ANÁLISIS DE DATOS EN SISTEMAS ELÉCTRICOS DE POTENCIA MEDIANTE TÉCNICAS BASADAS EN MÉTODOS DE DESCOMPOSICIÓN TENSORIAL

Abstract

This thesisexplorestheadvantagesoforganizingpowersystemsdataintensorar- raysandusingtensordecompositionstoprocesstheinformationoverthetraditional wayin2Darrays(matrices).Todoso,thisresearchworkreformulatessomewell- knownpowersystemproblems,suchasdataclustering,datavisualizationanddata compression forsmartmeterandPMUdata,ScreeningContingencyandCoherence Identication.This,undertheassumptionthatmanymeasureddatainthepower system canhaveamultidimensionalrepresentationintoatensor,whichcananal- ysed usingmethodsbasedontensordecomposition.Todemonstratetheadvantages of workingwiththesemethods;alongthisthesisfourdataminingmethodologies are proposedtounderstandingofthepowersystemdatacollectedbythemoderns monitoring systems: 1)An unsuperviseddataminingalgorithmisproposedforhomogeneoussmartgrid data, particularlyforelectricalloadprolesusingParallelFactorAnalysis(PARAFAC) tensor decomposition.Sincetheproposedtensorrepresentationallowstoassigna givendimensiontoaparticularvariableinvolved;datareduction,datacompression, data visualizationanddataclusteringarearchivedseparatelyforeveryvariable. 2)Data CompressionofPowerSystemsInformationfromAdvancedMonitoringIn- frastructure suchasPhasorMeasurementUnits(PMUs)andtheSmartMeter(SMs) is proposedusingTuckertensordecompositiontoachieveahighcompressionratio and alowreconstructionerror. 3)Screening ContingencyandCoherencyIdenticationarefacedsimultaneously ii based onPARAFAC-2tensordecomposition.Todoso,thetemporalandgeographical information isextractedfromatensorarraywithmultiplescontingencies. 4)A hybridclusteringalgorithmforelectricalloadprolesconsideringweather variablesisproposed.Theproposedalgorithmisbasedontheideaofunifyingcost functions resultingfromthedimensionalityreductionformulationofmultidimensional arrays. The resultsinthefourproposedmethodologiesshowthatthereformulationof the problemsyieldsmoreinformationabouttheeventstudiedthanwithtraditional matrix-based approaches.Thusdemonstratingthattensordecompositionisatool with greatpotentialfordataanalysisinelectricpowersystems.