基于神經(jīng)網(wǎng)絡算法的證券預測應用研究.pdf

1、桂林理工大學碩士學位論文基于神經(jīng)網(wǎng)絡算法的證券預測應用研究姓名：侯桂云申請學位級別：碩士專業(yè)：計算機應用技術(shù)指導教師：陳曉輝20090401桂林理工大學碩士學位論文ABSTRACTWiththeeconomicgrowthandtheconversionofpeople’Sinvestmentconsciousness，stockinvestmenthasbecomeanimportantpartofpeople’Slifeinmode

2、mtimeStockforecasthasgreatlybeenoneoffocusesofpublictopicTheproceedsofstockinvestmentalwaysequaltheriskThatmeansthegoodproceedsisbasedonthehigllriskoffailureSoestablishingastockforecastingmodelwhichhashigheroperationrate

3、andprecision，hastheoreticalsignificanceandapplicablevalueThecomplexityofinsidestructureandlevityofexteriorcomplicationinsystemofstockmarketmakestockmarketpredicationacomplexproblemThetraditionalmethodsandtoolshavenotmeti

4、tsrequirementThisthesisanalysesthemainproblemsbeingexistentintheprocessofstockmarketpredictionandcomparesvariousstockforecastingmethodsThefeasibilityofforecastingstocktrendbyusingBP(BackPropagation)neuralnetworkisdiscuss

5、edStockmarketisaverycomplexnonlineardynamicsystem，whichthetrendlookslikedisorderlybutithasinternaldisciplinarianactuallyandneuralnetworkhasthecapabilityofapproximatinganynonlinearsystemandthespecialityofselflearningandse

6、lfadaptingInordertoavoidlocalminimumandpromoteconvergencespeed，LevenbergMarquardtBPalgorithmhasbeenadoptedTheexperimentprovesthatthemethodofmodelingstockmarketusingneuralnetworkhasasatisfyingresultinstockpredictionThethe

7、siscombinegeneticalgorithmwithBPneuralnetworkandfindthemodelofgeneticBPneuralnetworkjustbecausetheshortcomingsofbasicBPalgorithmsincludingtheslowconvergencespeedandlocalminimumnegeneticBPalgorithmcarriesthroughwhole—spac

8、esearchusinggeneticalgorithmThegeneticalgorithmpaysattentiontounknownareasearch，whichhashi曲speed，relativelowprecision，CannotgetintolocalminimumTheBPalgorithmsearchestheareaincludingwholespaceoptimalvalueItcallimprovespee

9、dandprecisionTheoreticalanalysisandexperimentresultshowthatthemethodofstockpredictionusingneuralnetworkisfeasibleandefficientItprovesthatgeneticBPalgorithmCangreatlyimprovethereliabilityandspeed，whichhasfavorableforegrou

10、ndAtlast，thisthesisresearchesvariousparametersthatinfluenceontheresultsofstockprediction，whichparticularlydiscussestheproblemsincludingthetopologicalstructuresofnetwork，thenumberofhiddennodes，theselectionandpretreatmento