动态贝叶斯网络的灵敏性分析研究.pdf

《动态贝叶斯网络的灵敏性分析研究.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、计算机研究与发展DOI：10．7544／issn1OOO一1239．2014．20120609JournalofCo~mputerResearchandDevelopment5l(3)：536—547，2014动态贝叶斯网络的灵敏性分析研究姚宏亮张一鸣李俊照王浩(合肥工业大学计算机与信息学院合肥230009)(dmicyhl@163．com)ResearchonSensitivityAnalysisforDynamicBayesianNetworksYaoHongliang，ZhangYiming，LiJunzhao，

2、andWangHao(SchoolofComputerandformation，HefeiUniwersityoJTechnology，Hefei230009)AbstractSensitivityanalysisisanimportantmethodforresearchingthecharacteristicofcomplexsystem．Thepresentalgorithmsofdynamicsensitivityanalysisareusedtodealwithsomespecialkindsofdynami

3、cBayesiannetworkandtheircomputationcomplexityisveryhigh．Therefore，inordertohandlethesensitivityanalysisongeneraldynamicBayesiannetworkeffectively，anewalgorithmispresentedbasedonjunction—treealgorithm(DSAJT)．Thefunctionrelationsbetweenparametersand—conditionsprob

4、abilitydistributionofobjectnodeareestablishedbymessage～propagationonjunction—tree．DSA_JTcanlowerexponentialtimeandreducecalculationsignificantly．Butitscomputationcomplexityisstillhigh．Inordertofurtherimprovethecomputationalperformance，DSABKis—introducedbasedonDS

5、A_JTalgorithm．DSA—BKintroducesthefactorideawiththeproductoftheprobabilityofthesubsystemstoapproximatetheprobabilityofthewholesystem．DSABKreduces—thesizeoftherootbythelocalmarginalisationofinterfaceandupdatingthejointprobabilitydistributionofmode1．ComparedwithDSA

6、JT，DSABKcanlowerthecomputationalexponential——timesfurtherandsignificantlyimprovesthecalculationefficiency，andtheerrorisboundtobeproved．Then，basedonabstractingtheprocessofthetwoalgorithms，thecomputationformulasofdynamicsensitivityfunctionareproved，anditisshowntha

7、tthetwoalgorithmscaneffectivelydealwithsensitivityanalysisofgeneraldynamicBayesiannetwork．Finally，theeffectivenessisprovedintheexperimentonthenetworkoftheShanghaiStock．KeywordsdynamicBayesiannetwork；sensitivityanalysis；DSABKalgorithm；DSAJTalgorithm；——dynamicsens

8、itivityfunction摘要灵敏性分析是研究复杂系统特性的一种重要方法．现有动态灵敏性分析方法都是针对特定类型的动态贝叶斯网络且计算复杂度高．为了对一般动态贝叶斯网络的灵敏性进行有效分析，提出了一种基于联合树的动态灵敏性分析算法(DSA_JT)，DSA_JT算法构建动态网络的联合树，通过消息传播建立参数与目标结点的条件概率分