改進(jìn)的模糊C均值法在負(fù)荷特性統(tǒng)計(jì)數(shù)據(jù)聚類中的應(yīng)用畢業(yè)論文-

改進(jìn)模糊C-均值法在靜態(tài)負(fù)荷特性數(shù)據(jù)聚類中的研究改進(jìn)模糊C-均值法在靜態(tài)負(fù)荷特性數(shù)據(jù)聚類中的研究改進(jìn)的模糊C均值法在負(fù)荷特性統(tǒng)計(jì)數(shù)據(jù)聚類中的應(yīng)用摘要電力負(fù)荷是整個(gè)電力系統(tǒng)的安全穩(wěn)定運(yùn)行中較活躍的一部分。建立符合實(shí)際的動(dòng)態(tài)負(fù)荷模型對(duì)電力系統(tǒng)規(guī)劃、設(shè)計(jì)和運(yùn)行等諸方面均有十分重要現(xiàn)實(shí)意義。本文采用實(shí)用化負(fù)荷建模思想，對(duì)負(fù)荷特性進(jìn)行聚類，從而為變電站建立合適的負(fù)荷模型打下基礎(chǔ)?；谀壳柏?fù)荷建模方面存在的問(wèn)題，使用模糊C均值法，對(duì)同一地域不同地點(diǎn)變電站的負(fù)荷統(tǒng)計(jì)數(shù)據(jù)進(jìn)行聚類分析。針對(duì)湖南電網(wǎng)48個(gè)變電站，對(duì)模糊C均值法實(shí)施改進(jìn)后對(duì)其進(jìn)行聚類，并與未改進(jìn)的模糊C均值法的聚類結(jié)果進(jìn)行比較，以說(shuō)明改進(jìn)方案的有效性。關(guān)鍵字：電力負(fù)荷；負(fù)荷特性；聚類；模糊C均值法APPLICATIONOFIMPROVEDFCMTOELECTRICLOADCHARACTERISTICSOFSTATISTICALDATACLUSTERINGABSTRACTThepowerloadisanactivepartinthesecurityandstableoperationoftheentireelectricalpowersystem.Itissignificantlyimportanttomakesuitableloadmodelforthepowersystemplanning,designandoperation.Inthispaperthepracticalloadmodelingmethodisemployed,andtheloadcharacteristicsisclusteredtoestablishtheactualloadmodelforsubstations.Basedonthecurrentproblems,FCMwithhierarchicalclusteringisusedtoperformtheclusteringoftheloadcharacteristicsdataofthedifferentsubstationsonthesamearea,theimprovedmethodisappliedfortheclusteringofHunangridsubstation.Theclusteringresultshowsthattheimprovedmethodiseffectivecomparingwiththeunimprovedmethod.KeyWords:powerload;loadcharacteristic;cluster,FCM目錄第一章緒論11.1研究背景11.2發(fā)展及研究現(xiàn)狀21.2.1發(fā)展21.2.2研究現(xiàn)狀電力負(fù)荷建模的總體原則電力負(fù)荷建模的基本概念分類51.2.3實(shí)用化負(fù)荷建模思想統(tǒng)計(jì)綜合法總體測(cè)辨法71.3聚類分析在負(fù)荷特性分析中的應(yīng)用現(xiàn)狀81.4本文主要研究?jī)?nèi)容9第二章聚類分析102.1聚類分析的基本概念102.2聚類方法112.3系統(tǒng)聚類法152.3.1最小張樹聚類法162.3.2基于密度的聚類算法162.3.3基于網(wǎng)絡(luò)的聚類方法162.3.4基于模型的聚類算法162.3.5基于劃分的聚類算法162.4各算法優(yōu)缺點(diǎn)比較17第三章模糊C均值在負(fù)荷特性聚類中的應(yīng)用實(shí)例193.1聚類在電力系統(tǒng)中的應(yīng)用綜述193.2模糊C均值聚類算法203.2.1硬C均值聚類算法（HCM）203.2.2模糊C均值聚類223.2.3程序流程圖233.3對(duì)模糊C均值法的改進(jìn)253.3.1改進(jìn)的各方案比較253.3.2最終改進(jìn)方案的選定263.4聚類實(shí)例273.4.1原始數(shù)據(jù)聚類數(shù)據(jù)273.4.2未改進(jìn)的模糊C均值法在實(shí)例中的應(yīng)用273.4.3改進(jìn)的模糊C均值法在實(shí)例中的應(yīng)用283.4.4兩種算法的比較293.5結(jié)果分析31第四章結(jié)語(yǔ)33參考文獻(xiàn)34致謝35附錄36附錄A原始聚類數(shù)據(jù)36附錄B系統(tǒng)聚類法所得的聚類中心生成的隸屬度矩陣38附錄C改進(jìn)的模糊C均值法源程序41附錄D類間距離計(jì)算源程序47附錄E類內(nèi)距離計(jì)算源程序48第⑥負(fù)荷聚類為沒有安裝布測(cè)點(diǎn)的變電站建立實(shí)用模型提供了有效途徑，從而克服了廣域電力系統(tǒng)為變電站均安裝負(fù)荷測(cè)辨裝置的不可行性，所帶來(lái)的負(fù)荷模型建立的種種困難。負(fù)荷測(cè)辨裝置安裝在每一類的典型變電站中，記錄下實(shí)測(cè)數(shù)據(jù)并為理想安裝測(cè)點(diǎn)的典型變電站建立相應(yīng)的負(fù)荷模型，而對(duì)于同類中未布測(cè)點(diǎn)則可以通過(guò)數(shù)學(xué)方法確定該類中其它沒有安裝測(cè)辨裝置的變電站的綜合負(fù)荷模型。這個(gè)過(guò)程考慮了負(fù)荷時(shí)變性特性，為綜合測(cè)辨法應(yīng)用于實(shí)用化負(fù)荷建模工作開辟了一條嶄新的途徑。結(jié)語(yǔ)1.隨著電力系統(tǒng)規(guī)模和容量的不斷擴(kuò)張，電力系統(tǒng)安全經(jīng)濟(jì)運(yùn)行的重要性和復(fù)雜性愈發(fā)顯著地表現(xiàn)出來(lái)，安全性和經(jīng)濟(jì)性是相互關(guān)聯(lián)又互相矛盾的兩個(gè)方面，只有充分了認(rèn)識(shí)到了電力系統(tǒng)的運(yùn)行規(guī)律，才能找到即安全又經(jīng)濟(jì)的運(yùn)行方式。2.本文基于目前負(fù)荷建模方面存在的問(wèn)題，采用實(shí)用化負(fù)荷建模思想，認(rèn)識(shí)到屬于同一類的負(fù)荷具有相對(duì)穩(wěn)定的性質(zhì)和共同的特點(diǎn)，引入聚類分析法對(duì)同一地域不同地點(diǎn)電站的負(fù)荷統(tǒng)計(jì)數(shù)據(jù)進(jìn)行聚類分析。本文對(duì)模糊C均值法加以改進(jìn)后再對(duì)該數(shù)據(jù)進(jìn)行聚類，并與未改進(jìn)的模糊C均值法加以比較。3.本文研究?jī)?nèi)容及成果如下：查閱有關(guān)文獻(xiàn)，總結(jié)了近年來(lái)負(fù)荷建模的研究現(xiàn)狀，幾種常用的聚類算法；深入學(xué)習(xí)了模糊C均值聚類算法，掌握了其基本理論和算法特點(diǎn)；了解了MATLAB軟件及其工具箱。對(duì)傳統(tǒng)的模糊C均值法做了改進(jìn)，使其克服了對(duì)初值敏感，易受孤立點(diǎn)影響的不足。對(duì)改進(jìn)算法和未改進(jìn)算法的聚類結(jié)果進(jìn)行了比較，結(jié)果表明，兩項(xiàng)指標(biāo)均有改善，改進(jìn)后的模糊C均值法較改進(jìn)以前有了一定的優(yōu)勢(shì)。模糊聚類計(jì)算結(jié)果中的聚類中心矩陣給我們對(duì)建模數(shù)據(jù)的采樣點(diǎn)的選擇即負(fù)荷測(cè)辨裝置的安裝地點(diǎn)提供了可靠依據(jù)，我們可根據(jù)聚類結(jié)果，將每類做成一個(gè)通用的模型，并且推廣到同類的其他變電站中，負(fù)荷特性聚類是綜合測(cè)辨法實(shí)用負(fù)荷建模過(guò)程的一個(gè)基礎(chǔ)部分，因此，本文研究具有一定的實(shí)用價(jià)值。參考文獻(xiàn)[1]孫即祥.現(xiàn)代模式識(shí)別[M].國(guó)防科技大學(xué)出版社，2001.[2]高惠璇.應(yīng)用多元統(tǒng)計(jì)分析[M].北京大學(xué)出版社,2005.[3]肖健華.智能模式識(shí)別方法[M].華南理工大學(xué)出版社,2005.[4]章健.電力系統(tǒng)負(fù)荷建模方法的研究[D].華北電力大學(xué)博士學(xué)位論文,1997.[5]張紅斌.電力系統(tǒng)負(fù)荷模型結(jié)構(gòu)與參數(shù)辨識(shí)的研究[D].華北電力大學(xué)博士文,2003.[6]鞠平.綜合負(fù)荷特性的分類綜合方法及應(yīng)用[J].電力系統(tǒng)自動(dòng)化.2004,28(1):64-68[7]賀仁睦.電力系統(tǒng)負(fù)荷模型的分類與綜合[J].電力系統(tǒng)及其自動(dòng)化.1999,23(19):12-16.[8]熊傳平.廣域電力系統(tǒng)基于負(fù)荷分析與分類的負(fù)荷建模研究[D].河海大學(xué)碩士論文.2006.[9]金艷.綜合動(dòng)態(tài)負(fù)荷特性的分類與綜合研究.河海大學(xué)碩士論文.2002.[10]賀仁睦，魏孝銘，韓民曉.電力負(fù)荷動(dòng)特性實(shí)測(cè)建模的外推和內(nèi)插[J].中國(guó)電機(jī)工程學(xué)報(bào),1996(3):151-154.[13]李力，朱守真，沈善德等.負(fù)荷動(dòng)態(tài)模型集結(jié)[J].電力自動(dòng)化設(shè)備.Vol.19,No.4,Aug,1999:6-10,19.[14]陳柔伊，張堯，武至剛，陳澤懷改進(jìn)的模糊聚類算法在負(fù)荷預(yù)測(cè)中的應(yīng)用.電力系統(tǒng)及自動(dòng)化學(xué)報(bào)2005.617卷.[15]王天行,張澤.多元統(tǒng)計(jì)分析學(xué)[M].成都:成都科技大學(xué)出社.1992:199～214.[16]吳元奇，馮榮揚(yáng).聚類分子計(jì)算方法的理論及結(jié)果比較.廣東：湛江海洋大學(xué)學(xué)報(bào).2002.2.[17]王成山，曹旌，陳光遠(yuǎn).基于聚類分析的電力系統(tǒng)暫態(tài)穩(wěn)定故障篩選.電網(wǎng)技術(shù),2005.[18]徐南榮，宋文忠，夏安邦.系統(tǒng)辨識(shí).南京:東南大學(xué)出版社，1991.[19]鞠平,馬大強(qiáng)。電力系統(tǒng)負(fù)荷建模[M].水利電力出版社，1995.致謝本論文是在導(dǎo)師王進(jìn)副教授的悉心指導(dǎo)下完成的，半年來(lái)，導(dǎo)師的教誨和關(guān)懷使我受益匪淺，特別是導(dǎo)師淵博的學(xué)識(shí)、嚴(yán)謹(jǐn)?shù)摹⒖茖W(xué)的治學(xué)態(tài)度和平易近人的工作作風(fēng)是我終生學(xué)習(xí)的楷模，借此論文完成之際，謹(jǐn)向?qū)煹男燎谂囵B(yǎng)表達(dá)我最誠(chéng)摯的謝意！同時(shí)感謝徐超學(xué)長(zhǎng)和王文生學(xué)長(zhǎng)在我設(shè)計(jì)期間所給予的指導(dǎo)和無(wú)私幫助！最后向所有曾給予過(guò)幫助和關(guān)心的學(xué)長(zhǎng)、親人、朋友表示誠(chéng)摯的感謝和深深的祝福！謝謝你們！XXX2010.6附錄A原始聚類數(shù)據(jù)注：系統(tǒng)聚類法得到的冰柱圖程序：Y=pdist(data);Z=linkage(Y,'average');H=dendrogram(Z,0);附錄B系統(tǒng)聚類法所得的聚類中心生成的隸屬度矩陣U=Columns1through110.02540.05710.04250.01880.13790.03640.03930.09610.09760.64120.01400.81510.59980.72610.82690.03800.60260.02050.03210.03060.01080.01570.02460.05790.02540.02260.08550.05050.09330.05750.08110.00870.01360.01540.03780.02320.01120.16230.02150.03140.05210.10450.23480.00690.02260.05080.02940.01980.35180.04220.68900.46070.48500.02780.02600.06650.12330.10280.07370.09430.18840.06720.17490.08330.02410.89460.03040.07330.05060.02690.13020.05830.05920.12660.11790.05270.0291Columns12through220.00810.05770.01920.07780.05990.20950.01860.03070.19190.00190.19610.01150.04600.02720.23030.43320.04660.00840.03020.03150.00250.06930.91570.10800.76640.13040.08740.04240.02610.69500.04540.97960.05910.00620.03490.01360.04750.03560.18610.01350.02510.11850.00140.09480.02430.27490.06930.11600.07310.13350.87110.11400.34700.00640.18670.02230.34560.07270.29240.23400.10550.03280.06360.10760.00540.21090.01180.13290.03150.10560.07680.27650.02950.04160.15800.00280.1832Columns23through330.31290.13120.07080.01440.02160.07740.13820.08670.08770.17520.07770.03570.01690.01920.00270.79170.30980.01880.17230.04040.03680.14400.03150.01750.01650.00210.01890.06590.01960.06350.02850.04870.04240.15540.68460.04760.00660.01160.04630.64410.05030.05560.20360.03560.11340.05030.04980.00720.02010.07730.05780.08520.07140.16420.07270.10340.03220.05450.00940.10560.26480.03650.31960.10470.09190.51590.24760.06740.74150.95760.03050.15860.08500.22250.61180.27950.1117Columns34through440.12570.39480.77630.88740.15150.74530.05460.16030.04770.24480.01760.04520.06420.01460.00430.01010.01700.05180.26740.12530.01500.87820.03480.02810.00860.00340.00890.00980.30490.04990.10400.01580.01290.04560.08960.08820.05760.74160.10010.03690.07950.02710.55710.01020.12130.06950.02380.01190.02650.02660.26250.07480.08580.05700.01370.42930.16020.02970.01050.01970.03380.20220.21070.53440.03490.04540.19810.19370.05880.02490.04170.06740.08720.15720.07580.07530.0220Columns45through480.21930.14000.37740.17240.01100.05400.02620.01540.00900.03670.02680.01270.66290.07190.13010.67180.02620.09830.14260.03360.02160.15930.09100.02810.05000.43990.20600.0660附錄C改進(jìn)的模糊C均值法的源程序function[center,U,obj_fcn]=fcm(data,cluster_n,options)%FCMDatasetclusteringusingfuzzyc-meansclustering.%%[CENTER,U,OBJ_FCN]=FCM(DATA,N_CLUSTER)findsN_CLUSTERnumberof%clustersinthedatasetDATA.DATAissizeM-by-N,whereMisthenumberof%datapointsandNisthenumberofcoordinatesforeachdatapoint.The%coordinatesforeachclustercenterarereturnedintherowsofthematrix%CENTER.ThemembershipfunctionmatrixUcontainsthegradeofmembershipof%eachDATApointineachcluster.Thevalues0and1indicatenomembership%andfullmembershiprespectively.Gradesbetween0and1indicatethatthe%datapointhaspartialmembershipinacluster.Ateachiteration,an%objectivefunctionisminimizedtofindthebestlocationfortheclusters%anditsvaluesarereturnedinOBJ_FCN.%%[CENTER,...]=FCM(DATA,N_CLUSTER,OPTIONS)specifiesavectorofoptions%fortheclusteringprocess:%OPTIONS(1):exponentforthematrixU(default:2.0)%OPTIONS(2):maximumnumberofiterations(default:100)%OPTIONS(3):minimumamountofimprovement(default:1e-5)%OPTIONS(4):infodisplayduringiteration(default:1)%Theclusteringprocessstopswhenthemaximumnumberofiterations%isreached,orwhentheobjectivefunctionimprovementbetweentwo%consecutiveiterationsislessthantheminimumamountofimprovement%specified.UseNaNtoselectthedefaultvalue.%%Example%data=rand(100,2);%[center,U,obj_fcn]=fcm(data,2);%plot(data(:,1),data(:,2),'o');%holdon;%maxU=max(U);%%Findthedatapointswithhighestgradeofmembershipincluster1%index1=find(U(1,:)==maxU);%%Findthedatapointswithhighestgradeofmembershipincluster2%index2=find(U(2,:)==maxU);%line(data(index1,1),data(index1,2),'marker','*','color','g');%line(data(index2,1),data(index2,2),'marker','*','color','r');%%Plottheclustercenters%plot([center([12],1)],[center([12],2)],'*','color','k')%holdoff;%%SeealsoFCMDEMO,INITFCM,IRISFCM,DISTFCM,STEPFCM.%RogerJang,12-13-94,N.Hickey04-16-01%Copyright1994-2002TheMathWorks,Inc.%$Revision:1.13$$Date:2002/04/1422:20:38$ifnargin~=2&nargin~=3, error('Toomanyortoofewinputarguments!');enddata_n=size(data,1);in_n=size(data,2);%Changethefollowingtosetdefaultoptionsdefault_options=[2; %exponentforthepartitionmatrixU 100; %max.numberofiteration 1e-5; %min.amountofimprovement 1]; %infodisplayduringiterationifnargin==2, options=default_options;else %If"options"isnotfullyspecified,paditwithdefaultvalues. iflength(options)<4, tmp=default_options; tmp(1:length(options))=options; options=tmp; end %Ifsomeentriesof"options"arenan's,replacethemwithdefaults. nan_index=find(isnan(options)==1); options(nan_index)=default_options(nan_index); ifoptions(1)<=1, error('Theexponentshouldbegreaterthan1!'); endendexpo=options(1); %ExponentforUmax_iter=options(2); %Max.iterationmin_impro=options(3); %Min.improvementdisplay=options(4); %Displayinfoornotobj_fcn=zeros(max_iter,1); %ArrayforobjectivefunctionU=initfcm(cluster_n,data_n); %Initialfuzzypartition%Mainloopfori=1:max_iter, [U,center,obj_fcn(i)]=stepfcm(data,U,cluster_n,expo); ifdisplay, fprintf('Iterationcount=%d,obj.fcn=%f\n',i,obj_fcn(i)); end %checkterminationcondition ifi>1, ifabs(obj_fcn(i)-obj_fcn(i-1))<min_impro,break;end, endenditer_n=i; %Actualnumberofiterationsobj_fcn(iter_n+1:max_iter)=[];Function[U_new,center,obj_fen]=stepfcm(data,U,cluster_n,expo)%STEPFCMOnestepinfuzzyc-meanclustering%[U_NEW,CENTER,ERR]=STEPFCM(DATA,U,CLUSTER_N,EXPO)%performsoneiterationoffuzzyc-meanclustering,where%%DATA:matrixofdatatobeclustered.(Eachrowisadatapoint.)%U:partitionmatrix.(U(i,j)istheMFvalueofdatajinclusterj.)%CLUSTER_N:numberofclusters.%EXPO:exponent(>1)forthepartitionmatrix.%U_NEW:newpartitionmatrix.%CENTER:centerofclusters.(Eachrowisacenter.)%ERR:objectivefunctionforpartitionU.%%Notethatthesituationof"singularity"(oneofthedatapointsis%exactlythesameasoneoftheclustercenters)isnotchecked.%However,ithardlyoccursinpractice.%%SeealsoDISTFCM,INITFCM,IRISFCM,FCMDEMO,FCM.%RogerJang.11-22-94.%Copyright1994-2002TheMathWorks,Inc.%$Revision:1.13$$Date:2002/04/1422:21:02$Mf=U,^expo;%MFmatrixafterexponentialmodificationcenter=mf*data./((ones(size(data,2),1)*sum(mf))`);%newcenterdist=distfcm(center,data);%fillthedistancematrixobj_fcn=sum(sum((dist.^2).*mf));%objectivefunctiontmp=dist.^(-2/(expo-1));%calculatenewU,supposeexpo!=1U_new=tmp./(ones(cluster_n,1)*sum(tmp));Functionout=distfcm(center,data)%DISTFCMDistancemeasureinfuzzyc-meanclustering.%OUT=DISTFCM(CENTER,DATA)calculatestheEuclideandistance%betweeneachrowinCENTERandeachrowinDATA,andreturnsa%distancematrixOUTofsizeMbyN,whereMandNarerow%dimensionsofCENTERandDATA,respectively,andOUT(I,J)is%thedistancebetweenCENTER(I,:)andDATA(J,:).%%SeealsoFCMDEMO,INITFCM,IRISFCM,STEPFCM,andFCM.%RogerJang,11-22-94,6-27-95.%Copyright1994-2002TheMathWorks,Inc.%$Revision:1.13$$Date:2002/04/1422:20:29$out=zeros(size(center,1),size(data,1));%f