HarnessingArtificialIntelligencetoAcceleratetheEnergyTransitionWHITEPAPERSEPTEMBER2021IncollaborationwithBloombergNEFandDeutscheEnergie-Agentur(dena)ContentsPrefaceExecutivesummary1WhyAIisneededfortheenergytransition2ApplicationsofAIfortheenergytransition3“AIfortheenergytransition”principles4RecommendationsandoutlookContributorsAcknowledgementsEndnotes34591521222324Images:GettyImages,Unsplash©2021WorldEconomicForum.Allrightsreserved.Nopartofthispublicationmaybereproducedortransmittedinanyformorbyanymeans,includingphotocopyingandrecording,orbyanyinformationstorageandretrievalsystem.HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition2PrefaceItisincreasinglyclearthatspeedingupthetransitiontoalow-carboneconomyisnotonlyessentialbuturgent,andtheenergysectorisattheheartofthischallenge.Thiswhitepapercomesatacriticalmoment.ThefirstinstalmentoftheIPCC’ssixthAssessmentReportissuedinAugust2021andincreasinglyvisiblesignsofachangingclimateoverrecentyears,includingheatwaves,floodsandfires,havefocusedthemindsofpolicy-makers,corporatesandinvestorsalike.Newemissionreductiontargetsareemergingeveryweekandmanyorganizationsaremovingexistingtargetstonearertimeframes.Somecompaniesareevensettingcarbon-negativetargets.Withthe26thUNClimateChangeConferenceoftheParties(COP26)looming,weexpecttherateofclimatetargetannouncementstocontinueatpace.Whatislessclearishowdifferentstakeholdersandtheworldoverallcanactatthespeedandscalerequiredtomeetclimategoals.Theaimofthiswhitepaperistogatherconsensusandprovidepublicrecommendationsforaddinganotherpowerfultooltotackletheclimatechallenge–howtouseartificialintelligence(AI)tobestenactafast,equitableandlowest-costenergytransition.TheWorldEconomicForum’sGlobalFutureCouncil(GFC)onEnergyTransition,BloombergNEF(BNEF)andtheDeutscheEnergie-Agentur(dena),theGermanEnergyAgency,ranaseriesofroundtablesfromMarchtoMay2021forleadingexpertsfromtheenergyandAIsectorstoacceleratetheuptakeofAIforenergy.Thiswhitepapercontainsasynopsisofthediscussionsandrecommendationsfromthoseroundtables,namely,themostimportantapplicationsofAIforacceleratingtheenergytransition(Section2),asetofnine“AIfortheenergytransition”principlesthatwerecommendareadopted(Section3),andrecommendedactionsforkeystakeholdersinthepublicandprivatesectors(Section4).Thiswhitepaperaimstocontributetoenhancinganunderstandingof,aswellastrustin,AItechnologyfortheenergyindustry.Thenine“AIfortheenergytransition”principlesaimatcreatingacommonunderstandingofwhatisneededtounlockthepotentialofAIacrosstheenergysectorandhowtosafelyandresponsiblyadoptAItoacceleratetheenergytransition.Wehopetheseprinciplescaninspirethedevelopmentofacollaborativeindustryandpolicyenvironment.WewanttotakethisopportunitytothankalltheparticipantsfortheirtimeandcontributiontotheroundtablesandfortheirvaluableinputtothiswhitepaperandtheAIprinciples.AndreasKuhlmann,ChiefExecutiveOfficer,DeutscheEnergie-Agentur(dena)-GermanEnergyAgencyJonMoore,ChiefExecutiveOfficer,BloombergNEFEspenMehlum,HeadofEnergy,MaterialsandInfrastructureProgram-BenchmarkingandRegionalAction,WorldEconomicForumHarnessingArtificialIntelligencetoAcceleratetheEnergyTransition3ExecutivesummaryTheeffortstodecarbonizetheglobalenergysystemareleadingtoanincreasinglyintegratedandelectrifiedenergysystem,withmuchmoreinteractionbetweenthepower,transport,industryandbuildingsectors.Themovetodecarbonizetheenergysupplyisalsoleadingtohighlevelsofdecentralizationinthepowersector.Thiswillrequiremuchhigherlevelsofcoordinationandflexibilityfromallsectorplayers–includingconsumers–inordertomanagethisincreasinglycomplexsystemandoptimizeitforminimalgreenhousegasemissions.AIhastremendouspotentialtosupportandaccelerateareliableandlowest-costenergytransition,withpotentialapplicationsrangingfromoptimizingandefficientlyintegratingvariablerenewableenergyresourcesintothepowergrid,tosupportingaproactiveandautonomouselectricitydistributionsystem,toopeningupnewrevenuestreamsfordemand-sideflexibility.AIcouldalsobeacrucialacceleratorinthesearchforperformancematerialsthatsupportthenextgenerationofcleanenergyandstoragetechnologies.However,despiteitspromise,AI’suseintheenergysectorislimited,withitprimarilydeployedinpilotprojectsforpredictiveassetmaintenance.Whileitisusefulthere,amuchgreateropportunityexistsforAItohelpacceleratetheglobalenergytransitionthaniscurrentlyrealized.Thenine“AIfortheenergytransition”principles(seebelow)aimatcreatingacommonunderstandingofwhatisneededtounlockthepotentialofAIacrosstheenergysectorandhowtosafelyandresponsiblyadoptAItoacceleratetheenergytransition.Theprinciplesaresplitintothreeareas:thosethatgovernAIuse,thosethatwillhelpdesignAItobefitforpurpose,andthosethatenableAI’sdeploymentandareaimedathelpingtocreatecollaborativeindustryandpolicypractices.StandardsRiskmanagementResponsibility“AIforenergytransition”principlesGoverningDesigningEnablingSustainabilityAutomationDesignIncentivesDataEducationHarnessingArtificialIntelligencetoAcceleratetheEnergyTransition4WhyAIisneededfortheenergytransition1Globalenergysystemsareintransformation,andseveralkeytrendsaredrivingAI’spotentialtoaccelerateenergytransition.Theglobalenergysystemiscurrentlyundergoingamassivetransformation,andinthedecadesahead,itwillcontinuetobecomemoredecentralized,digitalizedanddecarbonized.Toreachthecommitmentsmadeunderthe2015ParisAgreement–limitingtheglobaltemperaturerisetowellbelow2°C–thistransitionmustaccelerate.Inrecentyears,theenergysectorhasbecomeincreasinglydigitalanditisclearthatfurtherdigitalizationwillbeakeyfeatureoftheenergytransitionandanessentialdriverofthesector’sprogresstowardsambitiousclimategoals.Theenergytransitionmustbeswiftandcoordinated;digitalizationisneededasanenablerDecarbonizingthepowersectoristhestartingpointforfull-systemdecarbonizationToachievedeepdecarbonization,itwillbenecessarytoshiftswiftlytoanenergysystemwithnoorverylittlecarbondioxideemissions.Theeffortstodecarbonizeourenergysystemareleadingtoanincreasinglyintegratedandelectrifiedenergysystemwithmuchmoreinteractionbetweenthepower,transport,industryandbuildingsectors,andasystemthatwillconsistofinterdependentenergyandtelecommunicationnetworks.Toacceleratetheshifttowardsawidespread,affordable,low-carbonenergysupply,thereisaneedforgreateroptimizationofeveryaspectofthisenergysystem,aswellasgreatercoordinationandcooperationbetweeneachcomponent.Thisrequiresabetterunderstandingof,andbettermechanismstomonitorandcontrol,thewaysinwhichpowergrids,buildings,industrialfacilities,transportnetworks,andotherenergy-intensivesectorsintegrateandinteractwithoneanother.Thisiswheredigitalizationcomesin:itisthekeytolinkingthedifferentsectorsintothemostreliable,affordableandcleanestsystempossible.Optimizingeachsectorseparatelywouldexcludeflexibility-generatingoptionsandreducethescopeforsystem-widetransformationprocessesthatwouldmaximizethebenefitsofdigitaltechnologyforthefullenergysystem,aswellasmorebroadlyfortheeconomy,theenvironmentandsociety.1Digitaltechnologiesalreadyautomatecomplexprocesses,orchestratedisparatesystems,andfacilitateinformationsharingintheenergysector,andsoftwarealreadyplaysasignificantroleinmanagingourenergysystems.Withtheexplosionintheavailabilityofdata,andasperformancecontinuestoimprove,digitaltechnologieswillplayanincreasinglycentralroleindrivingaswiftandcost-efficientenergytransition.Thesetechnologieswillfacilitateperformanceimprovementsandcostsavingsthroughacombinationofautomation,optimization,andtheenablingofnewbusinessandoperationalmodelsbothwithinandbeyondthetraditionalvaluechainofgeneration,transmission,distribution,tradeandconsumption.Thetransformationoftheenergysystemwillincludearapidexpansionoftherenewablepowersupplyandvastcleanelectrificationofheat,industryandtransport.Aselectricvehicle(EV)adoptiongrows,batterystoragecostsdecline,andbuildingsandheavyindustryturntonet-zeroelectricity,theshareofglobalenergydemandmetbyelectricityisprojectedtogrowby60%from2019to2050.Electricitywillincreasinglybeusedtoprovideheatingandcooling(e.g.heatpumps),transport(e.g.EVs)andevenrawmaterialssuchashydrogen(electrolyzers).Aselectricitysuppliesmoresectorsandapplications,itwillbecomethecentralpillaroftheglobalenergysupply.Thiswillcreatebothnewopportunitiesforvaluecreationandputnewpressuresonourcurrentsystemsofpowergeneration,transmission,tradeanddistribution.Toreachthecommitmentsmadeunderthe2015ParisAgreement–limitingtheglobaltemperaturerisetowellbelowtwodegreesCelsius–theenergytransitionmustaccelerate.HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition5ThistransitionrequiressignificantinvestmentThefuturepowersystemlookshighlydecentralizedThecomplexityofmanagingthepowersystemwillincreasesignificantlyInBNEF’sNewEnergyOutlook2020,along-termeconomictransitionscenarioonthefutureoftheenergyeconomy,56%ofpowergenerationcouldbeprovidedbysolarandwindin2050–amassive7.6TWofsolarand4.6TWofwind.2Thisscenarioassumesnofurtherpolicysupportfromtoday’slevelsandreflectsthefactthatfavourablesolar,windandstorageeconomicshavebecomeasignificantdriversintherapiddecarbonizationofthepowersector,evenwithoutstrongcarbonpricesornet-zerotargets.AccordingtoBNEFestimates,thisEconomicTransitionScenariowouldrequire$15.1trillioninvestmentinsolar,windandbatteries,and$14trillionpowergridinvestmentby2050.3Evenwiththesehistoricinvestments,thescenariooutlinedabovewouldlikelyleadtoanestimated2.2°Cglobalwarmingby2050.Toachieveglobalnet-zeroemissionsby2050,everysectoroftheenergyeconomyneedstoeliminateemissionscompletely.This,accordingtoBNEF’snet-zeroscenario,wouldrequireinvestmentsinenergyinfrastructuretototalbetween$92trillionand$173trillionbetween2020and2050.Themovetowardsgreaterproportionsofrenewableenergygenerationhastwomainpracticalconsequences:thefuturepowersystemwillhostmorepowerfromintermittentpowergenerators(sincesolarpanelsonlyproducewhenthesunisshining,andwindturbineswhenthewindisblowing),anditwillbemoredecentralized.InBNEF’sEnergyTransitionScenario,13%ofallglobalpowercapacityin2050willcomprisedistributedsmall-scalephotovoltaic(PV)energyandbatteries,upfrom4%today.4Thiswillaccelerateanongoingtrendofshrinkingmedianpowerplantsize,withBNEFexpectingthemedianpowerplantsizetoshrinkover80%,from944MWtoday(whichcorrespondstoalarge,naturalgas-firedpowerplant)to158MWin2050.Extrapolatingfromcurrentdeploymenttrendsandtakingintoaccountdecarbonizationtargets,itisclearthatinthefuturetherewillbevastlymorephysicalassetsconnectedtothepowergridand,inparticular,thedistributiongrid,wherepowerflowsarebecomingincreasinglydynamicandmultidirectional(seeFigure1).Theseassetsmightgeneratepowertosellbacktothegrid(e.g.solarhomes),uselargeamountsofpoweratonce(e.g.EVfastcharging)causingdemandtospike,ortheassetsmightbeconnectedtothegridwithoutgridoperatorsbeingaware(e.g.smarthomedevices).Thesedynamicscouldchallengethegrid’sstabilityandperformance,leadingtoissuessuchaspowerfrequencyimbalances,blackoutsandbrownouts,andsignificantcapacityoverbuild.Withoutreal-timedata,advancedanalyticsandautomation,theincreasinglycomplexpowerandenergysystemsofthefuturewillbecomeimpossibletomanage.Theproportionof2050’spowergenerationthatcouldbesmall-scaledistributedrooftopsolarTheshrinkageinmedianpowerplantsizeby205013%-83%…favourablesolar,windandstorageeconomicshavebecomeasignificantdriveroftherapiddecarbonizationofthepowersector.HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition6TransmissiongridPowerplantsLarge-scaleindustryRenewablesCommercialandindustryPublictransportSMEsBuildingsInterconnectionsDistributiongridTransmissiongridDistributiongridPowerplantsLarge-scaleindustryRenewablesPower-to-gasCommercialandindustryPublictransportElectricvehiclesBatteriesSMEsBuildingsInterconnectionsiIncludesbattery,electricandplug-inhybridpassengercarsonly(excludescommercialcarsandtwo-orthree-wheelers).iiSolarandwindonly(excludesotherrenewableenergysources).iiiIncludesutility-scaleandbehind-the-meterlithiumionbatterystorage.Source:Adaptedfromdena(2020).FiguresBNEF(2020).2020LowCouplingofsectors12millionElectricvehiclesi1.5TWRenewablesiiLow11GWHeat-pumpdemand<1%GlobalshareofflexibleloadBatterystorageiiiHighCouplingofsectors836millionElectricvehiclesi12TW1.3TWRenewablesiiHighHeat-pumpdemand8%GlobalshareofflexibleloadBatterystorageiii2050TransformationofglobalpowersystemsFIGURE1HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition7Artificialintelligencecansubstantiallycontributetoacceleratingtheenergytransition5AIreferstothebroaderconceptofmachinesbeingabletocarryouttasksnormallyrequiringhumanintelligence(suchasimageandspeechrecognition,decision-makingetc.).AIisnotasingletechnologyorproduct,butratherasetoftechniques,mathematicalmodelsandalgorithmswiththeabilitytoextractinsightsfromlargedatasets,identifypatternsandpredicttheprobabilitiesofpotentialoutcomesofcomplex,multivariatesituations.6AIisoftenconfusedwithautomation,butthetwoaredistinct(albeitrelated):automatedsystemsperformrepetitivetasksfollowingaprogrammedsetofrules,whileAIidentifiespatternsandinsightsindataand“learns”todothismoreaccuratelyandeffectivelyovertime.Data,softwareandautomationalreadyplayasignificantroleintheenergysector;however,AIexceedsthecapabilitiesoftraditionalsoftware.SomeusecasesofAIalreadyexistwithintheindustry,butinordertoaddressthechallengesoutlinedabove,webelievethatAItechnologieswillneedtobedeployedatamuchlargerscaleandatamuchfasterpacetospeeduptheenergytransitionandlowertheassociatedcostsifwearetorapidly,safelyandeconomicallytransitionawayfromfossilfuels.TheeconomicvalueofAIfortheenergytransitionisdifficulttoestimate,giventhatithasthepotentialtobewidelyadoptedacrosstheenergyvaluechaintoenableentirelynewrevenuestreamsthroughnewbusinessmodels,andgiventhatsomeofitsbenefitswillcomeintheformofavoidedcosts(e.g.loweringequipmentreplacementcoststhroughthepredictivemaintenanceofexistingassets).Consideringthelevelsofinvestmentrequiredtodelivertheenergytransition,evenifAIweretoreducetherequiredinvestmentorshavepeakenergydemandbyasmallpercentage,thiswoulddrivebillionsofdollarsinsavingsfortheindustryandconsumersalike.SelectedexamplesofthevalueofAIfortheenergytransitionBOX1trillion$1.3Thereducedcleanenergypowergenerationinvestment,2020-2050,resultingfromevery1%ofdemand-sideefficiency,accordingtoBNEF’snet-zeroscenario.7AIcouldachievethisbyenablinggreaterenergyefficiencyandhelpingtoflexdemand.billion$188Withoutintervention,increasedairtemperaturesduetoclimatechangecouldreducethelifetimeofpowergridequipmentandcouldcutthelifetimeoftransformersby10years,accordingtoscientificstudies.8BNEFanalysisshowsthatthiscouldcauseanadditional$188billioninreplacementcostsbetween2020and2050globally.9AIcanhelpoperatorsavoidthisadditionalcostbykeepingtransformerswithinoptimaloperatingranges.6-13%Acrossarangeofdevelopedmarkets,BNEFfoundthatthelackofintelligentflexibilitywouldincreasepowersystemcostsby6-13%in204010(studybasedonGermany,SpainandtheUK).AIcanhelpcontrolandbalancethisintelligentflexibility,forexample,withAI-enabledelectricbatterychargingandbatteries,therebyreducingsystemcosts,optimizingsystembuild-outwhileminimizingcurtailment,andreducingrelianceonfossilfuelplantsforbackup.EvenifAIweretoreducerequiredenergytransitioninvestmentbyafewpercent,thiswoulddrivebillionsofdollarsinsavings.HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition8ApplicationsofAIfortheenergytransition2AIisapowerfultoolwhichcanmanagethecomplexityoftheglobalenergytransitionandachievegreatersystemefficiency,thereforeloweringcostsandincreasingthespeedofthetransition.AItechnologyhasthepotentialtorapidlyacceleratetheenergytransition,particularlyinthepowersector.Inthissection,weidentifysomeofthemostpromisingAIapplicationsfortheenergytransitionacrossfourfocusareas:renewablepowergenerationanddemandforecasting,gridoperationandoptimization,energydemandmanagement,andmaterialsdiscoveryandinnovation.AIapplicationscanbefurtherclassifiedbasedonthedatainputstheyuse.AIcanusemanyformsofinputdata:audio,speech,images,videos,datagainedfromsensors,datacollectedmanuallyorrobotically,etc.Accordingtodena’s2020analysis11,12offieldsofapplicationsforAIintheenergyindustry(seeFigure2),themajorityofAIapplicationsfallinthefollowingcategoriesofdata:Market,commodityandweatherdataUsingdatacollectedfromvarioussources,e.g.electricityconsumptiondata,electricitypricedata,andweatherdata,AIisemployedtorecognizepatternsand/orprovideprobabilisticpredictionsoffutureoutcomesbasedonpatternsidentifiedwithinthedata.Thedatausedisoftentimeseriesdata,whichisaseriesofdatapointsthathavebeencollectedoverregulartimeintervalsandorderedchronologically.ImagesandvideosUsingimageandvideodataforAItorecognizeobjectsorconditionsinimages(e.g.usingsatelliteimagestodeterminecloudcoverpatternsand,inturn,predicttheoutputofasolarplant).EquipmentandsensordataUsinginputdatafromequipmentand“smart”devicesthatcombinesensorswithcommunicationsandnetworkingcapabilitiestoenablereal-timedigitalconnectivityandcoordinationofphysicalassets.Thesesystemsofsensinganddevice-levelcontrolareaprerequisitefortheintelligentcoordinationandautomationoftheenergysystemusingAI.Figure2showsthemostpromisingapplicationsofAIfortheenergytransitionandclassifiesthemaccordingtothetypeofinputthatisused.Aspecificapplicationcanuseseveralinputtypes.Theapplicationsarethenexplainedinmoredetailinthefollowingsections.HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition9ApplicationsofAIfortheenergytransition,classifiedbytypeofdataFIGURE2RenewablepowergenerationanddemandforecastingSitingofsolarandwindfarmsConstructionofpowerplantsImprovementofproductdesignPredictionofassetfailuresandoutagesOptimizationofmaintenanceschedulesPowerproductionforecastsPowerdemandforecastsGriddesignandplanningEquipmentoperationandmaintenanceMonitoringofgridperformanceIntelligentmanagementofdistributedrenewablesanddevicesOptimizationofelectricityconsumptionofequipment/buildingsOperationofvirtualpowerplants(VPP)AutonomousmaterialsdiscoveryAutomatedmaterialsynthesisandexperimentationRenewablepowergenerationanddemandforecastingManagementofenergydemandanddistributedresourcesMaterialsdiscoveryandinnovationGridoperationandoptimizationImagesandvideosMarket,commodityandweatherdataEquipment/sensordata36711123456789101112141513881010991111121315121324455Source:denaanalysis(2020)1Asrenewablepowergenerationgrows,bothinabsolutetermsandasashareofthepowersupply,itwillbecomeessentialtobetterpredictsolarandwindpowergeneration,toimprovecapacityfactorsandproductionuptimeatsolarandwindplants,andtoaccuratelyforecastpowerdemand.Frompowerplantsitinganddesignthroughtopowerschedulinganddispatch,AIhasaroletoplay.HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition10The1sitingofsolarandwindfarmshasamajorinfluenceonthecapacityfactorofthesepowerplants.CompaniesarealreadyusingAItoidentifysiteswiththemostfavourablesunandwindresourcesandthebestaccesstoexistinggridinfrastructure.When2powerplantconstructionbegins,AIcanalsobeusefulinmanagingandacceleratingcostlybuildschedulesby,forexample,optimizingthesequencingofequipmentdeliverytositesorusingcomputervisiontoidentifyinefficientordangerousworksiteprocesses.AIcanevenhelp3improveproductdesignfornewergonomicwindturbineblades,PVpanels,orpowerelectronicsandcontrolsystems.Whenpowerplantsstartgeneratingelectricity,operatorsarerequiredtocarryoutregularmaintenance.Thiscancostanythingfrom1%ofpowergenerationcosts(e.g.forutility-scalesolar)to20%(foroffshorewindfarms),accordingtoananalysisfromBNEF.13Failingtocarryoutmaintenancecanleadtosystemmalfunctionsandfailures,resultingindowntimeandadditionalrepaircosts.AIisincreasinglybeingintegratedintooperationsandmaintenanceprocessestoimprovetheirefficacyby4predictingfailuresandoutages,helpingtoreduceunnecessarymaintenanceandoptimizelifetimemaintenance,andavoidingordelayingcostlyequipmentreplacement.AIcanalso5optimizemaintenanceschedules,whichcansavesignificantcostsforremotesitessuchasoffshorewindfarms.Usingsensorswhichmonitortheassets’healthinrealtime,AItriggersanalarmifanomaliesaredetected.Itisrelativelydifficulttopredictwhenandhowmuchpowerwillbegeneratedfromsolarandwindfarms.Toeffectivelyintegratethemostrenewableelectricityintothegridandtocapturelucrativepowerpurchaseagreements,operatorsareusingAItobetter6forecastpowerproductionfromsolarandwindfarms.AIcanpredictthepoweroutputofsolarandwindassetsbylearningfromhistoricalweatherdata,real-timemeasurementsofwindspeedandglobalirradiancefromlocalweatherstations,sensordata,andimagesandvideodata(forexample,satelliteimagesofcloudcover).Theseshort-termpowergenerationforecastscanthenbefedintooperatingsystemsthatschedulelocalbatterystorageplantcharginganddischargetohelpreducecurtailmentofsolarandwindfarms.7Forecastingpowerdemandiscomplexandwhenbadlydoneitleadstoblackoutsorrenewablecurtailment.AIisgoodatspottingcomplexpatternsand,usinghistoricconsumptiondata,itcanbehelpfulinpredictingconsumerdemand,bothattheindividualandaggregatelevel.Thereisgreatpotentialincombiningbettergenerationforecastswithpowerdemandforecasts,tooptimizebothshort-termandlonger-termsystemplanningandoperation–whetherforshort-termpowergenerationschedulingorlong-termgridinvestments.Robustdemandandgenerationforecastsonalltimescales(weekly,day-ahead,hourly,andintra-hour)arecriticaltoreducingthedependencyonfossil-fuelledstandby“buffer”plantsandproactivelymanagingthegrowthofvariableenergysourcesandincreasingthevariablerenewablecapacitythatcanbeaccommodatedinthegrid.Gridoperationandoptimization2Currentplanstoreachnetzerobymid-centuryimplyamassiveincreaseinrenewablegenerationcapacityandexpansionoftransmissioninfrastructurewithinarelativelyshortperiodoftime.Duetothelongleadtimestoplanandcommissionnewtransmissioninfrastructure(leadtimesofasmuchastenyearsforanewtransmissionlinehavebeenreportedfortheUS),thedeploymentofnewtransmissioncapacitymightbecomeabottleneck.UsingAItooptimizegridoperationandenhancingthecapacityofexistingtransmissionanddistributionlines,aswellasextendingthelifetimeofexistingequipment,willbekeytosupportingtheenergytransition.Inaddition,inanintegratedanddecentralizedenergysystem,responsibilityforsystemoptimizationhappensatboththehigherandlowervoltagelevels,distributiongridsbecomemoreimportant,andmaintaininggridstabilityandensuringthesecurityofsupplybecomemorecomplex.AIcanbehelpfulingridplanningtooptimizeinfrastructurebuildbyextendingthelifetimeofexpensivegridequipmentandkeepingthewholegridsystemstable,evenasmorerenewablesareintegrated.Thereisgreatpotentialincombiningbettergenerationforecastswithpowerdemandforecasts,tooptimizebothshort-termandlonger-termsystemplanningandoperation.HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition11BNEFfoundthatatleast$14trillionneedstobeinvestedinnewgridinfrastructure/gridreplacementsby2050tosupporttheelectrificationofbuildings,industryandtransportandtostrengthenthedistributiongridsothatitcanintegrateandcommunicatewithmanymorerenewableenergysources.AIhasanimportantroletoplayinstrategicdecision-makingon8griddesignandplanning.Itcanusehistoricgriddataandelectricitygenerationanddemandforecaststobestdecidewhatgridequipmenttobuildwhere,andhowtosizethetransformersandwiresmostefficiently.AIcanalsouseclimatechangedataandforecaststoadviseonwhichpartsofthegridshouldbereinforcedormovedtobestavoidorminimizedisruptionsandblackouts,includingthosecausedbyclimatechangeintheformofextremeweatherevents(heat,cold,storms,floods,etc.).Whenthegridisinoperation,AIcanbeusedforarangeofimportant9equipmentoperationsandmaintenanceactivities.Ifthegridistogrowbymillionsofkilometres,asisexpectedinafuturelow-carbonintegratedenergysystem,thecostofregularmanualinspectionswillgrowcommensurately.Computervisionandroboticscanenableremoteinspectionofthegridbyanalysingvideofootagetakenbyhelicoptersorunmanneddrones.Thesesystemscanbetrainedtospotrottingpoles,birdnestsonwires,andovergrownvegetation,directingmaintenancecrewstothespotsthatrequirecondition-basedinterventions.Machinelearningcanalsohelpoperatorsunderstandtheperformanceoftransformersandpredictanomaliesandfailures,savingtimeandmoney.Risingglobaltemperature,extremeweathereventsandincreasingforestfirerisksareincreasingthecostofgridmaintenanceandthefrequencyofblackouts.Withincreasingoccurrenceofextremeweatherphenomena,AImaysupportthemitigationoftheseeventsbyidentifyingcriticalconditionsforequipmentearly,basedonweatherforecastandhistoricalgridperformances.Risingtemperatureswillreducethelifetimeofgridequipment,cuttingtransformerlifetimesbyuptotenyears.AIcanuseclimatedataandtransformeroperationsdatatodesignoptimaloperationrangesfortransformers,keepingthemwithinsaferparametersandavoidingover-utilization,thereforeextendingthelifetimeoftransformersandotherequipment.Beyondequipmentmaintenance,AIwillplayanimportantrolein10monitoringgridperformanceandhelpingtooperatethegridmoreefficiently.Monitoringthegridinrealtimeusinga“digitalgridtwin”,withAIhelpingtoidentifypatternsandmodellingthebehaviourofpowerlines,couldsignificantlyfacilitatethepenetrationandintegrationofrenewableelectricity.Thousandsofsensorsarealreadyinstalledontransmissiongridstounderstandhowtheyareperforming.Doingthesamefordistributiongridswouldbeveryexpensive,butmuchmoredistributiongriddataisnecessaryifitistobecomeamajorpartofanintelligentsystem.AIoffersanalternativewayofmonitoringsystemstabilitybymodellingelectricitycharacteristicsthroughprovidingmissinginformationbasedonwhatsensordataisavailable.AIcanalsohelpenhancetheutilizationofthetransmissioncapacityofapowerlinebyrespondingtoreal-timetemperaturemeasurementstodeterminetheupperlimitthatthelinecansafelycarry(insteadofusingstaticlimitsbasedontheoreticalandconservativetemperatureassumptions).Thiscouldincreasethetransmissioncapacityofexistinginfrastructuresignificantly.TomaximizethebenefitsofAIfordistributiongridsystemcontrol,severalobstacleswillhavetobeovercome,includingalackofreadilyavailabledatasetswiththequalityandquantityofdatarequiredfortrainingAImodels.Thisisinpartduetowidespreadsensordeploymentbeingarelativelyrecentphenomenon,butitisalsotheresultofalackofdataaccessduetocompetitiveorprivacyreasons,aswellasalackofsufficientlyaccuratedatalabelling.Thestandardizationandpoolingofhigh-qualitytrainingdata(e.g.imagesofdefects)canhelpremedythissituation.Anotherkeychallenge,aswithanyotherdata-drivenapplication,isbalancingdataprivacyversusdatausage(e.g.smartmeterdata)andensuringcompliancewithapplicabledataprotectionregulations(e.g.GeneralDataProtectionRegulation).TheproposaloftheEuropeanCommissionfortheArtificialIntelligenceAct,forexample,classifiesAIsystemsintendedtobeusedassafetycomponentsinthemanagementandoperationofcriticalinfrastructureashigh-risk.14Sincetheirfailuresmayaffecthumanlifeandhealthonalargescaleandconsiderablydisruptpubliclifeandeconomicactivities,thesupplyofwater,gas,heatingandelectricityareattributedtothisgroup,andAIapplicationsmustthereforemeethighstandardsbeforetheyarepermittedtobeused.WhilethisproposalishelpfulforclassifyingrisksandharmsfromAIimplementation,itcouldraisethecomplexityandcostofregulatorycompliancefortheenergyindustry.HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition12AIiseffectiveatspottingpatternsinlargedatasetsandoptimizingprocesses.Thisiswhatmakesitidealfor11intelligentlyintegratingdistributedrenewablesandbatteries.AIcouldplayasignificantroleinorchestratingtheinterplaybetweendistributedrenewablesandbatteries,andotherstoragedevices.Forexample,AIcanhelpahouseholdtooptimallyswitchbetweenbatterypower,on-sitesolargenerationandgridpower,therebysolvingthecustomer’sneeds,whetherthatbeminimizingcostsormaximizingself-consumption.AIisbeingusedbyfactoriesanddatacentrestohelp12optimizeelectricityconsumptionthroughlearninghowequipmentbehavesandidentifyingwaystoreduceelectricity.Inbuildings,AIcanoptimizetheelectricityusageofheatingandairconditioningunits,forexample,byusingsensordataandcomputervisiontodetermineoccupancylevelsandbetterunderstandabuilding’sthermalbehaviours.AIisusefulnotonlyinreducingpowerdemandbutalsoinshiftingittomatchtimesofhighrenewablesgeneration,allowingdemandtofollowsupply.Thiscouldreducethecarbonfootprintoftheconsumerandcouldbeanimportantenablerforconsumerstoswitchto24/7carbon-freeelectricity.AIcouldalsohelpabsorbsolarandwindpowerthatmightotherwisebecurtailed.Hyperscaledatacentresareparticularlyactiveinwhattheyrefertoas“renewablesmatching”.AIcanplayanimportantrolenotonlyinreducingorshiftingenergydemandbutalsoinopeninguptheenergyservicesmarkettoarangeofconsumerandindustrialdevices.Thefutureenergysystemwillworkbest,andwillhavethelowestcost,ifdistributedenergyconsumingandgeneratingdevicesgettoplayapartingridbalancingandpowerqualityoptimization.Today,largeindustrialequipmentinfactoriesisalreadyparticipatingindemandresponsemarkets,butoftenthisisarrangedmanuallybetweengridoperatorsandequipmentowners.Grid-scalebatteriescan,dependingonmarketdesign,contributetoancillaryservicessuchasfrequencycontrol,butsmallerbehind-the-meterassetsarerarelyabletoparticipate.DigitalizationandAIprovidetwonewopportunitiesfor13operatingdistributedenergyresourcesanddevicesas“virtualpowerplants”(VPP).First,theyoffertheabilitytoaggregateandorchestratesmallpowerplantsanddistributedenergyresourcestoprovidegridservicesotherwiseinaccessibletothem.Second,throughautomationandtheautonomyofsmalldistributeddevices(suchasfridgesorEVs),theyenableconsumerstohelpsupportthegridwhilemaintainingtheutilityoftheirdevicesand,insomecases,gaincompensationforthegridservicesprovidedbythem.VPPscanhelpsmall-scaleassetsaccessmarketsandservicesthattheyotherwisewouldnothaveaccesstoandcanincludeanycombinationofsmall-scalebatteries,windturbines,solarPVpanels,EVs,biogasgeneratorsandmore.Theircorecomponentissoftwareformonitoring,forecastinganddispatchingenergy,andAIisusefulherebecauseofitsabilitytoforecastassetperformance,energydemandandpowerprices.Itisalsohelpfulincreatingautonomoussystemscapableofmakingrapid,data-drivendecisionsaboutwhetheraphysicalassetshouldbidforagridservice.Otherdigitaltechnologies,includingblockchain,canalsohelpscaleVPPsbymakingiteasierfornewassetstoconnectautomaticallytoalocalVPPnetworkandbesecurelyidentifiedandpaidforgridservicesperformed.AIcouldalsoenablesmallerelectricity-consumingandgeneratingassetsinthehometobetterservegridoperators.OwnersofEVs,batteries,smartthermostatsorfridgesmaynotwanttocontributemicroservicestooperatorsby,forexample,helpingtoreducepowerdemandduringaheatwave,duetoconcernsitwouldimpacttheirdeviceavailability.AIcouldallowconsumerstoenabletheirequipmenttorampupordownwhilestillbeingfullyavailablewhentheirownerneedsthem.AndAIsystemsarenotonlyusefulintheplanningandoptimizationprocess;theycanalsomakemanydecisionsautonomously,fasterandmoreaccuratelythanhumans.RatherthanacustomermanuallyoperatingasystemtochargetheirEVwhenaskedtobythepoweroperator,anAI-enabledcontrolsystemmanagesthechargingrateandtimetobenefitboththeEVownerandthegridoperator.Forexample,AIrespondstotime-variedand/orlocationalmarginalpricesinthepowermarkettominimizecosts.Managementofenergydemandanddistributedresources3AIcanhelpincreasethepenetrationanduseofdistributedrenewablesandhasthepotentialtosignificantlyacceleratetheirdeployment.Itisalsobeingusedeffectivelyinimprovingenergyefficiencyinbuildings,factoriesanddatacentres.Beingabletoreduce,manage,aggregateandmanipulateenergydemandwillbeanimportantfactorinhoweffectivelyandcheaplytheenergysectorcandecarbonize.AIcanhelpahouseholdtooptimallyswitchbetweenbatterypower,on-sitesolargenerationandgridpower,therebysolvingthecustomer’sneeds,whetherthatbeminimizingcostsormaximizingself-consumption.HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition13AIcouldbecomeapowerfultooltogeneratenovelmolecularstructureswhichsatisfyspecificrequirementsforcertainapplications.Inaprocesscalled14autonomousmaterialsdiscovery,AIcanbeusedtoscreenpotentialmaterialsatthemolecularlevel,identifyinghigh-potentialcandidatesforagivenproblembypredictingthepropertiesofthesematerials.AIisalsobeingcombinedwithroboticsandusedfor15automatedsynthesisandexperimentationtotestthepropertiesofthesemoleculesandtheirperformanceunderarangeofconditions.Thefeedbackfromtheseexperimentsisthenusedtoinformthediscoverycycle.Acceleratingmoleculardiscovery,characterizationandutilizationprocessescouldsignificantlyreducethetimeandlowerthecostrequiredtodeploynewmaterials.Moreadvancedmaterialsinnovationisrequiredtodevelopcatalystsusedincarboncapture,utilizationandstorage(CCUS)technology.ThecostandenergyrequirementstoconvertCO2intoproductsarehighlydependentoncatalysts,oftenusingexpensiveorscarcemetals,whichprohibitscale-up.Otherpotentialinnovationareasincludeenergy-efficientmaterials(e.g.phase-changematerialsthatcanstoreandreleaseheat),thermoelectricmaterialsthatcanconvertheatintoelectricity,newsolarpanelmaterialscapableofimprovedsunlightenergy,conversionandnewbatterymaterialsandchemistriesthatimproveperformanceand/ordurability.FormaterialsusedinthemanufacturingofsolarPVpanels,ittypicallytakes25-35yearsfromthefirstreportofanovelmaterialuntilitismanufacturedforcommercialapplicationonalargescale.15AIcouldcutthattimeandgrowthepipelineofnewmaterialsmovingfromthelabtothemarket.Materialsdiscoveryandinnovation4Thedevelopmentofhigh-performance,low-costmaterialsforcleanenergygenerationandstoragehasbeenrecognizedasapriorityfortheenergytransition.However,theprocessofdiscovering,developinganddeployingadvancedmaterials,whichneedtosatisfycomplexperformancespecifications,ishighlycapitalintensiveandoftentakesyearstocomplete.HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition14“AIfortheenergytransition”principles3CommonguidingprinciplesareneededtounlockthefullpotentialofAIfortheenergytransition.Intheprevioussections,wesummarizedthesignificantpotentialthatAIofferstoacceleratetheenergytransition.Butthispotentialwon’tberealizedwithoutconcertedmultistakeholderaction.DuringaroundtableseriesconductedbetweenMarchandMay2021withleadingAIandenergyindustryexperts,participantshighlightedseveralcross-cuttingissuesthatarepreventingAIfrombeingadoptedrapidlyatscaleintheindustry.Basedonthesediscussions,wehaveestablishedthefollowingnine“AIfortheenergytransition”principles,keyconsensusprinciplesthat,ifadoptedbytheenergyindustry,technologydevelopersandpolicy-makers,wouldacceleratetheuptakeofAIsolutionsthatservetheenergytransition.ThefollowingprinciplesaimatcreatingacommonunderstandingofwhatisneededtounlockthepotentialofAIacrosstheenergysector,andhowtosafelyandresponsiblyadoptAItoacceleratetheenergytransition.WehopetheseprinciplescaninspirethedevelopmentofacollaborativeindustryandpolicyenvironmentaroundAIfortheenergytransition.HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition15“AIfortheenergytransition”principlesFIGURE3DataEstablishdata-sharingmechanismsandplatformstoincreasetheavailabilityandqualityofdataIncentivesCreatemarketdesignsandregulatoryframeworksthatallowAIusecasestocapturevaluebroughtaboutbythemRiskmanagementAgreeuponacommontechnologyandeducationapproachtomanagingriskspresentedbyAIStandardsImplementcompatiblesoftwarestandardsandinteroperableinterfacesAutomationDesigngenerationequipmentandgridoperationsforautomationandincreasedautonomyofAISustainabilityAdoptthemostenergyefficientinfrastructureaswellasbestpracticesaroundsustainablecomputingtolimitthecarbonfootprintofAIEducationEmpowertheenergyworkforcewithahuman-centredAIapproachandinvestineducationtomatchtechnologyandskilldevelopmentDesignFocusAIdevelopmentonusabilityandinterpretabilityResponsibilityEnsurethatAIethicsandresponsibleuseareatthecoreofAIdevelopmentanddeployment“AIfortheenergytransition”principlesGoverningEnablingDesigningSource:TheWorldEconomicForumHarnessingArtificialIntelligencetoAcceleratetheEnergyTransition16Principle1:Automation–designgenerationequipmentandgridoperationsforautomationandincreasedautonomyofAIThecomplexityofmanagingfutureenergysystemswillnotalwaysallowformanualhumancontrol.AIwillbeneededinincreasinglydecentralizedfutureenergysystemsthatintegratenotonlyelectricitybutalsoothersectors(mobility,heat,industry),andwhereagrowingnumberofcomplex,closetoreal-timedecisionshavetobemade.TheautonomyspectrumrangesfromaugmentedautomationwithAIsolelyassistinginhumandecision-making,tofullautonomywithAImakingdecisionsautonomouslywithouthumansupervision.ToenableAI’sfullbenefit,gridoperationsmustmovetowardsautomationandincreasedautonomyasstandard,andnewpowersystemequipmentmustbedesignedandsetupreadyforautomation.Thiswillrequireminimumtechnologystandardsfornewgrid-connectedinstallationsandupdatestogridoperationprocedures.Principle2:Sustainability–adoptthemostenergy-efficientinfrastructureaswellasbestpracticesaroundsustainablecomputingtolimitthecarbonfootprintofAIInthiswhitepaper,wehighlightthesubstantialpotentialforAItocontributetosustainabilitygoalsandacceleratetheenergytransition.WebelievethatthesustainabilityadvantagesofAIoutweighitsowncarbonfootprintbyfar.Nevertheless,trainingandrunningsomemachinelearningmodels(e.g.deeplearningmodels)canbecomputationallyintensive,requiringelectricitybothforcomputationandforcooling.AstheenergyindustrystartstoadoptAI,energyefficiencyandsustainabilitycriteriashouldbetakenintoconsideration,andtheenergyoremissioncostsofdeveloping,trainingandrunningmodelsshouldbereportedinastandardizedway.AsAIfindsitswayintotheenergysector,wemustinsistthatthesectoradoptsthemostenergyefficientinfrastructure,aswellasbestpracticeapproaches,toolsandmethodologiesforbuildingenergy-efficientmodelsandsustainablecomputing.Thisincludesrunningalgorithmsonhardwarepoweredbygreenelectricity,recyclingwasteheat,managingcomputingresourcesinanenergy-optimizedway,andusingbestpracticedevelopmenttechniques(e.g.forthetuningofhyperparametersinmodels)orrecyclingmodelsacrossdifferentapplicationsordomains.Thiswillleadtocontinuousimprovementsinenergy-efficientAI.Principle3:Design–focusAIdevelopmentonusabilityandinterpretabilityWhiletheindustrywillhiremoredatascientistsandcarryoutinternalworkforcetraining,thiswillnotbesufficientintheneartermifAIremainsaccessibletospecialistsonly.AImustbeeasytointerpretanduseforeveryonesoitcanbecomeanintegratedbaselayerforavarietyofoperationaltasks.ThiswillinvolvedevelopingAIalgorithmsthatexplainhowtheyweretrainedanddeveloped,designingAIalgorithmsthatexplaintheiractions,andbuildinglow-codetoolsthatareeasyandquicktouseandamendbynon-experts.DesigningHarnessingArtificialIntelligencetoAcceleratetheEnergyTransition17Principle4:Data–establishdatastandards,data-sharingmechanismsandplatformstoincreasetheavailabilityandqualityofdataToday,thepowersystemisnotsuitablymonitoredtoprovidereal-time,granularoperationsdata,particularlyatthedistributionlevel.SupervisoryControlandDataAcquisition(SCADA)data(thebasisofmostpowermonitoring)isnotsufficientlyfrequentordetailedtouseforadvancedAIalgorithms.Weneedmoresensorsandbettercommunicationsnetworkstobuildamoderndatainfrastructurethatwillallowforthefullbenefitsofdigitalizationtoberealized.Wheresufficientdataexists,itisoftenindifferentformats,notlabelledappropriately,storedon-site,andnotsharedwiththirdparties.Acriticalsteptoenablinggreaterdataexchangeistoagreeoncommondatastandardsacrosstheenergysector.Oncetheseareputinplace,secure,trust-basedsystemsfordatasharingwithintheenergysectorcanbeadopted,forexample,throughwidercloudandblockchainadoptionoranonymizeddataaggregationactivitiesmoderatedbyregulators.Attheregulatorylevel,thetrade-offsbetweenthebenefitsofusingdataandprotectingdataprivacyneedtobecarefullybalancedwithconcessionsmadewhennecessary(e.g.tomakeitanoptiontousesmartmeterdatawhenaggregatedandanonymized).Acriticalsteptoenablinggreaterdataexchangeistoagreeoncommondatastandardsacrosstheenergysector.Principle5:Incentives–createmarketdesignsandregulatoryframeworksthatallowAIusecasestocapturethevaluethattheycreateAIcanmakemoredecisionsandfasterdecisionsthanhumanscan,creatingnewopportunitiestomakeandcapturevaluewithinenergysystems.Intheabsenceofregulatoryframeworksthatadequatelyvaluebehind-the-meterdeviceflexibility,thereislittleincentivetoscaleuptheseusecases.AIapplicationswillonlyscaleoncethereareclearvaluepropositionsforcustomersandothermarketparticipants,andonlyregulatorscanestablishthefoundationalstructuresandframeworkstounlocktheseeconomicandsocietalvaluepropositions.ButthereiscurrentlynoagreementonhowtovalueAI-drivenapplications.Forinstance,thevalueofautomateddemand-responsebiddingfrombehind-the-meterdevicesorAI-assistedmicrotransactionsthatoptimizelocalbehind-the-meterconsumptioncannotcurrentlybefullycaptured.Principle6:Education–empowerconsumersandtheenergyworkforcewithahuman-centredAIapproachandinvestineducationtomatchtechnologyandskilldevelopmentForAItocontributemeaningfullytothefuturepowersystem,itneedstoearntrustfromtheengineers,employeesandmanagerswhorunit.EveryoneshouldfeelcomfortablewithAIbeingpartoftheirworkflows,eveniftheyarenotdevelopingtheAItoolsthemselves.Otherindustrieshavesucceededindoingthisbyredesigningteamstobringtogetherthenecessaryknowledgeandskillsets.Forthepowersector,thismightmeanteamscomprisingenergyengineersanddatascientists,wherethelatterdrivethedevelopmentofAIcapabilitiesandtheformerintegratetheoutputsofAIsystemsintogridmanagementprocessesandmaximizetheiroperationalvalue.ThesuccessfuldeploymentofAI-basedsolutionsbytheenduserwillalsoinvolveeducation.Educatingconsumersonhowtheirdataisusedinthesealgorithms,andwhatthelimitsofAIare,shouldhelpthembestinteractwithit.EnablingHarnessingArtificialIntelligencetoAcceleratetheEnergyTransition18Principle7:Riskmanagement–agreeuponacommontechnologyandeducationapproachtomanagingriskspresentedbyAIRecentEUregulationconsidersAIforenergyahigh-riskapplication.ForAIapplicationstoscalewithintheenergysector,regulatorsandindustryleadersneedtounderstandandmitigatepotentialrisksthatAImightpose.RegulatoryoptionsincludecommonqualitycontrolprocedureswhenbuildinganddeployingAI;designingdecentralizedcontrolstructures(ensuringthatonlyasmallpartisaffectedincaseofanincident);certifyingAIsystemsand/orsystemoperatorsforsafety;andconductingalgorithmicaudits.TechnologyoptionsincludeAI-basedsecuritylayers(i.e.usingAIsystemstodetectmanipulativebehaviourwithinthemarket)andautomatedloggingofAIsystems’activitiesanddecisions.WorkingoncommonapproachesaroundevaluatingandmanagingtherisksrelatedtoAIwillbecriticaltoestablishingtrustandtransparencyinalgorithms.Settingclearperimeterswithinwhichalgorithmsoperatecandecreaseriskswhencomparedtoconventionalhumanprocesses;however,theperceivedrisksarestilllowerwhenahumanisincontrol.EducationaroundAIrisksandAIriskmanagementwillbecriticalforregulators,policy-makers,energysectoremployeesandcitizens.Principle8:Standards–implementcompatiblesoftwarestandardsandinteroperableinterfacesTheincreasingnumberofintegrateddevicesandbehind-the-meterassetsmeansthatthesectorincreasinglyneedstoagreeuponstandardprotocolsforsoftwarecommunicationandmachineinterfaces.Today,therearemanydifferentstandardsandprotocolsfordifferentgeographiesandpartsoftheenergysystem(e.g.gridcommunications,smartmeters,EVchargers),whichresultsinalackofinteroperability.Thisfragmentationwillonlyincreaseasweattempttointegrateagrowingvarietyofappliancesandinstallationsintothegrid,leadingtosub-optimaloutcomesandagridthatisa“wholelessthanthesumofitsparts”.Allenergysystemstakeholdersandmarketparticipants,includingregulators,gridoperatorsandequipmentmanufacturers,shouldadoptcommonstandardsanddesignandinstallinteroperable“plugandplay”devices.Principle9:Responsibility–ensurethatAIethicsandresponsibleuseareatthecoreofAIdevelopmentanddeploymentAsAIadoptionhasacceleratedacrossindustrysectors,sotoohaveconcernsabouttherisksofunsafeorunethicalAI.Toensurebeneficialoutcomesandavoidsocietalharms,AIapplicationsintheenergysectormustadheretotheOECD’sfivecoreAIprinciples:inclusivity,fairness,transparency,robustness,andaccountability.Inpractice,thismeanstakingarisk-basedapproachwherebyAI’sgovernanceandriskmanagementpracticesareimplementedaccordingtothepotentialforharmofagivenusecase,withparticularattentiontohigh-riskusecases,sensitivepersonaldataandvulnerablepopulations.AIriskisbestmanagedwhenethicalconsiderationsarecoretothetechnologyandsystemdesignprocesses,whenAIsystemsarethoroughlydocumentedandrigorouslytestedpriortoandthroughoutimplementation,andwhenorganizationalprocessesaredesignedfortherapididentificationandmitigationofemergingissues.AstheindustryexpandsitsbroaderAItechnologyandmanagementcapabilities,itmustproactivelyensurethatAIethicsandresponsibleuseconsiderationsarefullyintegratedintoAIdevelopmentanddeploymentprocesses.WorkingoncommonapproachesaroundevaluatingandmanagingtherisksrelatedtoAIwillbecriticaltoestablishingtrustandtransparencyinalgorithms.GoverningHarnessingArtificialIntelligencetoAcceleratetheEnergyTransition19Recommendationsandoutlook4Companiesandpolicy-makersmustplayanactiveroleingoverningandshapingtheuseofAIintheenergysectorinaresponsibleway,andcreatinganenablingenvironmenttounlockAI’sfullpotential.Buildingonthe“AIfortheenergytransition”principles:Whatneedstohappennext?Howcantheseprinciplesbeoperationalized,andwhoneedstoact?TheenergyindustrywouldbenefitfromapproachingAI-relatedtechnologygovernanceinaproactiveandcollaborativeway.Thecomingyearswillbecrucialforencouraginginnovationinthisdomainanddemocratizingaccesstonewlow-carbontechnologiesacrosstheenergysystem.Asaprerequisiteforthis,anddigitalizationmorebroadly,theindustrywillhavetoadoptcommondatastandards,ifnotalreadyadopted.IncreasedcollaborationamongenergysectoractorscouldincludeR&Dcollaborations,sharingofbestpracticeapproachestooperationalizeAIprinciples,andshowcasingusecases.CollaborationcouldalsohelpbuildtrustbetweendevelopersandusersofAItechnologiesaswellaswithconsumersandregulatorsinterfacingwithAIsystems.Energycompanies/utilitiesexecutiveswillneedtothinkaboutwhetherandhowtheymakeuseofAI(e.g.whichchallengesAIcanhelpsolveand,therefore,whichprocesses,productsandserviceswillbenefitmostfromit).CompanyleadershipwillneedtobuildanunderstandingofwhattheAIprinciplesestablishedearlierinthiswhitepaper,andanyrelevantregulations,meanfortheirorganization,andhowtheycantranslatethisintoconcreteproductdesign,day-to-dayoperations,anddecision-makingprocesses.Companiescanstartbyexploringbestpracticesforknownusecases.ItwillbeastrategicdecisionforcompanieswhethertoadoptAIsolutionsbyprocuringthemfromexternalprovidersortodevelopthenecessarycapabilitiesandsolutionsin-house.Ineithercase,companieswillneedtoinvestincapacitybuildingtoensurethatstaffarecapableofmanagingtheintegrationofAIsystemsandrealizingtheirfullvalue.Asthemanagementandoperationofgridsbecomesincreasinglycomplex,inparticularonthedistributiongridlevel,gridregulatorsandoperatorsmustreviewthepotentialofarangeofdigitaltechnologies(e.g.machinelearning,quantumcomputing,blockchaintechnology,etc.)toaugmentthewaygridsareoperated.Asthepowersystemdecarbonizesanddecentralizes,thereisaneedtorethinkgridmanagementandanopportunitytoconsidernewandmoredecentralizedarchitecturesforgridaccess,operationandmanagementdecisions.Suggestionsincludemovingawayfromthetraditionalmanualcommand-and-controlmanagementapproach(withacentralsystemoperator),towardstechnology-enableddecentralizeddecision-making,whichwillallowforfasterdecision-makingandtheautomaticadditionofsmallerdistributedassetstothegrid(e.g.usingblockchain,digitalidentityandsmartcontracts).Policy-makersandsystemoperatorswillneedtoreviewexistingmarketdesignsandcreateadvancedelectricitymarketsthatrewardbothvariablelow-carbongeneration,aswellasflexibledemand.Todothis,atrulylevelplayingfieldfordistributedgenerationvis-à-vislarger-scalepowergenerationunitsneedstobecreatedandregulatoryhurdlesremoved.AsmanyAIusecasesintheenergysectorrelatetosmall-scaledistributedenergyresources,theseneedtohaveunrestrictedaccesstotheenergymarketsandthecorrespondingvaluepools,tomarketAI-assistedflexibility,forexample.Inregionalandnationalenergysystemmodellingandinfrastructureplanning,plannersshouldconsidertherole(s)thatAI-enabled,intelligentlydistributedenergyresourcescouldplay.Todate,energymodellingoftenignoresdistributiongridsandoverlooksthepotentialforthemtoactasasourceofgridflexibilityandbecomevaluableparticipantsinthegridmanagementprocess.Integratingthesesources,andgettingabetterunderstandingofhowtheycansupportthetransition,canleadtoamoreinformeddecisiononinfrastructureinvestmentssuchasgridextensionandmodernization,orthedeploymentofnew,centralizedpowergenerationunits.HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition20Nationalgovernmentsshouldconsiderbuildingclearerregulationsforenergydata(e.g.howitshouldbeprotectedandwhohastherighttouseit)andmakesurethataccesstothisdataisequitableandfair.Ifdataistobecomeacommodityfortheenergytransition,thengovernmentsshouldlayoutclearandsimpledesignrulestomakeitquicktocollect,safetostore,easytouse,andequitablydistributed.Whendesigningnewregulations,itiscriticaltoconsiderthelevelofbureaucracythatthisadds,asthismightcreatesignificantentrybarriersforstart-upsandsmallerplayers.Aspartofthisequitabledistributionofdata,governmentscoulddirectorincentivizeindustryorganizationsandpublicentitiestomanageandfundcentraldatabasesofindustrydata.Whensecure,anonymizedandaggregated,thesedatasetswouldenableAIalgorithmtrainingandpotentiallyreducealgorithmbiasesthatoftenresultfrompoorqualityorlimitedquantitiesofdata.HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition21ContributorsClaireCurryHeadofTechnologyandInnovation,UKJonMooreChiefExecutiveOfficer,UKBloombergNEFDeutscheEnergie-Agentur(dena)-GermanEnergyAgencyWorldEconomicForumLindaBabilonExpertDigitalizationandEnergySystems,GermanyAndreasKuhlmannChiefExecutiveOfficer,GermanyPhilippRichardDirector,DigitalTechnologiesandNetworks,GermanyMarkCaineProjectLead,ArtificialIntelligenceandMachineLearning,SanFranciscoDominiqueHischierProgramAnalystandEnergySpecialist,Energy,MaterialsandInfrastructurePlatform,GenevaEspenMehlumHeadofEnergy,MaterialsandInfrastructureProgram-BenchmarkingandRegionalAction,GenevaHarnessingArtificialIntelligencetoAcceleratetheEnergyTransition22AcknowledgementsTheWorldEconomicForum,BNEFanddenawouldliketothankthefollowingfortheirparticipationinaroundtableseriesconductedMarch-May2021andfortheirsignificantinputtothispaper.EdAbboPresidentandChiefTechnologyOfficer,C3AIGiuseppeAmorosoHeadofDigitalStrategyandGovernance,EnelArefBoualwanGroupManager,DigitalTransformation,ConsolidatedContractorsCompany(CCC)HendrikBrakemeierHeadofAIJourney,AppliedAINicoleBüttner-ThielFounder,MerantixLabsElleCarberryAssociatePartnerandGo-to-MarketLeader,IBMBeatrizCrisóstomoMerinoGlobalHeadofInnovation,IberdrolaJohn-PeterDolphinDirectorDataManagementandAnalytics,PG&ESimonEvansDigitalEnergyLeader,ArupGroupJamieExonDirectorofDigitalandCustomerServices,SanDiegoGasandElectricArantzaEzpeletaChiefTechnologyandInnovationOfficer,AccionaHeatherFeldmanDirectorInnovation,ElectricPowerResearchInstituteMadeleineGleaveChiefDataScientist,NithioPaolaJ.GranataHeadofSustainableImpact,SplightArtificialEnergyDavidGoddardChiefDigitalOfficer,HitachiABBPowerGridsGeorgeHackfordDirectorofStrategicAccounts,EMEAandAmericas,SparkBeyondRainerHoffmannSeniorManagerDataandAI;DigitalTransformation,EnBWEnergieBaden-WürttembergChristianJacobssonVicePresidentDataScienceandAnalytics,FortumLarshJohnsonChiefTechnologyOfficer,StemHannsKoenigHeadofCommissionedProjects,AuroraEnergyResearchEmmanuelLagarrigueExecutiveVice-President,ChiefInnovationOfficer,SchneiderElectricGavinMcCormickExecutiveDirector,WatttimeMiguelMoreiradaSilvaManagingPartner,WiimerMarcoMorettiHeadofDigitalInnovationandSustainability,EnelAmitNarayanFounderandChiefExecutiveOfficer,AutogridSystemsTitiaanPalazziCo-FounderandChiefOperativeOfficer,MystAIColinParrisSeniorVicePresidentandChiefTechnologyOfficer,GEDigitalMarcPetersCTOEnergy,EnvironmentandUtilitiesEurope,IBMRajeshRamachandranChiefDigitalOfficerIndustrialAutomation,ABBAlexRobartEnergyIndustryStrategyLeader,MicrosoftGiacomoSilvestriGroupHeadofDigital,EniLucasSpreiterHeadofWorkingGroupAIandClimateChange,KIBundesverbandMaheshSudhakaranChiefDigitalOfficer,IBMAndreasUlbigProfessorforActiveEnergyDistributionGrids,RWTHAachenUniversityJean-SimonVenneChiefTechnologyOfficer,BrainBoxAIIoannisVlachosDirectorEnergyMarkets,EnergyWebFoundationStephenWoodhouseDirector,AFRYMarziaZafarHeadofCustomerStrategyandPolicy,Kaluza,TheOVOGroupHarnessingArtificialIntelligencetoAcceleratetheEnergyTransition23Endnotes1.WorldEconomicForum,SystemValueFramework,2020,https://www.weforum.org/projects/system-value,(linkasof9/8/21).2.BloombergNEF,NewEnergyOutlook2020,https://about.bnef.com/new-energy-outlook/,(linkasof9/8/21).3.Ibid.4.Ibid.5.ThiswhitepaperfocusesontheapplicationsofAIthatwereconsideredtohaveahighpotentialtosupporttheenergytransition.Theintentionisnottogiveacomprehensiveoverviewofallapplicationsthroughouttheenergysector,e.g.forconventionalpowergenerationtechnologies.6.Strictlyspeaking,machinelearningisacurrentapplicationorsubsetofAIthatallowsmachinestolearnfromdatawithoutbeingprogrammedexplicitly.Inthiswhitepaper,thetwotermsareusedinterchangeably.7.BloombergNEF,NewEnergyOutlook2021.8.Fantetal.,“ClimatechangeimpactsandcoststoU.S.electricitytransmissionanddistributioninfrastructure”,Energy,15March2020,vol195,https://www.sciencedirect.com/science/article/pii/S0360544220300062,(linkasof9/8/21).9.BloombergNEF,FlexibilitySolutionsforHigh-RenewableEnergySystems,2018/2019,https://data.bloomberglp.com/professional/sites/24/2018/11/UK-Flexibility-Solutions-for-High-Renewable-Energy-Systems-2018-BNEF-Eaton-Statkraft.pdf,https://assets.bbhub.io/professional/sites/24/Flexibility-Solutions-for-High-Renewable-Energy-Systems-Spain-Outlook.pdf,(linksasof25/08/2021).10.BNEFPowerGridLongTermOutlook,2021’https://www.bnef.com/insights/2559111.DeutscheEnergie-Agentur(dena)-GermanEnergyAgency,ArtificialIntelligence-fromHypetoRealityfortheEnergyIndustry,2020,https://www.dena.de/fileadmin/dena/Publikationen/PDFs/2020/dena_ANALYSIS_Artificial_Intelligence_-_from_Hype_to_Reality_for_the_Energy_Industry.pdf(linkasof25/08/2021)12.DeutscheEnergie-Agentur(dena)-GermanEnergyAgency,ArtificialIntelligencefortheIntegratedEnergyTransition,2019,https://www.dena.de/fileadmin/dena/Publikationen/PDFs/2019/dena-REPORT_Artificial_Intelligence_for_the_Integrated_Energy_Transition.pdf(linkasof25/08/2021).13.BloombergNEF,1H2021WindO&MPriceIndex,https://www.bnef.com/login?r=%2Finsights%2F26811,(linkasof9/8/21).14.ArtificialIntelligenceAct(21April2021),ProposalforaRegulationoftheEuropeanParliamentandoftheCouncil,layingdownharmonisedrulesonArtificialIntelligence(ArtificialIntelligenceAct)andamendingcertainUnionlegislativeacts,2021,EuropeanCommission,EUR-Lex-52021PC0206,https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELLAR:e0649735-a372-11eb-9585-01aa75ed71a1,(linkasof9/8/21).15.Ibid.HarnessingArtificialIntelligencetoAcceleratetheEnergyTransition24WorldEconomicForum91–93routedelaCapiteCH-1223Cologny/GenevaSwitzerlandTel.:+41(0)228691212Fax:+41(0)227862744contact@weforum.orgwww.weforum.orgTheWorldEconomicForum,committedtoimprovingthestateoftheworld,istheInternationalOrganizationforPublic-PrivateCooperation.TheForumengagestheforemostpolitical,businessandotherleadersofsocietytoshapeglobal,regionalandindustryagendas.
VIP
