基于神經(jīng)遺傳學習算法的模型優(yōu)化研究.pdf

1、天津大學碩士學位論文X849001基于神經(jīng)遺傳學習算法的模型優(yōu)化研究StudyoftheModelingOptimizationBasedonNeuralNetworkandGA專業(yè)：管理科學與工程研究生：杜健指導教師：李波副教授天津大學管理學院2005年1月ABSTRACTTheEnsembleLearningisahotspotintheintelligentlearningfieldIthasbeenpaidattentionfr

2、omits’verybeginningandhasbeenusedsuccessfullyinmanyfieldsbecauseofits’goodperformanceandusabilityThecombinationofEnsemblealgorithmsandunstablelearnershasshownimpressiveeffectbothinaccuracyandstabilityAtthesametime，parthe

3、nogeneticalgorithmisahotspotduetoits’maturityItisadoredbySOmanyresearchersduetoits’abilitytosolvecombinationproblemsespeciallyproblemsinlargescaleBasedonthealgorithmsdiscussedabove，thisarticledoessomeusefulstudyinsomeasp

4、ectsSomethingWasdoneasbelow：(1)SomeexperimentsweredonetotestifythevalidityoftheBoostingalgorithmbasedonthestudyoftheEnsemblealgorithmTheexperimentswerebasedontheUCIdatabase，andthepurposeistoprovethegoodperformancesofBoos

5、tinginsomeaspectsas：Serviceability；StabilityTheabilityofparameteroptimizationAtlast，wediscusstheServiceabilityand“OverfittingProblem”fromthepointofMargin，andshowitinintuitionisticgraphics(2)Basedonthestudyabove，weputforw

6、ardanewmethodtoulteriorlyoptimizeparametersofBoostingalgorithmHerethemethodsweusedare：GeneticalgorithmandLSEalgorithmAndthesimulationresultsshowthatbothmethodscarlimprovetheperformancemo∞(3)Astoparthenogeneticalgorithm，w

7、ebringforwardanimprovedalgorithmbasedonparthenogeneticalgorithmtosolvetheCSPproblemAnewmethodwasbroughtforwardtosimpli匆thecodingprocess，atthesametime，anoperatorwasintroducedtoheIptofindthebestglobalsolutionAndthealgorith

8、mhasbeentestifiedtobeaexcellentonethroughthetestinmorethanoneindustrialenterprisesAllproofsshowthatitisagoodmethodtosolvetheCSPproblembothintheCuttingAccuracyandCalculating—EfficiencyKeyWords：Ensemblelearning，ParthenoGen