Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or...

35
a project by Manuel Wetzel, Frieder Borggrefe 21.09.2016 DLR Benefits of Decomposition Methods to Speed-up Energy System Modelling and Application to Stochastic Optimization

Transcript of Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or...

Page 1: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

a project by

ManuelWetzel,FriederBorggrefe21.09.2016

DLR

BenefitsofDecompositionMethodstoSpeed-upEnergySystemModellingandApplicationtoStochasticOptimization

Page 2: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

2

1. Introduction• TheproblemofuncertaintyinEnergySystemModelling

2. Currentimplementationofstochasticoptimization• Fromdeterministictostochasticmodelling• ImprovingconvergencebyEnhancedBendersapproaches

3. Challengesandpossibleimprovements• Currentchallenges• Application topathoptimization

Content

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 3: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

TheproblemofuncertaintyinEnergySystemModelling

Page 4: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

4

Theproblemofuncertaintyinenergysystemmodelling

• Longterm capacity expansion planning requires making decisions now,which have animpact onthe energy system for several decades.

• Context scenarios can give ageneral direction of development,however alargenumber of consistent sub-scenariosare possible.Thecurrentapproach usually considers only the main scenario and evaluatesrobustness by sensitivity analysis of sub scenarios.

Robust scenario High

Robust scenario Base

Robust scenario Low

Sensitivity analysis

Sensitivity analysis

Sensitivity analysis

Sensitivity analysis

Sensitivity analysis

Sensitivity analysis

0

1000

2000

3000

4000

5000

6000

1 2 3 4 5

HighUp

HighLo

MidUp

MidLo

LowUp

LowLo

2014 2020 2030 2040 2050

High

Base

Low

Model result Low

Model result Base

Model result High

Pathdevelopment of anuncertain parameter

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 5: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

5

Bendersdecompositionseparatingcapacityexpansionplanning andeconomicdispatch

Bendersdecompositionfortwo-stagestochasticoptimization(L-shapedMethod)[Benders1962andVanSlyke 1969]

Masterproblem:• Decidenowwhichcapacities

shouldbeexpanded

Subproblem:• Decidelateroneconomic

dispatchtosatisfyelectricaldemandwithcapacitiesgivenbymasterproblem

1stStage:Capacity expansion –Installed capacities

2ndStage:Dispatch –Generationvariables

Linkingvariables:Installed capacities,connecting the capacityexpansion problem and the dispatch problem

Mathematicalformulation inLPtable

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 6: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

6

Bendersdecompositionseparatingcapacityexpansionplanning andeconomicdispatch

1.Two-stagedecisiontree

Stage1(Masterproblem)Investmentdecision

Stage2(Subproblem)Stochasticeconomicdispatchscenarios

2.Multi-stagedecisiontree

Stage2Dispatch&Investmentdecision2020

2016 2050

20252020

Stage1Investmentdecision2016

Dispatch

Investment 2045

Stage3Dispatch&Investmentdecision2025

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 7: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

Current implementation of stochasticprogramming

Page 8: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

8

REMix SIMPLE

TheREMix Model

• Capacity expansion planning andeconomic dispatch

• High spatial,temporalandtechnological resolution

• Modularapproach to include heatsector,electromobility,DSMandothers

• Simplified version of REMix,usedas adevelopment platform

• Scalability of model dimensionsby generic data generation

Ø Modified for demonstratingimplementation approaches forstochastic optimization

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 9: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

9

Implementationof EnhancedBendersdecompositionapproaches

GAMSSimpleModel:• Capacity expansionplanning

and economic dispatch

1.Implementationof aBendersdecomposition approachSeperating capacity expansionand economic dispatch

Masterproblem:Investmentdecision

Subproblem:Economicdispatch

Suggested solutionCapacities for installed plantcapacities,electric storages and linksfor each region

Optimality CutInformation about the marginals of the firststage decision variables

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

DeterministicversionofTheGAMSSIMPLEmodel

Page 10: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

10

• Each subproblem is weighted by aprobability and represents aspecific setof assumptions for uncertain parameters

Implementationof EnhancedBendersdecompositionapproaches

1.Adaptationof the Regions:• Seperate regions and type1.Implementationof aBendersdecomposition approachSeperate regions and type

2.Stochasticmodel withscenarios• Each scenario has a

probability• Stochasticmodel is stillsingle

threat

Masterproblem:Investmentdecision

Subproblem:Economicdispatchscenarios

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Setof uncertain parameters• Electrical demand• Fuelprices• Wind timeseries• Solartimeseries• E-mobility usage

Page 11: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

11

Types of optimality cuts

Singlecut• One optimality cut periteration

is submitted to master problem

Multicut• Each subscenario submits an

optimality cut each iteration

Clustercut• Each cluster submits an

optimality cut each iteration

Subproblem:Dispatch

Master:Investment

Weightedbyscenario probability

Weightedbyscenario probability

• Multicutsgive detailed information butincrease complexity [Birge 1988]• Singlecutsaggregate information thereforemore iterations are necessary• Dynamicswitching between cut-types is possible [Skar 2014]

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 12: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

12

Implementationof EnhancedBendersdecompositionapproaches

• SmallESMwith 3powerplants,3storages,2node linksand 8760h• Formulation as Deterministic Equivalent for small problems faster than

Bendersdecomposition,butleads tomemory restrictions inlargemodels

1.Adaptationof the Regions:• Seperate regions and type

2.Stochasticmodel withscenarios

4.Grid computing• Scenariosand parts of the

modell can be placed inseveral threats

5.DeterministicModel• One modell formaster and

subproblem

3.SimpleComputing• Scenariosare processedone

afteranother

Testing withModelnodes:3Scenarios:100=40GBRAM

Testing withModelnodes:3Scenarios:100=3-4GBRAM

Testing withModelnodes:3Scenarios:100<<3-4GBRAM

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 13: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

13

BendersDecompositionMaster Subproblem

• Subproblems canbesolvedinparallel• Memorydemandscales withnumber

ofparallelsolveprocesses

Implementationof EnhancedBendersdecompositionapproaches

DeterministicEquivalent

• SizeofLPincreaseswithnumberofscenarios,solvedbySIMPLEX/Barrier

• OutofmemoryfortypicalREMixproblemswithscenariodimension

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 14: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

14

Improvement to achieve asychronity

t

Synchronous Model Asynchronous Model

tIteration1 Iteration2 Iteration3 Iteration4Iteration1 Iteration2

• NewMastercan be startedas soon as afraction of scenarios are solved,new iteration has only partial information therefore the number ofiterations increases [Linderoth2003]

• Trade-off betweenmore iterations and parallelisation

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 15: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

15

Algorithm parameters

Number ofclusters

Number ofscenarios

Number ofconcurrentmasters

Number ofnodes

Number oftimesteps

Deletingoptimalitycuts?

Clustersubmissionby GUSS

Multiplecuts perscenario?

Number oftechnologies

Typeofoptimality

cuts

Fractionofsolved

scenarios

• Largenumber of parameters to optimize performance of EnhancedBendersAlgorithm making systematic testingnecessary

SIMPLEModel

Optimality Cuts

GUSSusage

Asynchronity

Solvelinkand Solver

Solveroptions

Page 16: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

16

Method Clustercut Multicut Multicut +TR MC+ TR+GUSS

Iterations 138 77 54 61

Timeto solve 5:54h 4:42h 3:14h 2:51h

CPUload max 300% 449% 466% 134%

CPUload avg. 66.9% 70.8% 56.2% 18.6%

RAMusage max 0.58GB 0.58GB 0.48GB 0.67GB

RAMusage avg. 0.11GB 0.16 GB 0.11GB 0.26GB

Computational test results

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

• Modelbuilding timelimiting constraint,models are solved faster thangenerated therefore low average CPUload (100%equals 1core outof 32)

• Exchangeof information between GAMSand solver can be accelerated byusing shared memory and running GAMSand CPLEXindifferentthreads,this comes atthe cost of increased memory demand

Page 17: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

Challenges and possible improvements

Page 18: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

18

• Reducing computational overhead by using methods which simplify orreduce model building time(Usage of GUSS,externalizing model building)

• Balancing CPUload for migration to highperformance computing

• Generatingbetter cuts - insufficient electrical generation willcause allplanttypes inaregion to be built

• Deleting unused cuts inorder to keep the masterproblem as small aspossible while retaining important information

• Improving the starting point for the Trust-Regionapproach

• Combining stochastic optimization with decomposition onthe scenariolevel (decoposition inmodel nodes or timesteps)inaNested Bendersapproach

Challenges

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 19: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

19

Challenges

0

5

10

15

20

25

30

35

40

0

500

1000

1500

2000

2500

3000

3500

4000

4500

16:09:46

16:34:28

16:59:01

17:23:33

17:48:14

18:12:35

18:37:17

19:01:38

19:26:24

19:50:43

20:15:27

20:39:52

21:04:28

21:29:02

21:53:28

22:18:10

22:42:29

23:07:16

23:31:37

23:56:15

00:20:48

00:45:16

01:10:02

01:34:19

01:59:02

02:23:22

02:48:03

03:12:32

03:37:05

04:01:46

04:26:03

04:50:48

05:15:06

05:39:46

06:04:19

06:28:46

06:53:28

07:17:47

07:42:35

08:06:54

08:31:34

08:56:13

30.08.2016 31.08.2016

Summevon%CPU

Summevon%MEM

%CPU

%MEM

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

• Largescale scenario with typical synchronous behavior• Gapsindicate solving master while peaks represent parallelsubproblems• Highcorrelation between CPUand memory load indicating scalability with

number of parallelprocesses

Page 20: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

20

Application to path optimization

0

1000

2000

3000

4000

5000

6000

1 2 3 4 5

HighUp

HighLo

MidUp

MidLo

LowUp

LowLo

2014 2020 2030 2040 2050

High

Base

Low

Motivating examplefrom introduction

Result:3Context scenariopathways including3subscenarios each

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 21: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

21

Application to path optimization

2016 205020302020 2040 205020302020 2040

• Application of Bendersdecomposition incapacity expansion planning andeconomic dispatch leads to alargenumber of subproblems which can besolved inparallel

Masterproblem Subproblems

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 22: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

ProjectBEAM-ME

a project by

22

Page 23: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

23

• J.Benders,"Partitioningproceduresforsolvingmixedvariablesprogramming problems,"NumerischeMathematik,vol.4,pp.238–252,1962.

• R.M.VanSlyke andR.Wets,"L-shapedlinearprogramswithapplicationstooptimalcontrolandstochasticprogramming," SIAMJournalonAppliedMathematics,vol.17,no.4,pp.638–663,1969.

• J.R.Birge andF.V.Louveaux,"Amulticut algorithmfortwo-stagestochasticlinearprograms,"EuropeanJournalofOperationalResearch,vol.34,no.3,pp.384–392,1988.

• C.Skar,G.Doorman and A.Tomasgard,"Large-scale powersystem planning using enhancedBendersdecomposition,"PowerSystemsComputation Conference(PSCC),pp.1-7,2014,,doi:10.1109/PSCC.2014.7038297

• J.T.Linderoth andS.J.Wrigth,"Implementingadecompositionalgorithmforstochasticprogrammingonacomputationalgrid," ComputationalOptimizationandApplications,vol.24,pp.207–250,2003.

References

Page 24: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

Backup

Page 25: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

25

GAMSGUSS

7.GUSSimplementation

1.Adaptationof the Regions:• Seperate regions and type

2.Stochasticmodel withscenarios

4.Grid computing• Scenariosand parts of the

modell can be placed inseveral threats

GUSS

• Gathersdatafromdifferentsources/symbolsincollectionofmodels

• Updatesabasemodelinstancewiththisscenariodata

• Solvestheupdatedmodelinstanceand

• Scattersthescenario resultsto

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 26: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

26

GUSS

• Gathersdatafromdifferentsources/symbolsincollectionofmodels

• Updatesabasemodelinstancewiththisscenariodata

• Solvestheupdatedmodelinstanceand

• Scattersthescenario resultsto

UsingGUSS

• GUSSisaGAMSfacilitythatpermitssolutionofasetofscenariosforaGAMSmodelmodifyingdatatoruneachscenario.

• GUSSisnotreallyasolver• organizesandpassesdata to

theothergamssolvers• Collectionofmodelsaresolved

inasinglepasswithoutneedingLOOPovermultiplesolves.

• fasterthanloopbecauseonlyonemodelisbuild

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 27: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

27

Challenges

• Increasenumberofcutshanded fromonescenarioinoneiteration

• Furtherimplementationsnecessarytoevaluateexpectedperformanceofstochasticoptimization inREMix andaccelerateconvergenceo Deletionofunnecessaryoptimalitycuts,o Differentparametrisationsofalgorithm,o Dynamicadaptationoftrustregion,o Improvementofstartingpoint

• FurtherdevelopmenttowardsCPUloadbalancing necessaryformigration tohighperformancecomputing

• UsingREMix:ModularityofREMix environmentcomplicatestheimplementationofBendersdecompositionInformationaboutallactivevariables,equationsandmarginals necessary

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 28: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

Summary

• ModelledinGAMSandsolvedinCPLEXtypicallywithBarrierAlgorithm• Formulationasdeterministicequivalentimpossibleduetomemory

restrictions

• Staringpointbyregional presolve withmeanoverstochasticparameters• UtilisationoftheGAMSGridComputingenvironmentincombinationwith

GUSSinordertosubmitclustersofsubscenarios tobesolvedsimultanously• Optimalitycutsareaddedtothemasterproblemeitherassinglecuts,

clustercuts orfullmulticut• Currentlynocut-deletionprocedureimplemented

Page 29: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

29M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

• ModelledinGAMSandsolvedinCPLEXtypicallywithBarrierAlgorithm• Formulationasdeterministicequivalentimpossibleduetomemory

restrictions

• Staringpointbyregional presolve withmeanoverstochasticparameters• UtilisationoftheGAMSGridComputingenvironmentincombinationwith

GUSSinordertosubmitclustersofsubscenarios tobesolvedsimultanously• Optimalitycutsareaddedtothemasterproblemeitherassinglecuts,

clustercutsorfullmulticut• Currentlynocut-deletionprocedureimplemented

Page 30: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

30

• WelchenEinflusshatdieWahleinesDekomponierungsverfahrens aufdieWahldesSolvers?MachtesdabeieinenUnterschiedobnachz.B.Zeit,KnotenoderErzeugungundZubaudekomponiertwird?

• SindSlackvariablen wieBalanceEnsure undLoadEnsure fürDekomponierungsverfahren immernoch sinnvoll?OderkannalternativfrüherabgebrochenundeineneueIterationgestartetwerden(branch andcut,analog zuMIP)?

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 31: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

31

• BenötigteZeitfürdenBaudesGAMSModells– mussreduziertwerden,– bei100SzenarienundmehrerenSekundenModellbauzeitgehteinGroßteilderZeitin

denOverhead.– GUSSscheintinderHinsichtgutanwendbardaModellnureinmalgebautwerdenmuss.– FürIterativeVerfahrenwäreoptimalwennsicheinvorhergebautesGUSSModellmit

neuenParameternupdatenließe(spartModellbauzeitinspäterenIterationen

• CPUsmüssensich(nahezu)vollständigauslastenlassen,– d.h.wennTeiledesClustersfertigsinddürfendiesenichtaufneueAufgabenwarten

müssen.– ->AsynchronesVerfahrennötig!!!– Problem:EssolltentrotzdemkeineneuenModellegebautwerden.

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 32: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

32

• HöhereGranularitätmusserreichtwerden.Wenn80kCoresgleichzeitiganeinemProblemrechnensollen,dieAnzahlsinnvollerSzenarienjedocheherimBereichvon30-100liegt(ohneZufallsvariablen/vollständigeKombinatorik)müssendieSubproblemeweiterunterteiltwerden(z.B.nachKnoten/Zeitschritten).

⇒ ZusammenspielderKopplung von1.)Dekomponierung einesstochastischenProblems und2.)Dekomponierung innerhalb einesSubproblems

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 33: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

33

• HoherAnpassungsgradderstochastischenElemente:verschiedeneProjektewerdenunterschiedlichestochastischeParameterverwenden,wiekanndasGanzebenutzerfreundlichimplementiertwerden?

• HoheModularitätvonREMix erfordertautomatischesAnpassenderBendersDekomposition (WelcheVariablenwerdenwannentschieden,welcheGleichungensindaktiv,welcheMarginals müssenfürdieOptimalitätscutszurückgeführtwerden)

• Modellsollweiterhindeterminitisch verwendetwerdenkönnenohneinallenFälleneinestochastischeDimensionberücksichtigenzumüssen(Optionalität führtzuKomplexitätdesGAMSCodes)

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

Page 34: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

34

References

M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

0

1000000

2000000

3000000

4000000

5000000

6000000

7000000

8000000

1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101105109113117121125129133137

Series1

Series2

Series3

Series4

Series5

Series6

Series7

Series8

Page 35: Benefits of Decomposition Methods to Speed-up Energy ... · level (decoposition in modelnodes or timesteps) in a NestedBenders approach Challenges M. Wetzel andF. Borggrefe (DLR)

35M.Wetzeland F.Borggrefe(DLR) Application of Stochastic Optimization inESM 21.09.2016

2500000

2550000

2600000

2650000

2700000

2750000

2800000

2850000

2900000

2950000

3000000

1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101105109113117121125129133137

Series5

Series6

Series7

Series8