資源描述:
《基于非等時距加權(quán)灰色模型與神經(jīng)網(wǎng)絡(luò)的軌道不平順預測-論文.pdf》由會員上傳分享,免費在線閱讀,更多相關(guān)內(nèi)容在行業(yè)資料-天天文庫。
1、第36卷第1期鐵道學報Vol_36NO.12014年1月JOURNALOFTHECHINARAILWAYSOCIETYJanuary2014文章編號:10018361(2014)01—0081一O7基于非等時距加權(quán)灰色模型與神經(jīng)網(wǎng)絡(luò)的軌道不平順預測韓晉,楊岳,陳峰,吳湘華(中南大學交通運輸工程學院,湖南長沙41O075)摘要:對軌道不平順的發(fā)展趨勢進行有效預測,可以提高鐵路線路養(yǎng)護的維修效率,保障行車安全。根據(jù)軌道不平順的發(fā)展特性,提出一種基于非等時距加權(quán)灰色理論和神經(jīng)網(wǎng)絡(luò)法的組合預測方法。該方法通過構(gòu)建非等時距加權(quán)灰色預測模型,將原始TQI序列的平
2、均值作為累加序列初值,將連續(xù)累積函數(shù)的積分面積作為背景值,對累加序列進行加權(quán)處理,較好地反映了時間序列對軌道不平順預測結(jié)果的貢獻。在此基礎(chǔ)上,引入BP神經(jīng)網(wǎng)絡(luò)模型對TQI預測的殘差序列進行修正,較好地克服了單一模型預測精度偏低的不足。分別對滬昆線上行兩段線路的軌道不平順進行預測,結(jié)果表明該預測方法相對誤差平均值分別為2.76和2.O8,預測結(jié)果的后驗差比值分別為0.121和0.151,精度等級達到l級。關(guān)鍵詞:軌道不平順;神經(jīng)網(wǎng)絡(luò);非等時距;灰色模型.力口權(quán);殘差修正中圖分類號:U212.24+6文獻標志碼:Adoi:10.3969/j.issn.1
3、001—8360.2014.01.013PredictionofTrackIrregularityBasedonNon。。equalIntervalWeightedGreyModelandNeuralNetworkHANJin,YANGYue,CHENFeng,WUXiang—hua(SchoolofTrafficandTransportati0nEngineering,CentralSouthUniversity,Changsha410075,China)Abstract:Effectivepredictionofthetrackirregular
4、itydevelopmenttrendcanimprovetheefficiencyofrailwaylinemaintenance&repairsandSOensuretrafficsafety.Accordingtotrackirregularitydevelopmentcharacter—istics,thecombinationpredictionmethodbasedonthenon—equalintervalweightedgreytheoryandneuralnet—workmethodwasproposed.Withthismetho
5、d,byconstructingthenon—equalintervalweightedgreypredictionmodel,theaverageoftheoriginalTQIsequencewasregardedasthecumulativesequenceinitialvalue,theinte—gralareaofthecontinuousaccumulationfunctionwasusedasthebackgroundvalue,thecumulativesequencewasprocessedbyweighting.Therefore
6、,thecontributionofthetimesequencetothetrackirregularitypredic—tionresultswasbetterreflected.Onthisbasis,theBPneuralnetworkmodelwasintroducedtoamendtheTQIpredictionresidualssequenceandtoovercomethedrawbackoflowpredictionaccuracybyasinglemode1.ThetrackirregularitiesoftWOup—direct
7、ionsectionsoftheShanghai—KunmingLinewerepredictedrespectively.Thepredictionresultsindicatethatthemeansoftherelativeerrorsare2.76and2.08respectively,theposteri—orerrorratiosare0.121and0.151respectively,andtheaccuracylevelreachesA.Keywords:trackirregularity;neuralnetwork;non—equa
8、linterval;greymodel;weight;residualmodification利用軌道不平順