


PhysicaA363(2006)481–491
doc.guandang.net
Theapplicationofneuralnetworkstoforecastfuzzytimeseries
KunhuangHuarng,TiffanyHui-KuangYuÃ
DepartmentofInternationalTrade,FengChiaUniversity,100WenhwaRd.,Seatwen,Taichung407,Taiwan,ROCDepartmentofPublicFinance,FengChiaUniversity,100WenhwaRd.,Seatwen,Taichung407,Taiwan,ROC
Receivedinrevisedform21March2005Availableonline6September2005
Abstract
Fuzzytimeseriesmodelshavebeenappliedtohandlenonlinearproblems.Toforecastfuzzytimeseries,thisstudyappliesabackpropagationneuralnetworkbecauseofitsnonlinearstructures.Weproposetwomodels:abasicmodelusinganeuralnetworkapproachtoforecastalloftheobservations,andahybridmodelconsistingofaneuralnetworkapproachtoforecasttheknownpatternsaswellasasimplemethodtoforecasttheunknownpatterns.ThestockindexinTaiwanfortheyears1991–2003ischosenastheforecastingtarget.Theempiricalresultsshowthatthehybridmodeloutperformsboththebasicandaconventionalfuzzytimeseriesmodels.r2005ElsevierB.V.Allrightsreserved.
Keywords:Backpropagation;Forecasting;Nonlinear;Stockindex
1.Introduction
Fuzzytimeseriesmodelshavebeenappliedtohandlenonlinearproblems,suchasenrollment[1–5],stockindexforecasting[6–10],andtemperature[11],etc.Afuzzytimeseriesessentiallyconsistsofstepssuchasfuzzi cation,theestablishmentoffuzzyrelationships,anddefuzzi cation.Somestudieshavefocusedonthefuzzi cationalone[7,9,12,13].Nevertheless,manystudieshavefocusedontheestablishmentoffuzzyrelationships[1–6,8,10,14,15]becauseitisdirectlyrelevanttoforecasting.Therefore,thisstudyalsotargetstheestablishmentoffuzzyrelationships.
Arti cialneuralnetworkapproaches(hereafterreferredtomoresimplyasneuralnetworks)havebeensuccessfullyappliedtovariousapplications[16–18].Thenonlinearstructuresofneuralnetworkshave,forinstance,beenveryusefulinforecasting[19].Inordertotakeadvantageofitsnonlinearcapabilities[20],thisstudyhaschosenaneuralnetworktoestablishfuzzyrelationshipsinfuzzytimeseries,whicharealsononlinear.Hence,therationaleandmotivationforapplyinganeuralnetworkareself-explanatory.
Wehavefollowedthesuggestionsregardingconstructinganeuralnetworkforforecasting[21],includingdatapreparation,thenetworksetup,andmodelselectionandevaluation.Toinvestigatetheforecastingcapabilitiesoftheneuralnetworkapproach,weproposetwomodels,abasicmodelandahybridmodel.Thebasicmodelsimplyusesaneuralnetworkapproachtoforecastalloftheobservations,whilethehybridmodel
Correspondingauthor.Tel.:+886424517250ext.4309.
E-mailaddresses:khhuarng@fcu.edu.tw(K.Huarng),hkyu@fcu.edu.tw(T.H.-K.Yu).0378-4371/$-seefrontmatterr2005ElsevierB.V.Allrightsreserved.doi:10.1016/j.physa.2005.08.014
fuzzy time series and particle swarm optimization I-Hong Kuo a, Shi-Jinn ... Huarng, 2001a, 2001b; Huarng & Yu, 2004; Hwang et al., 1998; Li ...
[12] Huarng, K., & Yu, H. K., "Ratio-based lengths of intervals to improve fuzzy time series forecasting, " IEEE Transactions on Systems, Man, ...
Huarng and Yu (2005) proposed a Type-2 fuzzy time series model, in which extra data are used to enrich or to re?ne the fuzzy relationships obtained...
3. A bivariate neural network-based fuzzy time series model. T.H.-K. Yu, K.-H. Huarng / Expert Systems with Applications 34 (2008) 2945–2952 ...
[8]; Chen and Hwang [3]; Huarng [6] worked on various models for fuzzy time series forecasting to improve the forecast. ? Tel.: +91 5422307435. ...
For the step (1), (Huarng, 2001; Huarng and Yu, 2004) argued that length of intervals greatly affects forecasting results in fuzzy time series and ...
a simple method of forecasting based on fuzzy time seriesa simple method ...Huarng [5] presented a heuristic model for time series forecasting using ...
(2001) 387–394 www.elsevier.com/locate/fss E ective lengths of intervals to improve forecasting in fuzzy time series Kunhuang Huarng Department of ...
Data compression, Time series analysis and ...Automotive Fault Diagnosis,fuzzy evaluation,Robot ...[3] 《人工神经网络的模型及其应用》. 张立明编著...
因 此如 何有 效地 从 T S 模糊神经 网络 中抽取 Fuzzy 规则 是本文研究的重点. 本 文在 应用 改进 的 T S 模糊 神经 网络 学 习算 法的 基础上,...
热门文档