非线性时间序列组合预测模型研究

《非线性时间序列组合预测模型研究》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、分类号UDC密级y853684学位论文非线性时间序列组合预测模型研究作者姓名徐香坤指导导师姓名林树宽副教授东北大学计算机软件与理论研究所申请学位级别硕士学科类别工学学科专业名称计算机应用学位授予日期答辩委员会主席可每砌评阅人聚i净岣寸棚老}东北大学2005年12月东北大学硕士学位论文AbstractTheStudyofNon-linearTimeSeriesCombinationForecastModelAbstractNon—lineartimeseriesiSanattentionaIresearchfieldinrecentye

2、arsanditiSprovidedwithforegroundwhetherinmathematics，physics，lifescience，informaticsorineconomics，geoscience，astronomy．Todayabundantproductionisacquiredbystudyingonnon—lineartimeseries．AttheSfllTletimepeoplealsogetmanyforecastmethods．Butforcomplicatedapplicationcircumst

3、ance，itisunsatisfactoryusingsinglemethodtoforecast．Itisimportantquestiontobesolvedenhancingmodelprecisionofnon-lineartimeseriesundercomplicatedcircumstances．ThispaperissupportedbytheNatualScienceFoundationofLiaoningProvince，namely‘‘forecastmethodstudyofcomplicatedindust

4、ryobject'’，researchingcombinationforecastmodelinordertoenhanceforecastabilityundercomplicatedcircumstances．BPneuralnetworkhassuchcharacteristicsaslargescaleparallelprocessing，distributedinformationstorage，anti-jamming，self-organizationandself-adaptation，accordinglyithas

5、strongnon-linearapproachingandclassifyingidentifyability．Butitconvergesslowlyandexistslocalminimum．Thresholdauto—regressivemodelcanavailablydescribenon—lineardynamicsystemsthatexistcomplexphenomena．Infact，thresholdauto-regressivemodelhashigherapproachingprecision．Butiti

6、sunsatisfactoryforforecasting．Sothepaperproposesacombinationforecastmodelofnon—lineartimeseries．Firstly,constructsmonomialmodelsbasedonneuralnetworkandthresholdauto-regressionrespectively．Then，constructstheircombinationforecastmodel．Inthecourseofmodeling，mainlystudysthe

7、combiningmethodsofmonomiaimodels，includinglinearcombination，non—linearcombinationandhowtoascertaincombiningweights．Thepaperexperimentsontwotypicaltimeseriesandcomparestheresultsofthetwomonomialmodels，linearcombinationmodelandnon-linearcombinationmodel．Theexperimentalres

8、ultsshowthatcombinationmodelshavehigherprecisionsthanmonomialmodels，andtheaccuracyofnon—linearcombinationmodel