支持向量机算法研究及其在智能交通系统中的应用

《支持向量机算法研究及其在智能交通系统中的应用》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、I￡yIflZo⋯≈⑨天津大謦中目*代第一所^{硕士学位论文__用啊啊N日

2、喁t舟I■啊a__学科专业：系统工程作者姓名：韦忠升指导教师：马寿峰教授干津十堂{#窑毕瞬2006年6月ABSTRACTTheTraditionalStatisticsisagradualtheorywhichtheamountofspecimenistendingtoinfinite，whileStatisticalLearningTheory(SLT)isatheoryattheconditionofsmallspecimenamount．Since

3、thespecimenamountisalwayslimitedinreality,SLThasitstheoreticaladvantage．BasedonSLTandStructure砌skMinimum(SgM)principle，SupportVectorMachine(SVM)isnotonlysimpleinstructure，butalsohas900dgeneralizationability．SoSVMhasgraduallybecomeoneofhotspotissuesinthefieldofmachine

4、learning．Withtherapiddevelopmentofcommunication．informationscience&electronicengineeringandcomputertechnology,IntelligenceTransportSystem(ITS)hasbeenattachedextensiveimportanceincreasingly,thusputforwardagreaterdemandontechnologiesinmanyfields，suchasrecognitionofvehi

5、cle-typeandvehicle—plate，predictionoftrafficflowete．Thepurposeofthisarticleisto，basedontheSVM，conductresearchonbasicmethodsofmodelidentificationandregressiveanalysis，andtheirapplicationstoITS．Thisarticleincludesthefollowingmaintasksandexistingachievements：I．Basedonth

6、eProximalSVM，thisthesisdrewupontheideaofincrementallearningalgorithmandthesolutionofimbalanceintrainingspecimenamount缸differentsorts，laterdefinedanewkindofweiightedcoefficientmatrix，putforthamulti—classclassificationalgorithmofbalancedandincrement．Experimentalresults

7、haveshowedthatthisalgorithmhasfairlyhighstabilityandclassificationaccuracy．2．Byconductingresearchonseveralmulti-classclassificationalgorithms，summarizingtheirmeritsanddemeritsandcombiningwithtechnologyofHuffmanTree，thisdissertationbroughtforwardamulti-classclassifica

8、tionalgorithmwhichisbasedonHuffmanDecisionTree．Lateronconductedsimulativeexperimentsonmanyopeneddatabases，haveprovedthatalgorithmisnotonlywithrelativelyhighclassificationaccuracy,butalsowithveryhightrainingefficiency．3．Firstly,weconductedrelativelysystematicresearcho

9、ntheoryandrealizationofSupportVectorRegression，thenanalyzedthemeritsanddemeritsofv—svg,andfurtherputfonllthev-ESVRalgorithm．Finally