基于油色谱数据的变压器可靠度评估方法

更新时间:2023-05-09 07:43:01 阅读量: 实用文档 文档下载

说明:文章内容仅供预览,部分内容可能不全。下载后的文档,内容与下面显示的完全一致。下载之前请确认下面内容是否您想要的,是否完整无缺。

第29卷第7期2013年7月

电力科学与工程

ElectricPowerScienceandEngineering

V01.29.No.7

Jul..2013

基于油色谱数据的变压器可靠度评估方法

张钰宁1,程养春1’2,田野1,陈凯1,宋安琪1,李成榕1’2,袁帅3,杨宁3(1.华北电力大学高电压与电磁兼容北京市重点实验室,北京102206;

2.华北电力大学新能源电力系统国家重点实验室,北京102206;

3.中国电力科学研究所,北京100192)

摘要:变压器是电力系统的枢纽设备,其运行的可靠性直接关系着电力系统的安全运行。为此,提出了一种基于色谱数据统计规律的变压器可靠度评估方法。首先对正常状态以及过热和放电异常状态时变压器的油色谱数据进行分布规律分析,采取曲线拟合的方式求取各状态下气体含量的概率密度函数以及累积分布函数。根据异常状态时数据的分布规律,利用变压器的可靠性与故障的累计概率之间的关系,得出各气体含量达到国标注意值时变压器的可靠度。最后按照国家安全预警体系将变压器异常状态划分为5个严重等级。

关键词:变压器;气相色谱分析;正常状态;故障;可靠度

中图分类号:TM855文献标识码:ADOI:10.3969/j.issn.1672—0792.2013.07.001

0Introduction

Atpresent,judgingwhetherthetransformerbreaksdownandthefaulttypemainlyusestheabso-lutevalueandtheratioofthedissolvedgasesinthefield㈨].Nationalstandard[3]andIEC-60599

4]pro-videtheattentionvalueofdissolvedgas,butreference[3]onlylistsouttheattentionvalueofH2,C2H2andtotalhydrocarbon,whichisnotverycomprehensive.Althoughreference[4]liststhetypesofdissolvedgascompletely,itsanalysisobjectisallDGAdataanditdoesn’tmakeadistinctionbetweenthenormalandab—normalofthesampleswithsimpleassumption90%dataintonormaldata.Besides,thestatisticaldataoftransformersinRef.[4]areglobal,SOitdoesn’tpro—videtheattentionvalueofthetransfotiBerinourcoun—try.Finally,neitherofthemgivesamethodfordeter—miningthereliabilityof

transformers.

Forthis,thisarticlecollectsatotalof18307setsofthedataofdissolvedgascontentinoilfrom12500kVnormallyoperationtransformersinNorthChina;altogether3001setsofdatafrom10220kVtrans—foFillersinNorthChinaand1959setsofdatainallfrom3110kVtransformersintheNorthChinaarea.

Inaddition,thistextalsocollects266groupsoftypi—caldissolvedgasdataof186overheatingfaulttrans—formersbeforeaccidentand233

groupsoftypicaldis—solvedgasdataof178dischargefaulttransformersbe—foreaccidentfromtherelevantliterature.Thispaperplanstoanalyzethestatisticaldistributionruleofdis—

solved

gas

inoilofnormal

transformeranddefectivetransformer,andthenprovidesthequantitativerelia—bilityassessmentrulesof

transformertomakethediag-nosisresultaccurate.

Asthereliabilityoftransformeriscloselyrelatedtothefaultcumulativeprobabilityoftransformer,andthecumulativeprobabilitycanbeobtainedfromtheprobabilitydensity,。5。thispapermakesdistributiontestonthenormalandfaultdissolvedgasdatasamples

收稿日期:2013—05—06。

基金项目:国家重点基础研究发展计划项目(973计划)(2009CB724508)。

作者简介:张钰宁(1988一),女,硕士研究生,从事高电压与绝缘技术等方面的研究,E—mail:zynlsc@163.con。

瓦西甬疆■西蕊磊耋NGtNEERING

2电力科学与工程2013焦

andcurvefitting【6jinthegaswhichdoesnotobey

any

knowndistribution.Throughthestatisticsandanaly—

sis,thispapercalculatesthereliabilityofthetrans—

former.Theresultnotonlyhasthe

advantagesofchar—

acteristicgasmethod,intuitiveandconvenient,but

alsohastheconceptofquantityandeasytoguideen—

gineeringpractice.

1Analysisofthedistributionofdissolvedgas

inoilofnormaltransformer

Thissectionanalyzesallnormaldissolved

gases

(H2,CH4,C2H6,C2H4,C2H2,COandtotalhydro—

carbon)datasamplesofthetransformersofthreevolt—

ageclasses.

Thehistogramisusedto

represent

thefrequency

distributionofthecontinuousdataforcheckingthe

shapeanddistributionofsampledata.Therectangular

areaofhistogramindicatesthefrequency,andthesum

oftherectangularareaisequivalenttothesumofeach

frequency….Therefore,throughthehistogramwecan

haveanintuitiveunderstandingoftheprobabilitydis—

tributionofsampledata.Typicalhistogramsofthedis—

solvedgasesareshowninfigure1.Thecurveinthe

figureisthenormaldistributioncurve.

墨厂—————一

×2

04080

gascontent/(I.tL?L1)

(a)H:o

__

×

lO

02040

gascontent/(1aL?L-1)

(b)C:H。

Fig.1I-mtogrmofso眦dissolvedgases

FromFigure1wecanseeexactlythatthedistri—butionofH2andC2H6dissolvedintransformeroildoesnotobeythe

normaldistribution.ThediagramsalsoshowthattheH2distributionissimilartotheWeibull

distributionandthesamplesofC2H6contain

alargenumberofzeros.Likewise,otherdissolvedgasesalsodonotobeynormaldistribution,andC2H4andC2H2

alsohavelotsofzeros.ThezeroproportionofeachgasisshowninTable1.

Tab.1Zeroproportionofnormaldata(%)

Fromtable1,wecan

seethatthezeroproportionofC2H6,C2H4andC2H2isverylarge,sowecan’tig—notetheeffectofzero.BecauseSPSS(StatisticalProductandServiceSolutions)softwaremakesWeibulldistributiontestonthesampleexcludingze—ro,theabovethreekindsofgasesarenotsuitableforusingSPSSfor

Weibulldistributiontest.Thisarticle

willcarry

onthecurvefittingforthem.Thezeropro—portion

oftheremaining

gasesissmaller,soitcanbeignoredfortaking

Weibulldistributiontest.

Todeterminewhetheravariableissubjecttosomeprobabilitydistributionneedstousedistribution

test.SPSSmainlyprovidestwo

waystocarrydistribu—

tiontestonsample:oneisKolmogorov—Smirnov

test,andtheotherisP—Pplot(probability—probabilityplot)orQ—Qplot(quantile—quantileplot).Assume

thatsingle-?sampleK——S

inspection

confidenceintervalis1一a.Ifthetestofsignificanceisgreaterthana,thesampleobeystheassumeddistribution,otherwiseitdoesn’tobeythedistributionrule.Asingle—sample

K—_Stestonthesamplecannot

processsampledataatthesametime,whiletheP—Pdiagramcananalyzethedistributionandcarrycorrelationprocessingonda—taatthesametinle[8。.

ThissectionwilluseP—Pplotfortest.P—Pplotisdrawnaccordingtotherelationshipbetweenthecu—

mulative

proportionofvariable

andthecumulative

pro—portionofthegiven

distribution.Thetwostandardsfortestingwhetherthedatafitwelltothegivendistribu—tionare:(1)thecontactratioofdatapointsandlineontheP—Pplot;(2)whetherthepointsontheP—PtrendplotfluctuateinsmallrangeaboutlineY=0.

Fromthehistogram,somedissolved

gasesmayo—bey

Weibulldistribution,soweuseP—P

plotforthesampledatatotest

Weibulldistribution.The

probabil—

第7期

ELECTRICPOWERENGtNEERING张钰宁,等基于油色谱数据的变压器可靠度评估方法3

itydensityfunctionofWeibull‘distribution八戈)is¨?showninformula(1),andthetestresultsofH,are()E¨nshowninfigure2.嚣;

FromtheP—Pplotinthefigure2,wecansee…

¨1thatthedatacanbewellcontactwithline.andinthe

P—Ptrend

plot,thefluctuationofdatapointissmall.

sowecanthinkthatH,obeyWeibulldistribution[9。.

Similarly,itisknownthatCH4,COandtotalhydro—

carbonalsoobeyWeibulldistribution,andthe

gas

probabilitydensityanddistributionparametersarelis—

tedintable2.

八戈)x/A'),zI>0

(1)

Wherekisthe

shapeparameterandAisthescaleparameter.

Tab.2ParametersofWeibullofdissolvedgases

GasH2CH4Total

HydrocarbonCOShapeparameterk0.05320.12360.13260.0042ScaleparameterA1.0041.15881.0270.9848

Thissectionwilltakecurvefitting

waytocalcu—

latethedistributionfunctionfor

C2H6,C2H4andC2H2

duetothelargenumberofzeros.Their

probabilitydensityfunction(PDF)f(戈)issimulatedwithfor—mula(2)andcumulativedistributionfunction(CDF)F(z)issimulatedwithformula(3).Typicalfittingresultsareshowninfigure3.

八戈)=Y0+ate一下+02e一下(2)F(戈)=儿+a1(1一e一∥“)+u2(1一e一加2)(3)

WhereY0,al,u2,tlandt2arethevaluesthatare

tobeset

00

004

OOO

O-8

eprobabiIit)

Fig.2ResultsofWeibulldistributiontestofH

。矧r

铡/

’1’1’’’’—’T’。’’’‘r。。’‘‘‘r’’’‘’’‘’+●’+1一

Fig?3FittingresultsofPDF&CDFofC2H6

2Statisticalrule

analysisofdissolvedgasinoil

ofdischargedefectsandoverheatingfault

Thisarticlealtogethercollects232

groupsofdis—

chargefaultdata,including20groupsofpartialdis.

chargedata,65groupsoflowenergydischargedata

and99

groupsofhighenergydischargedata.Because

partialdischargesampledataaretoosmallandtheer—

rorofstatisticalruleisbig,soseparatestatisticala—

nalysisisnotcarriedout.Accordingtothedivision

standardinRef.[10],overheatisdividedinto

an-

der700oCoverheatandabove700oCoverheat.

Firstofall,takingnormaldistributionK—Stest

for

everysample(dischargetotalsample,lowenergy

andhigh

energydischargesample,overheatingsam—

pie)respectively.Seta=0.05,thecalculationre—

suitsshowthatthedoublesideprogressive

significance

arelessthan0.05,soallthe

samplesdonotobevthe

normaldistribution.Whendatado

notobevthenor—

maldistribution

ornormaldistributionfittinglevelis

relativelypoor,afeasiblemethodofdataprocessingis

dataconversion.Inthispaper,the

samplestakethe

logarithmbeforethedistributionruleanalysis.Be—

causethezerovalueofthesamplecannotbetaken

logarithm,wecansetitasalackofsystem.Thezero

proportionofeachgasisshownintable3.

Tab.3ZeroProportionoffaultdata(%)

H2CH4C2H6C2H4C2H2

hydrocarbon

Overheat2.630.001.130.3836.090.00

Discharge0.000.433.020.8616.380.00

Throughtable3,itisknownthatthezeropropor一分

后P叽

一4

|o出

括m

瞄唧山

廿

¨~

帅咖

暖i:丫聂l百—≯石西萎R

ENGiNEERiNG

电力科学与工程

2013焦

tionofthefivekindsofgasesis

small,which

can

betreated

asa

lackofsystembeforelogarithmictreat—

ment.When

overheatinganddischarge,the

zero

pro—

portionofC2H2isbiggerandnotsuitableforlogarith—

micnormaldistributiontest.Thispaperuses

theCHIVe

fitting

method(formula(2)andformula(3))to

ob—

tainthedistilbutionrule.

TakingK—Stest

forH2and

so

on,allprogres—

sivesignificanceofdischargedefectsampleisgreater

than0.05,indicatingthattheyobeythe

lognormal

distribution.So

at

the

differentdegrees

ofdischarge,

allthegasesobeythelognormaldistributionexcept

C2H2.Foroverheatingfaultsample,onlytheprogres—

sivesignificanceofCH4intheoverheatingtotalsample

is

lowerthan0.05anddoes

not

obeythe

lognormal

distribution.Thus,intheoverheatingtotal

sample,

allthe

gasesobey

thelognormaldistributionexcept

CH4andC2H2,thedistributionparameters

are

shown

intable4.Tothegaswhichdoesnotfollow

lognormal

distribution,thispaper

uses

formula(2)andformula

(3)for

curve

fitting.Theprobabilitydensityfunction

f(戈)oflognormal

distributionis:

八戈)=ilxor贯一掣

。己+n

(4)

Where/.t

istheaverageofthelogarithmofthe

variableand

or

is

thestandard

deviationofthelog—

arithmofthevariable.

擒爹一

40

20

Logarithmofthegas

conten

(b)CH。

Fig.4

Histogramofsomedissolvedgasesinlogs

Typicalhistogramsaftersamplestakinglogarithm

are

showninfigure4.Figure4denotesobviouslythat

thesamplesobeynormal

distributionaftertakinglog—

arithm.

Tab.4

Parametersofiognormalofdissolvedgases

Reliabilityanalysis

Reliability

istheabilityoftheproductperforming

required

functionunder

specifiedconditions

and

withintherequiredperiod,sothedegreeof

reliability

referstothe

probability

oftheproductcompletingpre—

scribedfunctionwithintheprescribedtimeandcondi—tions,whichisusuallyrecorded

as

R(£).

R(t)=P(T>t)

(5)

Reliability

is

closelyrelated

to

cumulativefaultprobabilitywhichsaysthe

probability

oftheequipment

that

can’t

completerequiredfunctionunderspecified

conditionsandwithintheprescribedperiodoftime,

recorded

as

F(t).

According

to

the

probabilitycomplementarytheo—

rem,we

can

get:

R(t)+F(£)=1

(6)

Ifthefaultdatadistributionis

known,F(£)can

beobtainedand

socan

R(t).

According

to

theaboveanalysis,we

canobtain

theCDFdiagramsofthedissolvedgases

at

different

stateswhich

are

shownin

figure5.Wherein,A

repre—

sents

normal,BrepresentsoverheatingfaultandC

re—

presentsdischarge

fault.The

Xcoordinateis

sample

aftertaking

logarithm.Formthefigure5,itisknown

thatthe

normaldata,overheatingfaultdata

anddis—

chargefaultdataofthedissolvedgasoverlapeachoth—er;theoverheatingdataofothergases

are

more

con—

centratedthan

dischargedata

inadditiontothetotal

hydrocarbon.Nationalstandardsprovidetheattention

1e

,‘

6弘H

第7期

张钰宁,等基于油色谱数据的变压器可靠度评估方法

ELEC下R:0P0诲拦R垂KGl魏薹基建}NG

valueofH2is

150

cL舻L

anCPtotalhydrocarbonisalso

150斗L/L.Theattention

valueofC2H2ofabove330

kVrunningtransformeris1灿L/L,andbelow330kv

is5斗I/L.The

reliabilitycorresponds

toattention

valueiSshownintable5.

0?8

言0.6

u04

0.20

lO

ln(x)

(a)H

0.8t臣O.6?o0.4.

O.2O一

—丛习1

10

ln(x)c)C!H

1_7

O.8l7

kO6

U04

0.20

0510

ln(x)fe)C,H,

0.4O.20

0.8kO.6o

U0.4

lO

ln(x)(b)C:H;

10

ln(x1(d)CH。

1厂1∥1

j,\

剧1

10

nCx

Fig.5CDFofdissolvedgasesTab.5

ReliabilityofMannvalue

Fromtable5,we

can

see

thatthereliability

cor—

iesponds

to

theattentionvalueofeachdissolvedgasisalmostmorethan0.5

securitywhichisshownintable6.

Tab.6

Faultseverity

肛L/L

According

to

theaboveanalysis,weconclude:

(1)Thedataof

characteristic

dissolvedgases

H2,CH4,COandtotalhydrocarbonofnormalopera—

transformerobey

Weibull

distributionwhileC2H6,

C2H4andC2H2disobeyanyknowndistributiondueto

lots

of

zero

values.Thecharacteristic

gasesH2,

C2H6,C2H4andtotalhydrocarbonofdischargeandO—

verheatingfaultobey

thelognormaldistribution,and

CH4ofdischargeobeysthelognormaldistribution,but

CH4ofoverheatingdoes

not

obeythelognormaldistil—

bution,C2H2disobeysanyknowndistributiondue

to

largenumberof

zero

values.

(2)According

to

the

earlywarningsystem

of

national

security,transformerabnormal

state

can

be

If

group

ofDGAdataisobtainedinthefield.dividedintofiveseriouslevels

or

bedividedintodif-

thecumulativeprobabilityofthefaultcan

befoundin

ferentlevelsoffailure‘severityifrequiredon

thebasis

figure5.Thenaccordingto

formula(6),the

reliabil—

ofthe

abnormaldatadistribution.

ityofthisvaluecanbeobtained.Accordingtothe

re—

liabilitywe

can

judge

whether

transformer

breaksdownReferences

andtheseverityofthefault.Thispaperdividesrelia—bilityaccording

to

theearlywarningsystemofnational

[1]徐志钮,律方成,赵丽娟.基于RPROP算法的变压器

碉川

彩历

EkE0亨曩iSP0嚣薹塞耋鬻0iNE基疑;裴0

6电力科学与工程2013年

油中溶解气体分析故障诊断[J].电力科学与工程,2005,(2):70—72.XuZhiniu,LuFangcheng,ZhaoLijuan.TransformerDGAfaultdiagnosisbasedonrpropalgorithm[J].ElectricPowerScienceandEngineering,2005,(2):70—72.[2]胡资斌,朱永利,董卓.基于遗传程序设计的变压器油中溶解气体含量预测[J].电力科学与工程,2011,27(7):7—11.HuZibin,ZhuYongli,DongZhuo.Predictionofgasescon—tentdissolvedinpowertransformeroilbasedongeneticpro?gramming[J].ElectricPowerScienceandEngineering,2011,27(7):7—11.[3]GB/T7252--2001.变压器油中溶解气体分析和判断导则[s].[4]IEC60599--2007.Mineraloil—impregnatedelectricale-quipmentinservice—guidetotheinterpretationofdissolvedandfreegasesanalysis[S].[5]王有元.基于可靠性和风险评估的电力变压器状态维修决策方法研究[D].重庆:重庆大学,2008.[6]陈光,任志良,孙海柱.最小二乘曲线拟合及Matlab实现[J].兵工自动化,2005,24(3):107—108.ChenGuang,RenZhiliang,SunHaizhu.CurvefittinginLeast-SquaremethodanditsrealizationwithMatlab[J].OrdnanceIndustryAutomation,2005,24(3):107—108.[7]包红,徐娜,李新,等.利用SPSS统计软件分别绘制以频数和频率为纵坐标的直方图[J].数理医药学杂志,2010,23(6):717—719.

BaoHong,XuNa,LiXin,eta1.MakingugeofSPSStodrawhistogramwithfequenceandfrequencyrespectivelyasordinates[J].JournalofMathematicalMedicine,2010,23(6):717—719.

[8]武东,费鹤良,汤银才.PP图及其在可靠性中的应用[J].上海师范大学学报(自然科学版),2004,33(2):13—17.

WuDong.FeiHeliang.TangYincai.PPplotanditsappli—cationstothereliability[J].JournalofShanghaiTeachers

University,2004,33(2):13—17.

[9]凌丹.威布尔分布模型及其在机械可靠性中的应用研究[D].成都:电子科技大学,2011.

[10]IEEEStdC57104—2008.IEEEGuidefortheinterpreta—tionofgasesgeneratedinoil??immersedtransformers?-redline

『S].

AReliabilityEvaluationMethodforTransformer

BasedonDataofDissolvedGasAnalysis

ZhangYunin91,ChengYangchunl一,TianYel,ChenKail,SongAnqil,

LiChengron91”,YuanShuai3,YangNin93

(1.BeijingKeyLaboratoryofHighVoltageandEMC;

2.StateKeyLaboratoryofAlternateElectricalPowerSystemwithRenewableEnergy,

NorthChinaElectricPowerUniversity,Beijing102206,China;

3.ChinaElectricPowerResearchInstitute,Beijing100192,China)

Abstract:Thetransformeristhekeyapparatusinelectricpowersystem.Itsreliabilitydirectlyaffectsthesafeop-erationofpowersystem.Therefore,areliabilityevaluationmethodfortransformerbasedonthestatisticalregulari.tiesofthegaschromatographydataoftransformersisproposed.Firstly,byanalyzingthedistributionruleofthegaschromatographydataoftransformersundernormalconditionandthegaschromatographydataofdischargeandover—heatfaults.theprobabilitydensityfunctionandthecumulativedistributionfunctionofthegascontentindifferentconditionswereobtainedbycurve-fittingmethod.Secondly,accordingtothedistributionruleofthegaschromatog-raphydataofdischargeandoverheatfaults,thetransformerreliabilitywheneachgascontentreachedtheattentionvalueofGBwasobtainedusingtherelationshipbetweenthereliabilityandthecumulativeprobabilityofthefailureofthetransformer.Finally,accordingtothenationalsecuritywarningsystem,faultconditionoftransformerscanbedividedintofivegrades.

Keywords:transformer;dissolvedgasanalysis;normalcondition;fault

condition;reliability

本文来源:https://www.bwwdw.com/article/t8te.html

Top