基于模糊神经网络的船舶避碰研究

江苏科技大学硕士学位论文基于模糊神经网络的船舶避碰研究姓名：陈建华申请学位级别：硕士专业：控制理论与控制工程指导教师：陈红卫20080101(FNN)DCPATCPABP,,ANFISANFISANFISANFISI(FNN)DCPATCPAANFISANFISAbstractAsthevesseltrafficisbecomingmoreandmorepopular,theseaaccidentsalsohaveincreasedsharplyinrecentyears.Mostofthemareduetoman-madereason,forinstance,itmanipulatesinappropriately.Sotheintelligentresearchofvesselcollisionavoidancehaspracticalworthinessandimportantmeaning.Thefuzzyneuralnetwork(FNN)isemergingasoneoftheprimaryresearchareasrecently,whichistheorganicintegrationoffuzzyinferenceandtheneuralnetwork.Thefuzzyinferencesystemitselfdoesn’thavethelearningability,buttheartificialneuralnetworkcannotexpressinferencefunctionofhumanbrainwithfuzzylanguage.Thefuzzyneuralnetworkgivesthepossibleexplanationtoweightofneuralnetworkandformsstructureoftheneuralnetworkinadvance.Therefore,itcanmaketheneuralnetworkeasytobeunderstoodbypeople,andhasgoodself-adaptability,stronggeneralization,andbetterrobustness.Itiswidelyusedtotheprediction,theintellectualcontrol,thepattern-recognition,andsoon.Firstofall,weanalyzedthefundamentalsofvesselcollisionavoidance.Andwemainlyfinishedfollowingworks:analyzedtheprocessesofshipcollisionaccidentsanditsavoidance;threeencountersituationsaredividedquantitativelyandthesituationsofcollisionavoidanceactionarealsodivided;accordingtovaryinginformationoftargetshiporientationanddistanceandinformationofspeedandcourseinbothownshipandtargetship,webuiltthemathematicmodelsofDCPA(distanceoftheclosestpointofapproach)andTCPA(timetotheclosestpointofapproach);proposedcollisionriskconceptsandintroducedthefactorsthataffectshipcollisionrisk.Then,basedonfuzzyinferencetheoryresearch,wediscussedfuzzyqualityinshipcollisionavoidance,chosethefactorsthataffectshipcollisionrisk,confirmedmembershipsofinfluencingparameters,andusedexampletoilluminatetheapplicationofevaluatingcollisionrisk.Thirdly,discussedspeedinessBPneuralnetworkmethodofthecollisionrisk,andresultswereidealbythesimulation.Intheend,weintroducedfuzzyneuralnetworkapplicationsaboutvesselcollisionavoidance.Thedetailsareasfollowings:evaluationofcollisionriskbyfuzzyneuralnetwork.Weproposedakindoffuzzyneuralnetworkmodel.Aftertheexamplewassimulatedandcomputed,theprecisionofoutputresultswerehighandmatchedthepractice.Sothismethodhasmoreadvantagethansimplefuzzyinferencesystem.vesselcollisionavoidanceinferencesystembasedonfuzzyneuralnetwork.Weproposedakindoffuzzyneuralnetworkmodelthatconsistthreenetworks.researchonANFISmethodaboutshipcollisionrisk.Wenotonlycomparedwithsimplethreelayerneuralnetworkandknewitsadvantage,butalsoasortofimprovedANFISmethodwasproposedandcametosomefurthercontentedresults.Theresearchresultsindicatedthatfuzzyneuralnetworkcouldbeeasilyusedinthevesselcollisionavoidance.Besides,itnotonlypromotesthedevelopmentofintelligentcollisionavoidancesystem,butalsoprovidesagoodkindofmethodforcollisionavoidanceresearch.Keywords:vesselcollisionavoidance,fuzzyneuralnetwork,vesselcollisionrisk,fuzzyinference,ANFIS.IIAbstractAsthevesseltrafficisbecomingmoreandmorepopular,theseaaccidentsalsohaveincreasedsharplyinrecentyears.Mostofthemareduetoman-madereason,forinstance,itmanipulatesinappropriately.Sotheintelligentresearchofvesselcollisionavoidancehaspracticalworthinessandimportantmeaning.Thefuzzyneuralnetwork(FNN)isemergingasoneoftheprimaryresearchareasrecently,whichistheorganicintegrationoffuzzyinferenceandtheneuralnetwork.Thefuzzyinferencesystemitselfdoesn’thavethelearningability,buttheartificialneuralnetworkcannotexpressinferencefunctionofhumanbrainwithfuzzylanguage.Thefuzzyneuralnetworkgivesthepossibleexplanationtoweightofneuralnetworkandformsstructureoftheneuralnetworkinadvance.Therefore,itcanmaketheneuralnetworkeasytobeunderstoodbypeople,andhasgoodself-adaptability,stronggeneralization,andbetterrobustness.Itiswidelyusedtotheprediction,theintellectualcontrol,thepattern-recognition,andsoon.Thedissertationmadeuseoftheadvantagesoffuzzytechnologyandneuralnetwork,combinedthemtogethertoconstructthefuzzyneuralnetworkmodelforshipcollisionavoidance,andcametosomefurthercontentedresults.Firstly,weanalyzedcollisionavoidancetheory,theprocessofvesselcollision,andthemathematicmodelsofDCPA(distanceoftheclosestpointofapproach)andTCPA(timetotheclosestpointofapproach).Secondly,weutilizedfuzzytechnologyandneuralnetworkrespectivelytodeterminethedegreeofcollisionrisk.Thirdly,itproposedafuzzyneuralnetworkmodelsforshipcollisionriskbycombiningtheadvantagesoffuzzytechnologyandneuralnetwork,andthemethodwasprovedtobeeffectivebyrealisticexperiments.Anditintroducedafuzzyneuralnetworkmodelsforshipcollisionavoidance.Finally,itbuiltanANFISmodelforshipcollisionriskandimprovedit.Theresearchresultsindicatedthatfuzzyneuralnetworkcouldbeeasilyusedintothevesselcollisionavoidance.Besides,itnotonlypromotesthedevelopmentofintelligentcollisionavoidancesystem,butalsoprovidesagoodkindofmethodforcollisionavoidanceresearch.Keywords:vesselcollisionavoidance,fuzzyneuralnetwork,vesselcollisionrisk,fuzzyinference,ANFIS.11.11898191080%(ARPA)(AIS)1.21.2.121983DCPATCPA,[]]DCPATCPA[2]