毕业设计外文翻译.docx
《毕业设计外文翻译.docx》由会员分享,可在线阅读,更多相关《毕业设计外文翻译.docx(15页珍藏版)》请在冰豆网上搜索。
毕业设计外文翻译
AComparisonofPowerFlowbyDifferentOrderingSchemes
Abstract—Nodeorderingalgorithms,aimingatkeepingsparsityasfaraspossible,arewidelyusedtoday.Insuchalgorithms,theirinfluenceontheaccuracyofthesolutionisneglectedbecauseitwon’tmakesignificantdifferenceinnormalsystems.While,alongwiththedevelopmentofmodernpowersystems,theproblemwillbecomemoreill-conditionedanditisnecessarytotaketheaccuracyintocountduringnodeordering.Inthispaperweintendtolaygroundworkforthemorerationalityorderingalgorithmwhichcouldmakereasonablecompromisingbetweenmemoryandaccuracy.Threeschemesofnodeorderingfordifferentpurposeareproposedtocomparetheperformanceofthepowerflowcalculationandanexampleofsimplesix-nodenetworkisdiscusseddetailed.
Keywords—powerflowcalculation;nodeordering;sparsity;accuracy;Newton-Raphsonmethod;linearequations
I.INTRODUCTION
Powerflowisthemostbasicandimportantconceptinpowersystemanalysisandpowerflowcalculationisthebasisofpowersystemplanning,operation,schedulingandcontrol[1].Mathematicallyspeaking,powerflowproblemistofindanumericalsolutionofnonlinearequations.Newtonmethodisthemostcommonlyusedtosolvetheproblemanditinvolvesrepeateddirectsolutionsofasystemoflinearequations.ThesolvingefficiencyandprecisionofthelinearequationsdirectlyinfluencestheperformanceofNewton-Raphsonpowerflowalgorithm.Basedonnumericalmathematicsandphysicalcharacteristicsofpowersysteminpowerflowcalculation,scholarsdedicatedtotheresearchtoimprovethecomputationalefficiencyoflinearequationsbyreorderingnodes’numberandreceivedalotofsuccesswhichlaidasolidfoundationforpowersystemanalysis.
Jacobianmatrixinpowerflowcalculation,similarwiththeadmittancematrix,hassymmetricalstructureandahighdegreeofsparsity.Duringthefactorizationprocedure,nonzeroentriescanbegeneratedinmemorypositionsthatcorrespondtozeroentriesinthestartingJacobianmatrix.Thisactionisreferredtoasfill-in.Iftheprogrammingtermsisusedwhichprocessedandstoresonlynonzeroterms,thereductionoffill-inreflectsagreatreductionofmemoryrequirementandthenumberofoperationsneededtoperformthefactorization.Somanyextensivestudieshavebeenconcernedwiththeminimizationofthefill-ins.Whileitishardtofindefficientalgorithmfordeterminingtheabsoluteoptimalorder,severaleffectivestrategiesfordeterminingnear-optimalordershavebeendevisedforactualapplications[2,3].Eachofthestrategiesisatrade-offbetweenresultsandspeedofexecutionandtheyhavebeenadoptedbymuchofindustry.Thesparsity-programmedorderedeliminationmentionedabove,whichisabreakthroughinpowersystemnetworkcomputation,dramaticallyimprovingthecomputingspeedandstoragerequirementsofNewton’smethod[4].
Aftersparsematrixmethods,sparsevectormethods[5],whichextendsparsityexploitationtovectors,areusefulforsolvinglinearequationswhentheright-hand-sidevectorissparseorasmallnumberofelementsintheunknownvectorarewanted.Tomakefulluseofsparsevectormethodsadvantage,itisnecessarytoenhancethesparsityofL-1byorderingnodes.Thisisequivalenttodecreasingthelengthofthepaths,butitmightcausemorefill-ins,greatercomplexityandexpense.Counteringthisproblem,severalnodeorderingalgorithms[6,7]wereproposedtoenhancesparsevectormethodsbyminimizingthelengthofthepathswhilepreservingthesparsityofthematrix.
Uptonow,onthebasisoftheassumptionthatanarbitraryorderofnodesdoesnotadverselyaffectnumericalaccuracy,mostnodeorderingalgorithmstakesolvinglinearequationsinasingleiterationasresearchsubject,aimingatthereductionofmemoryrequirementsandcomputingoperations.Manymatriceswithastrongdiagonalinnetworkproblemsfulfilltheaboveassumption,andorderingtoconservesparsityincreasedtheaccuracyofthesolution.Nevertheless,iftherearejunctionsofveryhighandlowseriesimpedances,longEHVlines,seriesandshuntcompensationinthemodelofpowerflowproblem,diagonaldominancewillbeweaken[8]andtheassumptionmaynotbetenableinvariably.Furthermore,alongwiththedevelopmentofmodernpowersystems,variousnewmodelswithparametersundervariousordersofmagnitudeappearinthemodelofpowerflow.Thepromotionofdistributedgenerationalsoencourageustoregardthedistributionnetworksandtransmissionsystemsasawholeinpowerflowcalculation,anditwillcausemoreseriousnumericalproblem.Allthosethingsmentionedabovewillturntheproblemintoill-condition.Soitisnecessarytodiscusstheeffectofthenodenumberingtotheaccuracyofthesolution.
Basedontheexistingnodeorderingalgorithmmentionedabove,thispaperfocusattentiononthecontradictionbetweenmemoryandaccuracyduringnodeordering,researchhowcouldnodeorderingalgorithmaffecttheperformanceofpowerflowcalculation,expectingtolaygroundworkforthemorerationalityorderingalgorithm.Thispaperisarrangedasfollows.ThecontradictionbetweenmemoryandaccuracyinnodeorderingalgorithmisintroducedinsectionII.NextasimpleDCpowerflowisshowedtoillustratethatnodeorderingcouldaffecttheaccuracyofthesolutioninsectionIII.Then,takinga6-nodenetworkasanexample,theeffectofnodeorderingontheperformanceofpowerflowisanalyzeddetailedinsectionIV.ConclusionisgiveninsectionVI.
II.CONTRADICTIONBETWEENMEMORYANDACCURACY
INNODEORDERINGALGORITHM
Accordingtonumericalmathematics,completepivotingisnumericallypreferabletopartialpivotingforsystemsoflineralgebraicequationsbyGaussianEliminationMethod(GEM).Manymathematicalpapers[9-11]focustheirattentiononthediscriminationbetweencompletepivotingandpartialpivotingin(GEM).Reference[9]showshowpartialpivotingandcompletepivotingaffectthesensitivityoftheLUfactorization.Reference[10]proposesaneffectiveandinexpensivetesttorecognizenumericaldifficultiesduringpartialpivotingrequires.Oncetheassessmentcriterioncannotbemet,completepivotingwillbeadoptedtogetbetternumericalstability.Inpowerflowcalculations,partialpivotingisrealizedautomaticallywithoutanyrow-interchangesandcolumn-interchangesbecauseofthediagonallydominantfeaturesoftheJacobinmatrix,whichcouldguaranteenumericalstabilityinfloatingpointcomputationinmostcases.Whileduetoroundingerrors,thepartialpivotingdoesnotprovidethesolutionaccurateenoughinsomeill-conditionings.Ifcompletepivotingisperformed,ateachstepoftheprocess,theelementoflargestmoduleischosenasthepivotalelement.Itisequivalenttoadjustthenodeorderinginpowerflowcalculation.Sothenoderelatetotheelementoflargestmoduleistendtoarrangeinfrontforthepurposeofimprovingaccuracy.
Thenodereorderingalgorithmsguidedbysparsematrixtechnologyhavewildlyusedinpowersystemcalculation,aimingatminimizingmemoryrequirement.Inthesealgorithms,thenodeswithfeweradjacentnodestendtobenumberedfirst.Theresultisthatdiagonalentriesinnodeadmittancematrixtendtobearrangedfromleasttolargestaccordingtotheirmodule.Analogously,everydiagonalsubmatricesrelatetoanodetendtobearrangedfromleasttolargestaccordingtotheirdeterminants.Sotheresultsobtainedformsuchalgorithmswilljustdeviateformtheprinciplefollowwhichtheaccuracyofthesolutionwillbeenhance.Thatiswhatwesaythereiscontradictionbetweennodeorderingguidedbymemoryandaccuracy.
III.DIFFERENCEPRECISIONOFTHESOLUTIONUSINGPARTICALPIVOTINGANDCOMPLETEPIVOTING
Itissaidthatcompletepivotingisnumericallypreferabletopartialpivotingforsolvingsystemsoflinearalgebraicequations.Whenthesystemcoefficientsarevaryingwidely,theaccuracyofthesolutionwouldbeaffectbyroundingerrorshardlyanditisnecessarytotaketheinfluenceoftheorderingontheaccuracyofthesolutionintoconsideration.
Fig.1DCmodelofSample4-nodenetwork
Asanexample,considertheDCmodelofsample4-nodesystemshowninFigure1.Node1istheswingnodehavingknownvoltageangle;nodes2-4areloadnodes.Followingtheoriginalnodenumber,theDCpowerflowequationis:
Tosimulatecomputernumericalcalculationoperations,foursignificantfigureswillbeusedtosolvetheproblem.ExecutingGEMwithoutpivotingon
(1)yieldsthesolution[θ2,θ3,θ4]T=[-0.3036,-0.3239,-0.3249]T,whosecomponentsdifferfromthatoftheexactsolution[θ2,θ3,θ4]T=[-0.3,-0.32,-0.321]T.Amoreexactsolutioncouldbeobtainedbycompletepivoting:
[θ2,θ3,θ4]T=[-0.3007,-0.3207,-0.3217]T,andtheorderofthenodeafterrowandcolumninterchangesis3,2,4.Sothisisamorereasonableorderingschemeforthepurposeofgettingmorehighaccuracy.
IV.THEINFLUENCEOFNODEREODERINGONTHEPERFORMANCEOFNEWTON-RAPHSONPOWERFLOWMETHOD
Fig.2Sample6-nodenetwork
Onthebasisoftheabove-mentionedanalysis,theschemefornodereorderingwillnotonlyaffectmemoryrequirementbutalsotheaccuracyofthesolutioninsolvinglinearsimultaneousequations.SoperformanceofNewton-Raphsonpowerflowmethodwillbedifferentwithvariousnodeordering.Inthissectionthreeschemesoforderingfordifferentpurposewillbeappliedtoasample6-nodenetworkshowninFig2tocomparetheinfluenceofthemontheaccuracyofthesolution,theconvergencerate,thecalculatedamountandthememoryneededinpowerflowcomputation.ThedetailoftheperformanceisshownintableIV.
A.Puropse1SavingMemo