q, L).
k/14: |/)| q)|lsshoppineccntcr|/) ln|urcu|q,r),/)| 
/ppcnJ|q, L).
k/z0: |LIp
0
,p
1
, ,p
n
|)|r)|tnl)|ln|klv£k,r) lstwosiJc|r, p
i
,
p
i÷1
) ln|!UNN£L,tnl) lscrossrivcrtun-
ncl|tnl) ln|tunnclcntry| tnl),[q
1
,q
2
|) p
i
q
1
< r
tunncl
|whcrc r
tunncl
÷ 5kilomctcrs) q
2
p
i÷1
< r
tunnc
; |lnscrt|p
i
,
q
1
, L) lnscrt|q
1
, q
2
, L)|.
k/J1: |L)|s
i
)|h)|ln|L, s
i
) ln|HlS!DklCSl!£,h) ln|urcu|s
i
,
50 m), h)| ; Dclctc|s
i
, L).
k/Jz: |LIp
0
, , p
i
, p
i÷1
, , p
n
|)|q)|h) | |ln|L,q) h
[HlS!DklCSl!£| ln|linc|p
i
p
i÷ 1
,20m),h) |ln|linc-
|p
i
q,20m),h)) |ln|linc|qp
i÷1
,20m),h)| ;lnscrt|p
i
, q, L).
The second caIegorv oI ruIes is used Io evaIuaIe Ihe quaIiIv oI a
sIaIe, such as IoIIowing ruIes:
k811: |/)|L)|. s
i
)|lsrcsiJcntiulurcu|/) ln|L, s
i
) ln|urcu|s
i
,
r), /)| j
1
L) ÷ j
1
|L) ÷ 0.6 Jcn|urcu|s
i
,r)).
k81z: |/)|L)| s
i
)|lsworkineurcu|/) ln|L,s
i
) ln|urcu|s
i
,r),
/)| j
1
|L) ÷ j
1
|L) ÷ 0.7 Jcn|urcu|s
i
r)).
k81J: |/)|L)| s
i
)|lsshoppineccntcr|/) ln|L,s
i
) ln|urcu|s
i
,
r),/)|  j
1
|L) ÷ j
1
|L) ÷ 0.8 Jcn|urcu|s
i
,r)).
k841: |r)|ln|klv£k, r) wiJth|r) < 500 m|  
i
÷ J.
k84z: |r)|ln|klv£k, r) 500 m < ÷ wiJth|r) < 1000 m; 
i
÷ 5.
k84J: |r)|ln|klv£k, r) wiJth|r) 100 m| 
i
÷ 7.
k844: |q)|ln|klv£k, q)| 
i
÷ 4.
k844: |q)|ln|klv£k,q)| 
i
÷ 6.
k851: |b)|ln|Pk!C!N8LD1,b)| runk|b) ÷ 5.
k85z: |b)|ln|Pk!C!N8LD2,b)| runk|b) ÷ J.
k85J: |b)|ln|Pk!C!N8LDJ,b)| runk|b) ÷ 1
k851: |b)|ln|Pk!C!N/k£/1,/)| Jpt|/) ÷ 5.
k85z: |b)|ln|Pk!C!N/k£/1,/)| Jpt|/) ÷ J.
k85J: |b)|ln|Pk!C!N/k£/1,/)| Jpt|/) ÷ 1.
k854: |b)|ln|k/kl!YPL/N!,u)| 
i
÷ 5.
k855: |b)|ln|lMPDk!/N!PL/N!,u)| 
i
÷ J.
k855: |b)|ln|NDkM/LPL/N!,u)| 
i
÷ 1.
The Ihird caIegorv oI ruIes is used Io deIermine Ihe weighIs oI
six IacIors in Ihe evaIuaIion IuncIion, such as kC11: 0.8, 0.8, 0.6,
0.5,0.2,0.6 .
NoIe IhaI Ihe ruIes can be revised, added and deIeIed. The ruIe
base can be updaIed easiIv. This keeps Ihe svsIem in accordance
wiIh Ihe IaIesI inIormaIion.
4.4. Hcuristic scurchine uleorithm
II Ihe number oI candidaIe soIuIions is noI suIñcienI, Ihe opIi-
maI soIuIion couId be missed. However, excessive candidaIes
mav cause Iow searching eIñciencv. WhiIe searching, Ihe seIecIion
oI candidaIe soIuIions wiII be conñned wiIhin a cerIain range. Sim-
iIar Io Ihe pruning meIhod in searching Iheorv oI AI, severaI condi-
Iions are seI Io IerminaIe sIaIes expanding IurIher and hence
improve Ihe searching eIñciencv. For exampIe, iI Ihe expense ex-
ceeds Ihe budgeI, or Ihe IeveI oI environmenI disrupIion is serious,
H. Wci ct ul. / KnowlcJec-8uscJ Systcms z4 |z011) 54z-55J 547
Ihe searching aIong Ihis direcIion wiII be sIopped. Two sIraIegies
are proposed Io acceIeraIe Ihe searching process, and Iheir Irigger-
ing condiIions are adjusIabIe according Io Ihe demand.
|a) ConIroI Ihe number oI middIe sIaIions.
The disIance beIween Iwo neighboring sIaIions shouId be no
Iess Ihan a consIanI MlND and no greaIer Ihan M/XD. In order
Io ensure Ihe newIv produced candidaIe poinIs converge Io Ihe Ier-
minaI sIaIion s
n
, an assigned area / is shown in Fig. 5 as Ihe scope
oI searching Ior Ihe middIe sIaIion s
i 1
, and s
i 1
s
n
OK. II can be
proved IhaI Ihe newIv added sIaIion wiII incIine Iowards Ihe Ier-
minaI sIaIion graduaIIv.
As Ihere are so manv poinIs on Ihe map, iI is neiIher necessarv
nor possibIe Io check aII oI Ihem. Since each sIaIion has an eIñ-
ciencv radius, as Iong as Ihe sIaIion is IocaIed wiIhin Ihis area,
Ihe posiIion wiII noI aIIecI iIs IransporIaIion IuncIion signiñcanIIv.
So Ihe program Iakes a smaII area as Ihe basic searching uniI, in-
sIead oI a pixeI in Ihe map. In Fig. 5 Ihe searching area is divided
inIo manv smaII secIors, and p is Ihe represenIaIive oI Ihem. CuI-
Iing Ihe map inIo manv smaII pieces can augmenI Ihe granuIariIv
oI searching and IaciIiIaIe Ihe searching speed.
Suppose IhaI sIaIions s
i 1
and s
i
are Ihe Iwo sIaIions. A new sIa-
Iion p is obIained bv searching Irom sIaIion s
i 1
. AIIerwards, iI
needs Io be considered wheIher iI is necessarv Io add a direcIion
conIroI poinI beIween sIaIion s
i 1
and p. II j4 decreases, Ihe conIroI
poinI shouId be added. The necessarv direcIion conIroI poinIs can
be added in accordance wiIh ruIes k/z0 and k/Jz. The raiIwav
shouId, Io Ihe greaIesI exIenI, avoid curving wiIh greaI curvaIure,
Ihus Ihe seIecIion oI desired poinIs is direcIion IimiIed. The range
is seI wiIhin an arc oI 90 . II direcIion poinIs s
i 1
and s
i
are aIreadv
seIecIed, Ihe nexI one s
i÷1
can onIv be seIecIed in Ihe area 8 shown
in Fig. 6, where s
i 1
s
i
SK, SOO ÷ 45 , OOT ÷ 90 , KOT ÷ 45 .
|b) TerminaIe Ihe improper sIaIes in Iime.
II everv vaIue oI a soIuIion's evaIuaIion IuncIion is beIIer Ihan
Ihe expecIed vaIue, Ihe soIuIion sIops expanding. For exampIe, iI
in one soIuIion iI requires a sIaIion Io be buiII under a deep river,
where Ihe expense oI buiIding Ihe sIaIion is greaIer Ihan Ihe bud-
geI, Ihe soIuIion is noI vaIid. The IoIIowing is a IisI oI Ihe condiIions
IhaI sIop searching immediaIeIv:
|1) S > M/XS;
|2) p
n 1
p
n
< 2 MlND;
|J) j
4
|L) > M/XMDN£YCDS!;
|4) 
i
> M/XCDNSD!Y;
|5) j
5
|L) > M/XCULCDS!;
|6) h [HlS!DklCSl!£| h urcu|s
i
,100m) s
i
L;
|7) j
6
|L) > M/X£NvCDS!.
These condiIions are divided inIo Iwo Ivpes. The ñrsI Ivpe in-
cIudes Ihe ñrsI Iwo condiIions, anv soIuIion saIisIving anv one oI
Ihem is caIIed a ''good soIuIion'', which can be accepIed as Ihe ñnaI
soIuIion. AII Ihe oIher condiIions beIong Io Ihe second Ivpe, and a
soIuIion under one oI Ihose condiIions is caIIed a ''bad soIuIion'',
which is ouI oI consideraIion. When more Ihan one candidaIe sIaIe
exisIs, Ihe besI one has Ihe prioriIv Io be expanded. One sIaIe can
generaIe manv subsequenI sIaIes. In Ihis case, Ihe sIaIes are sorIed
bv Iheir vaIues. The sorIing order is j
1
, j
4
, j
2
, j
J
, j
5
, j
6
.
Based on Ihe heurisIic searching aIgoriIhm, Ihe Iamous / aIgo-
riIhm in AI is empIoved Io generaIe a sIaIe Iree Ior Ihe meIro-Iine
pIanning.
5. lxpcrimcnts and rcsults analysis
Shanghai meIro svsIem currenIIv has esIabIished 12 meIro
Iines, wiIh 11 Iines currenIIv in service. The svsIem comprises
267 sIaIions wiIh over 410 km oI raiIwav II operaIes Irains Io carrv
4.5 miIIion cusIomers per dav on average and Ihe peak number has
exceeded 7 miIIion. A IurIher proposaI Io exIend Ihe svsIem has
been pubIished on Ihe oIñciaI websiIe, which cIaims IhaI Ihe IuIure
svsIem wiII achieve 21 Iines and 600 km bv 2020 |J5|. Faced wiIh
such a Iremendous raiIwav neIwork, iI is necessarv Io conducI Ihe
simuIaIion and opIimizaIion based on knowIedge in advance. Here
a soIIware svsIem was deveIoped Io IesI Ihe above-menIioned
ideas. AII experimenIs are based on Ihe acIuaI daIa oI Shanghai
in China. To achieve a cIear map, onIv a smaII parI oI Ihe daIa IhaI
is mosI concerned is dispIaved in Ihe maps.
5.1. £xpcrimcnts oj Jcsienine u sinelc mctro linc
Using our knowIedge-based designing svsIem, manv experi-
menIs were done. The ñrsI one was designing a new meIro-Iine be-
Iween residenIiaI area /1 and indusIriaI area /z. Fig. 7 is one oI Ihe
besI resuIIs. This meIro-Iine originaIes Irom /1 and passes Ihrough
indusIriaI area /J and commerciaI area /4. II inIersecIs MeIro-Iines
1, 2 and J aI sIaIions s1, sz and sJ respecIiveIv. SIaIions s4 and s5
are IocaIed in Ihe cenIer oI Ihe ciIv. Thev can greaIIv aIIeviaIe Ihe
IransporIaIion pressure. The sIaIions s5 and s5 are neighboring sIa-
Iions, which are separaIed bv Huangpu Kiver, and t1 and tz are
direcIion conIroI poinIs. II Ihe raiIwav is desired Io connecI s5
50
Terminal Station s
Station sn-1
|OP|=MÌN_D
|OQ|=MAX_
D
Area A
Possible new
station p
O
P
Q
R
lig. 5. Searching area Ior sIaIions.
45°
45°
Station s
i-1
Station s
i
Station s
i+1
¦OP¦=MÌN_D
P
Q
O
¦OQ¦=MAX_
D
Area B
R
S
T
lig. 6. AvaiIabIe area Ior avoiding a sharp Iurn.
548 H. Wci ct ul. / KnowlcJec-8uscJ Systcms z4 |z011) 54z-55J
and s5 in a sIraighI Iine, a new IunneI needs Io be buiII. BuI now
Ihe raiIwav goes Ihrough an exisIing IunneI t1tz, which greaIIv re-
duces consIrucIion diIñcuIIv and expense. For simpIiciIv, Ihis soIu-
Iion is considered as 'an OpIimaI SoIuIion'.
Fig. 8 is an exampIe oI a parI oI Ihe search Iree. The red nodes
and red Iines represenI Ihose sIaIes and rouIes IhaI are pruned.
The reason is IhaI Ihose sIaIes IaiIed Io saIisIv condiIions oI sIraI-
egies 1 or 2 in Ihe previous secIions. The green nodes and Iines are
Ihe parI oI process Io seek possibIe soIuIions, and Ihe IeaI nodes
are Ihe ñnaI soIuIions, such as s810, s84 and s85. The bIack nodes
are nodes IhaI have noI been expanded veI. For Ihe sake oI compu-
IaIionaI eIñciencv, onIv a IimiIed number oI nodes have a chance Io
be expanded.
The nexI Ihree experimenIs, Fig. 9|a)-|c), were done Io compare
Ihe resuIIs Ior appIving diIIerenI sIraIegies oI opIimizaIion. A ma-
jor diIIerence beIween Ihe soIuIions in Fig. 9|a) and Fig. 7 is IhaI
Ihe Iormer has seven sIaIions Iewer. ExcepI IhaI Ihe join beIween
s1 and sz is sIighIIv curved, Ihe resI oI Ihe raiIwav Iine is aImosI
sIraighI. The reason is IhaI iI did noI Iake inIo accounI Ihe maximi-
zaIion oI Ihe IransporIaIion eIñciencv and sIaIe evaIuaIion Iunc-
Iion j
4
perIorms as a dominanI index. So a soIuIion wiIh Iewer
sIaIions and shorIer IengIh requires Iower consIrucIion expense,
and consequenIIv iI Iends Io be expanded wiIh prioriIv. The soIu-
Iion in Fig. 9|b) did noI cross bv Ihe exisIing IunneI I1I2, buI joins
s5 and s6 direcIIv, wiIhouI consideraIion oI consIrucIion diIñcuIIv
and expense. The soIuIion in Fig. 9|c) Iocused on opIimizing Ihe
IraIñc neIwork and promoIing iIs surrounding area deveIopmenI.
The main diIIerence beIween Ihis soIuIion and Ihe opIimaI one is
IhaI mosI oI iIs sIaIions are seI around Ihe suburbs oI Ihe ciIv. Cen-
eraIIv, Ihe weighI oI j
1
is seI Io be Ihe IargesI one. In Ihis case, how-
ever, j
2
and j
J
deserve more weighI. Thus iI a sIaIe generaIes more
swiIch poinIs around buses or oIher meIro-Iines' sIaIions, or covers
areas wiIh Iow urbanizaIion IeveI, Ihe sIaIe is supposed Io be
ranked ahead and expanded ñrsI.
Fig. 10 gives Iwo meIro-Iines wiIh diIIerenI coIors. The bIue one
is Ihe acIuaI pIan made bv Ihe PubIic CommunicaIion Bureau oI Ihe
Shanghai MunicipaI CovernmenI. The red one is our experimenIaI
resuII. These Iwo Iines share a good superposiIion in mosI secIions.
Fig. 11|a)-|c) presenI Ihree good experimenI resuIIs. AII oI Ihem
share Ihe same IerminaI sIaIions and crosses Ihe enIire ciIv. The
vaIues oI j
1
|IransporIaIion eIñciencv) oI Fig. 11|a) and |b) are Iarge,
and Ihe vaIue oI j
J
|eIñciencv in IuIure) oI Fig. 11|c) is Iarge. These
Ihree resuIIs Iound bv our knowIedge-based designing svsIem aII
have Iheir evidenI raIionaIiIv.
lig. 7. ExperimenIaI resuII oI pIanning a new meIro-Iine.
lig. 8. An exampIe oI a parI oI Ihe search Iree.
lig. 9. ExperimenIaI resuIIs using diIIerenI sIraIegies.
H. Wci ct ul. / KnowlcJec-8uscJ Systcms z4 |z011) 54z-55J 549
In order Io evaIuaIe Ihe knowIedge-based search aIgoriIhm's
perIormance, some sIaIisIicaI daIa were coIIecIed wiIh respecI Io
Ihe number oI nodes, pruned nodes, average branches oI non-IeaI
nodes and Iime cosI. These daIa were based on 15 experimenIs
wiIh a condiIion oI 20 sIaIions buI wiIh diIIerenI IerminaIs. AIIer-
wards, one new sIaIion was added and anoIher 15 experimenIs
were conducIed.
From Ihe observaIion oI Ihe experimenIaI resuIIs, Ihe IoIIowing
sIaIisIics are acquired. A Iarge proporIion oI Ihe nodes in Ihe search
Iree were pruned Ior acceIeraIing Ihe searching speed. The raIio re-
mains sIabIe, wiIh an average oI 0.71. The IoIaI number oI nodes in
a search Iree is direcIIv proporIionaI Io iIs depIh, and Ihe average
vaIue is 16.46. During Ihe searching Ior new sIaIions each Iime, Ihe
number oI candidaIe sIaIions is beIween 8 and 10, and 8.97 on
average. This iIIusIraIes IhaI Ihe candidaIes are suIñcienI Ior being
seIecIed. Supposing a non-IeaI node has 6 oIIspring on average,
Ihen 5 oI Ihem can be pruned, and onIv one mav be IeII Io be ex-
panded. The average compuIaIionaI Iime oI 150 experimenIs is Iess
Ihan 2 min and even in Ihe worsI siIuaIion is no more Ihan J min.
Based on Ihese sIaIisIics, our aIgoriIhm can be considered Io be
eIñcienI.
5.z. £xpcrimcnts on mctro linc nctworks
AparI Irom doing experimenIs Ior a singIe meIro Iine, Ihe
auIhors oI Ihis paper have aIso simuIaIed and experimenIed on
aII Ihe exisIing meIro Iines in Shanghai. The aim is Io veriIv Ihe
lig. 10. ExperimenIaI resuII oI pIanning meIro-Iine 4.
lig. 11. Some experimenIaI resuIIs consisIenI wiIh parIs oI exisIed meIro Iines.
550 H. Wci ct ul. / KnowlcJec-8uscJ Systcms z4 |z011) 54z-55J
lig. 12. The required consIrainI knowIedge caIegories and prioriIies when Ihe seII-design error is Iess Ihan 20° Ior each meIro Iine |a) Ihe reIaIive densiIv oI Shanghai's
whoIe meIro-neIwork.
lig. 13. The inûuence oI knowIedge Ior design accuracv in diIIerenI meIro neIwork densiIv.
H. Wci ct ul. / KnowlcJec-8uscJ Systcms z4 |z011) 54z-55J 551
inûuence oI Ihe muIIi-domain knowIedge in Ihe heurisIic search
and Ihe diversiIv beIween Ihe IheoreIicaI caIcuIaIion and reaIiIv.
The design oI Ihe heurisIic search considers Ihe IoIIowing IacIors:
|1) EnvironmenI and ecosvsIem |F1), e.g. Io avoid Ihe weIIand
in Ihe urban area;
|2) PopuIaIion densiIv |F2);
|J) InIerchange amongsI meIro Iines |FJ);
|4) InIerchange beIween meIro Iines and oIher modes oI Irans-
porIaIion |F4), Ior insIance, Ihe raiIwav sIaIion and airporI
are oI uImosI signiñcance as connecIion IargeIs;
|5) KouIe opIimizaIion |F5), e.g. avoid sharp Iurns and overIap-
ping wiIh Ihe major IraIñc sIreams on Ihe surIace;
|6) ConnecIiviIv oI imporIanI nodes |F6), e.g. parks, cuIIuraI
IaciIiIies and Ihe commerciaI cenIer;
|7) ConsIrucIion |F7), Ior insIance, uIiIizaIion oI exisIing
bridges, IunneIs and raiIwavs;
|8) UrbanizaIion |F8), Ior insIance, IaciIiIaIing Ihe urbanizaIion
oI Ihe counIrv and dispersing popuIaIion Io Ihe suburbs;
|9) Coverage Io areas wiIhouI meIro Iine access |F9).
The experimenIs were impIemenIed according Io Ihe order oI
Ihe esIabIishmenI oI Ihe meIro Iines in Shanghai. The origin and
desIinaIion are conñgured in prior and one or more oI Ihe above
resIricIive IacIors were seIecIiveIv appIied. The aIgoriIhm chooses
and joins Ihe middIe sIaIions and evenIuaIIv Ihe errors in diverse
condiIions are caIcuIaIed. II Ihe IocaIions oI Ihe reaI meIro sIaIions
are considered as Ihe IargeI nodes, Ihe mean square error beIween
Ihe experimenIs and Ihe reaIiIv is Ihe ñIIing error. Fig. 12 presenIs
IhaI Ihe simiIariIv beIween Ihe simuIaIion and Ihe reaIiIv is above
80°. Considering Ihe exIensions oI Line 1, Line 4, Line 10 and Line
2, in order Io achieve beIIer design accuracv, Fig. 1J iIIusIraIes Ihe
phiIosophv oI using Ihe resIricIive ruIes, Ihe corresponding order
and Ihe consequenI inûuence in diIIerenI neIwork densiIies.
According Io Ihe experimenIaI resuIIs, a random search can bareIv
achieve Ihe desired resuIIs. In Ihe iniIiaI phase oI consIrucIion, due
Io Ihe Iow neIwork densiIv, iI requires enormous consIrainI
knowIedge Io acquire a resuII cIose Io Ihe human design. When
Ihe neIwork is verv dense, iI requires reIaIiveIv Iewer knowIedge
caIegories; hence Ihe compuIaIionaI speed is enhanced.
6. Conclusion
ArIiñciaI inIeIIigence IechnoIogv can be used comprehensiveIv
in CIS appIicaIions. Bv Iaking advanIage oI accuraIe posiIion-
reIaIed inIormaIion processing, based on Ihe CIS pIaIIorm, and
Ihe sophisIicaIed knowIedge represenIaIion and reasoning
meIhods in AI, Ihe enhanced inIeIIigence CIS is capabIe oI soIving
more compIicaIed probIems compared Io common CIS appIica-
Iions. The programming meIhod is adepI aI caIcuIaIing precise
maIhemaIicaI IormuIaIions, so a maIhemaIicaI modeI can be
esIabIished wiIh knowIedge inIegraIed inIo Ihe CIS daIabase and
probIem soIving can be achieved Ihrough a maIhemaIicaI
programming meIhod. Besides maIhemaIicaI programming, iI is
conceivabIe IhaI manv Iechniques and meIhods in AI can be
inIegraIed inIo CIS Ior beIIer adapIabiIiIv and greaIer depIh oI
embedding Io achieve beIIer perIormance, which is worIhv oI
sIudv in IuIure work.
Acknowlcdgmcnts
This work was supporIed bv Ihe 97J Program |ProjecI No.
2010CBJ27900).
kclcrcnccs
|1| I.B. Mena, AuIomaIic vecIorizaIion oI segmenIed road neIworks bv geomeIricaI
and IopoIogicaI anaIvsis oI high resoIuIion binarv images, KnowIedge-Based
SvsIems 19 |2006) 704-718.
|2| Iia-song Wang, Bao-qing Zhao, Chun Ye, De-qing Yang, Zhen Huang,
OpIimizing IavouI oI urban sIreeI canvon using numericaI simuIaIion
coupIing wiIh maIhemaIicaI opIimizaIion, IournaI oI Hvdrodvnamics 18 |J)
|2006) J45-J51.
|J| IoseI BenedikI, SebasIian Keinberg, LeopoId KiedI, A CIS appIicaIion Io enhance
ceII-based inIormaIion modeIing, InIormaIion Sciences 142 |2002) 151-160.
|4| Frank WiIIox, ExperI svsIems in Iand-use pIanning: an overview, ExperI
SvsIems wiIh AppIicaIions 29 |2005) 4J7-445.
|5| Amjad Waheed, HojjaI AdeIi, Case-based reasoning in sIeeI bridge engineering,
KnowIedge-Based SvsIems 18 |1) |2005) J7-46.
|6| S.D. Kirkbv, InIegraIing a CIS wiIh an experI svsIem Io idenIiIv and manage
drvIand saIinizaIion, AppIied Ceographv 16 |4) |1996) 289-J0J.
|7| Yao-Min Fang, Li-Yu Lin, Chua-Huang Huang, Tien-Yin Chou, An
inIegraIed inIormaIion svsIem Ior reaI esIaIe agencv-based on service-
orienIed archiIecIure, ExperI SvsIems wiIh AppIicaIions J6 |2009)
110J9-11044.
|8| Wann-Ming We, An inIegraIed experI svsIem|operaIions research approach
Ior Ihe opIimizaIion oI wasIe incineraIor siIIing probIems, KnowIedge-Based
SvsIems 18 |6) |2005) 267-278.
|9| Wei Hui, Xu Oing-xin, Bai Yu, Lai Loi Lei, A knowIedge-based creaIion oI
maIhemaIicaI programming Ior CIS probIem soIving, Ceographic InIormaIion
Sciences 11 |2) |2005) 97-112.
|10| E.A. EIIis, P.K.K. Nair, P.E. Linehan, H.W. Beck, C.A. BIanche, A CIS-based
daIabase managemenI appIicaIion Ior agroIoresIrv pIanning and Iree
seIecIion, CompuIers and EIecIronics in AgricuIIure 27 |2000) 41-55.
|11| Iohn D. SainI, I. Desachv, A carIographic probIem soIving supporI svsIem in
geographicaI inIormaIion svsIem, in: InIernaIionaI Svmposium on Ceoscience
and KemoIe Sensing, ICAKSS '9J, 'BeIIer UndersIanding oI EarIh EnvironmenI',
voI. J, 199J, pp. 1547-1549.
|12| AbduIrezak Mohamed, Tahir CeIik, KnowIedge based-svsIem Ior aIIernaIive
design, cosI esIimaIing and scheduIing, KnowIedge-Based SvsIems 15 |J)
|2002) 177-188.
|1J| P. KIungboonkrong, M.A.P. TavIor, A microcompuIer-based svsIem Ior
muIIicriIeria environmenIaI impacIs evaIuaIion oI urban road neIworks,
CompuIer, EnvironmenI and Urban SvsIems 22 |5) |1998) 425-446.
|14| Shiuan Wan, Tsu Chiang Lei, A knowIedge-based decision supporI svsIem Io
anaIvze Ihe debris-ûow probIems aI Chen-Yu-Lan Kiver, Taiwan, KnowIedge-
Based SvsIems 22 |2009) 580-588.
|15| IIIikhar U. Sikder, KnowIedge-based spaIiaI decision supporI svsIems: an
assessmenI oI environmenIaI adapIabiIiIv oI crops, ExperI SvsIems wiIh
AppIicaIions J6 |2009) 5J41-5J47.
|16| Mare P. Vavssieres, MeI K. Ceorage, Linda Bernherm, IuIie Young, Kichard E.
PIanI, An inIeIIigenI CIS Ior rangeIand impacI assessmenI. Proceedings oI
FourIh AnnuaI ConIerence on AI, SimuIaIion, and PIanning in High AuIonomv
SvsIems, 'InIegraIing VirIuaI KeaIiIv and ModeI-Based EnvironmenIs', 199J,
pp. 109-115.
|17| Kichard E. PIanI, Marc P. Vavssieres, Combining experI svsIem and CIS
IechnoIogv Io impIemenI a sIaIe-IransiIion modeI oI Oak WoodIands,
CompuIers and EIecIronics in AgricuIIure 27 |2000) 71-9J.
|18| A.M. Buis, K.A. Vingerhoed, KnowIedge-based svsIems in Ihe design oI a new
parceIing, KnowIedge-Based SvsIems 9 |5) |1996) J07-J14.
|19| W. Wen, A dvnamic and auIomaIic IraIñc IighI conIroI experI svsIem Ior
soIving Ihe road congesIion probIem, ExperI SvsIems wiIh AppIicaIions J4 |4)
|2008) 2J70-2J81.
|20| E. BaiIIv, S. HavaI, D. IoIIv, eI aI., Command and conIroI oI auIomaIed subwav
in mode oI disrupIed march using Iuzzv Iogic, IournaI oI InIeIIigenI & Fuzzv
SvsIems 6 |J) |1998) J29-J4J.
|21| A. Chun, H. Wai, D. Yeung, W. Ming, KuIe-based approach Io Ihe vaIidaIion oI
subwav engineering work aIIocaIion pIans, in: Proceedings oI InIernaIionaI
ConIerence on CompuIing, CommunicaIions and ConIroI TechnoIogies, voI. 5,
2004, pp. 275-280.
|22| DanieI CharIebois, David C. Coodenough, A.S. |PaI) BhogaI, SIan MaIwin, Case-
based Keasoning and SoIIware AgenIs Ior InIeIIigenI ForesI InIormaIion
ManagemenI|C|. in: InIernaIionaI Ceoscience and KemoIe Sensing
Svmposium, ICAKSS '96, 'KemoIe Sensing Ior a SusIainabIe FuIure', voI. 4,
1996, pp. 27-J1.
|2J| Sam Subrammaniam, A.C. Hobeika, Dan Schierer, A new hvbrid experI-CIS Ior
wide-area incidenI managemenI, IEEE InIernaIionaI ConIerence on SvsIems,
Man, and CvberneIics, 'Humans, InIormaIion and TechnoIogv', voI. 2, 1994, pp.
1710-1715.
|24| Mehdi FaIIah-TaII, The appIicaIion oI arIiñciaI neuraI neIworks Io anIicipaIe
Ihe average journev Iime oI IraIñc in Ihe viciniIv oI merges, KnowIedge-Based
SvsIems 14 |J-4) |2001) 20J-211.
|25| K. EIizondo, V. Parada, L. Pradenas, eI aI., An evoIuIionarv and consIrucIive
approach Io a crew scheduIing probIem in underground passenger IransporI,
IournaI oI HeurisIics 16 |4) |2010) 575-591.
|26| L. IngoIoIIi, P. Tormos, A. Lova, Domain-dependenI disIribuIed modeIs Ior
raiIwav scheduIing, KnowIedge-Based SvsIems 20 |2) |2007) 186-194.
|27| W. Wen, An inIeIIigenI IraIñc managemenI experI svsIem wiIh KFID
IechnoIogv, ExperI SvsIems wiIh AppIicaIions J7 |4) |2010) J024-J0J5.
552 H. Wci ct ul. / KnowlcJec-8uscJ Systcms z4 |z011) 54z-55J
|28| Semive Demircan, Musa Avdin, S. Savas Durduran, Finding opIimum rouIe oI
eIecIricaI energv Iransmission Iine using J muIIi-criIeria wiIh O-Iearning,
ExperI SvsIems wiIh AppIicaIions J8 |4) |2011) J477-J482.
|29| B. Krausz, K. Herpers, MeIroSurv: deIecIing evenIs in subwav sIaIions,
MuIIimedia TooIs and AppIicaIions 50 |1) |2010) 12J-147.
|J0| D. AuberI, F. Cuichard, S. BouchaIa, Time-scaIe change deIecIion appIied Io
reaI-Iime abnormaI sIaIionarv moniIoring, KeaI-Iime Imaging 10 |1) |2004)
9-22.
|J1| L.Y. Li, W.M. Huang, Y.H. Cui, eI aI., SIaIisIicaI modeIing oI compIex
backgrounds Ior Ioreground objecI deIecIion, IEEE TransacIions on Image
Processing 1J |11) |2004) 1459-1472.
|J2| N. Kim, H.S. Lee, K.I. Oh, eI aI., ConIexI-aware mobiIe service Ior rouIing Ihe
IasIesI subwav paIh, ExperI SvsIems wiIh AppIicaIions J6 |2) |2009) JJ19-
JJ26.
|JJ| AboIghasem Sadeghi Niaraki, Kvehvun Kim, OnIoIogv based personaIized
rouIe pIanning svsIem using a muIIi-criIeria decision making approach, ExperI
SvsIems wiIh AppIicaIions J6 |2009) 2250-2259.
|J4| Keechoo Choi, Tschangho Iohn Kim, A hvbrid IraveI demand modeI wiIh CIS
And experI svsIems, CompuIer, EnvironmenI and Urban SvsIems 20 |4|5)
|1996) 247-259.
|J5| The OIñciaI WebsiIe oI Shanghai MeIro. AvaiIabIe aI: <hIIp:||
www.shmeIro.com> |IasI accessed aI J1.12.2010).
H. Wci ct ul. / KnowlcJec-8uscJ Systcms z4 |z011) 54z-55J 55J

You might also like