資源描述:
《基于svm梅雨量預(yù)測(cè)的研究》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。
1、摘要㈣刪㈣『『f『刪Y18915Iltll1IIIf。17llUl近100多年來全球氣候變化日益劇烈,各種天災(zāi)接踵而至,夏季旱澇災(zāi)害是其中重要災(zāi)害之一。長(zhǎng)江中下游梅雨期資料對(duì)度量當(dāng)?shù)叵募竞禎呈且豁?xiàng)主要指標(biāo),因此對(duì)梅雨總量的預(yù)測(cè)研究對(duì)研究長(zhǎng)江中下游夏季旱澇情況有重要的意義。如何通過觀測(cè)有限個(gè)歷史樣本建立模型實(shí)現(xiàn)預(yù)測(cè)是尋找最優(yōu)預(yù)測(cè)模型的重要工作。統(tǒng)計(jì)學(xué)習(xí)理論是針對(duì)小樣本情況下的機(jī)器學(xué)習(xí)理論,其核心思想是通過控制學(xué)習(xí)機(jī)的復(fù)雜度實(shí)現(xiàn)對(duì)其推廣能力的控制。在這一理論下發(fā)展起來的支持向量機(jī)(SupportVectorMachines
2、,SVM)以VC維ⅣCDimension)和結(jié)構(gòu)風(fēng)險(xiǎn)最小化原貝JJ(StructuralRiskMinimization,SRM)為基礎(chǔ),解決了小樣本、過學(xué)習(xí)、非線性、高維數(shù)、局部小等許多實(shí)際問題。時(shí)間序列預(yù)測(cè)是智能計(jì)算中主要研究課題之一。本文主要研究的重點(diǎn)是根據(jù)近106a(1885.1990年)長(zhǎng)江中下游沿江梅雨期的梅雨總量數(shù)據(jù)和49a(1954.2002年)泰州地區(qū)梅雨量數(shù)據(jù),分別建立徑向基核函數(shù)、多項(xiàng)式核函數(shù)的時(shí)間序列支持向量機(jī)(SVM)回歸模型,并采用網(wǎng)格尋優(yōu)參數(shù)函數(shù)、遺傳算法、粒子群優(yōu)化算法對(duì)模型的參數(shù)分別進(jìn)
3、行優(yōu)化,然后對(duì)這六種模型的預(yù)測(cè)效果進(jìn)行比較,選擇出最佳的模型。關(guān)鍵詞:梅雨,支持向量機(jī),時(shí)間序列,網(wǎng)格尋優(yōu),遺傳算法,粒子群優(yōu)化AbstractInthepast100yearsormore,globalclimatechangeincreasingly,andkindsofnaturaldisasterscomeoneafteranother,andsummerdroughtsandfloodsareoneofthemajordisaster.TheinformationofMeiyuinmiddle—lowerre
4、achesofYangtzeRiverismajorindicatorstOmeasuredroughtsandfloodsinthemiddle—lowerreachesofYangtzeRiverinsummer.SopredictionofMeiyutOtalisimportantinstudyingdroughtsandfloodsinthemiddle-lowerreachesofYangtzeRiverinsummer.HowtOcreateapredictionmodelbyobservingafinit
5、enumberofhistoricalsamplesisanimportantworkofeconomicactivity.Statisticallearningtheory(SLC)focusesonthemachinelearningtheoryofsmallsamples.ItscoreistOcontrolthegeneralizationlearningmachinebycontrollingthecomplexityofmodels.Supportingvectormachine(SVM)isamethod
6、ofmachinelearningbasedonVCdimensionandstructuralriskminimizationprincipleofthestatisticallearningtheory.SVMhasadvantagesinsolvingsmallsamplesizeproblemsinpracticalapplications,suchassmallsample,noniinear,overlearning,nolinear,highdimensionalandlocalminimumpoint.
7、TimeseriesforecastingisoneofthemainI.esearchtopicsinintelligentcomputing.AccordingtOrecent106a(1885—1990)dataoftheMeiyuinmiddle—lowerreachesofYangtzeRiverand49“1954-2002)dataoftheMeiyuintaizhou,builtSVMregressiontimeseriesmodelbaseonPolyandRBF,andusedparameterfu
8、ncti—onofgridoptimization,GeneticAlgorithms(GA),ParticleSwarmOptimization(PSO)tOoptimizethemodelparameters,andthencomparativeeffectivenessofthesesixpredictionmodels,a