船舶横摇变参数LSSVM在线预报方法.pdf

《船舶横摇变参数LSSVM在线预报方法.pdf》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、第16卷第9期船舶力学Vol·16No-92012年9月业!!璺!堕墅里竺!塑坐!!!：婴!!———————————————————————————————————————————————————————————————_^————————————————————————一一一一ArticleID：1007—7294(2012)09—1024-11OnlinePredictionofShipRollingbasedonVaryingParametersLSSVMLIUSheng，YANGZhen(C011egeofAutoma

2、tion，HarbinEngineeringUniversity，Harbin150001，China)Abstract：1nordertoimprovetheaccuracyandreal—timenatureofpredictionmodelofshiproilingmotion。anonlinereal-timepredictionmethodispresented，whichcombineschaostheoryandleastsquaressuppoftvectormaehine(LSSVM)．Aimingatthe

3、problemthatthefixedparameterofforecast—ingmodelcannotbeadaptivelyadjustedwithdynamicchangeofshiprollingmotion，avaryingpa—rameteronlinemodelingmethodisproposedbasedonLSSVM．ThreeLSSVMsareusedtomodelpar—a11e11vandthewholepredictionprocessesaredividedintoaninitialstagea

4、ndseveralpredictionstages．ThenextpredictingLSSVMisselectedattheendofeachstage，atthesametime，thekernelDaran】etersoftheothertwoLSSVMsareresetaccordingtoheuristicrules，whichareusedascompar。ativeLSSVMsforthefollowingpredictingstages．Theexperimentsofshiprollingtimeseries

5、predic‘tionaremade．Thesimulationresultsindicatethatreal-timepredictionroot—mean—squareerroroftheproposedmethodisabout6．85％，whichhasbetteradaptabilitycomparedtofixedparameterpredic‘tinT1method．Keywords：ship；roll；chaos；leastsquaressupportvectormachine(LSSVM)；varyingpa

6、rameter；forecastingCLCnumber：U661．32Documentcode：A1IntroductionRecentlv，supportvectormachinetheorydevelopsrapidlyinapplicationofshipmotionrood—elingfield0<．Thereasonisthatsupportvectormachinecanobtainglobaloptimalsolution，pos—sessgoodgeneralizationabilityandprovidep

7、owerfulmethodsandtoolsforpracticalproblemswhicharesmallsampleandnonlinear．OnlinelearningalgorithmofLSSVM【qcantrackdynam。iccharacteristicsoftime—varyingnonlinear，whichisconsequentlyappliedtopredictionofshiDmotion．However，hyper—parametersofonlinetrainingalgorithmarese

8、tsubjectivelyandallthesamplesusethesamehyper—parameters，whichmeanthathyper—parameterscannotadjustautomaticallyassampleschange．Changecharac