铁路平交道口的事故预测模型

AccidentAnalysisandPrevention38(2006)

346–356

Accidentpredictionmodelforrailway-highwayinterfaces

JutaekOha, ,SimonP.Washingtonb,1,DooheeNama,2

b

TheKoreaTransportInstitute,Ilsan,Koyang-city,Kyeonggi-do411-701,SouthKorea

DepartmentofCivilandEnvironmentalEngineering,ArizonaStateUniversity,Tempe,AZ85287-5306,USA

Received15April2005;receivedinrevisedform29September2005;accepted14October2005

a

Abstract

Considerablepastresearchhasexploredrelationshipsbetweenvehicleaccidentsandgeometricdesignandoperationofroadsections,butrelativelylittleresearchhasexaminedfactorsthatcontributetoaccidentsatrailway-highwaycrossings.Between1998and2002inKorea,about95%ofrailwayaccidentsoccurredathighway-railgradecrossings,resultingin402accidents,ofwhichabout20%resultedinfatalities.Thesestatisticssuggestthateffortstoreducecrashesattheselocationsmaysigni cantlyreducecrashcosts.

Theobjectiveofthispaperistoexaminefactorsassociatedwithrailroadcrossingcrashes.Variousstatisticalmodelsareusedtoexaminetherelationshipsbetweencrossingaccidentsandfeaturesofcrossings.ThepaperalsocomparesaccidentmodelsdevelopedintheUnitedStatesandthesafetyeffectsofcrossingelementsobtainedusingKoreadata.

Crasheswereobservedtoincreasewithtotaltraf cvolumeandaveragedailytrainvolumes.Theproximityofcrossingstocommercialareasandthedistanceofthetraindetectorfromcrossingsareassociatedwithlargernumbersofaccidents,asisthetimedurationbetweentheactivationofwarningsignalsandgates.Theuniquecontributionsofthepaperaretheapplicationofthegammaprobabilitymodeltodealwithunderdispersionandtheinsightsobtainedregardingrailroadcrossingrelatedvehiclecrashes.©2005ElsevierLtd.Allrightsreserved.

Keywords:Railroadcrossings;Poisson;Gamma;Negativebinomial;Internationalcomparison

1.Introduction

Therelationshipsbetweenvehicleaccidentsandgeometricdesignofroadsections,suchashorizontalcurvature,verticalgrade,lanewidth,andshoulderwidth,havebeenextensivelystudiedusingregressionmodels(Abbessetal.,1981;Haueretal.,1988;PersaudandDzbik,1993;Kulmala,1995;PochandMannering,1996;Lord,2000;Ivanetal.,2000;Lyonetal.,2003;MiaouandLord,2003;Ohetal.,2003).Asimilarlylargeliteratureexistsforresearchonroadwayintersections.Relativelylittleresearch,however,hasbeencarriedouttounderstandandidentifyfactorsthatcontributetoaccidentsonrailway-highwaycrossings(RHXs).Becauseofthelackofdetailedinformationoncrossingelementdataandthefailureofselectingappropri-atetoolsforanalyzingthedata,statisticalmodelsexplainingtherelationshipsbetweenroadwaygeometry,gradecrossingchar-Correspondingauthor.Tel.:+82319103174;fax:+82319103235.

E-mailaddresses:jutaek@koti.re.kr(J.Oh),Simon.Washington@asu.edu(S.P.Washington),doohee@koti.re.kr(D.Nam).1Tel.:+14809652220,fax:+14809650557.2Tel.:+82319103092,fax:+82319103235.

0001-4575/$–seefrontmatter©2005ElsevierLtd.Allrightsreserved.doi:10.1016/j.aap.2005.10.004

acteristics,andcrossingaccidentfrequencieshaverarelybeendeveloped.Therefore,researchgapsremainregardingtheiden-ti cationoffactorsassociatedwithcrashesatRHXs.

Thispaperseeksto llsomeoftheknowledgegapsregardingcomplexrelationshipsbetweencrashesinvolvingmotorvehi-clesand/ortrainsthroughtheapplicationofstatisticalmodelstocrashdata.Thecontrastsandtradeoffsbetweenvariousproba-bilitymodelsarediscussedwiththemainfocustoprovidethegreatestinsightintoRHXrelatedcrashes,especiallyinKorea.InKorea,about95%ofrailroadaccidentsareassociatedwithRHXs,andatotalof402accidentsoccurredattheselocationsfrom1998to2002,ofwhichabout20%resultedinfatalities.Thesestatisticssuggestthateffortsareneededtodevelopeffec-tivecountermeasurestoreducecrossingaccidents.However,consideringthefactthatabout1800crossingsarelocatedinKorea,80accidentsperyearsuggestthatforaperiodof5yearsmostofthecrossingsobservedzeroaccidents.

Statisticalmodelsareusedtoexaminetherelationshipsbetweencrossingaccidentsandfeaturesofcrossings.However,manypaststudiesilluminatingthenumerousproblemswithlinearregressionmodels(JoshuaandGarber,1990;MiaouandLum,1993)haveledtotheadoptionofmoreappropriate