基于流的特征的P2P网络业务识别

《基于流的特征的P2P网络业务识别》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、问是加分免别行应用于神经网络模型，将表示网络结构复杂性的惩罚项引入目标函数中，在训练优化的过程中降低网络结构的复杂性，贝叶斯推理着眼于整个权空间中的概率分布，得出目标函数的最优化参数，达到提高泛化能力的目的。最后进行仿真，结果表明基于贝叶斯正则化BP神经网络的P2P业务识别方法的泛化能力要优于基于传统BP神经网络的P2P业务识别方法，可大大提高前者在实际中识别P2P业务的准确率。文章还对数据样本集大小对识别结果的影响进行了分析，得出如果减d,N练样本数，则同时会降低识别精确性。关键词：P2P，业务识别，流特征，神经网络，贝叶斯正则化，泛化能力▲pJ南京邮电大学

2、硕士研究生学位论文ABSTRACTCurrently,P2P的伍cisoccupyingalligherproportionintheIntemet．Accordingtothestatistics，P2PU'RtjfiCaccountsfor60％-70％innetworks盯vice．WhenP2PtlR伍cbringsconveniencetoIntemet嘲，italsobringsmanynegativeeffects，suchasbandwidth,copra班电securityproblemsand．TheidentificationofP2Pt

3、ral矗chasbocomeanmorourgentlyissue．ThegeneralP2Ptl斌icidentificationmethodsareportidentificationmethod,signaturematchingmethod．ButwiththedevelopmentofP2Ptechnology,P2Ptra伍cusingdynamicportandencryptiontechnology,mal【ingtheabovetwomethodsc趾notadapttoP2Ptechnology．P2Ptral矗cidentificatio

4、nbasedonfeaturesofnetworkflowsavoidsproblemsencounteredbytheabovetwomethods．ThismethodanalysisesfundamentaldifferencebdhwⅪnP2PtlR伍CandotherS训CCSinInternet，basedontheseessentialcharacteristicstoidentifyP2Ptl"R伍C．Firstly,thisarticleintroducestheconceptofP2Ptechnology，principlesandarch

5、itecUuqandgRIdicsthevarious懿i鲥丑gP2Ptrafficidentificationmethods．Thenintroduc鹤theneuralnetworksappliedfortheP2Ptra伍cidentification,andanalysisesthefactorsimprovingtheg嫩既ali刎onabmtyofneuralnetwork,ONthisbasisthisarticleputsforwardbayesianregularizationalgorithmtOimprovetheneuralnetwor

6、kgeneralizationability．Thisalgorithmjoinsthepenaltyitemindicatingnetworkcomplexityintheobjectivefunction，thepenaltyitemcallreducethecomplexityofnetworkstructureinthetrainingandoptimizingprocess，bayesiandeducefocusontheweightsprobabilitydistributioninthewholeweightsspace，theoptimalne

7、tworkparametersCanbeobtainedwiththemaximumposteriorprobabilitiesofthemodel．Therefore，theBayesianneuralnetworkimprovesthegeneralizationabilitytheoretically．Finally,simulationresultsrevealsthat：GeneralizationabilityofP2PtrafficidentificationmethodbasedonbayesianregularizedBPneuralnetw

8、orksuperiortotheP2P