基于小波神经网络的导航风险评估(IJEME-V1-N4-4)

I.J.EducationandManagementEngineering2011,4,20-27PublishedOnlineOctober2011inMECS()DOI:10.5815/ijeme.2011.04.04Availableonlineat’sRepublicofChina,Nanjing,ChinabSchoolofEnergyandPowerEngineering,WuhanUniversityofTechnology,Wuhan,ChinaAbstractFactorsthatinfluencethesecurityofnavigationhavethecharacteristicsofdynamics,randomness,uncertaintyandmutualinfluence,sothesystemofnavigationsecurityisatypicalnon-linearsystem.Duetothelimitationsoftraditionalmathematicalmethodssolvingthenon-linearsystem,thisarticlecombinesartificialneuralnetworkmodelandwaveletanalysistoconstructthewaveletneuralnetworkmodelforthenavigationriskassessment,andputsforwardtheindexsystem,weightscalculationandalgorithmwithexamples.Resultsshowsthatthewaveletneuralnetworkmodelforevaluatingcanbesolvedthedefectssuchassubjectivearbitrarinessandthefuzzyofconclusionsexistingintraditionalevaluationmethods,anditcandobetterthanthecommonartificialneuralnetworksinfittingaccuracyandconvergencerate,sotheapplicationofwaveletneuralnetworkmodelhasanimportantvalue.IndexTerms:Navigationsecurity;riskassessment;waveletneuralnetwork©2011PublishedbyMECSPublisher.Selectionand/orpeerreviewunderresponsibilityoftheInternationalConferenceonE-BusinessSystemandEducationTechnology1.IntroductionWiththeeconomicglobalizationandthegraduallyformingofinternationalmarket,seatransportationasaneconomictieconnectingnationsplaysagreatroleinpromotingtheprocessofglobaleconomicintegration.Thus,thedevelopmentofmaritimeindustry,theexpandingofports’scale,theconstructionofcross-sea(river)bridges,thegreatincreaseinthenumbersandthetonnageofships,andalsotheextremelyweathercausethesituationofnavigationsecurityincreasinglycomplex,andespeciallytheseriousriskeventsandpollutionhavethegreatpotentialtooccur.Sostrengtheningthestudyofthesecurityofnavigationcanimprovethenavigationorder,reducetheprobabilityofshipaccidents,regulatethebehaviorofnavigationandprovidethetheoreticalsupporttothefurtherimprovementofchannelthatiscrucialtoensurenavigationalsafetyandthepreventionofwaterpollution.Atpresent,themethodsofriskassessmentfornavigationaremainlyfocusedonthreeaspects:Firstly,determinethesecurityaccordingtotheriskassessmentcriteriaofnavigationriskfactors;secondly,analyzewhetheraparticularchannelispronetohaveaccidentsthroughtheanalysisofmaritimestatisticaldata;thirdly,TheNavigationRiskAssessmentUsingWaveletNeuralNetwork21judgewhetheraparticularchannelisdifficulttonavigateusingtheexperienceofthedrivingpersonnelandthepilot.But,infact,thesecurityofnavigationisamultiplesystemwhichisdecidedbymanyuncertainfactors,suchastheflowofship,thescaleofwaterwayandhydrologicalconditions,sousingonlytheoreticalanalysistosolvetheproblemcan’tachievesatisfactoryresults.Furthermore,thelackofstatisticaldatacouldresultintheimperfectionintheconstructionthemodelandthelackofaccurateestimateofartificialjudgment,wouldnotreallyreflecttheinfluencefactorsandwouldincreasesthedifficultiesinriskevaluation.Thenavigationsecurityisavaguesystem,andhowtocarryouttheeffectiveevaluationoftheriskofnavigationisstillinexploration.Navigationsecurityisdecidedbymanyfactorssuchasthechannelgeometryanditsdepth,thedensityoftraffic,theaidednavigationalequipments,thechannelenvironment,theweatherconditionandthehydrology,theships’typesandscale,thetypesofaccidentandthetrafficmanagementwhichhavethecharacteristicsofuncertainty,randomnessandfuzzy.Sothesystemofnavigationsecurityisatypicalnon-linearsystem.2.ModelSelectionandConstructionAstonon-linearDynamicsprocess,whetherthetraditionalsafetyevaluationmethods,Greysystemevaluationmodel,Fuzzycomprehensiveassessmentmodel,orAnalytichierarchyprocess,areunabletosolvethefollowingproblems:Firstly,thesemethodscan’tcarryoutthereal-timeanalysis;Secondly,thepre-determinedfactorsandtheresultsaremappedtobelinear,soitcan’treflectthecharacteristicsofstudysystem;Thirdly,oncetheweightsofevaluationfactorsaredetermined,theywouldnotvary,butthesystemisinconstantchangeanddevelopmentprocess;Fourthly,theyaredifficulttogetridoftheuncertaintyofsubjectiveunderstandingandambiguityofparticipatorsintheprocessofsafetyevaluation.ArtificialNeuralNetwork(ANN)modelcanovercometheinsufficienciestoacertainextent,butithasthedisadvantageofslowconvergencerate,andeasytofallintoaverysmallvalue.Accordingtotherelatedsimulationresults,theseriesaredecomposedbyWavelethavingabilitytotranslateandstretch,andcangetmoreapproximationofnon-linearfunctions.Therefore,thecombinationofWaveletandNeuralNetworkcoulddobetterintheabilitytoapproximation,analysisofnon-stationarysignals,faulttolerance,andconstructionofnon-linearfunctionmodelthannormalNeuralNetwork.ThisarticleattemptstobuildtheWaveletNeuralNetwork(WNN)modelbyMorletWaveletsBase.ThefollowingprocessesarecrucialtotheconstructionofWaveletNeuralNetworkmodel.A.ConstructionofthetransferfunctionofWaveletneuralnetworkIfasquareintegralfunctionsatisfiestheconditions[1]