conservation voltage reduction econometric impact analysis · 2018-02-02 · conservation voltage...

Conservation Voltage Reduction Econometric 

Impact Analysis

PresentedtoAESPSpringConference

PresentedbySanemSergici,TheBrattleGroup

(incollaborationwithPepcoMDteamledbySteveSunderhauf andBasilAllison)

May 11, 2016

AgendaBackgroundDataOverviewMethodology SelectingControlGroups ConservationAnalysis PeakAnalysis

Results ConservationAnalysis PeakAnalysis

Background: What is Conservation Voltage Reduction?

ConservationVoltageReduction(CVR)isareductioninfeedervoltagewhichresultsinareductioninenergyconsumption

Keyengineeringprincipal:VoltagecanbekeptonlowerendofAmericanNationalStandardInstitutestandardvoltagebandof114‐126volts

PepcoMaryland’simplementationofAdvancedMeteringInfrastructurehasenabledPepcotomonitorandvaryvoltagelevelswhileremainingwithinspecifiedstandards

Background and Objectives PepcoMDinitiatedtheCVRpilotprogramonAugust1,2013.Itencompasses7substations Approximately45,000residentialcustomers Approximately4,000non‐residentialcustomers

Thevoltagelevelswerereducedby1.5%atthosesubstationsparticipatinginthepilot

Theobjectiveofourstudywasto: QuantifytheconservationimpactoftheCVRprogramforresidentialandnon‐residentialcustomers

QuantifythepeakdemandimpactoftheCVRprogramforresidentialandnon‐residentialcustomers

BackgroundOverview of Previous Research‐ I

Moststudieshavebeenengineeringstudiesasopposedtoeconometricanalysis,andhavenotestimatedapeakdemandvsenergyconservationimpact,oraresidentialvsnon‐residentialimpact

SeveralstudieshavedemonstratedthattheimplementationofCVRleadstodecreasedconsumption,butthereisnoconsensusfora“CVRfactor”(energyreduction/voltagereduction) StudiesindicatearelativelywiderangeofCVRfactors,generallyrangingfrom.5to1

BackgroundOverview of Previous Research‐ II

Residentialandnon‐residentialloadmayresponddifferentlytheCVRasnon‐residentialloadgenerallyhasalargershareofmotorload,whichmaymitigatetheeffectofCVR

CVRasanideahasbeenaroundfordecades,buthasrecentlygainedmoreattentionasitisbecomingmorecost‐effectiveandalsoeasiertocontrol/monitorduetothedeploymentofAMI

BackgroundReview of Select Previous Studies‐ I

WestPennPowerCompany(2014) Studyreducedvoltageby1.5%doinga“onforaday,offforaday”approach

SimilartoPepcoMDstudyinthatitusesdifference‐in‐differencesmethodology

RangeofCVRfactorsbutaverageis0.86 IndianapolisPower&LightCompany(2013)

Studyturned“on”CVRforafewshortperiodsin2012and2013,andcompareddropinusageduringthoseperiodstopredictanimpact

StudyestimatedaCVRfactorof0.7‐0.8

BackgroundReview of Select Previous Studies‐ II

DominionVirginiaPower(2012) Studycomparedbaselinepre‐CVRperiodtoconsumptionduringperiodafterCVRwasimplemented

Impactcalculateusingaday‐pairingmethodinsteadofdifference‐in‐differences

Day‐pairingmethodmatchesdayfromthepre‐treatmentperiodtodaysinthepost‐treatmentperiodtocalculateCVRimpact

StudyfoundaCVRfactorof0.92

BackgroundReview of Select Previous Studies‐ III

PacificNorthwestNationalLaboratory(2010) EstimatedimpactofCVRon24modeledfeedersbyrunningaone‐yearsimulationofsystemandre‐runningwithreducedvoltagelevels

Resultswerevaried,butalmostallfeedersexperiencedsomereductioninbothpeakdemandandenergyconsumption

NorthwestEnergyEfficiencyAlliance(2007) StudymeasuredCVRimpactbycomparing24hoursonand24hoursoff,insteadofusingasetcontrolgroup

StudyfoundCVRfactorsforpeakdemandrangingfrom0.55‐1.12andforenergyrangingfrom0.3‐0.86

AgendaBackgroundDataOverviewMethodology ConservationAnalysis PeakAnalysis

Results ConservationAnalysis PeakAnalysis

Data OverviewThefollowingdatasetswereutilizedforthisanalysis

Billingdata Hourlyconsumption Weatherdata(dewpointanddrybulb temperatures) Advancedmeteringinfrastructure(AMI)activationdate ParticipationinDemandSideManagementprograms RecipientsofOpower HomeEnergyReports Netenergymetering(NEM)status

Data Overview Forthepeakanalysis,theprimarydatasetwashourlydatafromAMIforJune‐August2013and2014,hours‐ending15‐19

Fortheconservationanalysis,theprimarydatasetwasmonthlybillingdatafromSeptember2012throughAugust2014 MonthlydatausedbecausehourlydatawasonlyavailableforsummerbeforeCVRimplementationasAMIactivationstartedinearly2012butwasnotcompleteduntilmid‐2013

AgendaBackgroundDataOverviewMethodology SelectingControlGroups ConservationAnalysis PeakAnalysis

Results ConservationAnalysis PeakAnalysis

MethodologySelecting Control Groups

PepcoMD’sCVRprogramwasnotdesignedasarandomizedcontroltrial

PepcoMarylandengineeringandloadexpertsmatchedeachsubstationwhichreceivedCVRtreatmentwithacontrolsubstationwhichdidnotreceiveCVRtreatment

Tomatchtreatmentandcontrolsubstations,theexpertsconsideredcustomerandloadcharacteristicsandensuredthattreatmentandcontrolpairingsaregenerallyadjacent

Allpairingsareinasinglejurisdiction,manyfactorswhichaffectconsumption(e.g.,economicfactorsandweather)aresimilarbetweenpairings

MethodologySubstation Pairings

TreatmentSubstations ControlSubstations

KensingtonSub.193 LindenSub.156

LongwoodSub.192 WoodAcresSub.154

MontgomeryVillageSub.56 GaithersburgSub.31

BranchvilleSub.69 GreenbeltToaping CastleSub.173

RiverdaleSub.4 BladensburgSub.175

CampSpringsSub.72 St.BarnabasRd.Sub.59

Wildercroft Sub.178 LanhamSub.149

MethodologyValidating Control Group Wecarried‐outafter‐the‐factcomparisonofPepcoMaryland’scontrol‐treatmentpairingstovalidatethecontrolgroup

BelowarecomparisonsofcontrolandtreatmentconsumptionusinghourlyAMIdataforthepeakanalysis

MethodologyValidating Control Group We find that the residential customer load profiles are verysimilar to each other in terms of their shape and level for thetreatment and control groups This implies that the residential control group customersrepresent the but‐for usage of the residential treatmentcustomers fairly well

For the non‐residential customer load profiles, we find thatthey are very similar to each other in terms of their shapebut they differ in terms of the level of usage between thetreatment and control groups Treatment customers are slightly larger than the control groupcustomers, on average. This difference will be accounted for bythe fixed effects estimation routine

Methodology: Difference‐in‐Differences through Panel Data AnalysisWe carried out a Difference‐in Differences (DID) analysis through apanel data regression analysis to estimate the CVR impact Regression model compares the usage of the treatment and control

group customers before and after the CVR treatment, while accountingfor other factors that could potentially confound the estimated impactsuch as weather conditions, DSM program participation, AMI activation,and calendar dummies

The Fixed Effects (FE) estimation routine was used to ensure that theestimated coefficients from the resulting model are unbiased. FEestimation assumes that the unobservable factor in the error term isrelated to one or more of the model’s independent variables. Therefore,it removes the unobserved effect from the error term prior to modelestimation using a data transformation process

MethodologyCVR Impacts estimated in this Study

Impact Dataset AnalysisVariable

Pre‐treatmentPeriod(*)

Post‐treatmentPeriod

Peak Hourly AMIDataset HourlyUsage

June–August2013

June–August 2014

Conservation

MonthlyBillingData

Average DailyUsage

Sept.2012–August2013

Sept.2013–May2014

(*) The CVR program has begun on August 1st, however the CVR activation for the last treatment substation was on August 12, which is the effective start date of the CVR program for our analysis. For that reason, August 2013 is partially a pre-treatment month

MethodologyConservation Model SpecificationConservationmodelmeasuresaverageenergysavingsfromCVR

∗ ∗ ∗ ∗

∗ ∗ ∗ ∗

Where:Averagehourlyconsumptionforhouseholdi indayt.FlagindicatingthatthestartofthetreatmentperiodFlagindicatingthatthecustomerhasreceivedtheCVRtreatmentImpactofTemperatureHumidityIndexonusageFlagindicatingthatacustomer’sAMImeterhasbeenactivatedMonthspecificimpactcommontoallhouseholds

∗ MonthspecificimpactoftheTemperatureHumidityIndexIndicatorthatacustomerisparticipatinginDSMprogramCustomerfixedeffectiid errorterm,clusteredbyhousehold

Methodology Peak Impact Model PeakimpactmodelmeasurespeakdemandsavingsfromCVR Asthepeakimpactanalysisisfocusedonquantifyingthesavingsduringsystempeakconditions,weundertakeouranalysisusingdataonthehottestdaysoftheyear Wedefinepeakashoursending15‐19(usingPJM’scapacitymarketpeakdefinitionforsummer)

WedefinehottestdaysasthosewithaveragepeakTHIsgreaterthan77,whichequatestoroughly85°F)

Werunthepeakimpactmodelforweekdays,weekendsandalldaystogaugewhetherthepeakimpactvariesduetodifferentpeakloadcharacteristicsduringthesedays

MethodologyPeak Impact Model Specification

∗ ∗ ∗

∗ ∗ ∗

∗ ∗ Where:

Averagehourlyconsumptionforhouseholdi indaytFlagindicatingthestartofthetreatmentperiodFlagindicatingthatthecustomerhasreceivedtheCVRtreatmentImpactofTemperatureHumidityIndexonusageMonthspecificimpactcommontoallhouseholds

∗ MonthspecificimpactoftheTemperatureHumidityIndexIndicatorthatacustomerisparticipatinginDSMprogramgroupkCustomerfixedeffectiid errorterm,clusteredbyhousehold

AgendaBackgroundDataOverviewMethodology SelectingControlGroups ConservationAnalysis PeakAnalysis

Results ConservationAnalysis PeakAnalysis

ResultsConservation ImpactResidentialCustomersA1.5%reductioninvoltageisestimatedtoresultina1.4%reductioninconsumption Significantatthe1%level ImpliedCVRfactorof.93whichiswithinrangesuggestedbypreviousstudies

ResultsConservation ImpactNon‐ResidentialCustomers1.5%reductioninvoltageisestimatedtoresultina0.9%reductioninconsumption Notstatisticallysignificant,thoughstillanunbiasedestimateofthemeanimpact

ImpliedCVRfactorof0.6whichiswithinrangesuggestedbypreviousstudies

Insignificantresultlikelydrivenbysmallersamplesizeandalsoheterogeneityofcustomers

ResultsPeak ImpactResidentialCustomersA1.5%reductioninvoltageisestimatedtoresultina1.1%reductioninpeakconsumption Significantatthe1%level ImpliedCVRfactorof.73whichiswithinrangesuggestedbypreviousstudies

ResultsPeak ImpactNon‐ResidentialCustomers1.5%reductioninvoltageisestimatedtoresultina2.5%reductioninpeakconsumption Significantatthe1%level ImpliedCVRfactorgreaterthan1isbeyondexpectedrangeforCVRimpact

Highimpactimpliesthatthereareotherunobservableeffectswhichwewerenotabletocaptureinthisanalysis,likelyduetoheterogeneityofnon‐residentialcustomers

ResultsPeak ImpactResidentialpeakresultsarerobustacrossdaysandhours

Hour Ending All DaysWeekends & Holidays Only

Weekdays

(% Impact) (% Impact) (% Impact)

Hour 15 ‐1.13% ‐1.67% ‐0.90%Hour 16 ‐1.02% ‐1.23% ‐0.90%Hour 17 ‐1.02% ‐1.08% ‐1.00%Hour 18 ‐1.21% ‐1.16% ‐1.20%Hour 19 ‐1.17% ‐1.10% ‐1.20%

15‐19 Pooled ‐1.11% ‐1.28% ‐1.08%

ResultsConclusion

Residentialimpactisrobust PepcoMaryland’sCVRpilotprogramhasbeensuccessfulinleadingtoadecreaseinresidentialconsumptionduringpeakhoursandalsoyear‐round

Theresultsarestableacrossmultipleeconometricmodels

Non‐Residentialimpactismoredifficulttoquantifyusingeconometricmethodsduetoheterogeneityandsamplesizeissues Inthefuture,largerdatasetswithlargersamplesizemayresultinstatisticallysignificantresults