Graph database and graph computing for power system analysis renchang dai download pdf

Page 1


Graph

Visit to download the full and correct content document: https://ebookmass.com/product/graph-database-and-graph-computing-for-power-syst em-analysis-renchang-dai/

More products digital (pdf, epub, mobi) instant download maybe you interests ...

Graph Data Science with Neo4j: Learn how to use Neo4j 5 with Graph Data Science library 2.0 and its Python driver for your project Scifo

https://ebookmass.com/product/graph-data-science-withneo4j-learn-how-to-use-neo4j-5-with-graph-data-sciencelibrary-2-0-and-its-python-driver-for-your-project-scifo/

Modern Applications of Graph Theory 1st Edition

https://ebookmass.com/product/modern-applications-of-graphtheory-1st-edition-zverovich/

Graph Spectral Image Processing Gene Cheung

https://ebookmass.com/product/graph-spectral-image-processinggene-cheung/

Interval Methods for Uncertain Power System Analysis

Alfredo Vaccaro

https://ebookmass.com/product/interval-methods-for-uncertainpower-system-analysis-alfredo-vaccaro/

Graph Data Science (GDS) For Dummies®, Neo4j Special Edition Amy Hodler

https://ebookmass.com/product/graph-data-science-gds-for-dummiesneo4j-special-edition-amy-hodler/

Fuzzy Graph Theory: Applications to Global Problems

John N. Mordeson

https://ebookmass.com/product/fuzzy-graph-theory-applications-toglobal-problems-john-n-mordeson/

Graph neural networks for efficient learning of mechanical properties of polycrystals Jonathan M. Hestroffer

https://ebookmass.com/product/graph-neural-networks-forefficient-learning-of-mechanical-properties-of-polycrystalsjonathan-m-hestroffer/

Power System Analysis and Design 6th Edition J. Duncan

Glover

https://ebookmass.com/product/power-system-analysis-anddesign-6th-edition-j-duncan-glover/

Introduction to Graph Theory 2, 2002 reprint Edition

Douglas B. West

https://ebookmass.com/product/introduction-to-graphtheory-2-2002-reprint-edition-douglas-b-west/

IEEEPress

445HoesLane Piscataway,NJ08854

IEEEPressEditorialBoard

SarahSpurgeon, EditorinChief

JónAtliBenediktssonBehzadRazaviJeffreyReed AnjanBoseJimLykeDiomidisSpinellis

JamesDuncanHaiLiAdamDrobot AminMoenessBrianJohnsonTomRobertazzi DesineniSubbaramNaiduAhmetMuratTekalp

GraphDatabaseandGraphComputing forPowerSystemAnalysis

RenchangDai

PugetSoundEnergy Bellevue,WA,USA

GuangyiLiu

EnvisionDigital SanJose,CA,USA

IEEE Press Series on Power and Energy Systems Ganesh Kumar Venayagamoorthy, Series Editor

Copyright©2024byTheInstituteofElectricalandElectronicsEngineers,Inc.Allrightsreserved.

PublishedbyJohnWiley&Sons,Inc.,Hoboken,NewJersey. PublishedsimultaneouslyinCanada.

Nopartofthispublicationmaybereproduced,storedinaretrievalsystem,ortransmittedinanyformorbyanymeans, electronic,mechanical,photocopying,recording,scanning,orotherwise,exceptaspermittedunderSection107or 108ofthe1976UnitedStatesCopyrightAct,withouteitherthepriorwrittenpermissionofthePublisher,or authorizationthroughpaymentoftheappropriateper-copyfeetotheCopyrightClearanceCenter,Inc.,222Rosewood Drive,Danvers,MA01923,(978)750-8400,fax(978)750-4470,oronthewebatwww.copyright.com.Requeststothe PublisherforpermissionshouldbeaddressedtothePermissionsDepartment,JohnWiley&Sons,Inc.,111River Street,Hoboken,NJ07030,(201)748-6011,fax(201)748-6008,oronlineathttp://www.wiley.com/go/permission.

Trademarks:WileyandtheWileylogoaretrademarksorregisteredtrademarksofJohnWiley&Sons,Inc.and/orits affiliatesintheUnitedStatesandothercountriesandmaynotbeusedwithoutwrittenpermission.Allother trademarksarethepropertyoftheirrespectiveowners.JohnWiley&Sons,Inc.isnotassociatedwithanyproductor vendormentionedinthisbook.

LimitofLiability/DisclaimerofWarranty:Whilethepublisherandauthorhaveusedtheirbesteffortsinpreparingthis book,theymakenorepresentationsorwarrantieswithrespecttotheaccuracyorcompletenessofthecontentsof thisbookandspecificallydisclaimanyimpliedwarrantiesofmerchantabilityorfitnessforaparticularpurpose. Nowarrantymaybecreatedorextendedbysalesrepresentativesorwrittensalesmaterials.Theadviceandstrategies containedhereinmaynotbesuitableforyoursituation.Youshouldconsultwithaprofessionalwhereappropriate. Further,readersshouldbeawarethatwebsiteslistedinthisworkmayhavechangedordisappearedbetween whenthisworkwaswrittenandwhenitisread.Neitherthepublishernorauthorsshallbeliableforanylossofprofitor anyothercommercialdamages,includingbutnotlimitedtospecial,incidental,consequential,orotherdamages.

Forgeneralinformationonourotherproductsandservicesorfortechnicalsupport,pleasecontactourCustomerCare DepartmentwithintheUnitedStatesat(800)762-2974,outsidetheUnitedStatesat(317)572-3993orfax(317) 572-4002.

Wileyalsopublishesitsbooksinavarietyofelectronicformats.Somecontentthatappearsinprintmaynotbe availableinelectronicformats.FormoreinformationaboutWileyproducts,visitourwebsiteatwww.wiley.com.

LibraryofCongressCataloging-in-PublicationData:

Names:Dai,Renchang,author.|Liu,Guangyi(Scientist),author.

Title:Graphdatabaseandgraphcomputingforpowersystemanalysis/ RenchangDai,GuangyiLiu.

Description:Hoboken,NewJersey:Wiley-IEEEPress,[2024]|Includes index.

Identifiers:LCCN2023023393(print)|LCCN2023023394(ebook)|ISBN 9781119903864(cloth)|ISBN9781119903871(adobepdf)|ISBN 9781119903888(epub)

Subjects:LCSH:Graphdatabases.|Electricpowersystems.

Classification:LCCQA76.9.D32D332024(print)|LCCQA76.9.D32(ebook) |DDC005.75/8–dc23/eng/20230602

LCrecordavailableathttps://lccn.loc.gov/2023023393

LCebookrecordavailableathttps://lccn.loc.gov/2023023394

CoverDesign:Wiley

CoverImage:©AlejandroMendozaR/Shutterstock

Setin9.5/12.5ptSTIXTwoTextbyStraive,Pondicherry,India

Contents

AbouttheAuthors xiii

Preface xv Acknowledgments xvii

PartITheoryandApproaches 1

1Introduction 3

1.1PowerSystemAnalysis 6

1.1.1PowerFlowCalculation 6

1.1.2StateEstimation 6

1.1.3ContingencyAnalysis 7

1.1.4Security-ConstrainedAutomaticGenerationControl 7

1.1.5Security-ConstrainedED 8

1.1.6ElectromechanicalTransientSimulation 9

1.1.7PhotovoltaicPowerGenerationForecast 10

1.2MathematicalModel 10

1.2.1DirectMethodsofSolvingLarge-ScaleLinearEquations 10

1.2.2IterativeMethodsofSolvingLarge-ScaleLinearEquations 11

1.2.3High-DimensionalDifferentialEquations 11

1.2.4MixedInteger-ProgrammingProblems 11

1.3GraphComputing 12

1.3.1GraphModelingBasics 13

1.3.2GraphParallelComputing 14 References 14

2GraphDatabase 17

2.1DatabaseManagementSystemsHistory 17

2.2GraphDatabaseTheoryandMethod 18

2.2.1GraphDatabasePrincipleandConcept 18

2.2.1.1DefiningaGraphSchema 19

2.2.1.2CreatingaLoadingJob 20

2.2.1.3GraphQueryLanguage 21

2.2.2SystemArchitecture 25

2.2.3GraphComputingPlatform 25

2.3GraphDatabaseOperationsandPerformance 26

2.3.1GraphDatabaseManagementSystem 26

2.3.1.1ParallelProcessingbyMapReduce 27

2.3.1.2GraphPartition 29

2.3.2GraphDatabasePerformance 35 References 38

3GraphParallelComputing 41

3.1GraphParallelComputingMechanism 41

3.2GraphNodalParallelComputing 44

3.3GraphHierarchicalParallelComputing 46

3.3.1SymbolicFactorization 47

3.3.2EliminationTree 51

3.3.3NodePartition 56

3.3.4NumericalFactorization 57

3.3.5ForwardandBackwardSubstitution 58 References 59

4Large-ScaleAlgebraicEquations 61

4.1IterativeMethodsofSolvingNonlinearEquations 61

4.1.1Gauss–SeidelMethod 61

4.1.2PageRankAlgorithm 62

4.1.2.1PageRankAlgorithmMechanism 63

4.1.2.2IterativeMethod 66

4.1.2.3AlgebraicMethod 67

4.1.2.4ConvergenceAnalysis 69

4.1.3Newton–RaphsonMethod 72

4.2DirectMethodsofSolvingLinearEquations 75

4.2.1Introduction 75

4.2.2BasicConcepts 76

4.2.2.1DataStructuresofSparseMatrix 76

4.2.2.2MatricesandGraphs 78

4.2.3HistoricalDevelopment 80

4.2.4DirectMethods 81

4.2.4.1SolvingTriangularSystems 81

4.2.4.2SymbolicFactorization 82

4.2.4.3Fill-ReducingOrdering 82

4.3IndirectMethodsofSolvingLinearEquations 83

4.3.1StationaryMethods 83

4.3.1.1JacobiMethod 83

4.3.1.2Gauss–SeidelMethod 85

4.3.1.3SORMethod 86

4.3.1.4SSORMethod 86

4.3.2NonstationaryMethods 88

4.3.2.1CGMethod 88

4.3.2.2GMRES 89

4.3.2.3BCG(bi-CG) 90 References 91

5High-DimensionalDifferentialEquations 95

5.1IntegrationMethods 95

5.1.1AnOverviewofIntegrationMethodsandtheirAccuracy 95

5.1.1.1One-StepMethods 96

5.1.1.2LinearMultistepMethods 99

5.1.2IntegrationMethodsforPowerSystemTransientSimulations 100

5.1.3TransientAnalysisAccuracy 100

5.1.4TransientAnalysisStability 101

5.1.4.1AbsoluteStability 101

5.1.4.2StiffStability 102

5.2TimeStepControl 103

5.2.1AdaptiveTimeStep 104

5.2.1.1ChangebyIterationNumber 105

5.2.1.2ChangebyEstimatedTruncationError 105

5.2.1.3ChangebyStateVariableDerivative 106

5.2.2MultipleTimeStep 106

5.2.3BreakPoints 109

5.3InitialOperationCondition 110

5.4Graph-BasedTransientParallelSimulation 115

5.5NumericalCaseStudy 117

5.6Summary 123 References 124

6OptimizationProblems 125

6.1OptimizationTheory 125

6.2LinearProgramming 125

6.2.1TheSimplexMethod 127

6.2.1.1BasicFeasibleSolution 127

6.2.1.2TheSimplexIteration 128

6.2.2Interior-PointMethods 132

6.3NonlinearProgramming 138

6.3.1UnconstrainedOptimizationApproaches 139

6.3.1.1LineSearch 140

6.3.1.2TrustRegionOptimization 141

6.3.1.3Quasi-NewtonMethod 141

6.3.1.4DoubleDoglegOptimization 142

6.3.1.5ConjugateGradientOptimization 143

6.3.2ConstrainedOptimizationApproaches 145

6.3.2.1Karush–Kuhn–TuckerConditions 145

6.3.2.2LinearApproximationsofNonlinearProgrammingwithLinearConstraints 145

6.3.2.3LinearApproximationsofNonlinearProgrammingwithNonlinearConstraints 147

6.4MixedIntegerOptimizationApproach 147

6.4.1Branch-and-BoundApproach 148

6.4.2MachineLearningforBranching 150

6.5OptimizationProblemsSolutionbyGraphParallelComputing 151

6.5.1SimplexMethodBasedonGraphParallelComputing 151

6.5.2Interior-PointMethodBasedonGraphParallelComputing 154 References 156

7Graph-BasedMachineLearning 159

7.1StateofArtonPVGenerationForecasting 159

7.2GraphMachineLearningModel 160

7.3ConvolutionalGraphAuto-Encoder 162

7.3.1Auto-Encoder 162

7.3.2Auto-EncoderonGraphs 163

7.3.3ProbabilityDistributionFunctionApproximation 164

7.3.4ConvolutionalGraphAuto-Encoder 167

7.3.5GraphFeatureExtractionArtificialNeuralNetwork(R(G)) 169

7.3.6Encoder(Q)andDecoder(P) 170

7.3.7Estimationof P(V ∗ π ) 171

References 171

PartIIImplementationsandApplications 175

8PowerSystemsModeling 177

8.1PowerSystemGraphModeling 177

8.2PhysicalGraphModelandComputingGraphModel 178

8.3Node-BreakerModelandGraphRepresentation 180

8.4Bus-BranchModelandGraphRepresentation 189

8.5Graph-BasedTopologyAnalysis 190

8.5.1Substation-LevelTopologyAnalysis 190

8.5.2System-LevelNetworkTopologyAnalysis 196

References 198

9StateEstimationGraphComputing 199

9.1PowerSystemStateEstimation 199

9.2GraphComputing-BasedStateEstimation 201

9.2.1StateEstimationGraphComputingAlgorithm 201

9.2.1.1BuildNode-BasedStateEstimation 201

9.2.1.2Graph-BasedStateEstimationParallelAlgorithm 203

9.2.2NumericalExample 209

9.2.3Graph-BasedStateEstimationImplementation 215

9.2.3.1Graph-BasedStateEstimationGraphSchema 215

9.2.3.2NodalGainMatrixFormation 216

9.2.3.3BuildRHS 219

9.2.4Graph-BasedStateEstimationComputationEfficiency 220

9.3BadDataDetectionandIdentification 223

9.3.1Chi-SquaresTest 224

9.3.2AdvancedBadDataDetection 224

9.3.3BadDataIdentification 228

9.3.3.1NormalizedResidual 228

9.3.3.2LargestNormalizedResidualforBadDataIdentification 229

9.4Graph-BasedBadDataDetectionImplementation 229 References 231

10PowerFlowGraphComputing 233

10.1PowerFlowMathematicalModel 233

10.2Gauss–SeidelMethod 234

10.3Newton –RaphsonMethod 242

10.3.1BuildJacobianGraph 245

10.3.2Graph-BasedSymbolicFactorization 247

10.3.3Graph-BasedEliminationTreeCreationandNodePartition 249

10.3.4GraphNumericalFactorization 251

10.3.5BuildRight-HandSide 253

10.3.6GraphForwardandBackwardSubstitution 254

10.3.7Graph-BasedNewton–RaphsonPowerFlowCalculation 255

10.4FastDecoupledPowerFlowCalculation 257

10.4.1BuildB_PandB_PPGraphs 259

10.5Ill-ConditionedPowerFlowProblemSolution 261

10.5.1Introduction 261

10.5.2DeterminetheFeasibilityofthePowerFlow 262

10.5.3ProblemFormulationforDeterminingtheFeasibilityofPowerFlow 263

10.5.4PowerFlowFeasibilityVerification 264

10.5.5FindaFeasibleSolutionforthePowerFlowProblem 266 References 271

11ContingencyAnalysisGraphComputing 273

11.1DCPowerFlow 273

11.2BridgeSearch 276

11.3ConjugateGradientforPostcontingencyPowerFlowCalculation 282

11.4ContingencyAnalysisUsingConvolutionalNeuralNetworks 294

11.4.1ConvolutionalNeuralNetwork 295

11.4.2ConvolutionalNeuralNetworkComponents 297

11.4.2.1ConvolutionalNeuralNetworkInput 297

11.4.2.2ConvolutionalNeuralNetworkOutput 297

11.4.2.3ConvolutionalNeuralNetworkConvolutionalLayer 297

11.4.2.4CNNPoolingLayer 298

11.4.2.5CNNFullyConnectedLayer 299

11.4.3EvaluationMetrics 299

11.4.3.1Accuracy 299

11.4.3.2Precision 300

11.4.3.3Recall 300

11.4.4ImplementationofConvolutionalNeuralNetwork 300

11.5ContingencyAnalysisGraphComputingImplementation 302 References 306

12EconomicDispatchandUnitCommitment 309

12.1ClassicEconomicDispatch 309

12.1.1ThermalUnitEconomicDispatch 309

12.1.2HydrothermalPowerGenerationSystemEconomicDispatch 315

12.2Security-ConstrainedEconomicDispatch 320

Contents

12.2.1GenerationShiftFactorMatrix 323

12.2.2Graph-BasedSCEDModeling 325

12.2.3Graph-BasedSCED 327

12.2.3.1BuildupSimplexGraph 328

12.2.3.2Graph-BasedSimplexMethod 331

12.2.3.3UpdatePowerFlow 331

12.2.3.4Graph-BasedSCEDImplementation 333

12.3Security-ConstrainedUnitCommitment 334

12.3.1SCUCModel 334

12.3.2Graph-BasedSCUC 335

12.4NumericalCaseStudy 336

12.4.1Graph-BasedSCEDModeling 336

12.4.2BasicFeasibleSolution 340

12.4.3EconomicDispatchOptimalSolution 342 References 342

13AutomaticGenerationControl 345

13.1ClassicAutomaticGenerationControl 345

13.1.1SpeedGovernorControl 345

13.1.2SpeedDroopFunction 347

13.1.3FrequencySupplementaryControl 353

13.1.4FundamentalsofAutomaticGenerationControl 355

13.2NetworkSecurity-ConstrainedAutomaticGenerationControl 358

13.3Security-ConstrainedAGCGraphComputing 361 References 364

14Small-SignalStability 365

14.1Small-SignalStabilityofaDynamicSystem 365

14.2SystemLinearization 366

14.3Small-SignalStabilityMode 367

14.4Single-MachineInfiniteBusSystem 367

14.4.1ClassicalGeneratorModel 367

14.4.2Third-OrderGeneratorModel 369

14.4.3NumericalCaseStudy 373

14.4.3.1StableCase 373

14.4.3.2InstableCase 376

14.5Small-SignalOscillationStabilization 378

14.6EigenvalueCalculation 379

14.6.1Graph-BasedSmall-SignalStabilityAnalysis 382

14.6.2BuildupSmall-SignalStabilityGraph 383

14.6.3NumericalExample 383 References 388

15TransientStability 391

15.1TransientStabilityTheory 391

15.1.1StabilityRegionandBoundary 391

15.1.2EnergyFunctionMethod 391

15.1.2.1ControllingUEPMethod 392

15.1.2.2Stability-Region-BasedControllingUEPMethod 393

15.2TransientSimulationModel 393

15.2.1GeneratorRotorModel 393

15.2.2GeneratorElectro-MagneticModel 394

15.2.3ExcitationSystemModel 394

15.2.4GovernorModel 396

15.2.5PSSModel 397

15.3TransientSimulationApproach 397

15.3.1TransientSimulationAlgorithm 398

15.3.2Steady-StateEquilibriumCondition 398

15.3.3GeneratorInjectionCurrent 400

15.4TransientSimulationbyGraphParallelComputing 401

15.4.1TransientSimulationGraph 401

15.4.2LoadingDataintoGraph 403

15.4.3Graph-BasedTransientSimulationImplementation 406

15.5NumericalExample 406

15.5.1PowerFlowData 406

15.5.2DynamicData 406

15.5.3PowerFlowResults 409

15.5.4Steady-StateEquilibriumPoint 410

15.5.5GeneratorInjectionCurrentCalculation 415

15.5.6CalculateBusVoltage 416

15.5.7SimulationResults 416 References 421

16Graph-BasedDeepReinforcementLearningonOverloadControl 425

16.1Introduction 425

16.2DDPGAlgorithm 426

16.2.1Terminology 426

16.2.2QFunction 427

16.2.3QValueApproximation 427

16.2.4PolicyGradient 428

16.3BranchOverloadControl 429

16.3.1States 429

16.3.2Actions 430

16.3.3Rewards 430

16.4Graph-BasedDeepReinforcementLearningImplementation 430 References 433

17Conclusions 435 Appendix 437

Index 481

AbouttheAuthors

RenchangDai,PhD, isaconsultingengineerandprojectmanageratPugetSoundEnergy. HereceivedhisPhDdegreeinelectricalengineeringfromTsinghuaUniversity,China,in2001.

Dr.Daihasworkedonavarietyofpowersystemproblems,includingpowersystemplanning, operations,andcontrol.HewastheprincipalengineerandgroupmanagerforGlobalEnergyInterconnectionResearchInstituteNorthAmerica,whereheledateamofengineersinresearchingand developinggraphdatabaseandgraphcomputingtechnologiesforpowersystemplanningand operations.

Dr.DaiwasateamleaderforGEEnergy.InGEEnergy,hedesigned,developed,andimplementedEnergyManagementSystem.HewasalsoafoundingmemberoftheGEEnergyConsulting SmartGridCenterofExcellence,whereheconsultedonsmartgriddeploymentandrenewable energygridintegrationprojects.In2005,whenhewasaleadscientistinGEGlobalResearch, hewasawardedtheGEGlobalTechnicalAwardforhiscontributionstothedevelopmentofwind turbinegeneratorfaultridethroughtechnology.

Dr.DaiisaseniormemberoftheIEEE.Hehasworkedintensivelyongraph-basedpowersystem analysisandhaspublishedover100papersininternationaljournalsandconferences.

GuangyiLiu,PhD, isachiefscientistatEnvisionDigital.Heisleadingateamofengineersto developpowersystemapplicationsoftwarethatisbasedongraphdatabaseandgraphcomputing technologies.HereceivedhisPhDdegreeinelectricalengineeringfromtheChinaElectricPower ResearchInstitute,China,in1990.

Dr.Liuhasworkedonavarietyofpowersystemresearchfields,includingEnergyManagement System(EMS),DistributionManagementSystem(DMS),ElectricityMarket,ActiveDistribution Network,andBigData.HewastheprincipalengineerandchieftechnologyofficerforGlobal EnergyInterconnectionResearchInstituteNorthAmerica,whereheledateamofengineersin researchinganddevelopinggraphdatabaseandgraphcomputingtechnologiesforpowersystem calculation,analysis,andoptimization.

Dr.LiuisaseniormemberoftheIEEEandafellowoftheChineseSocietyofElectrical Engineering.Hehasworkedintensivelyonpowersystemanalysisandoptimization,andhehas publishedover200papersininternationaljournalsandconferences.

Westartedtoworkonpowersystemanalysisdecadesago.Improvingpowersystemanalysis computationefficiencyisanongoingtaskandachallengeforonlineapplicationsandoffline analysesinthepowerindustry.Thetirelessandremarkableeffortshavebeenendeavoredby researchersandengineerstryingtoachievereal-timesteady-state,dynamic,andoptimization applications.Keepingthisambitioninmind,theideaofgraphcomputingwasinspiredatanoccasionalconservationwithProfessorShouchengZhangfromStanfordUniversityin2015whenhe introducedastart-upcompanyandtheirworkongraphdatabasestous.Theperfectmatchofgraph natureandpowernetworkstructuresparkedthelongexplorationandjourneyofresearchingand developinggraphcomputingtheory,algorithms,methods,approaches,andapplicationsforyears. Thisbookisacomprehensivesummaryandknowledgesharingofourresearchandengineering workongraphcomputingforpowersystemanalysis.

Thisbookisdividedintotwoparts.PartIdevotesthefirstsevenchapterstohighlightingthetheoreticalmethodsandapproaches.PartIIiscomposedofChapters8–17onpracticalimplementationsandapplications.PartIservesprerequisitesofgraphcomputingwithbasicsandadvancesof graphdatabases,graphparallelcomputing,andknowledgeofsolvingalgebraicequations,optimizationproblems,differentialequations,andtheircombinations.PartIIprovidesacomprehensive illustrationofgraph-basedpowersystemmodeling,analysisapproaches,andimplementations withdetailedgraphqueryscripts.TheimplementedapplicationspresentedinPartIIcoverpower systemtopologyanalysis,stateestimation,powerflowcalculation,contingencyanalysis,securityconstrainedeconomicdispatch,security-constrainedunitcommitment,automaticgeneration control,small-signalstability,transientstability,anddeepreinforcementlearning.

Currently,thepracticeofpowersystemmodelingfocusesonusingrelationaldatabases.Inrelationaldatabases,dataareorganizedandmanagedintables.Therelationshipsbetweentablesare connectedbyseparatedtablesorbyusingajoinoperationtosearchcommonattributesindifferent tablestofindtherelationships.Inthismechanism,itischallengingtomaintainandmanipulate alargedatasetinarelationaldatabase.

Contrarytotherelationaldatabase,agraphdatabaseusesgraphstructuresforsemanticqueries withnodesandedgestostoredata.Theessentialdifferencesbetweengraphdatabasesandrelationaldatabasesarethattheedgedirectlydefinesthedatarelationshipandgraphdatabasesare designedforparallelcomputing.Graphcomputingmodelspowersystemsasagraphwhichisconsistentwiththefactthatpowersystemphysicallyisagraph – busesareconnectedbybranchesasa graph.Thegraphdatastructuretellsthetopologyofthepowernetworkandtherelationsofpower systemcomponentsnaturally.Thegraphcomputingmechanismbyusingqueriesonnodesand graphpartitions,promotesparallelcomputingforpowersystemapplications.

Graphdatabasesarenewtothepowerindustry.Graphcomputingisnoveltoresearchersand engineers.Themainobjectiveofthisbookistoprovidearoadmapandguidancetoreaderstolearn thealternativeandinnovativeapproachestomodelingandsolvingpowersystemproblemsfrom scratch.Forthispurpose,theprocessesofdefiningvertex,edge,andgraphschema,creatingaloadingjob,anddevelopingdetailedgraphqueriesaredemonstrated.Thescriptsforeachpowersystem analysisareprovidedandexplainedindetailtofacilitatereaderstograduallyandcomprehensively mastergraphcomputing.Tomakethisbookareferencetograduatestudents,illustrativeproblems arepresentedandhands-onexperiencesingraphcomputingdesignandprogrammingareprovided bydetailedscripts.

Thegraphcomputingresearchactivitiesarestillprogressing.Webelievethatgraphcomputing hasgreatpotentialinvariouspowersystemanalyses.Wehopethisbookcaninviteandinspire researchersandengineerstostudyandresearchgraphdatabasesandgraphcomputingandapply thegraphcomputingtheoryandapproachestodeveloppowersystemapplications.

RenchangDai GuangyiLiu

Acknowledgments

Thisbookisaresultofthefascinatingjourneyofourstudyandresearchworkfordecades.Wefirstly acknowledgeourresearchadvisors,ProfessorMingDingfromtheHefeiUniversityofTechnology, ProfessorBomingZhangfromTsinghuaUniversity,andProfessorErkengYufromtheChina ElectricPowerResearchInstitute,forusheringusintotheworldofpowersystemanalysisand leadingusintotheresearchareaofpowersystemreal-timeapplications.Theknowledgeand experiencetheysharedwithusalongwiththeiradvice,influenceourresearchtothisday.

Wealsoacknowledgeourresearchteamfortheircontributionsandsupportonthischallenging andrewardingworkformanyyears.AgreatthanksgoestoDr.TingChen,Mr.HongFan,Dr.Chen Yuan,Dr.JingjinWu,Dr.YitingZhao,Dr.JunTan,Dr.JiangpengDai,Dr.YongliZhu,Dr.Longfei Wei,Dr.XiangZhang,Dr.PengWei,Dr.YachenTang,Mr.KewenLiu,Mr.WendongZhu, Mrs.TingtingLiu,Mrs.BowenKan,Mr.HaiyunHan,Mr.LetianTeng,Dr.KaiXie,Mr.Zhiwei Wang,Dr.XiChen,Mrs.ZiyanYao,Dr.WeiFeng,Dr.YijingLiu,Mrs.JingHong,Mr.Huaming Zhang,Dr.SaeedD.Manshadi,Dr.MarianaKamel,Dr.YaweiWang,Mr.YananLyu,andmany othercolleaguesandinterns,fortheircontributionsontheresearchanddevelopmentofthematerialpresentedinthisbookandwonderfulideasaboutthegraphcomputingrelatedresearchtopics.

WearegratefulandthankfultoProfessorFranLifromtheUniversityofTennessee,Knoxville, ProfessorJianhuiWangfromSouthernMethodistUniversity,ProfessorHsiao-DongChiangfrom CornellUniversity,andProfessorYinyuYefromStanfordUniversityforpartneringwithusonthe researchanddevelopmentworkandfantasticdiscussionongraphcomputing.

Lastbutnotleast,wewouldliketothankourfamilies.Weunderstandwritingthismanuscriptis notaneasytaskfromdayone.Therealjourneyisevenharderthanexpected,withunexpected detours.Weexpressourheartfeltthankstothemfortheirsupportandunderstandingoverthepast severalyears.

RenchangDai GuangyiLiu

Introduction

Theelectricalpowersystemhasbeenrevolutionizingoverthedecadesintoahighlyinterconnected, large,andcomplexrenewablesystem.Populationsandeconomicgrowthgloballydemandhigher electricity.Transactionscrossinglargeareasareencouragedtomakemoreeconomicandenvironmentalsenseandresultinlargepowerflowingoverawidearea.High-voltagetransmissiontechnologiesboostedvoltagelevelsto1000kVUltra-High-VoltageAlternating-Current(UHVAC)and ±800kVUltra-High-VoltageDirectCurrent(UHVDC)totransmitpoweroverthousandsofmiles [1].Advancedpowerelectronicdevicesenableflexiblealternatingcurrenttransmissionsystem (FACTS),forinstance,staticvarcompensatorandvoltagesourceconverter-basedSTATCOM, beingadaptedtocontrolpowerflowagilelyandaccuratelyinelectricpowergrids[2,3].

Theacceleratingdecarbonizationofenergy systemspromotedandpromisedworldwide requiresarisingpenetrationlevelofrenewable energy,distributedenergy,andenergystorage makingthepowersystemeverlargeandmorecomplex.Tomakethelarge,complex,and dynamicpowersystemsecureandcost-effective,real-timemonitoring,operating,andcontrol arecrucial.Accurateandfastcalculation,combinedwithintelligentdecisionsonpowersystems ismorevitalthanever,shiftingfromtheanalyticalEnergyManagementSystem(EMS)tothe intelligentEMS[4].

Whilethepowersystemhasbeenevolvingtobebigger,thepowersystemisgettingtobemore intelligent.Inthepowerindustry,thesmartandintelligentgridisdevelopedbasedonthefollowing technologies:bigdata,deeplearning,andhigh-performancecomputing[5,6].Bigdataanddeep learningusuallyinvolveintensivecomputingefforts,thushigh-performancecomputingisthekey tomakingsmartdecisionsontime.Intelligentreal-timeanalysisbasedonmulti-sourcebigdata analysis,deeplearningtechniques,andhigh-performancecomputing,asthetrend,makesthe powersystemadaptiveandpredictivepossible.

Comparedwiththetraditionalgrid,themodernpowersystemisoperatedundermoreuncertaintybecauseoftheintermittentrenewableenergyandpowermarkettransactionsasdescribed above.Intelligentreal-timeanalysisandcalculationneedtoadopttheseuncertaintiesmorequickly toanticipateextremeeventstomakebetterdecisionstimely.Thenaturalreflectionofhuman beingsincomplexoperatingenvironmentsrequiresmoreaccurateandintuitivedatamodels andpowerfulcalculationmethods.Forexample,toprovidepowerfultechnicalsupportforintelligentreal-timeanalysisandcalculation,introducingthelatestnumericalcalculationtechnology intothepowersystempowerflow,stateestimation,contingencyanalysis,security-constrained automaticgenerationcontrol,security-constrainedunitcommitment,andfaster-than-real-time transientsimulationrequirenoveldatastructureandcalculationapproach.Thecurrentdata

processingandcalculationapproachesoftheexistingpowersystemapplicationsfacethefollowing threechallenges:

1)DataManagementandAnalysis:Itisarequirementandchallengetodevelopdataacquisition, processing,andstoragetechnologiesthatcansimultaneouslymeettheneedsofgridonlineanalysisandofflineplanningforconvergedmulti-sourcebigdatasets.

2)MathematicalMethodsandComputation:Itisnecessaryandchallengingtodevelopnewmathematicaltoolsandalgorithmstoachievefaster-than-real-timegridsimulation.

3)ModelsandSimulations:Theparalleldynamicsimulationframeworkofpowersystemsisnot currentlydevelopedenoughtosupportreal-time,wide-areaprotection,andcontrol.

ThetrendsoftheelectricpowersystemarechallengingtheexistingEMSandMarketManagementSystem(MMS)intheircomputationallyintensiveapplications,suchaspowerflow,stateestimation,contingencyanalysis,multi-timepointnetworkanalysis,security-constrainedautomatic generationcontrol,security-constrainedunitcommitment,security-constrainedeconomicdispatch,andfaster-than-real-timetransientsimulation.Next-generationEMS/MMSisrequiredto beevolvingtoaccommodatelargerscale,highercomplex,moreconstrained,anduncertainpower systemswithafasterthanreal-timemannerorevenlook-aheadcapabilitywithfuturesituational awareness[7].

Tomeettheabovechallenges,therequirementsandgoalsofparallelanalyticaltechnologiesand toolsareurgentlyneededtosupportnewdatamanagementtoolsandrapidcomputationalanalysis methods.

Duetotheirefficientdatamanagementandrapidcomputationalanalysiscapabilities,graph databaseandgraphcomputingtechnologiesaregainingmoreandmoreattention.Theyarepromisingtechnologiestoeffectivelysolvetheproblemofbigdatarapidanalysisandprocessing.Inthe fieldofe-commerce,graphcomputingtechnologybasedonagraphdatabaseplaysakeyrolein real-timetradingandreal-timeanalysissuchasanti-moneylaundering,badtransactiondetection, intelligentnavigation,andotherfields.ManyInternetcompanieshavealsodevelopedtheirgraph computingtechnologiesandproducts,suchasPregel,agraphcomputingsystemdevelopedand designedbyGoogle.TheTrinityprojectofMicrosoftResearchisaboutgraphdatabaseandgraph computingprojects.Google’sPregelproductshavebecomeoneoftheindustryexamplesofsuccessfulgraphdatabaseapplications.

Inthefieldofanalysisofpowergrids,usinggraphdatamanagementandcomputationtechnologyisnew.Thisbooksummarizestherecentresearchanddevelopmentachievementsonthistopic. Graphdatabasearchitecturetopowersystemsisfirstintroduced,whichisneededtosupport fine-grainedparallelcomputingtoimprovepowersystemcomputationefficiency.Thetraditional relationaldatabaseisreplacedbyagraphdatabasetomodelpowersystemsandimplement applications.Usingthegraphdatabase,theprogramtosolvelarge-scalealgebraicequations, high-dimensionaldifferentialequations,andoptimizationproblemsisreconfiguredandredesigned toaccommodategraphparallelcomputinginthisbook.Byusinggraphdatabaseandgraphcomputing,thecomputationalmodelcanbeintegratedwiththegridmodel,datastorage,andnumerical calculation,whilemakingfulluseofbigdatatechnologiessuchasmemorycomputing,distributed parallelcomputing,anddecompositionaggregation.Thetechnologyhasthesignificantadvantagesof largescale,highspeed,andhighefficiencyofcomputingdata,andprovidesatechnicalsolutionwith greatpotentialfordatamanagementandanalysisandcalculationofagiantpowergrid.

Withoutfastandaccuratecalculations,atimelyresponsetoreal-timeeventsisimpossibleand thesystemisrunningatrisk.AnalysisoftheNorthAmericanblackoutof14August2003 showsthatdelayedandmissedresponsesarethemainreasonsforthewide-areablackout[8].

Sequence-of-eventsrecordsshowthatthesystemexperiencedtwoandahalfminutesofdisturbancesfromtheinitialeventtothesystemcollapse.Intheinitialcriticalnineseconds,thehundreds ofgeneratorsthattrippedofflinewerenotfullycapturedbytheEMSsystem,sincetheEMSupdate cycle(typicallyinminutes)ismuchlongerthanthecriticaleventsinterval.Clearly,theopportunity totaketimelymeasurestopreventtheblackoutwasmissedandthepossibilityofamoretimely responsewouldhavebeenenhancedbyanEMSwithafastercycletime.

ThetimehascomewhenitiscriticaltoimprovethecomputationefficiencyofEMSapplications toaccommodatemodernpowersystemsthatareofever-increasingsizeandoperationalcomplexity. ThegoaltodayistoprovideacycletimeequivalenttotheSupervisoryControlandDataAcquisition (SCADA)cycletimeorfastertoprovidelook-aheadcapabilitywithfuturesituationalawareness usingforecastedandscheduledinformationofloadforecasting,unitcommitment,andoutage schedule.Theanalyticalprocessingtimeneedstobereducedfromtensofsecondstosubseconds [7].Tomeetthisrequirement,technologiesforthenext-generationofhigh-performanceEMSare beingstudied[9–11].However,thecomputationcapabilitytocompletethecoreEMSapplications, suchasstateestimation,powerflow,andcontingencyanalysisataSCADAsamplingratehasnot yetbeenachieved.

ToachieveEMScomputationcycletimesthatarefasterthantheSCADAsamplingrate,anovel databasearchitecturealongwithfastcomputationalmethodsarepresentedinthisbook.Among thevariouscomputationaltoolsavailabletoimprovecomputationefficiency,parallelcomputing isapromisingtechnology,providingabundantstoragealongwithmultipleprocessingpaths [12,13].In[12],parallelstateestimationusingapreconditionedconjugategradientalgorithm andanorthogonaldecomposition-basedalgorithmisproposed.Theproposedalgorithmcansolve stateestimationproblemsfasterusingparallelcomputing,butitisinfeasibletodealwithalarge conditionnumberofagainmatrix.AlvesandMonticelliin[13]proposedanapproachtosolving contingencyanalysisbyparallelcomputeranddistributednetwork.Toutilizethelinearityofthe powersystemcomponent,currentbalanceequationssubjectedtoKirchhoff’scurrentlawareused tomodelpowerflowproblemsin[14].Theresultshowsareductionincomputationaltimebyover 20%whenusingthecurrentbalanceequations.

Thestate-of-the-artcommunicationtechnologiesandcomputationtechnologiesininformation systemsarebrilliantlyshowingpowersystemengineersatechnicalsolutiontomeasure,monitor, andanalyzeelectricpowersystemswidelyandquickly.However,whentheexponentiallygrowing dataisacquiredatthecontrolcenter,thedatabaseandcomputationengineconsumesalongertime toprocesswhichdeterioratesthecomputationefficiency.Toachieveanalysiswithhighcomputationefficiency,novelsystemarchitecture,andfastcomputationalalgorithmsareneededtoassist operatorstoensureareliable,resilient,secure,andefficientelectricpowergridpromptly.Among thevariouscomputationalalgorithms,parallelcomputingisapromisingtechnologytoimprove computationefficiencytakingadvantageofmoderncomputationtechnology,abundantstorage space,andparallelcapabilityofdatabaseandGPU.Multiple-coreCPUsandGPUsareavailable nowadaysasaffordablehardwareconfigurationstofacilitateparallelcomputing.However,the state-of-artEMS/MMSdoesnoteffectivelyharnessthemulti-threadedparallelizationcapability intheirapplications[12,13]forthereasonofthetraditionalrelationaldatabaseandcomputation algorithmsappliedbyEMS/MMSwerenotdesignedforparallelcomputing.

Toaccommodateparallelcomputing,boththedatabaseandcalculationapproachesfortheEMS/ MMSapplicationsneedtoberedesignedtofitintoaparalleldatabasemanagementsystemand parallelcomputing.Previousworksinvestigatedthefeasibilityofadoptinggraphcomputingon topologyprocessing,stateestimation,powerflowanalysis, “N 1” contingencyanalysis,and security-constrainedeconomicdispatch[15–21].Realizingthatparallelprocessingofthepower

systemapplicationsneededbyreal-timeoperationandlong-termplanningcanbeenhancedbytakingadvantageoftheembeddedgraphcharacteristicsofapowersystem,thisbookhasmarrieda graph-baseddatabasewithgraphcomputingtoachievehighcomputationaleffectivepowersystem analysistoaccommodatetheevolvingpowersystemsandpowermarket.

Toaccommodateparallelcomputing,databaseandmathematicalmodelforpowersystemcalculationneedtoberedesignedtofitintoparalleldatabasemanagement,parallelanalysis,andfast visualization.

Inthisbook,thecriticalpowersystemapplicationsarerevisited.Thecomputationalapproaches involvedintheseapplicationsareintroducedindetail.Theseapproachesareabstractedtobemathematicalproblemsinsolvinglarge-scalealgebraicequations,high-dimensionaldifferentialequations,andmixedintegerlinearoptimizationproblems.Graphdatastructureandgraphparallel computingareintroducedtomodelthepowersystemingraphandsolvetheproblemsinparallel.

1.1PowerSystemAnalysis

1.1.1PowerFlowCalculation

Powerflowcalculationisawell-knownapplicationinpowersystemanalysis.Theintention ofpowerflowcalculationistoobtainbusvoltagemagnitudeandangleinformation.Oncethe voltageinformationisknown,activepowerandreactivepowerflowoneachbranchcanbeanalyticallydetermined.Inmathematics,thepowerflowcalculationmodelisasetofhigh-dimensional nonlinearalgebraicequations.

Thereareseveraldifferentmethodstosolvenonlinearequations.Thewell-knownNewton–RaphsonmethodlinearizesequationsusingaTaylorserieswiththelineartermonly.IndustrygradeEMSalsousestheFast-decoupledpowerflowmethodtoapproximateactiveandreactiveflow equationsbydecouplingvoltagemagnitudeandanglecalculations.Althoughdecoupledpower flowmethodtakesafewmoreiterationsthanNewton–Raphsonmethodtoconverge,eachiteration takesmuchlesstime.Forreactance-dominatedtransmissionnetworks,decoupledpowerflow methodoutperformstheNewton–Raphsonmethodoncomputationefficiency.Thecostisthe approximationonJacobianmatricesbydecoupledpowerflowmethoddeterioratespowerflowconvergence.Usually,inindustry-gradeEMS,decoupledpowerflowmethodisconductedfirst,then Newton–Raphsonmethodsecondifdecoupledpowerflowmethoddiverges.Thisstrategypracticallyprovidessupportingevidenceofitseffectivenessforcontingencyanalysisforalarge-scalesystemwiththousandsofcontingencies.

1.1.2StateEstimation

Thepowersystemstateestimation(SE)isbasedonreal-timetelemetryfromSCADA.Thenetwork topologyconnectionofthepowersystemisdeterminedinreal-time,alongwiththereal-timeoperatingstateofthepowergridwhichformsthebasisoftheonlineanalysissoftware.ItservestomonitorthestateofthegridandenablesEMSstoperformvariousimportantcontrolandplanningtasks suchasestablishingnearreal-timenetworkmodelsforthegrid,optimizingpowerflows,andbad datadetectionandanalysis.

ThereareatleastthreemajoraspectsofthefuturepowergridthatwilldirectlyimpactSE research.First,moreadvancedmeasurementtechnologieslikephasormeasurementunitshave offeredhopefornearreal-timemonitoringofthepowergrid.

1.1PowerSystemAnalysis

Second,newregulationsandmarketpricingcompetitionmayrequireutilitycompaniestoshare moreinformationandmonitorthegridoverlargegeographicalareas.Thiscallsfordistributedcontrol,andhence,distributedSEtofacilitateinterconnection-widecoordinatedmonitoring.

Lastly,tofacilitatesmartgridfeaturessuchasdemandresponseandtwo-waypowerflow,utility companieswillneedtohavemoretimelyandaccuratemodelsfortheirdistributionsystems.This callsforSEatthedistributionlevel,whichplacesmorestringentrequirementsonSEalgorithms.So far,utilitycompanieshavedonelittleinimplementingSEindistributionsystems,eventhoughSE hasbeendeployedextensivelyintransmissionsystemsfordecades.However,astheelectricpower gridbecomessmarter,moredistributionautomationwillbeneededandSEatthedistributionlevel willbecomemoreimportant.Thecontrolmechanisminthedistributionsystemwillmostlikelybe distributedandactiveinnature,andsowillbethecorrespondingSEfunctions.Thisnecessitatesthe developmentofnewdistributedSEalgorithmsthatavailthemselvesofthesubstantiallyincreased numberofreal-timemeasurements.

1.1.3ContingencyAnalysis

Itisachallengeandagoaltooperatealarge-scale,complex,anddynamicpowergridwithsafety andcost-effectiveness.Contingencyanalysisisoneoftheapplicationstosecurepowersystems operatingwithnoviolation.ContingencyanalysisusesbasecasepowerflowdrivenfromSEto assessthesecurityofpowersystemsunderthecontingencyofasingleequipmentoutageandtheir combinations.Contingencyanalysisisusuallyrunningperiodicallyeveryonetotwominutes.

Thecontingencyanalysisisusuallybasedonanonlinepowergridanalysistofigureouttheweak pointandsecurityrisksofthepowergridandissueanalarmwhenthesystemisrunningatrisk.It facilitatesdispatchingoperatorstodealwithpotentialoperationissuesintimetopreventcascading eventsandblackouts.

Contingencyanalysisistime-consumingasitinvolvesalargenumberofcomputationsofACload flow.Toreducethecomputationaltime,anautomaticcontingencyscreeningapproachisbeing adoptedwhichidentifiesandranksonlythoseoutageswhichcausethelimitviolationonpower flowinthelinesorvoltagesonthebuses.Practically,onlyselectedcontingencieswillleadtosevere conditionsinthepowersystem.Therefore,theprocessofidentifyingtheseseverecontingenciesis referredtoascontingencyselectionandthiscanbedonebycalculatingseverityindicesforeach contingency.Thisisimportanttotargetthevulnerablepointinalarge-scalepowersystemnetwork withaminimumtimerequirement.

Thepotentialofartificialneuralnetworksfornonlinearadaptivefilteringandcontrol,theirabilitytopredictsolutionsfrompasttrends,theirenormousdataprocessingcapability,andtheirability toprovidefastresponsesinmappingdatamakethemapromisingtoolfortheirapplicationtopower systems.

Lookingforward,real-timeandintelligenttechnologiesneedtobedevelopedforcontingency analysistopromotealookaheadandpredictivesecurityawareness.Thedevelopmentofbigdata andhigh-performancecomputingtechnologyisthekeytomakingthisgoalpossible.

1.1.4Security-ConstrainedAutomaticGenerationControl

Theautomaticgenerationcontrol(AGC)isusedtobalanceactivepowerandregulatetie-linepower flowwhileminimizingthepowergenerationcost.Inthepresentstateoftheart,theAGCbasepoint isdeterminedbytheeconomicdispatch(ED)andAGCregulatestheareacontrolerrortobezero andcontrolsthetie-linepowerflowtothedesiredcommand.

EDoptimizesgenerationundernetworksecurityconstraints.Therearetwocommonlyused methodsforactiveEDinpowersystems:(i)offlineEDand(ii)onlineED.

OfflineEDcalculatesunitcommitmentanddispatchesunitactivepoweroutputforthenextday orthenextfewdaysinatimeintervalofhoursbasedonthegenerationcapacity,gridnetworkconstraints,aswellastheforecastedload.

SinceofflineEDisbasedonloadforecasting,thegenerationdispatchmaynotaccuratelymeetthe actualload.Theoperatingconditionsofthepowersystemarechanging,andtheactivepoweroutputofthegeneratorsmaydeviatefromthescheduledpowergenerationsetpoint.Therefore,online EDadjuststhegenerationoutputsetbyofflineEDcontinuouslytosatisfythepowersystem’sactual operatingpointinashorttimeinterval(5–15minutes).

Inpractice,foramiddle-scalepowersystem,EDcalculationwithnetworksecurityconstraints takesminutes.DuetotheshortcycleofAGCcontrolconflictingwiththeextensivecomputation effortsofthesecurity-constrainedoptimization,networksecurityconstraintsarenotmodeledin AGCinreal-time.WhenpowersystemoperationchangessignificantlyintheEDcycle,AGCcommandscannotguaranteethatthenetworksecurityconstraintsaresatisfied.

ThepresentstateoftheartassumesthepowersystemoperationpointdoesnotchangesignificantlyenoughbetweentwoEDexecutionstopushtheAGCbasepointdeterminedbyEDtoviolate networksecurityconstraints.Thisassumptionisnotalwaystrueinpowersystemoperations.

Inthecaseofintermittentrenewableenergypenetratedpowersystemsandfast-responsepower electronics-basedgenerationintegratedtransmissionanddistributionpowernetwork,thesystem powerflowhasahighprobabilityofshiftingawayfromthebasepointwhichisoptimizedinthe cycleofED.AGCdoesnotoptimizepowerflowwithinnetworksecurityconstraints.Withthe presentAGCcommand,thepowerflowmayresultinviolations.Thus,theAGCwithoutnetwork securityconstraintspresentsrisksinpowersystemoperation.

WhenSEestimatesviolatedpowerflow,thepresentstateoftheartheuristicallychangesgenerationlimitsofgeneratorsforAGCregulationtoalleviatetheriskofoverflowunderAGCcommand. ThisapproachhasdrawbacksinthatitisheuristicandtheSEexecutioncycleinminute-scalecannotfitintoAGCexecutioninseconds.

Takingadvantageofthehighperformanceofgraphcomputing,thenetworksecurity-constrained AGCispotentiallyachievable.

1.1.5Security-ConstrainedED

Optimizationtheoryisanimportanttoolindecisionscienceandtheanalysisofpowersystemoperation.Insecurity-constrainedED,objectivesaresetasaquantitativemeasureoftheperformanceof thesystemunderstudy,whichcouldbethepowersystemgenerationcost,renewableenergy,ora combinationofquantities.Themainobjectiveistofindvaluesofthevariablesthatoptimizethe objective.Inaddition,thevariablesarerestrictedorconstrainedtomeetphysicallawsandsecurity requirements.Theprocessofidentifyingobjectives,variables,andconstraintsforagivenproblemis knownasmodeling.Oncethemodelhasbeenformulated,anoptimizationalgorithmcanbeused tofinditssolution.

Unconstrainedoptimizationapproachesarethebasisofconstrainedoptimizationalgorithms. Particularly,mostoftheconstrainedoptimizationproblemsinpowersystemoperationcanbe convertedintounconstrainedoptimizationproblems.Themajorunconstrainedoptimization approachesthatareusedinpowersystemoperationarethegradientmethod,linesearch,Lagrange multipliermethod,Newton–Raphsonoptimization,trust-regionoptimization,quasi-Newton method,doubledoglegoptimization,conjugategradientoptimization,andsoon.

Ageneralformulationofconstrainedoptimizationapproachescanbemodeledas:

where Ω = x ci x =0, i ; cj x ≥ 0, j Ι

Linearprogramshavealinearobjectivefunctionandlinearconstraints,whichmayincludeboth equalitiesandinequalities.Thefeasiblesetisapolytope,aconvex,connectedsetwithflat,polygonalfaces.Thecontoursoftheobjectivefunctionareplanar.

Powersystemoperationproblemsarenonlinear.Thus,nonlinearprogramming(NLP)-based techniquescanhandlepowersystemoperationproblemssuchastheoptimalpowerflowproblem andsecurity-constrainedEDwithnonlinearobjectiveandconstraintfunctions.

Thelinearprogrammingmodelsdiscussedabovehavebeencontinuous,inthesensethatdecisionvariablesareallowedtobefractional.However,fractionalsolutionsarenotrealisticinpower systemunitcommitment.Thisproblemiscalledtheinteger-programmingproblem.Itissaidtobea mixedintegerprogramwhensome,butnotall,variablesarerestrictedtobeintegers.

1.1.6ElectromechanicalTransientSimulation

Transientstabilityanalysisassessesthestateofthepowersystemafteraseveredisturbanceusing transientsimulations.Themathematicalmodelfordescribingpowersystemdynamicbehaviorisa nonlineardynamicsystemthatincludeshigh-dimensionalnonlineardifferentialequationsandlargerscalenonlinearalgebraicequations(DAEs).Whenthetimedomainmethodisappliedtopower systemtransientsimulation,differential-algebraicequationsaresolvedbyusingtime-consuming numericalintegrationmethods.Toimprovethecomputationefficiency,thechoiceofaproperstep sizeandparallelcomputingisessential.

Ifthestepsizeistoolarge,theresultwillbecomeinaccurateorevencompletelywrongwhenthe largestepsizeisnotwithintherangeofnumericalstability.Ifthestepsizeistoosmall,thetransient simulationwilltakelongerthannecessarytokeeptheaccuracy.Theadaptivetimestepisasolution thatusesthesmallestpossibletimesteptoobtainanaccurateresult,therebyincreasingthecalculationspeedwhileensuringthecalculationaccuracy.

Ineachtimestep,weneedtosolvehighorderhigh-dimensionaldifferential-algebraicequations. Tofurtherimprovethecomputationefficiency,aparallelcomputingalgorithmhasbeeninvestigatedinpowersystemtransientsimulation.Inthisbook,agraph-basedparallelcomputingmethod isdemonstratedwiththeadaptivetime-stepnumericalintegrationmethod.

Usingthesequentialmethod,ineachiteration,thenetworkequationsaresolvedtoupdatethe networkbusvoltageincludingthegeneratorterminalvoltage.Differentiateequationsusegeneratorterminalbusvoltageasaboundaryconditiontosolvethedynamicstatesofthegenerator, exciter,governor,PowerSystemStabilizer(PSS),andcurrentinjectionsfromgeneratorstonetworks.Theupdatedcurrentinjectionsareappliedtosolvenetworkequationsinthenextiteration untiltheconvergedsolutionisachieved.

Graphcomputingdemonstratedoutperformanceonpowersystemsteady-stateapplications wherethetechnologywillbeusedtosolvealgebraicequationsinthesequentialmethod-basedtransientsimulation.Inthesequentialmethod,sincethedifferentialequationsforeachgenerationsystemareindependentoncetheterminalvoltagesaresolvedbyalgebraicequations,thedifferential equationsetscanbesolvedbygraphparallelcomputingnaturallywhichwillbeaddressedinthis bookindetail.

1.1.7PhotovoltaicPowerGenerationForecast

Inrecentyears,therapidexhaustionoffossilfuelsources,environmentalpollutionconcerns,and theagingofdevelopedpowerplantsareconsideredcrucialglobalconcerns.Asaconsequence, renewableenergyresourcesincludingwindandsolarhavebeenrapidlyintegratedintotheexisting powergrids.Thereliabilityofpowersystemsdependsonthecapabilityofhandlingexpectedand unexpectedchangesanddisturbancesinproductionandconsumptionwhilemaintainingquality andcontinuityofservice.Thevariabilityandstochasticbehaviorofphotovoltaic(PV)powerare causedbyincludingvoltagefluctuations,aswellaslocalpowerqualityandstabilityissues[22]. Hence,accuratephotovoltaicpowergenerationforecastingisrequiredfortheeffectiveoperation ofpowergrids[23].

Thestudiesinsolarirradianceandphotovoltaicpowerforecastingaremainlycategorizedinto threemajorclasses:

1)Thepersistencemodelsseverasabaselinethatassumestheirradiancevaluesatfuturetime stepsareequaltothesamevaluesattheforecastingtime[22].

2)Physicalmodelsemployphysicalprocessestoestimatefuturesolarradiationvaluesusingastronomicalrelationships[24],meteorologicalparameters,andnumericalweatherpredictions (NWPs)[25].

3)Statisticalandartificialintelligencetechniquesestimateorregresssolarirradianceandphotovoltaicpowergeneration[26–32].

Toremovethestrongsmoothnessassumption,increasethegeneralizationcapability,and improvethecomputationefficiency,inthisbook,theproblemofspatio-temporalprobabilisticsolar radiationforecastingispresentedasagraphdistributionlearningproblem.Intheapproach,asetof solarmeasurementsitesinawideareaismodeledasanundirectedgraph,whereeachnoderepresentsasiteandeachedgereflectsthecorrelationbetweenhistoricalsolardataofitscorresponding nodes/sitestomodelthesolarradiationspatio-temporalcharacteristics.

1.2MathematicalModel

Ingeneral,powersystemanalysiscouldbetransformedtosolvealinearsystem Ax = b,differential equations dx dt = fx , t ,and/oroptimizationproblemsmin x Ω fx .Asafundamentalfunction,graph parallelcomputingapproachestosolvethethreetypicalmathematicalproblemsinvolvedinpower systemanalysis,optimization,andsimulationsarekeycomponentsinthisbook.

1.2.1DirectMethodsofSolvingLarge-ScaleLinearEquations

Directmethodsarewidelyusedtosolvelinearequationsbyafinitesequenceofoperations.Inthe absenceofroundingerrors,directmethodswoulddeliveranexactsolution.Besidestheirhighefficiencytosolvemoderate-sizelinearsystems,directmethodsarealsopopulartosolvelargesparse linearsystems,likepowerflow,SE,andotherpowersystemproblems.Sparsedirectmethodsarea tightlycoupledcombinationoftechniquesfromnumericallinearalgebra,graphtheory,graphalgorithms,permutations,andothertopicsindiscretemathematics[33].Andsuchproblemhasbeen extensivelystudied.

Thisbookfocusesondirectmethodsforsparselinearsystems,suchaslower–upper(LU),Cholesky,andotherfactorization,andtheimplementationbygraphparallelcomputing.Itfirst

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.