Citation
Control of Commercial Building HVAC Systems for Power Grid Ancillary Service

Material Information

Title:
Control of Commercial Building HVAC Systems for Power Grid Ancillary Service
Creator:
Lin, Yashen
Place of Publication:
[Gainesville, Fla.]
Florida
Publisher:
University of Florida
Publication Date:
Language:
english
Physical Description:
1 online resource (107 p.)

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Mechanical Engineering
Mechanical and Aerospace Engineering
Committee Chair:
BAROOAH,PRABIR
Committee Co-Chair:
KUMAR,MRINAL
Committee Members:
DIXON,WARREN E
MEYN,SEAN PETER
Graduation Date:
8/9/2014

Subjects

Subjects / Keywords:
Air flow ( jstor )
Climate models ( jstor )
Commercial buildings ( jstor )
Cooling ( jstor )
Environmental control systems ( jstor )
Heating ventilation and cooling ( jstor )
Modeling ( jstor )
Signals ( jstor )
Simulations ( jstor )
Temperature control ( jstor )
Mechanical and Aerospace Engineering -- Dissertations, Academic -- UF
ancillary -- building -- grid -- hvac
Genre:
bibliography ( marcgt )
theses ( marcgt )
government publication (state, provincial, terriorial, dependent) ( marcgt )
born-digital ( sobekcm )
Electronic Thesis or Dissertation
Mechanical Engineering thesis, Ph.D.

Notes

Abstract:
Ancillary services are used to correct the mismatch between power generation and consumption in a power grid. They are needed to maintain the grid's functionality and reliability. With more renewable energy sources being integrated into the power grid, more volatility is introduced, which increases the need for ancillary service. Recent years have seen a growing interest in utilizing demand side resources to provide ancillary service. The responsibility of such service providers is to vary their power consumption from that in normal operation so that the power deviation tracks a reference signal provided by the grid operators. In this work, we explore the potential of commercial building HVAC systems in providing ancillary service. Commercial buildings account for 40% of the total electricity consumption in the U.S. and their flexible loads are large untapped resources. One important constraint for demand side ancillary service providers is that their primary functionality has to be maintained while providing the service. In the case of commercial building HVAC systems, it is to ensure comfortable indoor climate. The key idea is frequency separation. The thermal dynamics of the building is slow due to its large thermal inertia. This feature makes it possible to make fast changes in the HVAC system power consumption without significantly affecting indoor climate. In this work, we consider two distinct actuation mechanisms to provide ancillary service. In the first part, we consider varying the supply air fan power to provide high-frequency ancillary service. Algorithms for controlling the fan power are developed. The algorithms are first tested in simulation and then implemented in a real commercial building -- Pugh Hall on University of Florida campus. Results show that supply air fan can be used to provide high quality ancillary service in the frequency range [1/(10 mins), 1/(30 secs)] without noticeable effect on the indoor climate. Baseline estimation is performed by using the separation between the frequency of reference signal and that of the building climate control system. In the second part, we consider utilizing the chiller to provide low-frequency ancillary service. We propose an indirect method to vary the chiller power: the supply air flow rate is controlled, which changes the heat exchange in the cooling coil. This in turn affects the chiller power consumption. The problem of determining the baseline power consumption is circumvented by scheduling the baseline ahead of time based on load and weather prediction. The scheduled baseline is updated periodically based on measurements to provide comfortable indoor climate and robustness to prediction error. Simulation studies demonstrate that high quality ancillary service in the frequency range [1/(1 hour), 1/(10 mins)] can be provided without significant effect on the indoor climate. ( en )
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis:
Thesis (Ph.D.)--University of Florida, 2014.
Local:
Adviser: BAROOAH,PRABIR.
Local:
Co-adviser: KUMAR,MRINAL.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2015-08-31
Statement of Responsibility:
by Yashen Lin.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Embargo Date:
8/31/2015
Resource Identifier:
969976978 ( OCLC )
Classification:
LD1780 2014 ( lcc )

Downloads

This item has the following downloads:


Full Text

PAGE 1

CONTROLOFCOMMERCIALBUILDINGHVACSYSTEMSFORPOWERGRID ANCILLARYSERVICE By YASHENLIN ADISSERTATIONPRESENTEDTOTHEGRADUATESCHOOL OFTHEUNIVERSITYOFFLORIDAINPARTIALFULFILLMENT OFTHEREQUIREMENTSFORTHEDEGREEOF DOCTOROFPHILOSOPHY UNIVERSITYOFFLORIDA 2014

PAGE 2

c r 2014YashenLin 2

PAGE 3

Tomyparentsandwife 3

PAGE 4

ACKNOWLEDGMENTS Firstandforemost,Iwouldliketoexpressmysinceregratit udetomyadvisorDr. PrabirBarooahforguidingmethroughmystudy.Thisdissert ationwouldhavenot beenpossiblewithouthissupportandadvice.Henotonlypro videdinsightfuladvice inresearchtopic,butalsohelpedmebecomearigorousandin dependentthinker.His emphasisandguidanceoncommunicationskillsisespeciall ybenecialtomeasan internationalstudent.Ifeelveryfortunatetohavetheopp ortunitytoworkwithhimandI wouldliketothankhimforeverythinghehasdoneforme. IwouldliketothankcommitteemembersDr.WarrenDixon,Dr. MrinalKumar, andDr.SeanMeyn,whohavebeenalwayssupportiveandhelpfu ltome.Iamgrateful fortheirconstructiveadviceandinspiringdiscussionswh ichguidedmythroughmy study.IalsowanttoextendmyspecialgratitudetoDr.Timot hyMiddelkoop,who providedcrucialhelpintheexperimentalpartofmyresearc h.Itisapleasuretothank Dr.HerbertA.Ingley,forsharinghisexpertiseinHVACsyst em.Hispatience,kindness, andknowledge,isextremelyhelpfulincompletingmywork.I wanttogivespecialthanks toPederWinkel,SkipRockwell,andothersfromUniversityo fFlorida(UF)Physical PlantDivisionwhosparedtimehelpingwithourexperiments thoughtheyhavemany othercommissions. Also,IwouldliketothankmycolleaguesHeHaoandChendaLia o,whoprovided memuchneededhelpandusefuladvice,bothinresearchandpe rsonallife.Ithank SiddharthGoyalforhishelpintheexperiments,whichwould nothavebeenpossible otherwise.Lastbutnottheleast,Iwouldliketothankmypar entsandwife.Iamheartily thankfulfortheirfaith,devotion,love,supportandencou ragement. 4

PAGE 5

Finally,IwouldliketothankNationalScienceFoundationw hosupportedmy researchthroughGrantNo.0931885(CPS)and0955023(CAREE R). 1 1 Anyopinions,ndings,andconclusionsorrecommendations expressedinthis materialarethoseoftheauthor(s)anddonotnecessarilyre ecttheviewsofthe NationalScienceFoundation. 5

PAGE 6

TABLEOFCONTENTS page ACKNOWLEDGMENTS .................................. 4 LISTOFTABLES ...................................... 8 LISTOFFIGURES ..................................... 9 ABSTRACT ......................................... 11 CHAPTER 1INTRODUCTION ................................... 13 1.1DemandSidePowerGridAncillaryService(AS) .............. 13 1.2CommercialBuildingHAVCSystemsinProvidingAncillar yService .... 14 1.3BackgroundonVAVHVACSystems ..................... 20 1.4Organization .................................. 24 2HIGH-FREQUENCYA.S.THROUGHFEED-FORWARDCONTROL ...... 25 2.1ControlArchitecture .............................. 25 2.2PerformanceMetrics .............................. 29 2.2.1TrackingMetrics ............................ 29 2.2.2IndoorClimateMetrics ......................... 31 2.3SimpliedHVACModelforControllerDesign ................ 32 2.3.1FanandDuctModel .......................... 32 2.3.2BuildingThermalModel ........................ 34 2.4SimulationStudy ................................ 36 2.4.1HighFidelityHVACModel ....................... 36 2.4.2SimulationResults ........................... 39 2.4.3PotentialofCommercialBuildingHVACSystemsintheU .S. .... 43 3EXPERIMENTALEVALUATIONOFHIGH-FREQUENCYA.S.THROUGH FEEDBACKCONTROL ............................... 45 3.1ControlArchitecture .............................. 45 3.2SystemIdentication .............................. 47 3.2.1TestBed ................................. 48 3.2.2Sine-sweepExperiments ........................ 50 3.3ControllerDesign ................................ 53 3.3.1AncillaryServiceControllerthroughFanSpeed(ASCF S) ..... 53 3.3.2AncillaryServiceControllerthroughAirFlow(ASCAF ) ....... 55 3.3.3SimulationStudy ............................ 55 3.3.3.1ReferenceSignals ...................... 56 3.3.3.2SimulationResults ...................... 56 3.4ExperimentalResults ............................. 57 6

PAGE 7

3.4.1ASCFSExperiments .......................... 58 3.4.2ASCAFExperiments .......................... 59 3.4.3FilteringtheMeasurementNoise ................... 61 3.4.4EconomicPotential ........................... 62 4LOW-FREQUENCYA.S. .............................. 64 4.1ControlArchitecture .............................. 64 4.1.1SmithPredictor ............................. 66 4.1.2KalmanPredictor ............................ 68 4.2ModelingtheHVACSystemwithChiller ................... 70 4.2.1CoolingCoilDynamics ......................... 71 4.2.2ChillerDynamics ............................ 73 4.2.3ZoneThermalDynamics ........................ 74 4.2.4IndoorClimateControllerandAirowDynamics ........... 75 4.2.5PowerConsumption .......................... 75 4.3SimulationStudy ................................ 76 4.3.1SimulationSetup ............................ 76 4.3.2Results ................................. 77 4.3.3PotentialofCommercialBuildingHVACSystemsintheU .S. .... 80 5LOW-FREQUENCYA.S.WITHBASELINESCHEDULING ........... 82 5.1ControlAlgorithm ................................ 82 5.2BaselineScheduling .............................. 85 5.2.1LoadPredictionModel ......................... 85 5.2.2BaselineSchedulingatt = 0 ..................... 88 5.2.3BaselineUpdate ............................ 90 5.2.4PTCDesign ............................... 92 5.3StabilityAnalysisofBASA ........................... 93 5.4SimulationStudy ................................ 94 5.4.1ReferenceSignals ........................... 95 5.4.2BASA .................................. 95 5.4.3Results ................................. 96 6CONCLUSIONANDFUTUREWORK ....................... 100 REFERENCES ....................................... 103 BIOGRAPHICALSKETCH ................................ 107 7

PAGE 8

LISTOFTABLES Table page 2-1Estimatedthermalparametersfor2outof11zones ............... 39 2-2Performancemetricsinfeed-forwardcontrolsimulatio ns. ............. 41 3-1Performancemetricsinfeedbackcontrolsimulations. ............... 57 3-2Performancemetricsinexperiments. ........................ 61 3-3Effectsofancillaryservicecontrollersonroomclimat e. .............. 61 3-4Performancemetricswithlteredmeasurements. ................. 61 4-1Zonedynamicsparameters ............................. 76 4-2Modelparameters .................................. 77 4-3Performancewithreferencesignalsofdifferentfreque ncy. ............ 80 4-4Comparisonofcomfortviolation. .......................... 80 4-5Effectofdelaymismatchinchillerpower. ..................... 80 5-1ExampleofcomputingB i ( T j oa ). ........................... 87 5-2BASAcontrollerparameters ............................. 96 5-3Performancemetricsinexperiments. ........................ 98 5-4Effectsofancillaryservicecontrolleronroomclimate . .............. 98 8

PAGE 9

LISTOFFIGURES Figure page 1-1Ancillaryservicesinterpretedasactuationinacontro lsystem. ......... 13 1-2ACEsignalfromPJMon05/04/2009. ....................... 16 1-3Providingancillaryservicefromcommercialbuildings . .............. 17 1-4SchematicofacommercialbuildingVAVHVACsystem. ............. 21 1-5Singlemaximumcontrollogicforindoorclimatecontrol . ............. 23 1-6OutsideviewofthetestbedPughHall. ...................... 24 2-1Schematicofproposedfeed-forwardcontrolarchitectu re. ............ 26 2-2Detailedschematicofproposedcontrolarchitecture. .............. 27 2-3ThermalcomfortenvelopespeciedbyASHRAE. ................. 31 2-4FanmodelvalidationwithPughHalldata. ..................... 34 2-5MagnitudefrequencyresponsecomparisonofH v r vandH v r T. .......... 36 2-6PughHallsectorusedinconstructingthehighdelitymo del. .......... 39 2-7Zone1modelvalidationwithPughHalldata. ................... 40 2-8Trackingperformanceofreferencesignalswithdiffere ntband-widths. ..... 41 2-9Fanspeedandzonetemperaturevariationsfrombaseline . ........... 43 2-10r Rvalueoftrackingwithplant-modelmismatch. .................. 43 3-1Schematicofproposedfeedbackcontrolarchitecture. .............. 45 3-2Twolocationstoinjectcontrolcommandinthelocalcont rolloop. ........ 46 3-3FloorplanoftheauditoriumservedbyAHU-2inPughHall. ........... 48 3-4DatafromatypicaldayduringnormaloperationfromAHU2inPughHall. .. 49 3-5Measurementnoisecharacterization. ....................... 51 3-6FrequencyresponseofH 1identiedfromsine-sweepexperiment. ....... 52 3-7FrequencyresponseofH 2identiedfromsine-sweepexperiment. ....... 52 3-8ClosedloopfrequencyresponseofASCFSwithproportion alcontroller. .... 54 3-9ClosedloopfrequencyresponseofASCAFwithlagcompens ator. ....... 55 9

PAGE 10

3-10TherawandlteredACEsignals. .......................... 56 3-11SimulationresultofASCFSwithproportionalcontroll er. ............. 57 3-12SimulationresultofASCAFwithlagcompensator. ................ 58 3-13ResultsoftheASCFScontrollertestinAHU-2ofPughHal l. .......... 59 3-14Aclose-up5-minutesliceofFigure3-13. ..................... 60 3-15FieldtestresultswithlagASCAF. .......................... 60 4-1Schematicoftheproposedcontrolarchitecture. .................. 66 4-2PredictedfuturereferencebyKalmanpredictorwithdif ferentdelaylengths. .. 69 4-3Predictionerrorsofreferencesignalwithdelaymismat ch. ............ 70 4-4Coolingcoilmodelvalidationwithelddatacollectedf romAHU-2inPughHall. 73 4-5Zonethermalmodelvalidation. ........................... 76 4-6Exogenousinputsforatypicalday. ......................... 78 4-7Performanceoftheancillaryservicecontroller,andco mparisonwithbaseline. . 79 5-1SchematicillustrationoftheBASAcontroller. ................... 83 5-2Schematicillustrationofthereceding-horizonupdate . .............. 84 5-3ArchitectureofthePTC. ............................... 92 5-4Loadmodelcalibrationandvalidation. ....................... 96 5-5Scheduledbaselineandancillaryservicereferencesig nal. ........... 97 5-6Powerdeviationtrackingperformance. ...................... 97 5-7Roomtemperatureandsupplyairowrateduringthesimul ation. ....... 98 10

PAGE 11

AbstractofDissertationPresentedtotheGraduateSchool oftheUniversityofFloridainPartialFulllmentofthe RequirementsfortheDegreeofDoctorofPhilosophy CONTROLOFCOMMERCIALBUILDINGHVACSYSTEMSFORPOWERGRID ANCILLARYSERVICE By YashenLin August2014 Chair:PrabirBarooahMajor:MechanicalEngineering Ancillaryservicesareusedtocorrectthemismatchbetween powergenerationand consumptioninapowergrid.Theyareneededtomaintaintheg rid'sfunctionalityand reliability.Withmorerenewableenergysourcesbeinginte gratedintothepowergrid, morevolatilityisintroduced,whichincreasestheneedfor ancillaryservice.Recent yearshaveseenagrowinginterestinutilizingdemandsider esourcestoprovide ancillaryservice.Theresponsibilityofsuchserviceprov idersistovarytheirpower consumptionfromthatinnormaloperationsothatthepowerd eviationtracksareference signalprovidedbythegridoperators. Inthiswork,weexplorethepotentialofcommercialbuildin gHVACsystemsin providingancillaryservice.Commercialbuildingsaccoun tfor40%ofthetotalelectricity consumptionintheU.S.[ 1 ]andtheirexibleloadsarelargeuntappedresources.One importantconstraintfordemandsideancillaryservicepro vidersisthattheirprimary functionalityhastobemaintainedwhileprovidingtheserv ice.Inthecaseofcommercial buildingHVACsystems,itistoensurecomfortableindoorcl imate.Thekeyideais frequencyseparation.Thethermaldynamicsofthebuilding isslowduetoitslarge thermalinertia.Thisfeaturemakesitpossibletomakefast changesintheHVACsystem powerconsumptionwithoutsignicantlyaffectingindoorc limate. Inthiswork,weconsidertwodistinctactuationmechanisms toprovideancillary service.Intherstpart,weconsidervaryingthesupplyair fanpowertoprovide 11

PAGE 12

high-frequencyancillaryservice.Algorithmsforcontrol lingthefanpoweraredeveloped. Thealgorithmsarersttestedinsimulationandthenimplem entedinarealcommercial building–PughHallonUniversityofFloridacampus.Result sshowthatsupplyair fancanbeusedtoprovidehighqualityancillaryserviceint hefrequencyrange[1=(10min) , 1=(30sec)]withoutnoticeableeffectontheindoorclimate.Baseline estimationisperformedbyusingtheseparationbetweenthe frequencyofreference signalandthatofthebuildingclimatecontrolsystem. Inthesecondpart,weconsiderutilizingthechillertoprov idelow-frequency ancillaryservice.Weproposeanindirectmethodtovarythe chillerpower:thesupply airowrateiscontrolled,whichchangestheheatexchangei nthecoolingcoil.Thisin turnaffectsthechillerpowerconsumption.Theproblemofd eterminingthebaseline powerconsumptioniscircumventedbyschedulingthebaseli neaheadoftimebasedon loadandweatherprediction.Thescheduledbaselineisupda tedperiodicallybasedon measurementstoprovidecomfortableindoorclimateandrob ustnesstopredictionerror. Simulationstudiesdemonstratethathighqualityancillar yserviceinthefrequencyrange[1=(1hour) , 1=(10min)]canbeprovidedwithoutsignicanteffectontheindoorclim ate. 12

PAGE 13

CHAPTER1 INTRODUCTION 1.1DemandSidePowerGridAncillaryService Inapowergrid,thegenerationanddemandhavetobebalanced atalltimes. Mismatchbetweensupplyanddemandcausesthefrequencyoft hepowertodeviate fromitsdesignvalue(60HzintheU.S.),whichdeteriorates thepowerquality.Large deviationcanevenleadtoablack-out.Ancillaryservicesa reneededtocorrectthe mismatchtoensurethefunctionalityandreliabilityofthe powergrid. Withimprovementintechnology,renewableenergysourcesh avebecomemore viablealternativesforpowergeneration.However,theses ources,suchaswindand solar,areusuallyvolatileinnature.Ifwewanttointegrat emorerenewablesintothe powergrid,weneedmoresourcestoprovideancillaryservic etohandlethevolatility. AsillustratedinFigure 1-1 ,differentresourcescanbeutilizedtoprovideancillary service.Traditionally,ancillaryserviceisprovidedbyt hegenerationside.Fastramping Figure1-1.Ancillaryservicesinterpretedasactuationin acontrolsystem. powergeneratorsrampupanddownaccordingtothegrid'snee d.However,tosatisfy thegrowingneedforancillaryservice,moregeneratorswil lhavetobebuilt,whichis expensiveandenvironmentallyunfriendly. Analternativeistoexplorethedemandsidewhichmayhavemu chlesscostinthe longrun.Recentresearchhasshownthataggregatedpopulat ionofthermostatically controlledloads(TCLs),suchasrefrigerators,aircondit ioners,andelectricwater 13

PAGE 14

heaters,canprovideancillaryservicewiththehelpofappr opriatecontrolalgorithms; seeforexample[ 2 – 6 ]andreferencestherein.Theseworksfocusonresidential loadssuchasA/Candrefrigeratorswithon/offcontrol.In[ 2 ],asimplelinearmodel isdevelopedforasingleload,thenasystemofcoupledFokke r-Planckequationsare constructedtodescribethedynamicsofapopulationofTCLs .In[ 3 ],adiscretetime LTIsystemisproposedtoapproximatethedynamicsofapopul ation.Thedead-band oftheTCLisdividedintobins,andthestatesofthesystemar ethenumbersofloads ineachbin.Controllawsforswitchingloadsaredevelopedb asedontheapproximated LTImodeltomaketheTCLpopulationtrackapowerreferences ignal.In[ 5 ],asimplied equivalentthermalparameters(ETP)modelisusedforasing leTCLandthecontrol actionofwhethertoswitchaloadisdeterminedbythestatus anddynamicsofthe particularload,insteadofthatofthepopulation.In[ 4 ],domesticheatpumpsare controlledbyanaggregatortoprovideancillaryservice;s imulationsareperformed basedondatafromaDanishdomesticheatpumpprojectandpow ergriddata. In[ 6 ],theset-pointofthethermostatiscontrolledtovarythep owerconsumptionof apopulationofTCLstoachieveloadfollowing. Otherdemandsidesourcesarealsostudiedtoprovideancill aryservice.Aeetof electricvehiclescanbeutilizedasaresourceforthegridb ymanipulatingthecharging anddischargingwhentheyareconnectedtothepowergrid;se eforexample[ 7 , 8 ]. Alargepopulationofpoolpumpscanalsoprovideancillarys ervicewithappropriate controloftheirruntime[ 9 ].Batteryenergystoragedevicesarealsogoodalternative sfor powergridancillaryservice;seeforexample[ 10 ]andreferencestherein. 1.2CommercialBuildingHAVCSystemsinProvidingAncillar yService Inthiswork,weexplorethepotentialofcommercialbuildin gHVACsystemsfor providingancillaryservice.Thereareseveraladvantages thatmakecommercialbuilding HVACsystemsgoodcandidates.Firstofall,commercialbuil dingsarelargeresources– theyaccountfor40%ofthetotalelectricityconsumptionin theU.S.andHVACsystems 14

PAGE 15

consumeabouthalfofit[ 1 ].Thealgorithmspresentedinthisworkareapplicableto variableairvolume(VAV)HVACsystems,whichserve30%ofal lcommercialbuilding oorspaceintheU.S.[ 1 ].Insuchsystems,thesupplyairowratecanbevaried continuouslybetweenalowandhighvalue.Thismakesthempa rticularlywell-suitedfor sophisticatedcontrol.Incontrasttotheon/offcontrolTC Lcommonlyusedinresidential buildings,wherealargenumberofloadsneedtobeaggregate dtoprovidecontinuously varyingancillaryservice,asinglecommercialbuildingHV ACsystemiscapableof trackingacontinuousreferencesignal.Moreover,manybui ldingsareequippedwith buildingautomationsystems(BASs),makingthetaskofimpl ementingancillaryservice controlinexpensive.Commercialbuildingsalsousuallyha vehighband-widthInternet connection,whichcanbeusedtocommunicatewiththeBAs.Fi nally,commercial buildingshavehighthermalinertia,whichcanbetranslate dtoeffectiveenergystorage muchlikealargebattery. Ancillaryservicescanbebroadlydividedintotwocategori es:(i)thoseused continuouslyduringnormaloperation;(ii)thoseusedinco ntingencysituations,such asfollowingthelossofagenerator.Thisworkfocusesonthe rsttype.Thebalancing authorities(BAs),suchasISOs/RTOs(IndependentSystemO perators/Regional TransmissionOrganizations),monitorthepowergrid.Asig nalthatindicatesthedesired powerdeviationfromthescheduledvalue,calledareacontr olerror(ACE),iscalculated bytheBAs[ 11 ].Referencesignalsarebroadcasttoancillaryservicepro viderssothat thetotalancillaryserviceprovidedtothepowergridtrack stheACE.Theindividual ancillaryserviceproviderisrequiredtovaryitspowergen eration/consumptionsothat thedeviationfromscheduledvaluetracksthereferencesig nalitreceives.Figure 1-2 showstheACEsignalforonedayfromPJMInterconnectionLLC (PJM),anISOinthe northeastoftheU.S. TheACEsignalhascomponentsoverawiderangeoffrequencie s,fromseconds tohours.Itisreasonabletodecomposethesignalintodiffe rentfrequencyrangesfor 15

PAGE 16

0 6 12 18 24 -1500 -1000 -500 0 500 1000 1500 2000 HoursMW Figure1-2.ACEsignalfromPJMon05/04/2009. differentancillaryserviceproviders.Forexample,PJMof ferstwotypesofreference signals–traditionalregulationsignal(RegA)anddynamic regulationsignal(RegD)– forserviceproviderstochoosefrom.RegAhaslowerfrequen cy,whichissuitablefor sourcesthatcanprovideslowlyvaryingservice;whileRegD hashigherfrequencywhich issuitableforsourcesthatcanprovidefastvaryingservic e.Alsonotethatthemeanof ACEsignalisclosetozero.Thismeansforademandsidesourc e,providingancillary servicewillnotchangeitstotalenergyconsumptionfromit snormaloperation. Aschematicofproposedarchitectureforprovidingancilla ryservicefromcommercial buildingsisshowninFigure 1-3 .Weassumeareferencesignalforpowerdeviation, P BA,istransmittedbytheBAtoalltheancillaryserviceprovid ers,includingsmart buildingsthatcanprovideancillaryservice,justasitisd onetodayforgenerators. DifferentBAshaveslightlydifferentpolicies,sothissig nalcanbeeitherACEor processedACE.Atabuilding,weassumethat P BAisband-passlteredtoobtaina referencesignal P r(inunitofpower)withmagnitudeandfrequencythatisappro priate forthatbuilding.Inthisworkweassumethateachbuildingi sfreetodesignitsown band-passltertoensurethatitsequipmentisnotdamageda nditsindoorclimateisnot 16

PAGE 17

Band-pass filter Ancillary Service Controller + Building HVAC system P BA P r P Figure1-3.Providingancillaryservicefromcommercialbu ildings. adverselyaffectedbytheancillaryserviceitprovides.Th issignal P risthereference forthepowerdeviationfromthebaselineofthebuilding'sH VACsystem.Thebaseline poweristhepowerthebuilding'sHVACsystemwouldhavecons umedifitwerenot providinganyancillaryservice.Theobjectivesofourcont rolalgorithmaretwo-fold:to varythepowerconsumptionoftheHVACsystemfromitsbaseli nesothatthedeviation Ptracks P r,whileatthesametimeensuringthattheindoorclimateisma intained withinacceptablelimits. Animportantconstraintfordemandsideancillaryservicep rovidersisthattheir primaryfunctionalitieshavetobemaintainedwhileprovid ingtheservice.Fora commercialbuildingHVACsystem,itistoensureacomfortab leindoorclimate.This includescomfortablezonetemperatureandacceptableindo orairquality(IAQ).Akey ideahereisfrequencyseparation.Commercialbuildingsha veverylargethermalinertia, sohigh-frequencyvariationintheHVACsystem'spowercons umptionwillnotchange theindoorclimatenoticeably.Recallthattheancillaryse rvicereferencesignaliszero mean.Thusitispossibletomaintainacomfortableindoorcl imatewhileproviding ancillaryservice. Itisimportanttonotethattheancillaryservicediscussed inthisworkisdifferent fromtraditionaldemandresponseservices,whichtypicall yfocusonreducingconsumption duringpeakhourstoavoidpowershortages.Commercialbuil dingshavebeenusedfor suchprograms;seeforexample[ 12 – 14 ].Thistypeofserviceusuallyhassomeadverse effectontheprimaryfunctionalityoftheload.Inthecaseo fcommercialbuildings,the indoorclimatemightbeaffected.Thus,itisusuallydeploy edforashortperiodoftime. Incontrast,theancillaryservicestudiedinthisworkhasb othupanddowncomponents 17

PAGE 18

inthereferencesignalandhaszeroenergychangeonaverage .Ourcontrolalgorithms donotadverselyaffecttheindoorclimateofthebuildingan dthuscanrunautomatically andcontinuously. Followingtheideathatdifferentsourcesprovideancillar ysignalindifferent frequencyranges,weutilizemultiplepiecesofequipmenti ntheHVACsystemfor ancillaryserviceindifferentfrequencyranges. Forhigh-frequencyancillaryservice,weconsiderthesupp lyairfanasthesource forpowervariation.Thefanspeediscontrolledthroughava riablefrequencydrive (VFD),whichhasfastdynamics.Thismakesitpossibletotra ckahigh-frequency signal.Therearethreeissues.First,theindoorclimateha stobemaintained.The buildingthermaldynamicsareusuallyveryslowduetothebu ilding'slargethermal capacitance.Thus,high-frequencyvariationinthesupply airfanspeedwillnot signicantlychangetheindoorclimate.Second,powercons umptionofotherequipment cannotbesignicantlyaffected.Althoughchangesinsuppl yairowratewillchangethe heatexchangeinthecoolingcoil,whichinturnaffectsthec hillerpower,thesedynamics arealsoslowerthanthatofthefan.Third,achallengeforde mandsidesourcesto provideancillaryserviceisevaluatingthebaselinepower consumption.Thisisimportant becausetheancillaryserviceademandsidesourceprovides isthepowerdeviation fromitsbaseline.Itishardtoevaluatethequalityofthese rviceasourceprovided withoutknowledgeofthebaseline.Theproblemisalsonottr ivialbecauseonlythetotal powerconsumptionisavailablewhenthesourceisproviding ancillaryservice[ 15 , 16 ]. Inthecaseofhigh-frequencyancillaryservice,thebuildi ngthermaldynamicsandits localclimatecontrolsystemaremuchslowerthanthevariat ioninthereferencesignal. Thisprovidesusanopportunitytoextractthebaselinebyl teringthetotalpower. Inthiswork,wearguethataslongasthereferenceisfasteno ughtobeabove theband-widthofthechillerandbuildingclimatecontrols ystem,supplyairfancanbe usedtoprovidehighqualityancillaryservice.Thishypoth esisisveriedexperimentally. 18

PAGE 19

Werstproposeafeed-forwardcontrolalgorithmwhichcomm andsthefanspeed.A simpliedmodeloftheHVACsystemisdevelopedforcontroll erdesign.Thealgorithm istestedinsimulation.Wethenimplementamorerobustfeed backcontrolalgorithmin arealcommercialbuilding–PughHallonUniversityofFlori dacampus.Thetotalpower measurementsarelow-passlteredtorecoverthebaselinep owerforimplementingthe controllerandevaluatingthequalityoftheancillaryserv iceweprovided.Bothsimulation andeldexperimentsshowgoodresults. Sincethisisarelativelynewresearcharea,notmuchlitera tureisavailableon thistopic.Inconcurrentwork[ 17 ],theauthorsproposedadirectandanindirect algorithmtovarysupplyairfanpowerforancillaryservice .Inthedirectmethod,fan speediscontrolled,whileintheindirectmethod,theroomt emperatureset-pointis controlled.Thealgorithmsaretestedinsimulation.Inthe simulation,thebaseline powerisassumedtobeconstantandknowntotheancillaryser vicecontroller.The effectoftheancillaryservicecontrollerandotherexogen ousinputsonthebaselineis notconsidered.In[ 18 ],thestaticpressureset-pointisusedascontrolinput.Al though partofthemodelisidentiedfromelddata,evaluationoft hecontrolstrategyisdone insimulation.Tothebestknowledgeoftheauthor,thiswork isthersttoimplement ancillaryservicecontrollerinarealcommercialbuilding .Therstpartofthisworkis reportedin[ 19 , 20 ]. Forlow-frequencyancillaryservice,thechillerpowerwil lbeaffectedbythechange ofsupplyairowrate.Thechillerusuallyconsumesmuchmor epowerthanthesupply airfans,whichprovidesusagoodopportunitytoextractmor eancillaryservicefrom theHVACsystem.Inthiswork,weproposeanindirectmethodt outilizethechiller.The supplyairowrateiscontrolledwhichaffectstheheatexch angeatthecoolingcoil, whichwewillcallcoolingcoilpowerinthesequel.Thisintu rnchangesthechillerpower consumption.Thispartofworkisreportedin[ 21 ]. 19

PAGE 20

In[ 22 ],theauthorsdevelopedadynamicalmodelforavariablespe edheat pumpandanalyzeditspotentialforprovidingancillaryser vice.Directcontrolonthe compressormotorisrequiredfortheproposedalgorithm,wh ichmaynotbeavailable. Thebaselineestimationproblemismoredifculttotacklei nthecaseoflow-frequency ancillaryservice,becausethefrequencyofthereferences ignalisnolongerhigherthan theband-widthofthebuildingclimatecontrolsystem.Thus thelow-passlteringidea isnotapplicable.Inthiswork,analgorithmisproposedtos olvethisproblem.Instead oftryingtoestimatethebaselinepowerconsumptionon-lin e,weschedulethebaseline aheadoftimebasedonloadandweatherprediction.Alowerle velcontrollerisdesigned fortrackingthescheduledbaselineandancillaryservicer eferencesignal.Thescheduler updatesthebaselineperiodicallybasedonindoorclimatem easurementstoensure thebuildingiskeptatcomfortableconditions.Thebenets ofthisnewalgorithmare two-fold:(i)thebaselineisclearlydenedwhichmakesimp lementingcontrollerand evaluatingtheserviceprovidedpossible;(ii)bysmartsch eduling,wecanminimizethe energyconsumptionoftheHVACsystem. 1.3BackgroundonVAVHVACSystems Inthissection,weprovideabriefintroductionofVAVHVACs ystem.Thetestbed wedevelopedfordatacollectionandeldexperimentsisals odiscussed. ThemainfunctionalityofanHVACsystemistoprovidetherma lcomfortand maintainIAQ.Coolingorheatingloadinabuildingisgenera tedbymanysources,for example,theoccupants,solarradiation,heatexchangethr oughbuildingenvelop,and etc.AnHVACsystemcirculatesairthroughthebuildingtopr ovidethecoolingorheating required,sothatacomfortableindoorclimatecanbemainta ined. InaVAVsystem,themassowrateofthesupplyaircanbevarie dcontinuously, incontrasttoaxedsupplyairowrateinaconstantairvolu me(CAV)system.This featureprovidesmoreexibilityinoperatingthebuilding HVACsystem.Themain componentsofaVAVsystemincludetheairhandingunits(AHU s),chillers,boilers,and 20

PAGE 21

nr r r r Figure1-4.SchematicofacommercialbuildingVAVHVACsyst em. VAVboxes.TheAHUsandVAVboxesconditionanddistributeth eairinthebuilding. Thechillersprovidechilledwatertocoolanddehumidifyth eair.Theboilersprovidehot watertoheattheair. AschematicofaVAVHVACsystemforacommercialbuildingiss hownin Figure 1-4 .Theairinsidethezonesiscollectedandrecirculatedasre turnair.Part ofthereturnairexitsthebuildingasexhaustair.Freshout doorairisbroughtinand mixedwiththerestofreturnair,whichensuresgoodIAQ.The mixedairthengoes throughacoolingcoilintheAHU,wheretheairiscooledandd ehumidied.Incoldand dryclimates,theremayalsobeareheatcoiltoincreasethet emperatureandhumidityof theair.TheconditionedairisthensenttotheterminalVAVb oxesbyasupplyairfanfor distributionintothezones.Thechilledwaterusedtocondi tiontheairinthecoolingcoil isproducedinachilleranddeliveredthroughpipes.Simila rly,aboilerproducesthehot waterusedtoheattheairintheheatingcoil. NormalbuildingoperationisperformedbyaBAS.Buildingop erationstatus,such asroomtemperatureandsupplyairowrate,aremeasuredbys ensors.Appropriate 21

PAGE 22

controlactionsarecomputedbytheBASbasedonthemeasurem ents.Therearemany localcontrollersfordifferentequipmentintheHVACsyste m.Herewebrieydiscussa fewthatarerelatedtothetopicofthiswork. Atthezonelevel,thereisanindoorclimatecontroller(see Figure 1-4 )whosetask istomaintainapre-determinedtemperatureset-pointT sp.ThezonetemperatureTismeasuredandusedtocomputecontrolactions.Asignaliss enttoasupplyairow controllerintheAHUtovarythesupplyairowrate.Dependi ngonthetypesofAHU, differentmechanismsareused.InsomesinglezoneAHUs,the desiredsupplyairow ratem d aiscomputeddirectlyandsenttothesupplyairowratecontr oller.Inothercases wherestaticpressureintheductismeasured,theindoorcli matecontrollerchanges thedamperpositionintheVAVboxes.Forinstance,whenthez oneistoohotandmore supplyairisdesired,somedampersopenwider,theductpres suredrops,andthe supplyairfanneedstoblowmoreairtomaintaintheductpres sureatapre-specied set-point.Althoughindirectly,changingthedamperposit ionisequivalenttosendinga desiredsupplyairowratesignaltothesupplyairowcontr oller.ThedesiredreheatQ rhisalsocomputedbytheindoorclimatecontroller.Itiscont rolledbymanipulatingthe valveintheheatingcoil. Indeterminingm d aandQ rh,theso-called“singlemaximum”logicisusuallyused, whichisshownisFigure 1-5 .Dependingonthezonetemperature,thecontrollerworks inthreemodes:cooling,heating,anddead-band.Asshownin Figure 1-5 ,theyare separatedbythecoolingset-pointandheatingset-point.I ncoolingmode,thereisno reheattothesupplyair,andthesupplyairowrateincrease saszonetemperature increases.Inheatingmode,theairowrateismaintainedat apredeterminedminimum value,whilethereheatincreasesaszonetemperaturedecre ases.Indead-bandmode, thereisnoreheatandthesupplyairowrateiskeptatminimu m.Inpractice,themode switchistriggeredafterthezonetemperaturestaysinthec orrespondingregionfora certainamountoftimetoavoidfastoscillation. 22

PAGE 23

Reheat Supply air flow rate Heating set point Cooling set point Heating mode Dead-band Cooling mode Temperature Figure1-5.Singlemaximumcontrollogicforindoorclimate control. AtAHUlevel,thesupplyairowcontrollerandcoolingcoilc ontrollerareimportant inthecontextofthiswork.Theformeroneistaskedtocontro lthesupplyairfanto achievethedesiredsupplyairowratem d acomputedbyindoorclimatecontroller (indirectlyinstaticpressurecase).Thelatterone–cooli ngcoilcontroller–maintains thedischargeairleavingthecoolingcoilatapre-determin edset-pointT sp dabyvarying thechilledwaterowrate.SinceT sp daisxed,changeinsupplyairowratechanges thecoolingcoilpower.Thischangewillbepickedupbythech illedwaterandaffectsthe chillerpowerconsumption.Moredetailaboutthecoolingco ilcontrollercanbefoundin Chapter 4 . Toperformeldexperimentsanddevelopmorerealisticmode lforacommercial building,wedevelopatestbedinPughHall,abuildinginthe UniversityofFlorida. Figure 1-6 showsapictureofPughHall.Finishedin2008,PughHallisa4 0,000sq.ft. facility,includingclassrooms,ofces,andateachingaud itoriumforlecturesandevents. IthasaVAVHVACsystemwiththreeAHUs.Manypre-installeds ensors,suchasroom temperature,VAVboxsupplyairtemperatureandmassowrat e,VAVdamperand reheatvalveposition,andetc.,areavailableinthebuildi ng. 23

PAGE 24

Figure1-6.OutsideviewofthetestbedPughHall. PughHallisequippedwithaSiemensAPOGEE TM BAS,whichsupportsBACnet protocolandhasafriendlygraphicalinterface.Althought heyareusefultools,itis noteasytoincorporatenewon-linealgorithmwiththem.Tor ecordon-linedataor implementmoderncontrolstrategy,weneedothermethods.W eutilizedacontrol systemsoftwarerecentlydevelopedbyDr.TimothyMiddelko op.Partialdetailsofthe softwarearedescribedin[ 23 ].Throughthesoftware,weareableto:(i)recordthe BASdataandaccesstheminrealtimeatafastrate;(ii)remot elycontrolthephysical devicesintheHVACsysteminrealtime.Moredetailaboutthe softwareandcontrol implementationisprovidedinChapter 3 . 1.4Organization Therestofthisworkisorganizedasfollow:inChapter 2 ,weproposeafeed-forward controllertoprovidehigh-frequencyancillaryserviceby utilizingthesupplyairfan;the controlleristestedinsimulation.InChapter 3 ,wedescribeafeedbackfancontroller andresultsofeldtestsinPughHall.Chillerisusedtoprov idelowerfrequencyancillary serviceinChapter 4 .Acontrollerisdesignedandtestedinsimulationassuming thebaselineisknown.InChapter 5 ,analgorithmisproposedtosolvethebaseline evaluationproblem.Finally,Chapter 6 concludesthisworkanddiscussesfuturework. 24

PAGE 25

CHAPTER2 HIGH-FREQUENCYA.S.THROUGHFEED-FORWARDCONTROL Inthischapter,wewilldescribeamethodofutilizingthesu pplyairfantoprovide ancillaryserviceinthehigh-frequencyrangewithfeed-fo rwardcontrol.Section 2.1 presentsourproposedcontrolarchitecture.InSection 2.3 ,wedevelopasimplied modelforcontroldesign.InSection 2.4 ,thesimulationstudythatexaminesthe performanceofthecontrollerispresented. 2.1ControlArchitecture Inthischapter,weconsiderthepowervariationinthesuppl yairfantobetheonly sourceforprovidingancillaryservice.RecallfromSectio n 1.2 ,ourgoalistovarythe powerconsumptionoftheHVACsystemfromitsbaselinesotha tthedeviation Ptracks thereferencesignal P r.Thesupplyairfanhasfastdynamics,whichmakesitpossibl e totrackahigh-frequencyreferencesignal.Weassumethatt hepowerconsumedby theboilersupplyinghotwaterforreheatingandthechiller providingchilledwaterto thecoolingcoiloftheAHUareindependentofthefanpower.I nmanyHVACsystems, theboilerusesnaturalgasinsteadofelectricity,whichju stiestherstassumption. Thesecondassumptionmayappearstrong–thepowerconsumed bythechillermay infactchangeifthefanspeedand,consequently,airowrat echanges.However, thedynamicalinterconnectionbetweentheAHUandthechill ercanbethoughtof asalow-passlterduetothelargemechanicalinertiaofthe chiller.Therefore,high frequencyvariationsinthefanpowerwillnotchangethepow erconsumptionofthe chiller.Thus,thedecouplingassumption–thatfanpowerva riationsdonotchange chillerpowerconsumption–holdsaslongasthevariationsa refastandofsmall magnitude.Inaddition,insomeHVACsystemschilledwateri ssuppliedfromawater storagetank.Forsuchsystems,thedecouplingassumptionh oldsnaturally. Weproposetoachievethecontrolobjectivesthroughafeedforwardcontroller.A sketchoftheproposedcontrolarchitectureisshowninFigu re 2-1 andamoredetailed 25

PAGE 26

versionisshowninFigure 2-2 .ThecontrolsignaluinFigure 2-1 couldbecommandfor anyactuatorsinthebuildingHVACsystemwhichcanchangeth epowerconsumption ofthesystem.Inthischapter,wechoosefanspeedcommandas ourcontrolvariableu, sinceitisthemostdirectcommandthesupplyairfanreceive sintheBAS. RecallfromSection 1.2 ,thesupplyairowcontrollersendscommandutothe fan.Ourancillaryservicecontrollerchangesthiscommand sothatthefan'spower consumptionischangedinsuchawaythatthedeviationincon sumption–bothpositive andnegative–tracksthereference P r.Thereference P ristransformedtoaancillary servicecommandu rbytheancillaryservicecontroller.Thiscommandisthenad dedto thebaselinefanspeedcommandu bproducedbythebuilding'sexistingfancontroller. SupposeP b ( t )isthebaselinepowerconsumptionofthefanduetothetherma lload onthebuilding,andP ( t )isthefanpowerconsumptionwiththeadditionalancillary servicecommand.Thenthedeviationinpowerconsumptionis P ( t ) = P ( t )P b ( t ). Clearly,changingthefanspeedfromitsbaselinevaluewill changetheairowthrough thebuilding.Wedesigntheancillaryservicecontrollerso that P ( t )tracks P r ( t )while causinglittlechangeinthebuilding'sindoorclimate. nr nnrr rnnr n r Figure2-1.Schematicofproposedfeed-forwardcontrolarc hitecture. Theband-widthofthereferencesignalsenttobuildingssho uldbechosenwiththe followingfactorstakenintoaccount.First,highfrequenc ycontentinresultingancillary servicecommandu r(seeFigure 2-2 )isdesirableuptoacertainlimit.Sincethethermal dynamicsofacommercialbuildinghavelow-passcharacteri sticsduetoitslargethermal capacitance,highfrequencychangesintheairowcauselit tlechangeinitsindoor temperature.Thestatementisalsotrueforindividualzone softhebuilding.Additionally, theVFDandfanmotorhavelargeband-widthsothathighfrequ encychangesinthe signalu rleadtonoticeablechangeinthefanspeedand,consequently ,fanpower. 26

PAGE 27

Figure2-2.Detailedschematicofproposedcontrolarchite cture. Botheffectsaredesirable,sincewewanttoaffectthefanpo werconsumptionwithout affectingthebuilding'stemperature.However,anextreme lyhighfrequencycontentinu risnotdesirableasitmightcausewearandtearofthefanmoto r.Likewise,u rshould nothaveverylowfrequencycontent.Otherwise,evenifthem agnitudeofu rissmall, itmaycausesignicantchangeincoolingprovidedoverlong periodsoftime,whichin turncanproduceanoticeablechangeinthetemperatureofth ebuilding.Furthermore,a largeenoughchangeinthetemperaturewillcausetheindoor climatecontrollerstotryto changeairowratetoreversethetemperaturechange.Ineff ect,thebuilding'sexisting controlsystemwilltrytorejectthedisturbancecausedbyu r.Beingafeedbackloop,this disturbancerejectionpropertyisalreadypresentinthebu ildingcontrolsystem.Ifthe controllersinthebuilding(fancontrolleraswellasthein doorclimatecontrollers)donot havehighband-width,theywouldnotrejecthighfrequencyd isturbance.Inshort,the 27

PAGE 28

frequencycontentofthe“disturbance”u r ( t )shouldlieinaparticularband[ f low , f high ], wherethegainoftheclosedlooptransferfunctionfromu rtofanspeedvissufciently largewhilethatofthetransferfunctionfromu rtotemperatureTissufcientlysmall. Recallthatweneedthefanpowerdeviation P ( t )totrack P r ( t ).Werstdiscuss howthefanpoweriscomputed.Thepowerconsumptionofafani sproportionaltothe cubeofitsspeed[ 24 ]:P ( t ) = c 1 ( v ( t )) 3 ,(2–1) wherec 1isaconstant,andvisthenormalizedfanspeedinpercentage.Forexample,100indicatesthatthefanisrunningatfullspeed,and50meansitisrunningathalf speed.Additionally,thesupplyairowrateisgivenbym a ( t ) = c 2 v ( t ),(2–2) wherec 2isaconstant.Letv bbethebaselinespeedwhichwouldbeobservedinthe absenceoftheancillaryservice.Weneedtocomputetheaddi tionalfanspeedv rso thatc 1 ( v b + v r ) 3c 1 ( v b ) 3 =P r ( t )(see( 2–1 )).Wecomputetheadditionalfanspeed usingarstorderTaylorseriesapproximation: P r ( t ) = 3 c 1 ( v b ( t )) 2 v r ( t ).Specically, theconverterblockinFigure 2-2 isastaticfunctionthatcomputesthecommandv raccordingtotheTaylorseriesapproximation:v r =P r ( t ) 3 c 1 ( v b ( t )) 2 .(2–3) Theresultingcommandv risthenpassedthroughapreltertoproducethe commandu r.ThefanspeedcommandthatissenttotheVFDisu b + u r.Theprelter isneededsothatthegainofthetransferfunctionfromv rtovintheband[ f low , f high ]iscloseto1,seeFigure 2-5 .Inthisgure,aswellasinsimulationstudies,wetake[ f low , f high ]tobe[1=600, 1=8]Hz.Thechoiceofthesevalueswillbeexplainedlater.The prelterisdesignedbycomputinganapproximateinverseof thetransferfunctionfrom 28

PAGE 29

u rtov,whichiscalculatedbasedonthediagraminFigure 2-2 usingthelinearized aggregatebuildingthermalmodel( 2–17 )describedinSection 2.3.2 . Toimplementthecontrolschemedescribedabove,anon-line estimateofthe baselinefanspeedv b ( t )isneeded.Theaccuracyoftheestimatehasastrongeffect onthetrackingperformanceofthereferencesignal.Weuset ime-scaleseparationto facilitateestimatingthebaselinefanspeed.Sincethebui ldingclimatecontrolsystem haslow-passcharacteristics,thebaselinefanspeedisalo wfrequencysignal.Aslong asthefrequencyofthereferenceliesabovetheband-widtho fthebuilding'sclimate controlsystem,thefanspeedcommandedbythebuilding'sco ntrolleru bwillnotreact tothechangestothefanspeedcommandedfromtheancillarys ervicecontrolleru r; andviceversa.Inthatcasethelow-frequencycontentofthe fanspeediswhatthe buildingwouldhavecommandedwithouttheancillaryservic e.Thatisthebaseline fanspeed.Wethereforeestimatethebaselinefanspeedbyl teringthemeasuredfan speedthroughalow-passlterwithcut-offfrequencylower thanf low.Figure 2-2 shows theoverallcontrolarchitecture.Theestimatedbaselinef anspeedisdenotedby^ v b,and theestimatedbaselinefanpoweristhereforegivenby^ P b = c 1 (^ v b ) 3,dueto( 2–1 ).The ancillaryservicecontrolleruses^ v b ( t )insteadofv b ( t )in( 2–3 ). 2.2PerformanceMetrics Therearetwoimportantaspectsinevaluatingthequalityof theancillaryservice providedbytheHVACsystem:referencetrackingperformanc eandeffectonindoor climate.2.2.1TrackingMetrics Fortherstcriterion,severalmeasuresofthetrackingerr orareconsidered.A commonlyusedstatisticalmeasureistherootmeansquare(R MS)value:E R =vuut 1 N NXi =1 e ( i ) 2(2–4) 29

PAGE 30

wheretheerroratsamplingtimeiis:e ( i ) :=P r ( i ) ^ P ( i ).Itismorereasonableto comparetheerrorstatisticswiththemagnitudeoftherefer encesignal,sowecompute thefollowingratio:r R = E R maxj P rj (2–5) WealsoconsiderthePJM'sperformancescore[ 25 ],whichtheyusetoevaluate thequalityofancillaryserviceasourceprovided.Thetota lscoreconsiststhreeparts: correlationscore,delayscore,andprecisionscore.Weden otethemasS c,S d,andS pcorrespondingly.S candS ddescribethedelaybetweenthereferencesignalandthe responseoftheresourceandtheyaredeterminedtogether.W edenethecorrelation coefcienttobe:R P () = cov (P r ( t ),P ( t +)) P r ( t ) P ( t +)(2–6) where isthestandarddeviationofthesignal.Thedelay isdenedasthetimeshift withwhichtheresponsehasthehighestcorrelationwiththe referencesignalwithina 5-minutewindow: = arg max 2[0, 5min] R P ()(2–7) ThescoresS candS darethendeterminedas:S c = R P ( ), S d = 5min 5min (2–8) TheprecisionscoreS pdescribestheaccuracyofthetrackingperformance.Itis calculatedbasedonthetrackingerrorasfollows:S p = 11 n nXi =1j P ( i ) P r ( i )j j P r aj (2–9) where P r aisthehourlyaverageofthereferencesignal,nisthenumberofsamples.The totalperformancescoreS tistheaverageofthethreeparts,i.e.,S t = 1 3 S c + 1 3 S d + 1 3 S p. 30

PAGE 31

2.2.2IndoorClimateMetrics Itisalsoimportanttomaintaintheindoorclimateatcomfor tablerangewhile providingancillaryservice.Inthisregard,weconsiderth eroomtemperatureand IAQ.Forquantifyingthethermaldiscomfortthecontroller maycause,wechoosethe temperatureviolationD Tdenedin[ 26 ]:D T ( t ) =8>>>><>>>>: T ( t ) + T low ,ifT ( t )T high 0,otherwise 9>>>>=>>>>;,(2–10) whereT low , T higharetheminimumandmaximumtemperatureallowedwhenthebui lding isoccupied.Figure 2-3 showsthecomfortenvelopefromASHRAEhandbook[ 27 ]. Accordingtothegureandotherspecicationsprovidedint hehandbook,wesetthe comfortableT lowto70FandT highto75F. Figure2-3.ThermalcomfortenvelopespeciedbyASHRAE. 31

PAGE 32

Thedischargeairisusuallysettoarelativelylowvaluesot hathumidityratioof thesupplyairissmalltokeepthezonescomfortable.Inmany commercialbuildings, thezonehumidityisnotmeasuredbytheBASandisnotusedind eterminingcontrol actions.Thusinthiswork,wedonotconsiderhumidityineva luatingtheindoorclimate. Also,thevariationinsupplyairowratehastobekeptsmall toavoidactuation saturationanddamagetotheequipment.Toquantifythevari ation,wedene: m a , avg = 1 N NXi =1m a ( i )m b a ( i ) m b a ( i ) (2–11) m a , max = maxm a ( i )m b a ( i ) m b a ( i ) (2–12) wherem aisthemeasuredsupplyairowrate,andm b aisthebaselinesupplyairow rate. 2.3SimpliedHVACModelforControllerDesign Thedynamicsofthecompleteclosedloopsystemofabuilding relatingzone temperaturestofanspeedcommandarequitecomplexduetoth einterconnectionof thezone-levelcontrolleddynamics,dynamicsofpressured istributionintheducts,and building-levelfancontroller.Forthepurposeofcontrold esign,weusesimpliedmodels ofsomeofthesecomponents,andtparameterstothosemodel sfromdatacollected fromPughHall.2.3.1FanandDuctModel Inpractice,thefanspeediscontrolledbytheVFDwhichalso acceleratesor deceleratesthefanmotorslowlyintheinterestofequipmen tlife.Becauseofthis rampingfeatureofVFD,weassumethetransferfunctionfrom thecontrolcommandto thefanspeedisofrst-order 1 dv ( t ) dt + v ( t ) = u ( t ),(2–13) 32

PAGE 33

where 1isthetimeconstant,andu ( t )isthefanspeedcommandsentbythefan controller.Thefanspeedcontrolleristypicallyaproport ional-integral(PI)controller. RecallfromSection 1.3 ,weassumethatthesupplyairowcontrollersensesthe desiredsupplyairowratem d adirectly.Afancontrollerchangesthefanspeedsothat theactualsupplyairowratetracksthisdesiredvalue.Thi sallowsustosidestepthe verychallengingproblemofmodelingtheductpressuredyna mics.Yet,theassumption isjustiedsincethatiswhatthefancontrolloopdoes,albe itindirectly.Takinginto accounttheairtransportationintheduct,hereweassumeth etransferfunctionfromthe desiredmassowratem d a ( t )tothedesiredfanspeedv d ( t )isofrstorder 2 dv d ( t ) dt + v d ( t ) = m d a ( t )=c 2 ,(2–14) where 2isthetimeconstant,andthedivisionofm d a ( t )byc 2isdueto( 2–2 ). Wenowestimatetheparametersc 1 , c 2and 1inthemodels( 2–1 ),( 2–2 )and( 2–13 ) fromdatacollectedfromPughHall.ThedatausedisfromAHU1inthebuildingwith a35KWratedfanmotorwhichsuppliesairto41zones.Usingleast -squaresmethod andarandomlychosen24hourlongdataset,theparametersar eestimatedtobec 1 = 3.3105KW,c 2 = 0.0964 kg=s,and 1 = 0.1 s.Figure 2-4 showsthemeasured versuspredicteddataforthethreevariables:fanpowercon sumptionP ( t ),airow ratem a ( t )andfanspeedv ( t ).Thetopplotdepictsmeasurementandpredictionof fanpowerP ( t )frommeasuredfanspeedv ( t )withestimatedc 1andmodel( 2–1 ).The middleplotshowscomparisonofmeasurementandprediction ofairowratem a ( t )frommeasuredfanspeedv ( t )withestimatedc 2andmodel( 2–2 ).Thebottomplot depictsmeasurementandpredictionoffanspeedv ( t )frommeasuredfaninputu ( t )withestimated 1andmodel( 2–13 ).Thedatausedforvalidationispickedfromanother randomlychosen24hourlongdataset.Weseethatthepredict edmodels( 2–1 ),( 2–2 ) and( 2–13 )withtheestimatedparametersaregoodtsfortheactualme asurements. Estimationofthetimeconstant 2in( 2–14 )ischallenging,sinceinthecurrentHVAC 33

PAGE 34

5 10 15 KW 4 6 8 kg/s 00 04 08 12 16 20 24 60 70 80 90 Time (h)% Measurement Prediction Fan Power Fan Speed Air Flow Rate Figure2-4.FanmodelvalidationwithPughHalldata. system,theductpressurerespondstothechangeofdesireda irowrateinaclosed loopmannerduetotheclosedloopfanspeedcontroller.Wema keaheuristicestimation thatthetimeconstant 2isapproximately10seconds.InSection 2.4.2 ,wewillconduct numericalexperimentstochecktherobustnessofthedesign edcontrollerbasedonthis heuristicallyestimatedvaluewhenthereisamodelplantmi smatch. 2.3.2BuildingThermalModel Inwhatfollows,asimpliedthermalmodelofthebuildingba sedontheaggregate buildingtemperatureT ( t ),whichcanbethoughtofastheaveragetemperatureofall zones,isdiscussed.Thissimplebutnon-linearthermalmod elrelatesthetotalmass owratetothebuildingtemperature. Considerthefollowingphysics-basedlumpedthermalmodel ofthebuilding([ 28 , 29 ])C dT dt = 1 R ( T oaT ) + C p m a ( T daT ) + Q ,(2–15) 34

PAGE 35

whereC , Rarerespectivelythethermalcapacitanceandthermalresis tanceofthe building,T oaistheoutsideairtemperature,C pisthespecicheatofair,m aisthesupply airowrate,Qistheheatgeneratedbyotherexogenoussources,suchasocc upants andsolarradiation. Toobtainatransferfunctionofthebuildingthermaldynami csfordesigningthe ancillaryservicecontroller,welinearizetheaggregatet hermaldynamics( 2–15 ).We dene~ Tand~ m aasthedeviationsofthebuildingtemperatureandsupplyair owrate fromtheirsteady-statevaluesT andma:T = T+ ~ T , m a = ma + ~ m a .(2–16) Substituting( 2–16 )into( 2–15 ),weobtainthelinearizedmodeloftheaggregatebuilding thermaldynamics:d ~ T dt =1 + C p Rma CR ~ T + c p ( T daT) C ~ m a ,(2–17) wherewehaveassumedT oaandQtobeconstantforthetimeunderconsideration. Weusethisassumptionandlinearizationonlyfordesignand testthedesignthrough simulationswithtimevaryingsignalsandhighdelitynonlinearmodelinSection 2.4.2 . Wenextaggregatetheeffectofalltheindoorclimatecontro llersintoonecontroller thatwecallthebuildingtemperaturecontroller,byimagin ingthatitcomputesthe desiredtotalmassowratem d a ( t )basedonthedifferencebetweenthedesired buildingtemperatureT spandactualbuildingtemperatureT ( t ),andthensignalsthe fancontrollertoprovidethismassowrate.Sinceeachofth eindoorclimatecontrollers incommercialbuildingsareusuallyPIcontrollers,wechoo sethebuildingtemperature controllertobeaPIcontrolleraswell.TheinputofthePIco ntrolleristhetemperature deviationfromitsdesiredvalue~ Tanditsoutputisthedesiredairowratem d a. ThemodelforthewholeHVACsystemisobtainedbyintegratin gtheindividual models.Nowwecanexaminethemagnituderesponsesoftwocru cialtransferfunctions, 35

PAGE 36

10 10 1 1 __ 8 0 -20 -40 -60 H v r T H v r v FrequencyMagnitude (dB)-3 0 __ 600 Frequency (Hz) Figure2-5.MagnitudefrequencyresponsecomparisonofH v r vandH v r T. whichareshowninFigure 2-5 :H v r visthetransferfunctionfromdisturbancetofan speedandH v r Tisthetransferfunctionfromdisturbancetotemperature.W eseefrom thegurethatwithinthepre-speciedband,H v r vhasarelativelyhighgainwhileH v r Thasanextremelylowgain.Thisconrmsthat,inproperfrequ encyrange,theancillary servicecontrollercanchangethefanspeedtoextractancil laryservicewithoutaffecting theroomtemperaturesignicantly. 2.4SimulationStudy 2.4.1HighFidelityHVACModel Althoughasimpliedthermalmodelisusedforcontroldesig npresentedin Section 2.3 ,weuseacomplexphysics-basedmodelinthesimulationstud iesaimed attestingthecontroller'sperformance.Thismodelisbrie ydescribednext;see[ 29 ]for details. Tocopewiththedifcultyofmodelingductpressuredynamic sthatcouplezone leveldynamicstothefandynamics,wemakethefollowingsim plication.Weassume thateachindoorclimatecontrollerdemandsacertainamoun tofairowrate,by generatingadesiredairowratecommandm d i ( t )inresponsetothemeasured temperaturedeviationfromthesetpoint:T d i ( t )T i ( t ).Thetotaldesiredsupplyair owrate,m d a ( t ),isthesumofthedesiredairowrateintoeachzonem d i ( t ):m d a ( t ) = nXi =1 m d i ( t ).(2–18) 36

PAGE 37

Thesignalm d a ( t )istheinputtocomputethedesiredfanspeedv d ( t )(see( 2–14 )).The actualtotalmassowrateism ( t ) = c 2 v ( t ),wherev ( t )istheactualfanspeed.Itis dividedamongthezonesinthesameproportionasthedesired airowrates:m i ( t ) =i m a ( t ),i = m d i=(Xj m d j ) .(2–19) Thebuilding'scontrolsystemeffectivelyperformsthisfu nction,althoughsignalingis performedthroughphysicalinteraction,insteadofthroug htheexchangeofelectronic signals. Thethermaldynamicmodelofamulti-zonebuildingisconstr uctedbyinterconnection ofRC-networkmodelsofindividualzonesandthecorrespond ingindoorclimate controllers.WeconsiderthefollowingRC-networkthermal modelforeachzonein thebuilding:C i dT i dt = T oaT i R i +Xj2Ni T ( i , j )T i R i , j + c p m i ( T daT ) + Q i ,(2–20)C ( i , j ) dT ( i , j ) dt = T iT ( i , j ) R ( i , j ) + T jT ( i , j ) R ( i , j ) .(2–21) Theaboveequationsaresimilarto( 2–15 ).Thedifferencesarethatthesecondterm ontherighthandsideof( 2–20 )representstheheatexchangebetweenzoneiandits surroundingwallsthatseparateitselffromneighboringzo nes,and( 2–21 )modelsthe heatexchangebetweenzonei,zonej,andthewallseparatingthem. The“singlemaximum”controllogicdescribedinChapter 1.3 isusedfortheindoor climatecontrollers.Recallthattherearethreeoperating modesinthiscontrolscheme: coolingmode,heatingmode,anddead-bandmode.Ifallthezo nesareintheheating ordead-bandmodesimultaneously,thesupplyfanwillbemai ntainedataminimum speedtosatisfytheventilationrequirement.However,thi sscenarioislesscommonin practice.Additionally,ourmethodchangesthefanspeedfa standwithsmallmagnitude. Wethereforeassumeallthezonesareinthecoolingmode.Int hismode,thereisno 37

PAGE 38

reheating,andthesupplyairowrateisvariedtomaintaint hedesiredtemperaturein thezone.TypicallyaPIcontrollerisusedthattakestemper aturetrackingerrorT d iT iasinputanddesiredairowratem d iasoutput. Thehighdelitymodelofamulti-zonebuilding'sthermaldy namicsisconstructed bycouplingthedynamicsofallthezonesandindoorclimatec ontrollers,withm i'sas controllableinputs,T oa , Q i , T daasexogenousinputs,andT i'sandm d i'sasoutputs.The commandm d a,computedusing( 2–18 ),isusedtocalculatethedesiredfanspeedv d ( t ), whichinturnservesasinputtothefancontroller,whoseout putisu b.Thetotalfan commandu b + u ristheinputtothefan,withoutputfanspeedv(whichalsodetermines thepowerconsumptionandmassowratethrough( 2–1 )and( 2–2 )).Themassow ratethrougheachzone,computedusing( 2–19 ),thenservesasinputstothebuilding thermaldynamics. Foroursimulationstudies,wecreateabuildingwith4storiesand44zones, whichresemblesAHU-1andthezonesitconditionsinPughHal l.Eachstoryhas11zonesconstructedbycuttingawayasectionofPughHall.The zonesservicedby eachoftheAHUsinPughHallarenotcontiguous,whichnecess itatessuchactitious construction.Figure 2-6 showsalayoutofthese11zones.TheHVACsystemofthis buildingconsistsofasingleAHUandindoorclimatecontrol lersforeachofitszones.We identifythemodelofeachofthese11zonesfromdatacollect edinPughHall.Model identicationconsistsofdeterminingtheRandC(resistance/capacitance)parameters inthemodel( 2–20 )-( 2–21 )forthezone.Theleast-squaresapproachdescribedin[ 29 ] isusedtotthemodelparameters.Datacollectedfromthezo nesduringnighttimeis usedformodelcalibrationtoreduceuncertaintyofsolarra diationandoccupantheat gains.TheoutsideairtemperatureT oaisobtainedfromhistoricaldata[ 30 ]. TheestimatedparametervaluesaresummarizedinTable 2-1 .Hereweonlyinclude thethermalcapacitancesandresistancescorrespondingto zone1and2.Figure 2-7 showsthemeasuredandpredictedtemperaturesforzone1,wherethepredictionsare 38

PAGE 39

Figure2-6.PughHallsectorusedinconstructingthehighd elitymodel. obtainedfromthecalibratedhigh-delitymodel( 2–20 )-( 2–21 ).Weseethatthemodel predictsthemeasuredtemperaturequitewell.Similarresu ltsareobtainedfortheother9zones. Table2-1.Estimatedthermalparametersfor2outof11zones ParameterEstimation(J= C)ParameterEstimation( C=W) C 1 6.210 5 R 1 0.0072 C (1,2) 3.810 5 R (1,2) 0.01 C (1,5) 2.1410 5 R (1,5) 0.004 C 2 5.710 5 R 2 0.0063 C (2,3) 3.810 5 R (2,3) 0.01 C (2,5) 2.410 5 R (2,5) 0.004 2.4.2SimulationResults Inthissection,wedescribesimulationexperimentswhicht esttheperformance oftheancillaryservicecontrollerdescribedinSection 2.1 fortrackingreferencesignal byvaryingthefanpower.ThebuildingmodeldescribedinSec tion 2.4.1 isusedfor thesimulationsreportedhere.Thereferencesignal P BAusedforconstructingthe reference P ristakenfroma4-hourlongsamplefromPJM[ 31 ]on05/04/2009.It isthenscaledsothatitsmagnitudeislessthanorequalto5KW–aconservative 39

PAGE 40

22 24 02 04 06 08 21 22 23 24 25 Hourso F Measurement Prediction Figure2-7.Zone1modelvalidationwithPughHalldata. estimateoftheancillaryservicecapacityoftheparticula rbuilding.Afth-order Butterworthlterwithpassband[ f low , f high ]isusedastheband-passlterwhile constructing P r.Thedesignofthepassband[ f low , f high ]willbediscussedsoon. Tounambiguouslydetermineperformanceofthecontrolsche me,weperform twosimulations.First,abenchmarksimulationiscarriedo utwiththeancillaryservice turnedoffsothatu r ( t )0.Thefanspeedisvariedonlybythebuilding'sclimate controlsystemtocopewiththetime-varyingthermalloads. Theobservedfanpower consumption-thetruebaseline-isdenotedbyP b ( t ).Second,anothersimulationis conductedwiththeancillaryserviceturnedonandalltheex ogenoussignals(heat gainsofthebuilding,outsidetemperature)beingidentica ltothoseinthebenchmark simulation.Theactualfanpowerdeviation, P ( t ),isdenedasthedifferencebetween thefanpowerconsumptionobservedinthesecondsimulation ,P ( t ),andthatinthe rst,P b ( t ).Inaddition,wedenetheestimatedfanpowerdeviation, ^ P ( t ),asthe differencebetweenthemeasuredfanpowerconsumptioninth esecondsimulationP ( t ), andtheon-lineestimateofthebaselinepower^ P b ( t )inthesamesimulation.Notethat^ P b ( t ) = c 1 (^ v b ( t )) 3and^ v b ( t )istheoutputofthelow-passltershowninFigure 2-2 . Thepassbandoftheband-passlterhaslargeeffectsonthet rackingperformance. ResultsoftwodifferentpassbandarepresentedhereinTabl e 2-2 andFigure 2-8 .As seeninthetable,referencesignalthatcanbesuccessfully trackedbytheproposed 40

PAGE 41

13 14 15 16 17 -5 0 5 HoursKW / P r / P / ^ P AGoodtrackingofareferencesignalwithband-width[ 1 600 , 1 8 ]Hz 13 14 15 16 17 -5 0 5 HoursKW / P r / P / ^ P BPoorertrackingofareferencesignalwithband-width[ 1 1200 , 1 600 ]Hz Figure2-8.Trackingperformanceofreferencesignalswith differentband-widths. Table2-2.Performancemetricsinfeed-forwardcontrolsim ulations. Passbandr RPJMperformancescore S c S d S p S t [1=600, 1=8]Hz0.0240.99710.9130.970[1=1200, 1=600]Hz0.0980.97810.6220.867 fanspeedcontrolmechanismisrestrictedinacertainbandwidththatisdeterminedby thebuildingclimatecontrolsystem.Figure 2-8 (A)showsifthereference P r ( t )has aband-widthhigherthan1=600Hz(correspondingtoperiodof10minutes),theactual fanpowerdeviation P ( t )cantrackthereferencesignalextremelywell.Additionall y, theestimatedfanpowerdeviation ^ P ( t )matchestheactualfanpowerdeviation P ( t )verywell.However,ifthereferencesignalcontainsfreque ncieslowerthan1=600Hz, theindoorclimatecontrollerscompensatefortheindoorte mperaturedeviationsin thezonesbymodifyingairsupplyrequirements,thusnullif yingthespeeddeviation commandoftheancillaryservice.Thisresultsinapoorertr ackingperformance,as evidencedbyFigure 2-8 (B),inwhichther Rvaluechangesfrom0.024toabout0.1andthePJMprecisionscoreS pdropsfrom0.913to0.622.Weestimatetheupper 41

PAGE 42

bandlimittobe1=8Hztoavoidstressonthemechanicalpartsofthesupplyfan.I n addition,sincetheACEdatafromPJMissampledevery4seconds,thedetectable frequencycontentinthisdataislimitedto1=8Hz.Thus,thepassbandoftheband-pass lterischosenas[1=600, 1=8]Hz;seeFigure 2-5 .Itisimportanttonotethatwehave neglectedcoolingcoilandchillerdynamics,whichhasatim econstanttypicallylarger than200seconds[ 32 ].Weexpectthatinpracticetheproposedcontrolarchitect urewill beabletosuccessfullytrackreferencesignalintheband[1=0 , 1=8],where 0 = 200(correspondingtoperiodofabout3minutes).Atfrequencieslowerthanthat,unmodeled dynamicsofchillers/coolingcoilsmayaffectperformance . Thebaseline,estimatedbaselineandactualfanspeedwitha ncillaryservice controlleraredepictedinFigure 2-9 (A).Weseethattheestimatedbaselinespeed matchesthetruebaselinefanspeedverywell.Thedeviation oftheactualfanspeed withancillaryservicecontrollerfrombaselinespeedissm all,morespecically, m a , avg = 2.0%and m a , max = 12.4%.Thisensuresthatthepressureintheductstaysnear designedvalues.Finally,Figure 2-9 (B)depictsthetemperaturedeviationsofthe individualzonesfromtheirsetpoints.Weobservethatthet emperaturedeviationsare verysmall–temperatureviolationD T = 0–whichisunlikelytobenoticedbythe occupants. InSection 2.4.1 ,wemadeaheuristicestimationofthetimeconstant 2in( 2–14 ) tobe10seconds,anddesignedtheancillaryservicebasedonthishe uristicestimation. Wenowtestthedesignedcontroller(assuming 210seconds)onsystemswhose actualtimeconstant 2variesbetween5secondsand40secondstocheckthe robustnessofourdesignedcontrollerwhenthereisamodelp lantmismatch.We seefromFigure 2-10 thatthedesignedcontrollerisrobusttouncertaintyonthe time constantestimation;ther Rvalueoftrackingforsystemwithdifferentvaluesoftime constantaresimilar,andallofthemaresmall. 42

PAGE 43

13 14 15 16 17 60 70 80 90 100 TimeFan speed (%) Baseline Measurement Estimation AFanspeedwithandwithoutancillaryservicecontroller. 13 14 15 16 17 -0.1 -0.05 0 0.05 0.1 Hourso F BZonetemperaturedeviationsfromset-point. Figure2-9.Fanspeedandzonetemperaturevariationsfromb aseline. 5 10 15 20 25 30 35 40 0.02 0.022 0.024 0.026 0.028 t 2 (seconds)r R Figure2-10.r Rvalueoftrackingwithplant-modelmismatch. 2.4.3PotentialofCommercialBuildingHVACSystemsintheU .S. Thesimulationresultsshowthatasingle35KWfancaneasilyprovideabout5KW ancillaryservicecapacity,whichisabout15%ofthetotalfanpower.InPughHallof UniversityofFlorida,therearetwootherAHUs,whosesuppl yfanmotorsare25KW and15KWrespectively.ThismeansPughHallbyitselfcouldprovid eabout11KW ancillaryservicecapacitytothegrid.Thereareabout5millioncommercialbuildingsin theU.S.,withacombinedoorspaceofapproximately72, 000millionsq.ft,ofwhich approximately30%oftheoorspaceisservedbyHVACsystemsthatareequippedw ith 43

PAGE 44

VFDs[ 1 ].Assumingfanpowerdensitypersq.ft.ofallthesebuildin gstobethesameas thatofPughHallwhichhasanareaof40, 000sq.ft.,thetotalancillaryservicereserves thatarepotentiallyavailablefromalltheVFD-equippedfa nsincommercialbuildingsin theU.S.areapproximately6.6GWinthefrequencyrange[1=(3min) , 1=(30sec)]. 44

PAGE 45

CHAPTER3 EXPERIMENTALEVALUATIONOFHIGH-FREQUENCYA.S.THROUGHFE EDBACK CONTROL Inthischapter,wewilldiscusstheimplementationoftheid eaofproviding high-frequencyancillaryservicefromsupplyairfandescr ibedinChapter 2 .Inpractice, thefeed-forwardcontrolstructureusedinChapter 2 mayleadtopoorperformancedue tomodelingerroranduncertainties.InSection 3.1 ,amorerobustfeedbackcontrol architectureisproposed.ModelsareidentiedfromPughHa lldataforcontrollerdesign inSection 3.2 .InSection 3.3 ,twocontrollersaredesignedfortwocontrolinputs.In Section 3.4 ,thecontrollersareimplementedinPughHall;theexperime ntalresultsare discussed. 3.1ControlArchitecture InthesimulationstudyinChapter 2 ,afeed-forwardcontrolalgorithmisusedto showcasethepotentialofutilizingthefantoprovideancil laryservice.Inthedesign procedure,theprelterisobtainedbyinvertingthetransf erfunctionfromthefanspeed tofanpower.However,inpractice,modelingerrorisunavoi dable,especiallyfora complexHVACsystem.Thustheperformanceofthecontroller canbedegradedina realbuilding.Experimentalresultconrmstheexpected. Toimprovethequalityoftheancillaryserviceprovided,af eedbackcontrolalgorithm willbedeployed.Figure 3-1 showstheproposedcontrolarchitecture.Thebuilding HVACandclimatecontrolsystemarestablebydesign.Adding thefeedbackloop, theclosedloopsystemcouldremainstablebyproperdesigno ftheancillaryservice controller.Asdiscussedbefore,therearemultipleactuat ionoptionsforuinFigure 3-1 . nr nnrr rnnr n r Figure3-1.Schematicofproposedfeedbackcontrolarchite cture. 45

PAGE 46

Thismeansthecontrolcommandcouldbeinjectedintosevera lplacesinthelocal controlloops.Inthischapter,wewillexploretwooptions-u 1(fanspeedcommand)andu 2(desiredsupplyairowrate)asshowninFigure 3-2 whichchangesthefanpower consumption.Theideaofusingu 1istheonediscussedinChapter 2 :theancillary nnrr r rr nnrnnrr rrnnnrrrnn Figure3-2.Twolocationstoinjectcontrolcommandinthelo calcontrolloop. servicecontrollersimplymodiesthefanspeedcommandu f.Ineffect,theancillary servicecontrolcommandenterstheexistingHVACcontrolsy stemasadisturbance (ignoringthefeedbackeffect).Foralowfrequencyreferen cesignal,thelocalfan controllerwillrejecttheancillaryservicecontrolcomma ndasdisturbance.Thisleadsto thesecondoption:u 2,whichisanadditiontothedesiredsupplyairowratem d a,orow rateset-point.Achangeinm d awillcauseachangeinthefanmotorpowerconsumption -albeitindirectly.Theindoorclimatecontrolloopisusua llylessaggressivethanthefan controlloop,andisthereforeunlikelytorejectthelow-fr equency“disturbance”u 2.In theexperimentsreportedhere,wevarytheairowratebycom mandingtheowrate set-point.InAHUswheretheairowrateiscontrolledindir ectlythroughstaticpressure set-point,controlcanbeexecutedbyvaryingthestaticpre ssureset-point. WewillcalltherstoptionASCFS(AncillaryServiceContro llerthroughFanSpeed command),andthesecondoptionASCAF(AncillaryServiceCo ntrollerthroughAirFlow set-point). 46

PAGE 47

Thedeviation Pisnotdirectlymeasurable;onlythemeasurementofpowerPis available.Toobtain P,wetakeadvantageofthefactthatreferencesignalhashigh er frequencythanthatofthebuildingclimatecontrolsystem. Thebaselinepowercan berecoveredbylow-passlteringP,whichtakesoutthehighfrequencycomponent introducedbyourancillaryservicecontroller.Then Pcanbeobtainbysubtractingthe baselinefromP.Thisisequivalenttoaddingahigh-passltershowninFigu re 3-2 .We denotethisestimatedpowerdeviationby ^ P. Theancillaryservicecontrollerdescribedheredoesnotov erridetheexisting HVACcontrolsystem,itmerelymodiesthecommandsintheHV ACsystem.Akey requirementinchoosingthecontrolinputsiseaseofimplem entation.Boththefan speedcommandandairowrateset-pointscanbemodiedthro ughsoftwareinAHU-2 ofPughHallthatisusedasourtestbed. 3.2SystemIdentication Designandanalysisoftheancillaryservicecontrollerreq uireamodelofhowthe commanduaffectsfanpower.InChapter 2 ,eachcomponentintheHVACsystemis modeledandintegratedtoformthefullmodel.However,dyna micsofHVACsystemsare highlyuncertainandinformationaboutsomeofthelocalcon trolloopsishardtoobtain. Fortunately,informationaboutallthecomponentsisnotne cessaryforcontrollerdesign. Inthissectionwedescribehowaninput-outputmodelistto dataobtainedatourtest bed. Itisknownthataphysics-basedmodelforHVACsystemdynami csisnon-linear. Forthepurposesofcontrol,itisfoundthatalinearinput-o utputmodelcanbettodata withinthefrequencyrangeofinterest.Theinputofthesyst emisthevariationinufrom itsbaselinevalue.Sincethereisnoancillaryservicecont rollerduringnormaloperation, theinputissimplyu.Theoutputofthesystemisthepowerdeviationfromitsbase line powerprole P.Sinceweconsidertwodifferentinputs,u 1andu 2,wewillidentifytwo input-outputmodels:u 1to P,whichwewillcallH 1,andu 2to P,whichwewillcallH 2. 47

PAGE 48

3.2.1TestBed TheeldexperimentsarecarriedoutinAHU-2ofPughHallonU niversityofFlorida campus.ThisAHUisdedicatedtoalargeauditoriuminthebui lding.Theauditorium is22 ft .high,hasaoorareaof6000 ft 2,andcanholdmorethan200occupants. Figure 3-3 showstheoorplanoftheauditorium.Figure 3-4 showsdatafroma24-hour periodduringnormaloperationcollectedfromAHU-2ofPugh Hall.NotethattheAHUis shutdownfrom11p.m.to6a.m.aspartofa“nighttimesetback ”strategyforreducing energyuse. Figure3-3.FloorplanoftheauditoriumservedbyAHU-2inPu ghHall. AsbrieydiscussedinSection 1.3 ,normaloperationiscontrolledbySiemens APOGEE TM BASinPughHall.Monitoringandcommandingcanbeachievedt hrough graphicaluserinterfaces(GUIs)ortheunderlyingPPCL[ 33 ]program.Thefollowing isanexcerptofPPCLcodeinPughHall'sBASsystem,whichspe ciesthataPID controllerisusedtomaintainthesupplyairowratetoitss et-point: 48

PAGE 49

0 6 12 18 24 0 5 10 15 Fan power (KW) 0 6 12 18 24 0 1000 2000 Fan speed (rpm) 0 6 12 18 24 0 50 100 Fan speed command (%) 0 6 12 18 24 70 72 74 76 Room temp (F) 0 6 12 18 24 0 2000 4000 6000 Time (hour)SA Flow (CFM) Hour Figure3-4.Datafromatypicaldayduringnormaloperationf romAHU-2inPughHall. 05230LOOP(128,"%A%AVG.SAIR-FLW","%A%SVFD-VO","%A%SA FL-STPT", 10,0.2,0,5,0,0,100,0)Here SAFL-STPT istheset-pointforthesupplyairowrate.Valuesofmany“p oints”, includingthefanspeedcommandandsupplyairowrateset-p oint,canbesetthrough PPCLcode.NewcontrollogiccanalsobecodedintotheBASvia PPCLcode. Inthiswork,weuseacontrolsoftwaredevelopedbyDr.Timot hyMiddelkoop[ 23 ] toimplementourcontrolalgorithminsteadofchangingthee xistingPPCLcode.One reasonisthattheexistingPPCLcodesaredesignedandteste dtomaintainthenormal operationofbuilding.Itisbenecialtobeabletogobackto thisstablecontrolsystem, incaseunexpectedeventsoccurduringtheexperiments.Ano therreasonisthatthe 49

PAGE 50

PPCLlanguageisarelativelylowlevelprogramminglanguag e,soitisnoteasytocode. Thesoftwaresystemweusesamplesallthebuilding“points” (sensor/actuator/set-point values)andmakethemavailableforreal-timecontrolandof f-lineanalysisinascalable andreliablemanner.Duetothemassivedatavolumesinvolve d,datacollectionoccurs onchangeofvalue(COV)eventstoavoidspikesinnetworktra fc.Thecontrolsoftware isdesignedtosimultaneouslysupportanumberofapplicati onsbyinteractingwith variousdatabases.Controlcommandsfromtheapplications arecommunicatedto thesystembyappendingarowtoatableinarelationaldataba se.Ascheduler,which checksforsuchupdateseverysecond,thencommunicatesthe senewcommandstothe equipmentcontrollersthroughBACnet[ 34 ].Becauseofitsarchitecture,thesoftware enablesbothopenloopoperationandclosedloopcontrolofa nyactuatorofthebuilding. MeasurementscollectedatCOVeventsmeanthatthesampling intervalforeach variableisdistinctandtimevarying.Foreaseofanalysis, inputsarecommandedat 4secondintervalsandoutputsareaveragedand/orinterpol atedtogeneratevalues sampledat4secondintervals.Thechoiceofthe4secondsamp lingintervalwas inspiredbythefactthatPJMbroadcastsitsreferencesigna lsevery4seconds. Itwasfoundthatmeasurementnoiseisnotnegligible.Figur e 3-5 showspower measurementsobtainedover10minuteswhilethefanspeedco ntrolcommand wassettoaconstantvalueof55%.Alsoshownisanestimateofthepowerspectral density,whichisconsistentwithwhitenoise.Thetopplots howsthetimedomainpower measurement;themiddleplotshowspowerspectraldensityo fthenoise;thebottomplot inthisgureisanestimateofthemarginaldistributionoft hemeasurementnoise,based onahistogramofobservations.3.2.2Sine-sweepExperiments Frequencydomainidenticationispreferabletottingali nearmodelwithout knowledgeofthemodelstructure.Duetothelargemeasureme ntnoise,thesine-sweep methodisusedforestimatingthefrequencyresponse[ 35 ]. 50

PAGE 51

0 100 200 300 400 500 600 2 2.2 2.4 Time (seconds)Power (KW) 0 20 40 60 80 100 120 -40 -20 0 Frequency (mHz)Power / Freq (dB / Hz) -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0 2 4 6 8 Noise (KW) Histogram pdf Figure3-5.Measurementnoisecharacterization. Sine-sweeptestsareconductedbymodifyingthefanspeed(o rmassowrate set-point)commandthroughthecontrolsoftwareinsuchawa ythatu 1(u 2)becomesa sinusoidalsignalwithagivenfrequency.Intheseexperime ntsthenominalfanpower isequalto2.5KW,andforeachfrequency,thenumberofdatap ointscollectedis approximately500. WerstconsiderthesystemH 1:fromu 1to P.Thefrequencyresponseestimates obtainedfromtheexperimentisshowninFigure 3-6 .Fromthegure,wesee thatthemagnitudeofthefrequencyresponsepeaksbetweent hefrequencyf2[1=(1min) , 1=(30sec)].Forlowerfrequency,theinputisrejectedbythelocalclim ate controller;forhigherfrequency,theresponsedecaysduet othedynamicsofthefanand motor.A“best-t”linearmodelobtainedfromthisexperime nthassecond-ordertransfer function,^ H 1 ( s ) = 0.0173 s + 1.7280106 s 2 + 0.0360 s + 0.0144(3–1) ItsfrequencyresponseisalsoshowninFigure 3-6 . 51

PAGE 52

10 -2 10 -1 10 0 -40 -20 0 Magnitude (dB) (KW / %) Experimental Fitted 10 -2 10 -1 10 0 -180 -90 0 90 Phase (degree)w (rad/s) H 1 Figure3-6.FrequencyresponseofH 1identiedfromsine-sweepexperiment. Similarly,sine-sweepexperimentsaredesignedtoidentif ythesystemH 2-fromu 2to P.ThefrequencyresponseofH 2estimatedfromthesine-sweepexperimentsis showninFigure 3-7 .Thetransferfunctionofthettedsecond-ordersystemis: 10 -2 10 -1 10 0 -100 -80 -60 -40 Magnitude (dB) (KW / CFM) Experimental Fitted 10 -2 10 -1 10 0 -180 -90 0 Phase (degree)w (rad/s) H 2 Figure3-7.FrequencyresponseofH 2identiedfromsine-sweepexperiment. ^ H 2 ( s ) = 1.809105 s 2 + 0.04455 s + 0.01205 ,(3–2) 52

PAGE 53

whoseBodeplotisalsoshowninFigure 3-7 .Thesystemhasanearlyconstantgainin therangef2[1=(10min) , 1=(1min)]. 3.3ControllerDesign Recallthefeedbackarchitectureforancillaryservicecon trollerinFigure 3-1 . Thegoalistoprovideancillaryservicetothegrid,whilere spectingindoorclimate constraints.Toensurethelatter,largevariationinthefa nspeedorairowrateisto beavoided.Suchvariationsareundesirablealsoduetothep ossibilityofequipment damageandviolationofindoorairqualitystandards[ 36 ].Inaddition,thefanspeed commandissubjecttomagnitudeconstraints,sinceittakes valuesbetween0and 100%. Forthesereasons,thefrequencyrangesforprovidingancil laryservicearechosen wherethesystemhaslargegain.Thefrequencyresponseiden tiedinSection 3.2 leads tothechoiceoff2[1=(1min) , 1=(30sec)]forASCFSandf2[1=(10min) , 1=(1min)]forASCAF. Thereferencesignalcannotbetooslowsincethechillerpow ermightbeaffected, asexplainedinSection 1.2 .Thetestbedwechooseisservicedbyanoff-sitechiller plantwithmassiveinertia,soweestimatedanyairowratev ariationuptotensof minuteswillnotaffectthechiller'spowerconsumption.Th erangeoffrequencyofthe referencesignalislimitedtohigherthan1=(10min). 3.3.1ASCFS Thebandoffrequenciesinwhichthecontrollerisrequiredt operformreference trackingisnarrow,f2[1=(1min) , 1=(30sec)],andthepassbandofthetransfer functionH 1isalsonarrow.Henceaproportionalcontrollerwasfoundto beadequatefor ASCFS. 53

PAGE 54

10 -2 10 -1 10 0 -20 -10 0 Magnitude (dB) (KW / KW) 10 -2 10 -1 10 0 0 Phase (degree)w (rad/s) H 1 ry 10 -2 10 -1 10 0 0 10 20 30 Magnitude (dB) (KW / KW) 10 -2 10 -1 10 0 0 Phase (degree)w (rad/s) H 1 ru Figure3-8.ClosedloopfrequencyresponseofASCFSwithpro portionalcontroller. LetH 1 ( j!)betheopen-loopfrequencyresponseobtainedfromthesinesweep experiments.Theclosedloopfrequencyresponseisgivenby :H ry 1 ( j!) = k p H 1 ( j!) 1 + k p H 1 ( j!)(3–3) wherek pistheproportionalgain.Anotherrelevantquantityisthef requencyresponse fromthereferencesignaltothecontrolcommand,whichdete rminestheamountof actuationrequired.Itisgivenby:H ru 1 ( j!) = k p 1 + k p H ( j!)(3–4) Thegainistunedsothat(i)H ry 1 ( j!)iscloseto1inthefrequencyofinterestforgood trackingperformance;(ii)H ry 1 ( j!)issmallinotherfrequencyfornoiseanddisturbance rejection;(iii)H ru 1 ( j!)issmallinthefrequencyofinteresttoavoidactuationsatu ration. Weassumethatforareferencesignalwithpeakmagnitude1KW,wecanvarythefan speedwithin10%ofitsnominalvalue.Thisisequivalentto jH ru 1 ( j!)j20 dBfor ! intherangewhereancillaryservicewillbeprovided.Thefr equencyresponsesofthe closedloopsystem,withk p = 15,areshowninFigure 3-8 ,whichsatisfythesecriteria. TheleftplotshowsH ry 1 ( j!);therightplotshowsH ru 1 ( j!);theverticallinesindicatethe frequencyrangeofinterest. 54

PAGE 55

10 -2 10 -1 10 0 -75 -50 -25 0 Magnitude (dB) (CFM / KW) 10 -2 10 -1 10 0 -270 -180 -90 0 Phase (degree)w (rad/s) H 2 ry 10 -2 10 -1 10 0 30 40 50 60 Magnitude (dB) (CFM / KW) 10 -2 10 -1 10 0 -90 0 Phase (degree)w (rad/s) H 2 ru Figure3-9.ClosedloopfrequencyresponseofASCAFwithlag compensator. 3.3.2ASCAF InthecaseofASCAF,thecriteriadescribedinlastsections tillapply.Inthiscase wechoosealagcompensator,whichhastheform:C lag ( s ) = k sz sp(3–5) Aftertuningthecontrollerparameters,weachievebothgoo dtrackingperformanceand smallactuation.Thepoleissettop =0.0017,zeroissettoz =0.6283,andgainis settok = 30.Theclosedloopfrequencyresponsewiththelagcompensato risshown inFigure 3-9 .TheleftplotshowsH ry 2 ( j!);therightplotshowsH ru 2 ( j!);theverticallines indicatethefrequencyrangeofinterest.Fromthegure,we seethatinourfrequency rangeofinterest,f2[1=(10min) , 1=(1min)],thetrackingisreasonable.Foractuation, weassumethatforareferencesignalofpeakmagnitude1KW,weareallowedto varythesupplyairowratewithin1000 CFMofitsnominalvalue.Thisisequivalentto jH ru 2 ( j!)j60 dB,whichisalsometinthefrequencyrangeofinterest. 3.3.3SimulationStudy Priortotestingonanactualbuilding,simulationsarecond uctedusingtheidentied models^ H 1and^ H 2.BasedonthediscussioninSection 3.2.1 ,measurementnoise(white andGaussian,withstandarddeviation = 0.1)isaddedtotheoutput. 55

PAGE 56

3.3.3.1ReferenceSignals Threetypesofreferencesignal P rareusedinthischapter.Twoofthemare obtainedbypassingtheACEsignalthroughaband-passlter ,inwhichtherawACE datawasfromPJMontheday05/04/2009.Theband-widtharech osentobef2[1=(1min) , 1=(30sec)]forASCFSandf2[1=(10min) , 1=(1min)]forASCAF. Afth-orderButterworthlterisusedtoobtain P rinthisfrequencyrange.Inthe sequel,wewillcalltherstoneFastACE,andthesecondSlow ACE.Theothertypeof referencesignalisRegD,whichisbroadcastbyPJMtoancill aryserviceparticipants.Its magnitudeisbetween-1and1.Notethatallreferencesignal sarescaledtohavinga maximummagnitudeof1KW.TherawACEsignalandthereferenc esignalsareshown inFigure 3-10 . 0 10 20 30 40 -800 -600 -400 -200 0 200 400 600 Time (minute)Power (MW) Raw ACE Filtered fast Filtered slow Figure3-10.TherawandlteredACEsignals. 3.3.3.2SimulationResults TheASCFSistestedusingtheFastACEsignal,andtheASCAFis testedwith SlowACEandRegD.Thereferencesignalsandtheresultingpo werdeviationswiththe twocontrollersareshowninFigure 3-11 and 3-12 .InFigure 3-11 ,thebottomplotisa 56

PAGE 57

Table3-1.Performancemetricsinfeedbackcontrolsimulat ions. SimulationReferenceS t r R ASCFSFastACE0.860.18ASCFS(Trueoutput)FastACE0.890.15ASCAFSlowACE0.790.25ASCAF(Trueoutput)SlowACE0.820.22ASCAFRegD0.850.18ASCAF(Trueoutput)RegD0.880.15 5-minuteclose-upofthetopplot.In 3-12 ,thetopplotshowsresultwithlteredACEas referencesignal;thebottomplotshowsresultwithRegDasr eferencesignal. 0 10 20 30 40 -2 0 2 Power (kw) Reference Simulated 30 31 32 33 34 35 -2 0 2 Time (minute)Power (kw) Figure3-11.SimulationresultofASCFSwithproportionalc ontroller. TheresultingperformancemetricsareshowninTable 3-1 .ThePJMperformance scoreS tareabovetherequiredthreshold0.75inallexperiments,wh ichindicates satisfactoryperformance.Insimulation,thetruepowerde viationwithoutthemeasurement noiseisalsoavailable.Theperformancemetricsrecalcula tedwiththetruevaluesare alsoshowninTable 3-1 . 3.4ExperimentalResults Inthissection,wewilldiscusstheeldtestresultsforthe ancillaryservice controllers.Eachexperimentisconductedovera40minutes timedurationusing 57

PAGE 58

0 10 20 30 40 -1 0 1 Power (KW) Reference Simulated 0 10 20 30 40 -1 0 1 Time (minute)Power (KW) ACE RegD Figure3-12.SimulationresultofASCAFwithlagcompensato r. AHU-2atPughHall.Fortyminutesisequaltothelengthofthe testrequiredbyPJM tomeettheirqualicationcriteria[ 25 ].Findingsfromexperimentssurveyedhereare consistentwithsimulationexperiments.3.4.1ASCFSExperiments TotesttheASCFS,weusetheband-passlteredACEsignalast hereference signal,withfrequencyoff2[1=(1min) , 1=(30sec)].Thetrackingperformance,room temperaturevariation,andsupplyairowrateduringthete stareshowninFigure 3-13 . Azoomed-in5-minutetrackingperformanceisalsoshowninF igure 3-14 . PerformancescoresareshowninTable 3-2 .ThePJMscore,0.74,ismarginally lowerthanthethreshold0.75.Thisscoreislowerthanthatinthesimulation,whichisnot unreasonablebecausetheestimatedmodel^ H 1isnotaveryaccuratedescriptionofthe realsystem. Also,theeffectofancillaryservicecontrollersonroomcl imateisshownin Table 3-3 .Thevariationinroomtemperatureissmall,whereD T = 0-meaning thereisnotemperatureviolationduringthetest.Thevaria tioninsupplyairowratecan becalculatedbysubtractingtheestimatedbaselinevaluef romthemeasureddata.As 58

PAGE 59

0 10 20 30 40 -2 0 2 Power (KW) Reference Measured 0 10 20 30 40 71 72 73 74 Room temp (F) 0 10 20 30 40 700 800 900 1000 1100 1200 Time (minute)Fan speed (rpm) Figure3-13.ResultsoftheASCFScontrollertestinAHU-2of PughHall. showninTable 3-3 ,theaveragevariationislessthan4%andthemaximumvariationis lessthan20%. 3.4.2ASCAFExperiments TotesttheASCAF,weusethesametwosetsofreferencesignal sthatareusedin thesimulationstudy:aband-passlteredACEsignalandPJM 'sRegD.Thetestresults areshowninFigure 3-15 .TheleftplotshowsresultwithlteredACEasreference signal;therightplotshowsresultwithRegDasreferencesi gnal;topplotsarethe referencetrackingperformance;middleplotsshowthetemp eratureevolution;bottom plotsarethesupplyairowrate.Theperformancemetricsar eshowninTable 3-2 . 59

PAGE 60

30 31 32 33 34 35 -1.5 -1 -0.5 0 0.5 1 1.5 Time (minute)Power (KW) Reference Measured Figure3-14.Aclose-up5-minutesliceofFigure 3-13 . 0 10 20 30 40 -1 0 1 Power (KW) Reference Measured 0 10 20 30 40 71.5 72 72.5 73 Room temp (F) 0 10 20 30 40 3500 4000 4500 5000 Time (minute)SA Flow (CFM) 0 10 20 30 40 72 73 74 Room temp (F) 0 10 20 30 40 3500 4000 4500 5000 Time (minute)SA Flow (CFM) 0 10 20 30 40 -1 -0.5 0 0.5 1 1.5 Power (KW) Reference Measured Figure3-15.FieldtestresultswithlagASCAF. Thecontrolsystem'sscoreexceedsPJM'sthresholdinbothc ases.Despitenoiseand modelingerror,theexperimentaldatashowgoodmatchwitht hesimulationresults. 60

PAGE 61

Also,asshowninTable 3-3 ,theeffectonroomclimateissmall:D Tiszero;the averagevariationislessthan4%andthemaximumvariationislessthan15%. Table3-2.Performancemetricsinexperiments. TestReferencer RPJMperformancescore S c S d S p S t ASCFSFastACE0.300.8010.410.74ASCAFSlowACE0.270.940.940.440.77ASCAFRegD0.190.940.950.600.83 Table3-3.Effectsofancillaryservicecontrollersonroom climate. TestD Tm a , avg(%) m a , max(%) ASCFS03.417.1ASCAF-ACE03.413.8ASCAF-RegD03.915.0 3.4.3FilteringtheMeasurementNoise ThelowprecisionscoreincaseoftheASCFSispartiallycaus edbythemeasurement noise.Theactualancillaryserviceprovidedtothepowergr idisbasedonthetruepower consumption,withoutthemeasurementnoise.Thus,itisrea sonabletorecoverthe noise-freepowerdeviationforperformanceevaluation.We dosobypassingthe measurementsthroughabi-directionallow-passlteroffline.Thelterusedisa5 thorderButterworthlterwithcutofffrequencyat1=(30 sec )forASCFSexperimentsand1=(60 sec )forASCAFexperiences.Thenoisesignalthatislteredouti sfoundtohave similarstatisticsasthenoisediscussedinSection 3.2.1 .Recalculatedwiththisltered powerdeviation,performancescoresimproved;seeTable 3-4 .Inparticular,allofthem exceedPJM's0.75requirementfortheprecisionscore. Table3-4.Performancemetricswithlteredmeasurements. Testr RPJMperformancescore S c S d S p S t ASCFS0.270.8310.490.77ASCAF-ACE0.240.980.940.500.81ASCAF-RegD0.140.990.940.730.89 61

PAGE 62

3.4.4EconomicPotential WhatistheeconomicvalueofPughHallorsimilarbuildingst oaBA?Anestimate oftherevenuethatcanbeearnedbyparticipatinginPJM'sma rketcanbecomputed basedontheirpubliclyavailablepolicymanuals[ 37 , 38 ]. Traditionally,frequencyancillaryserviceispaidbycapa city.Theserviceproviders whowanttoparticipateintheancillaryservicemarkethave topassqualication testesbeforedoingso.Oncequalied,theircompensationi sdeterminedbythe capacityassignedbytheISO,mostlyirrespectiveofhowwel lorpoorlytheirpower generation/consumptiontracksthereferencecommandbroa dcastbytheISO.However, thismethodologyhasbeencriticizedasbeingunfairtothos ewhoprovidemore accurateresponse.TheFERCorder(No.755[ 39 ])asksISOstotakeintoaccount theperformanceoftheserviceprovidedwhendeterminingth ecompensation. ThepaymentschemesadoptedbytheISOstoabidebyorder755a reinastate ofuxatthetimeofwriting.Inthiswork,weestimatetherev enuebasedonPJM's publiclyavailablepolicy[ 37 , 38 ].Therearetwomainstepsinthepaymentcalculation: determinetheclearingpriceandsettlingpayment.Inthecl earingpricedetermination process,eachparticipantsubmitsabidingpriceandthecap acityitcanprovideto PJM.PJMadjuststhebidsbytheprovidershistoricalperfor mance.PJMthenranks theadjustedpricesfromallbidders,andchoosesfromlowto high,untiltherequired capacitytobalancethegridissatised.Thepriceofthelas tpickedbidder(highestbid) istheclearingprice.Aftertheserviceisprovided,thepay mentissettledaccordingto theassignedcapacity,actualperformancescore,andthety peofreferencesignal.The actualpaymentcredithastwoparts:capacitypart-M candperformancepart-M p. Theyarecalculatedasfollows:M c =CAS PS RMCCPM p =CAS PS MR RMPCP (3–6) 62

PAGE 63

where CASisthecapacityassignedbyPJMtotheparticipant; PS istheperformance scoreoftheancillaryservice; RMCCP istheancillaryservicemarketcapacityclearing price; RMPCP istheancillaryservicemarketperformanceclearingprice ;and MR isthe mileageratioofthereferencesignal,whichiscomputedby:MR =PN i =1jReg ( i )Reg ( i1)j PN i =1jRegA ( i )RegA ( i1)j (3–7) whereReg ()istheactualreferencesignal,andRegA ()isthereferencesignalforthe traditionalancillaryserviceprovider,whichhaslowerfr equencythanRegDsignal. InthePughHallexperiments,AHU-2wasoperatingat2.5KW.R esultsshowthat AHU-2caneasilyprovide1KWcapacityofancillaryserviced uringitsoperationalhours: 6amto11pm.TherearetwootherAHUswhichnormallyoperates at7KWand5KW. Atthisoperatingcondition,PughHall'sHVACsystemcanpro vide8.7KWofcapacity. Sincethemarketclearingpricesaretimevarying,weusedat afrom2013publicly availablefromPJM'swebsite[ 40 ].Assumingthatthebuildingbidsfor8.7kWofcapacity ofRegDserviceforeachofitsoperationalhoursandeachoft hosebidsisaccepted,the yearlyrevenueforPughHalliscalculatedfrom( 3–6 )and( 3–7 )tobe$2135.Thetotal capacityoftheAHUsaremuchhigherthantheoperatingpoint usedinthecalculation, andmorerevenuecanbeproducediftheyoperateathigherpow er. 63

PAGE 64

CHAPTER4 LOW-FREQUENCYA.S. Inthischapter,wewillexpandthefrequencyoftheancillar yservicecanbe providedbyHVACsystemtolowerrange,byutilizingthechil lerpower.InChapter 2 and 3 ,thefrequencyofthereferencesignalishighenoughsothat theslowercooling coildynamicswillnotreacttoit.Ifwewanttoprovideancil laryserviceinlower frequencies,thechillerhastobeconsidered,sincethefre quencyofthereference signalfallsinthesamerangeasthecoolingcoildynamics.T hechillerusuallyconsumes muchmorepowerthanthesupplyairfans,whichprovidesusag oodopportunityto extractmoreancillaryservicefromtheHVACsystem. Whenthefrequencyofthereferencesignalbecomeslower,th eremaynotbeclear frequencyseparationbetweentheancillaryservicecontro lloopandthebuildingclimate controlloop.Thusthelow-passlterusedinChapter 2 and 3 toextractthebaseline on-lineisnolongerapplicable.Inthischapter,wedevelop amethodtoprovideancillary controllerfromchillerwiththeassumptionthatthebaseli neisknown.Thebaseline estimationproblemispostponedtoChapter 5 . Thefollowingsectionsareorganizedasfollow:inSection 4.1 ,wepresentthe architectureanddesignprocedurefortheproposedcontrol ler.Section 4.2 describethe dynamicalmodeloftheHVACsystem;itsmaindifferencefrom themodelinChapter 2 and 3 istheaugmentationofhumidityandcoolingcoildynamics.T hecontrolleristested insimulation,whichisdiscussedinSection 4.3 . 4.1ControlArchitecture SimilartothefancontrollerinChapter 2 and 3 ,ourobjectiveisstilltovarythe instantaneouspowerconsumptionoftheHVACsystemtotrack anancillaryservice referencesignalwhilemaintainingcomfortableindoorcli mate.Thedifferenceisthatwe considerthechillerpowerinthischapter. 64

PAGE 65

Coolingcoilistheplacewheretheaircirculatinginthebui ldingisconditioned.The airiscooledanddehumidiedbythechilledwaterwhentrave lingthroughthecooling coil.Thechilledwaterpicksuptheheat,andreturnstothec hiller.Powerisconsumed atthechillertocoolthereturnedchilledwaterandproduce thesupplychilledwaterto thecoolcoil.ThechillercanbelocatedfarawayfromtheAHU andisoftencontrolled separatelyinsteadofbeingpartofthebuildingHVACcontro lsystem.Thus,direct controlofthechillermaynotalwaysbeavailable.Inthiswo rk,weproposeanindirect methodtocontrolthechillerpower.Ourcontrolvariableis thesupplyairowrateinthe HVACsystem.Understandably,changesinsupplyairowrate willaffectthecooling coilpowersinceconditioningmoreairrequiresmorepower. Thevariationincoolingcoil powerisdeliveredtothechillerbythechilledwater.Since ittakestimeforthechilled watertoreturntothechiller,thereisatransportdelaybet weenthecoolingcoilpower andchillerpower.LetQ ccbethecoolingcoilpowerandP cbethepowerconsumptionof thechiller,weassumethatP c ( t ) = 1 C Q cc ( tt d ),(4–1) wheret disthedelayand Cisthechillerefciency. Thetwomajorchallengesinthisancillaryservicecontroll erdesignare:(i)complex non-lineardynamicsoftheHVACsystem;and(ii)transportd elayinchillerpower.We developthedynamicalmodelandlinearizethesystemindesi gnphasetotackle(i),and useSmithpredictor[ 41 ]andKalmanpredictor[ 42 ]todealwith(ii).Figure 4-1 shows aschematicrepresentationofthesignalowintheproposed controlarchitecture.It shouldbeemphasizedthattheproposedarchitecturedoesno treplacetheexisting buildingclimatecontroller.Itmerelymodiesthecommand edrateofairow. Thecontrolvariableischosentobethedesiredsupplyairo wrate,thesame asthatinASCAF(seeSection 3.1 ).Thereferencesignal P r ( t )isfedintoaKalman predictortopredictthefuturereferencesignal P r p =^ P r ( t + t d ),wheret disthe 65

PAGE 66

n rn nrr n r n rnr r n r rnr r r nrr!nnrr "#nrnnrr Figure4-1.Schematicoftheproposedcontrolarchitecture . transportdelayinthechillerpower.Let P ( t ) = P ( t )P b ( t )bethedeviationofthe measuredpowerconsumptionfromthebaselineP b ( t ).Ourancillaryservicecontroller computesthedesiredadditionalsupplyairowratem r a ( t )whichdrives ^ P r ( t ) P ( t )to0.Thesignalm r a ( t )isaddedtothedesiredsupplyairowratem d a ( t )determinedby thebuilding'sexistingindoorclimatecontroller.Thesum ,denotedbym c a ( t ),isthensent tothesupplyairowratecontroller.Incontrasttothefanp ower,whichisdetermined byfanspeed,thechillerpowerisalsoaffectedbythemixeda irtemperatureT ma ( t ), humidityratioW ma ( t ),andmassowratem a ( t ).Thesevariableswillbeincludedinour model.Moreaboutchillerpowercomputationispresentedin Section 4.2 . 4.1.1SmithPredictor First,weconsiderthedelayfreecase:P nd ( t ) = P f ( t ) + Q cc ( t ),whereP f ( t )is thefanpower.Wecombinedynamicsofallthecomponentsofth eHAVCsystemwithm d aasinputandP ndastheoutput.Theindividualcomponent'smodelispresente din Section 4.2 .Anequilibriumpoint( x, m da , w)ischosen,wherem daisthenominalmass owthatisobservedinnormaloperationandw isthenominalvalueofallexternal signalsincludingzonetemperaturesetpoints,outsidewea therconditions,andetc.The 66

PAGE 67

LTIapproximationisthenobtainedbylinearizationaround thisequilibriumpoint: _ x = Ax + Bm d a + EwP nd = Cx + Dm d a ,(4–2) where x = xx,m d a = m d am da ,w = ww,P nd = P ndPnd,andPndisthe equilibriumpowerconsumptionwhenm d a ( t )m da , w ( t )w . Standardloop-shapingmethodisusedtodesigntheancillar yservicecontroller.Let theplanttransferfunctionbeG ( s ),andthecompensatortransferfunctionbeC ( s ),we have:Y ( s ) = C ( s ) G ( s ) 1 + C ( s ) G ( s ) R ( s ) + G ( s ) 1 + C ( s ) G ( s ) D ( s )C ( s ) G ( s ) 1 + C ( s ) G ( s ) N ( s ),(4–3) whereR ( s )isthereferencesignal,D ( s )isthedisturbance,andN ( s )isthenoise.The sensitivityfunctionofthesystemisdenedas:S ( s ) = 1 1 + G ( s ) C ( s ) .(4–4) Equation( 4–3 )canbewrittenas:Y ( s ) = (1S ( s )) R ( s ) + G ( s ) S ( s ) D ( s )(1S ( s )) N ( s ).(4–5) Theancillaryservicecontrolleristhendesignedasacompe nsatorsothattheclosed loopsensitivityfunctionS ( j!)iscloseto0inthefrequencyofinterest,andcloseto1otherwise,sothatbothreferencetrackinganddisturbance rejectioncanbeachieved. OnceS ( s )isdetermined,thecontrollercanbecalculatedby:C ( s ) = 1S ( s ) G ( s ) S ( s ) .(4–6) Notethatambientenvironment,solarheatgain,andinterna lheatgainarealluncontrollable inputs,whichcanbeviewedasdisturbance,soitisimportan tforthecontrollertobe robusttodisturbances. 67

PAGE 68

Nowwewilladdressthetransportdelayshownin( 4–1 ).Toensurestabilityofthe closedloopinpresenceofthedelay,aSmithpredictorisuse d.ThedelaybetweenQ ccandP ccanbeestimatedfromtheowrateofchilledwaterandthegeo metryandlength ofthepipe,whichmakesSmithpredictorapplicable.TheSmi thpredictorisgivenas: C ( s ) = C ( s ) 1 + C ( s ) G ( s ) et d s ,(4–7) whereC ( s )isthecompensatordesignedinthedelayfreecase.Withthis compensator, wehavethetransferfunctionoftheclosedloopdelayedsyst emas: H ( s ) = H ( s ) et d s ,(4–8) whereH ( s )isthedesiredtransferfunctioninthedelay-freecase. 4.1.2KalmanPredictor Thesmithpredictordoesnotachieverealtimereferencetra ckingsincethedelay remainsinthepathfromthereferencetotheoutput.Ifthere ferencesignalcanbe predictedt dtimeunitsintothefuture,referencetrackingcanbeachiev edbylettingthe feedbackcontrolsystemwiththeSmithpredictoroperateon thepredictedreference.A Kalmanpredictorisusedtoobtainthepredictedreferences ignal^ t dtimeunitsintothe future,where^ t distheestimateddelayintheplant,andthispredictedrefer encewillbe usedbytheancillaryservicecontroller. TheKalmanpredictorusesadouble-integratormodel:_ x 1 ( t ) = x 2 ( t ) _ x 2 ( t ) =( t ),P r = x 1 ( t ) + v ( t )(4–9) where ( t ), v ( t )arenoiseterms.Theideabehindthismodelisthatsincether eference signalissmooth,itchangesatanapproximatelyconstantra teduringshorttime intervals.Notethatthismethodworkswellonlywhenthepre dictionperiodisrelatively 68

PAGE 69

shortconsideringthefrequencyofthereferencesignaltob eadvanced,wherethe approximationisreasonable. Thecontinuousdynamics( 4–9 )arerstdiscretized:x k +1 = Ax k +k(4–10) P r k = Cx k +k(4–11) where kand kareGaussianwhitenoises.ThenastandardKalmanpredictor isused tocalculatethereferencesignaln-stepsintothefuture[ 42 ]: ^ P r ( k + n ) = CA n ^ x ( kjk ),(4–12) where^ x ( kjk )isthestateestimatedattimekbytheKalmanlter. Theaccuracyofthispredictiondependsontheband-widthof theinputandthe lengthofdelay.Assumingon-sitechillerisused,thedelay isestimatedtobe30 sfor PughHallbasedonthegeometryofthepipeandowrateofthec hilledwater.We predictthefuturereferencewithdifferentdelays,andthe resultisshowninFigure 4-2 . Theerrorratiointhegureisdenedastheratioofpredicti onerrortorootmeansquare 0 0.2 0.4 0.6 0.8 1 -1.5 -1 -0.5 0 0.5 1 1.5 Time (h)Error ratio 30s 60s 90s Figure4-2.PredictedfuturereferencebyKalmanpredictor withdifferentdelaylengths. 69

PAGE 70

ofthereferencesignal.Fromthegure,weseethatinthe30 sdelaycase,theerrorratio islessthen20%.Thepredictorerrorgrowsasthedelayincrease. Inpractice,accurateknowledgeofthedelaymaynotbeavail able.Theuncertainty inthedelayintroducesamismatchbetweentheactualdelaya nddelayusedincontroller design.Westudytheeffectofthisuncertaintyonpredictio naccuracybyperforming simulationsinwhichthetruedelayis30 sbuttheKalmanpredictorusesadelayestimate of20 sand40 s,respectively.TheresultsareshowninFigure 4-3 .Theresultshowsthat delaymismatchincreasesthepredictionerror,asexpected ,butnotbyalot.Although 0 0.2 0.4 0.6 0.8 1 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 Time (h)Error ratio 30s (Exact) 20s 40s Figure4-3.Predictionerrorsofreferencesignalwithdela ymismatch. theerrorappearstobelargeatsomeinstances,itoccurswhe nthemagnitudeof thereferencesignalissmall.Theeffectofthiserroronref erencetrackingisfurther discussedinSection 4.3 ,whichshowsthattheresultingerrorinreferencepredicti onis acceptable. 4.2ModelingtheHVACSystemwithChiller Inthissection,wewilldescribethedynamicalmodelforthe HVACsystemwith chiller.Tosimplifytheproblem,wefocusonAHUdedicatedt oasinglezone.Itisnot uncommonthatonelargezone,suchasanauditoriumoraconfe renceroom,isserved 70

PAGE 71

byanindividualAHU.AHU-2inPughHallwillserveasaprotot ypeindevelopingthe model. FirstweaugmenttheHVACsystemwiththecoolingcoilcontro lloopandthechiller. RecallfromSection 1.3 thecoolingcoilcontrolsystemmaintainsthedischargeair temperatureatapre-speciedset-point,usually55 oF,byvaryingtheowrateofchilled waterpassingthroughthecoolingcoil.Theinlettemperatu reofthechilledwaterintothe coolingcoilisusuallyconstant,ataround44 oF.Thepowerconsumptionofthechilleris directlyaffectedbyvariationintheairowratesincecond itioningmoreairrequiresmore coolingenergy.Inthefollowingsections,wewilldiscusst hemodelofeachcomponent inmoredetail.4.2.1CoolingCoilDynamics Heatandmoistureareremovedfromairatthecoolingcoilatt heAHU.The dynamicsofacoolinganddehumidifyingcoilarecomplexwit hmanyunknown parameters[ 32 , 43 ].Inthiswork,weadopttheideaofaddingatimeconstantto asteadystatemodel–asdonein[ 44 ]–togetarstorderdynamicalmodelfor thecoolingcoil.Weusethesubscriptaforairside,wforwaterside,1forinlet conditions,and2foroutletconditions.Theinletandoutletwatermassowra te arethesame,i.e.,m w 1 = m w 2 = m w.Theinletandoutletairmassowratesare alsoassumedtobeequalsincethedifferenceduetowatervap orcondensationis small,i.e.,m a 1 = m a 2 = m a.Theinputsofthecoolingcoilaretheinletairandwater conditions:u cc = [ T a 1 , w a 1 , m a , T w 1 , m w ] T,outputsaretheoutletairandwaterconditions:y cc = [ T a 2 , w a 2 , T w 2 ] T. Supposethesteadystateinput-outputrelationsaregivenb yy cc = g ( u cc ), g :R5! R3,whichisdeterminedbytheparametersofthecoolingcoil.W ethenlinearize itaroundthedesignconditions,whicharedenotedbyuccandycc.Bydening~ u cc = 71

PAGE 72

u ccuccand~ y cc = y ccycc,weget~ y ccJ ~ u cc , J =@g @u ccjucc(4–13) Addingasingletimeconstanttothesteadystatemodel( 4–13 ),thecoolingcoil dynamicscanbewrittenas:~ y cc ( s ) = 1=cc s + 1=cc J ~ u cc ( s )(4–14) where ccisthetimeconstantoftheopen-loopcoolingcoildynamics. NotethattheJacobianJdenestheDCgainsofthetransferfunctionfrom~ u ccto~ y cc.ToestimateJinthecoolingcoilmodel,werstpickaparticularcoilmode l thatresemblesthecoilinAHU-2ofPughHall.Foragiveninle tconditions,theoutlet conditionsareobtainedfromDaikinMcQuayToolsSuite[ 45 ].TheJacobianisthen estimatednumerically.Theoutletconditionspredictedby themodelandthemeasured outletconditionsareshowninFigure 4-4 .Itcanbeseenfromthegurethatourmodel predictsbothT a 2andT w 2wellwithamaximumpredictionerrorlessthan2 o F.The predictedw a 2hasamaximumpredictionerrorlessthan1.5103.AlsonotethatT a 2andw a 2arenotusedinthepowerconsumptioncalculation,sotheywi llhavelesseffect ontrackingperformance. Inpractice,thecoolingcoilisunderclosedloopoperation .Theclosedloopcooling coilmodelisobtainedbyaddingaPIDcontrollerwhichcomma ndsthechilledwaterow ratetoachievedesiredconditionedairtemperature.Thecl osedloopcoolingcoilmodel isanLTIsystemwith2states,5inputsand3outputs. Thecoolingcoilpowercanbecalculatedfromthechangeinen thalpyofeither theairsideorthewaterside.Wewillchoosetousethewaters idesincethereisno humiditychangeinvolvedwhichmakesthecalculationeasie r.Thechilledwatergains 72

PAGE 73

0 2 4 6 8 10 12 52 54 56 58 T a2 (F) Measured Predicted 0 2 4 6 8 10 12 5 6 7 8 x 10 -3 w a2 0 2 4 6 8 10 12 58 60 62 64 66 68 Time (h)T w2 (F) Figure4-4.Coolingcoilmodelvalidationwithelddatacol lectedfromAHU-2inPugh Hall. heatQ cc ( t )fromtheair:Q cc ( t ) = m w ( t ) C pw ( T w 2 ( t )T w 1 ( t )).(4–15) 4.2.2ChillerDynamics Thechillerproducesthechilledwaterusedinthecoolingco il.Ignoringtheheat lossthroughthepiping,theheatpickedupbythechilledwat eratthecoolingcoilwillbe takenoutbythechiller.In[ 46 ],theauthorsusedarstordersystemtoapproximatethe 73

PAGE 74

dynamicsbetweenloadchangeinabuildingandthechillerpo wer.Thetimeconstant isabout200 s,whichisofthesamemagnitudeasthatofourcoolingcoilmod el.Inthis work,wedonotconsideraseparatedynamicalmodelforthech illerandassumethatthe coolingcoilmodelcapturesallthedynamicsfromthesupply airowtothechillerpower. Thechillerpowerisassumedtobethecoolingcoilpowerwith atimedelayandachiller efciencyfactor;see( 4–1 ). Weincludethetransportdelay–whichiscausedbythechille dwatertravelling betweenthecoolingcoilandthechiller,andcannotbeavoid ed–inourmodelbecause ofitsimportanceinthecontrolsystem.Withoutpropercont roldesign,thedelaycould leadtopoorperformanceoreveninstability.4.2.3ZoneThermalDynamics ThetemperaturedynamicsarepresentedinSection 2.4.1 .Wewillusethesame dynamicsdescribedin( 2–20 )and( 2–21 ). Wewillalsoaddthehumiditydynamics.Thehumiditychangei nthecirculated airhassignicanteffectonthecoolingcoilpower,whichma kesitimportantinthe discussionofthischapter.Dynamicsofthezonehumidityra tioisoneofmasstransfer, andisgovernedbyarstorderdifferentialequation:_ W = R g VP da Tn p!H 2 O + m a W sW 1 + W s (4–16) whereWandW sarethehumidityratioofroomairandsupplyair;Visthevolume oftheroom;n pisthenumberofpeopleintheroom; !H 2 Oistherateofwatervapor generatedbyapersonduetorespiration;R gandP daareconstantcoefcientsrelatedto airproperties.Moredetailscanbefoundin[ 47 ]. Thecoolingcoilpowerisalsoaffectedbytheconditionofmi xedairenteringthe coolingcoil.Certainamountofoutdoorairismixedwithexh austairrecirculatedfromthe zoneinamixingboxbeforeenteringtheAHUtomaintainaccep tableIAQ.Weinclude themixingboxinthezonedynamics.Let betheratioofoutdoorairowratetothe 74

PAGE 75

totalsupplyairowrate,thenwehave:T ma =T oa + (1 ) T , W ma =W oa + (1 ) W .(4–17) Theequations( 2–20 ),( 2–21 ),( 4–16 ),and( 4–17 )togetherdenethedynamical modelofthezone,asystemofcouplednon-lineardifferenti alequationswith3states(T,T w,W),8inputs(m a,w s,T s,Q s,Q i,Q rh,T oa,W oa),and4outputs(T,W,T ma,W ma). 4.2.4IndoorClimateControllerandAirowDynamics Again,the“singlemaximum”logicdescribedinSection 1.3 isusedforindoor climatecontroller.Inthischapter,wedealwithlow-frequ encyreferencesignal,sothe fastdynamicsinsidethefancontrolloopisnotimportant.T hus,weapproximatetransfer functionfromthedesiredsupplyowratem d atoactualsupplyairowratem abyarst ordersystem:m a ( s ) = 1=f s + 1=f m d a ( s ),(4–18) where fisthetimeconstantofthesystem.Thistimeconstantaggreg atesthedynamic effectoftheinertiaofthefananddynamicsofairowthroug hducts. 4.2.5PowerConsumption Weconneourselvestosystemswherereheatispoweredbyste am,sothatitdoes notcontributetotheelectricpower.Thetotalpowerconsum edP ( t )isthesumoffan powerandchillerpower:P ( t ) = P f ( t ) + P c ( t ).(4–19) Thefanpowerisrelatedtomassowrateofairas:P f ( t ) = c f m a ( t ) 3 ,(4–20) wherec fisaconstantcoefcientwhichcanbeestimatedfromdata;se eSection 2.3.1 . ThechillerpowerP cisgivenby( 4–1 )and( 4–15 ). 75

PAGE 76

4.3SimulationStudy 4.3.1SimulationSetup ThesubsystemsdescribedinSection 4.2 areintegratedtogetherandimplemented in Simulink .FielddataarecollectedfromAHU-2inPughHalltoestimate parametersin themodel.Zoneparametersareestimatedusingthemethodfr om[ 29 ];theyareshown inTable 4-1 .ThemeasuredzonetemperaturefromPughHallandthetemper ature predictedbythemodelareshowninFigure 4-5 . Table4-1.Zonedynamicsparameters ParameterValue C r 3.410 7 J=K C w 5.110 7 J=K R 1.3103 K=W 0 2 4 6 8 71 71.5 72 72.5 73 73.5 74 74.5 Time (h)Temp (F) Predicted Measured Figure4-5.Zonethermalmodelvalidation. OthermodelparametersarelistedinTable 4-2 .m a , miniscomputedfromtheoor areaandoccupantsofthezonetomeetASHRAEstandardforIAQ .Thevelocityof chilledwateriscomputedtobearound1.5 m=s.Weassumethatthepipeis45 mlong foranon-sitechiller,whichgivest d = 30 s.Thefanconstantc fiscalculatedfromc 1andc 2inSection 2.3.1 ,whicharecalibratedusingdatafromthesamebuilding.The otherparametersareobservedorestimatedfromrealdataco llectedfromPughHall. 76

PAGE 77

Table4-2.Modelparameters ParameterValueDescription m a , min 4000 CFMMinimumsupplyairowratem a , max 8000 CFMMaximumsupplyairowrateT CLG 73 o FZonecoolingsetpointT HTG 71 o FZoneheatingsetpoint 0.4OutdoorairratioT w 1 44 o FChilledwaterinlettemperatureT a 2, r 55 o FDischargeairtemperatureset-point f 30 sFan,damper,ducttimeconstant cc 100 sCoolingcoiltimeconstantt d 30 sDelayinchillerpowerc f 36.84 W=( kg=s )Fanpowercoefcient ThebaselinepowerP bisobtainedbysimulatingthesystemwithouttheancillarys ervice controller. Outdoorenvironmentandloadsinthezoneareexogenousinpu tsneedtobe speciedinthesimulation.Aninputproleforatypicalday isshowninFigure 4-6 . HourlyoutdoortemperatureT oaandhumidityW oainformationistakenfromthe website www.wunderground.com forGainesville,FLonthesamedayastheACE signal(2009-05-04).Twotypesofexogenousheatgainareco nsideredintermQ– thesolarloadQ sandinternalloadQ igeneratedbyoccupantsandequipment.Q sis computedbasedonthelocationandorientationofthebuildi ngandtheareaandthe shadingcoefcientofthewindows,usingthemethoddescrib edin[ 48 ].Q iisgenerated fromnormalscheduleofoccupantsandequipment.4.3.2Results ACEdatafromPJMon05/04/2009isusedas P BAinthesimulation.Afth-order Butterworthlterisusedtoobtain P r.Theltergainneedtobedeterminedrst. Theaverageairowvariation m a , avgdescribedinSection 2.2 isusedtoevaluate theoscillation.Wechoosethescalingfactortobe4105,inwhichcase m a , avg15%.Multiplepassbandsareusedinsimulationstoobtainrefer encesingleinvarious frequencyranges,whichwillbediscussedshortly. 77

PAGE 78

0 5 10 15 20 60 80 100 Temp (F) 0 5 10 15 20 0.005 0.01 0.015 Humidity ratio 0 5 10 15 20 0 5 10 x 10 4 Time (h)Load (W) Figure4-6.Exogenousinputsforatypicalday. Weuser RandPJMdescribedinSection 2.2 asperformancemetrics.The referencetrackingresultisshowninFigure 4-7 .Inthegure,twosimulationswith differencetimescaleinreferencesignalareshown:3to30m inutesforCase1and3to 60minutesforCase2.Thisisachievedbychangingtheband-p asslterfortheACE. Fromthegure,weseethatinbothcasesweareabletotrackth ereferencesignal withmaximumpowerofabout20KW.Thetemperaturedeviationfrombaselinecaseis largerinCase2thaninCase1,butbotharelessthan1 o F.Themaximumtemperature deviation T max ( o F )isshowninTable 4-3 .Theperformancescoreisevaluated,see alsoTable 4-3 .PJMrequiretheprovidertoreachascoreof0.75tobequaliedfor ancillaryservicemarket.Ourcontrollerperformswellabo vetherequirement. Wealsotestouralgorithmwithreferencesignalsofdiffere ntfrequencyranges byvaryingthepassbandoftheband-passlter.Thesimulati onresultsareshownin Table 4-3 .Thecontrollerdoesnotworkwellforhighfrequencyrange( 1to3minutes). 78

PAGE 79

10 11 12 13 14 -20 0 20 40 Power (KW) ~ P ~ P r 10 11 12 13 14 -20 0 20 40 Power (KW) ~ P ~ P r 0 5 10 15 20 70 75 80 T (F) Baseline Case 1 Case 2 10 11 12 13 14 2 4 6 8 Time (h)m a (kg/s) Baseline Case 1 Case 2 Case 1 Case 2 Figure4-7.Performanceoftheancillaryservicecontrolle r,andcomparisonwith baseline. Sincethetimeconstantofthecoolingcoilfallsinthatrang e,itcannotreactfast enoughtotrackthereferencesignal.Thetrackingperforma nceofthemiddleand lowfrequencyrangesarebothgood.However,thezonetemper aturevariationfromthe baselinecasegrowsasthereferencesignalbecomesslower. Thatisbecauseweare tryingtoover-conditionorunder-conditionthezoneforlo ngertimes,whichdrivesthe temperatureawayfromthedesiredvalue.Alsothecomfortvi olationandairowrate 79

PAGE 80

Table4-3.Performancewithreferencesignalsofdifferent frequency. Timescale(min)r RPJMperformancescore T max ( o F ) S c S d S p S t 1to30.430.670.85-0.950.500.243to100.040.9610.820.930.1010to300.060.9610.810.920.4430to600.060.950.950.800.901.693to30(Case1)0.050.9610.810.920.453to60(Case2)0.060.960.980.790.910.87 variationincase2areprovidedinTable 4-4 .Theaveragetemperatureviolation D Tshowsourcontrollerdoesnotcausemorediscomfortthanthe baselinecase. Table4-4.Comparisonofcomfortviolation. Simulation D Tm a , avg(%) Baseline 1.30 Withancillaryservicecontroller1.1621.2 Effectofchillerpowerdelaymismatchontrackingperforma nceisalsostudiedin simulation.Thetruedelayofthesystemis30 s.Table 4-5 showstheperformanceof thecontrollerwhenitisdesignedassumingthedelayis15 sand45 s,respectively.As Table4-5.Effectofdelaymismatchinchillerpower. TrueDelay(s)Delayindesign(s)r RPJMperformancescore S c S d S p S t 30150.050.9510.840.9330450.070.970.990.790.92 showninthetable,inbothcases,theeffectofchillerdelay mismatchonthedelayscoreS dissmall.Inthe45 scase,theprecisionscoreS pisworsecomparedtothecasewhen accuratedelayknowledgeisused.However,itisstillaccep table.Withthisgivenmodel andreferencesignal,ourcontrolstrategyisabletohandle50%ofdelaymismatchin bothdirections.Weobservedthatlargerdelaymismatchmay drivethesystemunstable. 4.3.3PotentialofCommercialBuildingHVACSystemsintheU .S. ThesimulationresultsshowthatAHU-2inPughHall,whichha saratedcooling capacityof95KW,couldprovide20KWofancillaryserviceinthefrequencyrange 80

PAGE 81

[1=(1hour) , 1=(3min)].PughHallhasanother2AHUsthathasacombined coolingcapacityof360KW.ThetotalancillaryservicecapacityofPughHall(40, 000sq.ft.)isestimatedtobeapproximately100KW.IntheU.S.,thetotaloorareaof commercialbuildingsisabout72, 000millionsq.ft.,about30%ofwhichisservedby VAVsystems[ 1 ].Assumingthatthecoolingpowerdensity(KWpersq.ft.)is similar amongthesebuildings,thecommercialbuildingsectorcoul dprovide60GWofancillary serviceserviceintheaforementionedfrequencyband. 81

PAGE 82

CHAPTER5 LOW-FREQUENCYA.S.WITHBASELINESCHEDULING Inthischapter,wewilldiscussanalgorithmtoprovidelowfrequencyancillary servicewithouttheassumptionusedinChapter 4 thatthebaselinepowerconsumption isknown.Thebaselineestimationproblemiscircumventedb yschedulingthepower consumptionaheadoftimebasedonloadpredictionobtained fromadata-drivenload model.Alowerlevelcontrollerisdesignedtotrackthebase lineandancillaryservice referencesignal. ThealgorithmisoutlinedinSection 5.1 anddetailsarepresentedinSection 5.2 .In Section 5.3 ,weanalyzethestabilityoftheproposedalgorithm.Thealg orithmistested insimulationanddiscussionsareprovidedinSection 5.4 . 5.1ControlAlgorithm Ourobjectiveremainsthesame:tovarytheinstantaneouspo werconsumption oftheHVACsystemtotrackanancillaryservicereferencesi gnalwhilemaintaining comfortableindoorclimate.Wewillstilluse P BAtodenotethesignalbroadcastfrom BAs,and P rtodenotethelteredreferencesignalforourHVACsystemto track.Inthe simulationstudyinChapter 4 ,weobservedthatthefanconsumeslessthan10%ofthe chillerpower,soforthepurposeofprovidingancillaryser vice,weignorethefanpower partandconsideronlythechillerpowerinthischapter. Inpractice,whenprovidingancillaryservicethemeasured chillerpowerP c ( t )isthe sumofthebaselinepowerandthepowerdeviation,i.e.,P c ( t ) = P b c ( t ) +P ( t )(5–1) Todetermine P ( t ),thebaselineP b c ( t )needstobedeterminedandsubtractedfromthe measuredpower.Estimatingthiscounter-factualbaseline powerconsumptionishighly non-trivial.Inthehigh-frequencyancillaryservicecase ,whichisdiscussedinChapter 2 and 3 ,alow-passlterisusedtogetthebaselinesincethebuildi ngclimatecontrol 82

PAGE 83

systemaremuchslowerthanthereferencesignal.However,t hisisnotapplicablein thelow-frequencycase,wherethefrequencyofreferencesi gnaloverlapswiththatof thebuildingclimatecontrolsystem.Toaddressthisproble m,we schedule thebaseline powerconsumptionaheadoftimebasedonthepredictedtherm alload,andcallitP r c. Thereferencecommandforthetotalpowerconsumptionisthe sumofthescheduled baselineandtheancillaryservicereferencesignal:P r = P r c +P r(5–2) Aschematicofthealgorithm,whichwewillcallBASA(buildi ngancillaryservice algorithm)inthesequel,isshowninFigure 5-1 .Ahighlevelschedulerschedulespower consumptionP r candQ r rhaheadoftime.Alowlevelcontroller,whichwewillcallPTC (powertrackingcontroller),variestheairowrateinorde rthatthepowerconsumptionof theHVACsystemP ctracksthereferenceP r c. Bandpass filter PTC Scheduler Building + + Figure5-1.SchematicillustrationoftheBASAcontroller. Theexistingbuildingclimatecontrolsystemthatisusedto varytheairowrateto maintainindoorclimateisoverriddenforpowertracking.I norderthattheindoorclimate isstillmaintained,thescheduledpowerconsumptionisper iodicallyupdated,usinga receding-horizonoptimizationbasedonloadpredictionan dperiodiccorrections,which isshownin 5-2 .Thestepsaredescribednext.Timeisdividedintoanumbero fslots 83

PAGE 84

ented Scheduled Figure5-2.Schematicillustrationofthereceding-horizo nupdate. forschedulingpurposesandforimplementationpurposes,w ithlength t Sand t I, respectively. 1. Atthebeginningofthek-thschedulingperiod TS k := ( k t I , k t I + t S ],predict thermalloadforthisperiod.Thepredictionalgorithmisde scribedinSection 5.2.1 . 2. Decidethedesiredbaselinepowerduring TS kbysolvinganoptimizationproblem: minimizecoolingandreheatingenergyduring TS kwhileensuringthatthermal comfortandIAQconstraintsaresatisedandactuationlimi tsarerespected.The desiredchillerpoweriscomputedfromthedesiredcoolingp ower. 3. Duringtheimplementationperiod TI k := ( k t I , ( k + 1) t I ],trackthereference chillerpower( 5–2 )byvaryingthesupplyairowrate.ThePTCisdesignedforth is purpose. 4. Attheendofimplementationperiod,updatetheloadestimat ionforthenext schedulingperiod TS k +1,andgobacktostep1. Notethattoensureacomfortableroomclimate,theschedule rcomputesreference signalsforbothchillerpowerandreheatpower.Sincewedon otconsiderthereheatas asourceforancillaryserviceinthiswork,thereisnoneedt odesignanewcontroller forthereheating.Instead,weassumethatthescheduledreh eatpowerwillbeprovided bylocalcontrollersintheBAS.Wewillfocusondesigningth ePTCforchillerpower tracking. 84

PAGE 85

5.2BaselineScheduling TheprimarytaskforaHVACsystemistotakecareofthetherma lloadina buildingtomaintaincomfortableindoorclimate.Thus,pre dictingtheloadisakey partofschedulingbaseline.5.2.1LoadPredictionModel Akeycomponentofthebaselineschedulingisamodelthatall owspredictingthe futureloadontheHVACsystembasedonambienttemperaturea ndtimeofweek. Themodelisborrowedfrom[ 49 ]:itdependsonambienttemperature-whichcanbe measured-andmanyotherquantities,suchasoccupantbehav iorandapplianceuse, thatarehardtomeasure.Thereforetimeofweekistakenasas urrogateforthose quantities. Beforedescribingthemodel,wehavetodenetheterm“therm alload”precisely anddescribehowtomeasureitsothatmeasurementscanbeuse dtotamodel. ConsiderthewholebuildingasasinglecapacitorC,andletTbetheindoortemperature. Thetemperaturedynamicsofthebuildingcanbeapproximate dby:C _ T ( t ) = Q rh ( t ) + Q oa ( t ) + Q s ( t ) + Q occ ( t )Q cc ( t )Q ea ( t )Q env ( t )(5–3) whereQ cc ( t )andQ rh ( t )aretheheatexchangeprovidedbytheHVACsystem(cooling andheatingcoils),andtheothertermsrepresentvarioushe attransferprocesses thattakeplacebetweenthebuildingandtheambient:Q oa ( t )istheheatenteringthe buildingthroughoutdoorair;Q occ ( t )istheheatenteringthebuildingfromoccupants andequipment;Q ea ( t )istheheatleavingthebuildingthroughexhaustair;andQ env ( t )istheheatleavingthebuildingthroughsurfaces,suchaswa llsandwindows. Weregroupthetermsin( 5–3 )as:C _ T ( t ) =Q cc ( t ) + Q rh ( t ) + Q l ( t )(5–4) 85

PAGE 86

wherethetermQ l ( t ) := Q oa ( t ) + Q s ( t ) + Q occ ( t )Q ea ( t )Q env ( t )(5–5) iscalledthe thermalload experiencedbytheHVACsystem.Itistherateofheatthe HVACsystemhastoremovefromthebuildinginordertomainta inindoortemperature ataconstantset-point,T sp,sothat_ T ( t ) = 0.Thus,assumingthattheconventional controllerisabletodoitsjobperfectly,thefollowingrel ationshipshouldholdduring timeswhentheconventionalHVACcontrollerisinuse:0 =Q cc ( t ) + Q rh ( t ) + Q l ( t )(5–6) ThisequationgivesusQ l ( t )frommeasurementsofQ cc ( t )andQ rh ( t ),whichare obtainedfromenergymetersinstalledinchilledwaterandh eatingsystems. Oncethequestionofgettingdatafortheloadisresolved,th enextquestionistting amodeltothisdata.Itismoreconvenienttousediscretetim esettingfortherestofthis section.LetNbethenumberoftimeslotsaweekisdividedinto.Forj = 1, 2,, N,the proposedmodelforthethermalloadis:Q j l =j + nXi =1i B i ( T j oa )(5–7) wherenisthenumberoftemperaturecomponentswhichwillbediscus sedsoon. Theterm jistheloadcorrespondingtotheoccupancyactivityattimej.Thesecond termontherighthandsidedescribestheloadduetoweather. Loadtendstochange differentlyindifferenttemperaturerange.Forexample,w henthetemperatureisvery highorverylow,theoccupantstendtostaylongerinside,wh ichincreasestheloadina non-linearfashionwithrespecttothetemperature.Tocapt urethisnon-lineareffect,we useapiece-wiselinearfunction.ThetemperatureT j oaisdividedintonbins.ThefunctionB i ( T j oa )returnsthetemperaturecomponentinthei-thbin.Eachcomponentisassigned acoefcient i.Thetotalloadduetoweatheristhesumof i B i ( T j oa ).Morerigorously, 86

PAGE 87

letb i , i = 1, 2,, n1betheboundsofthetemperaturecomponents,wehave:B 1 ( T j oa ) =8>><>>:T j oaifT j oab 1 b 1ifT j oa>b 1 ,andfori = 2,, n1, B i ( T j oa ) =8>>>>>><>>>>>>:0ifT j oab i1 T j oab i1ifb i1b i ,andB n ( T j oa ) =8>><>>:0ifT j oab n1 T j oab iifT j oa>b n1(5–8) AnexampleofcomputingB i ( T j oa )isgiveninTable 5-1 ,wheren = 5andb 1 = 60, b 2 = 70, b 3 = 80, b 4 = 90. Table5-1.ExampleofcomputingB i ( T j oa ). T j oa B 1 ( T j oa ) B 2 ( T j oa ) B 3 ( T j oa ) B 4 ( T j oa ) B 5 ( T j oa ) 5555000075601050095601010105 From( 5–7 ),weseethatgivenT j oa,Q j lislinearin jand i.Thus,wecan write( 5–7 )inregressionform:Q j l =T j (5–9) where = [1 ,,N ,1 ,,n ] Tisthevectorofparameterstobedetermined,and T jistheregressor.LetY =266664Q 1 l...Q N l377775, =266664 T 1... T N377775 87

PAGE 88

whereYandareknownfrommeasurements.ThenwehaveY = (5–10) Theparameters canbeestimatedbystandardleast-squaresmethod:^= ( T )1 T Y(5–11) Withthismodel,thethermalloadatsomefuturetimet-forwhichanambient temperatureprediction^ T oa ( t )isavailable(e.g.,fromweatherforecasts)-canbe predicted:^ Q l ( t ) =T p ( t ) ^ (5–12) where T p ( t )iscomputedfrom^ T oa ( t ). 5.2.2BaselineSchedulingatt = 0Thebaselinepowerconsumptionisscheduledbysolvinganop timizationproblem: minimizeenergyconsumptionduringthek-thschedulingtimeinterval TS kwhile satisfyingconstraintssuchasthermalcomfort,ventilati onrequirements,andequipment saturationlimits.Thealgorithmforupdatingthebaseline atsubsequentscheduling intervalsbuildsonthealgorithmusedattherstschedulin gperiod TS 0 = [0 t S ],sowe rstdescribethatindetail. Thepowerconsumptionweconsiderintheoptimizationisthe sumofcoolingcoil powerandreheatpower,i.e.,Q cc ( t ) + Q rh ( t ).Thereheatpowerisassumedtobe controlleddirectly,whichmakesQ rh ( t )adecisionvariable.Thecoolingcoilpoweris controlledindirectlybycontrollingthemassowrateofai rthroughthecoolingcoil. Theproblemofminimizingtheenergyconsumptioncanbestat edasfollows,fort2TS 0 = [0, t S ]: minimizem ( t ), Q rh ( t )Z TS 0 ( Q cc ( t ) + Q rh ( t )) dt(5–13) 88

PAGE 89

subjectto Q cc ( t ) + Q rh ( t ) + ^ Q l ( t ) = 0(5–14)m ( t )2[ m lb , m ub ] Q rh ( t )2[ Q lb , Q ub ](5–15) wherem lbandm ubaretheboundsforsupplyairowrate;Q lbandQ ubarethebounds forreheatpower;^ Q l ( t )isthepredictionofloadfromtheloadpredictionmodel( 5–12 ). InSection 4.2.1 ,wediscussedthatcoolingcoilpowerQ cc ( t )canbecomputedfrom eitherthechilledwatersideortheairside.Hereintheopti mizationproblem,theair owratem ( t )isthevariableofinterest,sowecomputeQ cc ( t )fromtheairside,whereQ cc ( t )equalstheproductoftheairowratem ( t )andenthalpydifferencebetweenthe airstreambeforethecoolingcoil( mixedair )andafterthecoolingcoil( dischargeair ):Q cc ( t ) = m ( t )( h ma ( t )h da ( t ))(5–16) whereh ma ( t ) = C pa T ma ( t ) + W ma ( t )( h we + C pw T ma ( t )) h da ( t ) = C pa T da ( t ) + W da ( t )( h we + C pw T da ( t )) T ma ( t ) =( t ) T oa ( t ) + (1 ( t )) T ( t ) W ma ( t ) =( t ) W oa ( t ) + (1 ( t )) W ( t )(5–17) whereC paisthespecicheatcapacityofair,C pwisthespecicheatcapacityofwater, andh weisthelatentheatofvaporizationofwaterat0 o C.Weassumethat(i)thecooling coilcontrolloopmaintainsthedischargeairatitstempera tureset-point;(ii)dischargeair humidityis:W da ( t ) = W ma ( t ) + c w ( T daT ma ),wherec wiscalibratedfromelddata;(iii) outsideairratioiskeptatitsset-point,and(iv)thatther oomtemperatureandhumidity staysattheirset-points,andalltheset-pointsareconsta nts.Undertheseassumptions, 89

PAGE 90

thespecicenthalpiesdonotdependontheactuationm ( t ),sotheenergyminimization isequivalenttopowerminimizationateveryinstantoverth eschedulingperiod. Thebaselinepowerscheduleiscomputedbysolvingthefollo wingoptimization problem:m r ( t ), Q r rh ( t ) =argminm ( t ), Q rh ( t ) ( Q cc ( t ) + Q rh ( t ))(5–18) foreveryt2TS 0,subjecttotheconstraintsspeciedby( 5–14 )and( 5–15 ).Thisisa linearprogramintwodecisionvariables,m ( t )andQ rh ( t ),andcanbesolvedeasily. ThescheduledbaselinecoolingpowerQ r cc ( t )canbereadilyobtainedfrom( 5–16 )oncem r ( t )isdetermined. Theproblemcouldbeinfeasiblewhentheloadistoohighorto olowsuchthateven fullcapacitycannotsatisfytheloadconstraint.Inthesec ases,fullcapacitywillbeused, whichisthebesttheHVACsystemcoulddoinsuchsituations.5.2.3BaselineUpdate Anumberofassumptionsaremadeinobtainingthebaselinesc heduleatthe rsttimeperiod,suchasaccuracyofloadprediction,perfe cttrackingperformance oflocalcontrolloops,etc.Sincetheseassumptionswillbe violatedinpractice,the scheduledpowerwillnotleadtoanindoortemperatureexact lyequaltotheset-point. Insubsequentschedulingperiods,theseuncertaintiesare accountedforbyexamining howfartheactualspacetemperaturevariesfromtheset-poi nt,andaddingcorrection terms.Thegoalsofthebaselineupdateare:(i)togetabette restimateofthethermal load^ Q l ( t )giventhepastmeasurementsothatthepredictionforthenex tperiodcan beimprovedover( 5–3 ),and(ii)tocompensatefortheover/undercoolingthebuil ding experiencedduringthepastperiod. Thebaselineisupdatedatanyschedulingperiod TS k,k1,bysolvingthe optimizationproblem( 5–18 ),withonlyonedifference:theloadconstraint( 5–14 )is 90

PAGE 91

replacedby: Q cc ( t ) + Q rh ( t ) + ^ Q l ( t ) + Q k c 1 ( t ) + Q k c 2 ( t ) = 0, t2TS k(5–19) whereQ k c 1 ( t ), Q k c 2 ( t )aretwocorrectiontermsin TS k.Wenowdescribehowthese correctiontermsarecomputedandtherationalebehindthei rcomputation. Theuncertaintiesinthetemperaturedynamics( 5–4 )arecapturedbyadisturbance term, Q l ( t ),so( 5–4 )becomes:C _ T ( t ) =Q r cc ( t ) + Q r rh ( t ) + ^ Q l ( t ) +Q l ( t )(5–20) wheretheterm Q l ( t )alsocapturesanydeviationoftheactualcoolingandheatin g powerfromtheirscheduledvalues.SincethescheduledQ r cc ( t )andQ r rh ( t )haveto satisfytheconstraint( 5–19 ),wehave:C _ T ( t ) =Q k c 1 ( t )Q k c 2 ( t ) +Q l ( t )(5–21) Notethatfortherstschedulingperiod,thecorrectionter msare0. Atthebeginningofthek-thschedulingperiod,i.e.,att = k t Ifork = 1, 2, ...,the disturbanceterm_ T ( k t I )isestimatedfromthemeasurementsofTuptothattime.The derivativeofTcanbeestimatedbyanumberofmeanssuchasa“dirtyderivati ve”ora Kalmanlter[ 50 ];wedonotdescribethemindetailhere. TherstcorrectiontermischosenasQ k c 1 ( t ) :=r1 _ T ( k t I ) e ( tk t I ) , t>k t I(5–22) where r1and aredesignparameters.Therationalebehindthistermistha t r1 _ Testimatesloadpredictionerror.Weexpectthiserrortobeo naverage0,sotheactual loadshouldreturntotheestimated^ Q l ( t ),i.e., Q l ( t )shouldgotozeroforlarget. Thesecondcorrectiontermisdesignedtoremovethetotalex traheatthathas remainedinthebuildingattheendofthethelast(k-1-th)im plementationperiod,which 91

PAGE 92

issimplyE end = C ( T ( k t I )T sp )ThecorrectiontermischosenasQ k c 2 ( t ) :=8>><>>: r2 E end t2( k t I , k t I + 1 r2 ] 0 t2( k t I + 1 r2 , k t I + t S ](5–23) where r2isadesignparameterwhichindicateshowfastwewouldliket hetemperature tobedrivenbacktotheset-point.Ifwewishtoreturnthezon etemperaturetoits set-pointquicker,alarger r2shouldbeused. 5.2.4PTCDesign TheobjectiveofthePTCistomakethechillerpowertrackthe totalreference signalP r ( t )duringeveryimplementationperiod:t2TI k.ThePTCisverysimilarto theancillaryservicecontrollerdiscussedinSection 4.1 .Ithastwomainparts:aSmith predictortoachievereferencetrackingwithadelayandaKa lmanpredictortopredictthe futurereferencefortheSmithpredictortouse.Thearchite ctureofthePTCisshownin Figure 5-3 . an predi predi Building PTC Figure5-3.ArchitectureofthePTC. ThemodeldevelopedinSection 4.2 isusedincontrollerdesign.ThecompensatorC PTCintheSmithpredictorisdesignedsothattheclosedloopsys temH cl(fromP rto measuredchilledpowerconsumptionP c)hasgoodtrackingperformance. IncontrasttothereferencesignalinChapter 4 ,thereferencesignalP rhereis consistoftwoparts;see( 5–2 ).ThescheduledbaselineP r cisknownaheadoftime,so 92

PAGE 93

advancingitby^ t disstraightforward.Theancillaryservicereferencesigna l P r,however, isreceivedfromtheBAandisthereforenotknownaheadoftim e.TheKalmanpredictor discussedinSection 4.1 isusedtopredictthispartofreferencesignal. 5.3StabilityAnalysisofBASA ThebaselineupdateprovidesafeedbackmechanisminBASA.C onsiderthecase wherethereisnoupdateorancillaryservicesignal:theloa d^ Q l ( t )ispre-calculated from( 5–12 )andthereferencesignalisthencalculatedfrom( 5–18 ).Thiscontrol strategyworksinafeed-forwardfashion,wheretheinput(Q r cc ( t )andQ r rh ( t ))isdecided aheadoftimeanddonotdependontherealtimeroomtemperatu remeasurement. Nowconsiderthecasewithupdate.Thedynamicsafteradding thetwocorrecting termsisgivenin( 5–21 ).Forconvenienceofanalysis,weletx 1 ( t ) = T ( t )T sp,andx 2 ( t ) = _ T ( t ).Notethat_ x 1 ( t ) = x 2 ( t ). Considerthek-thimplementationperiodt2TI k = ( k t I , ( k + 1) t I ].Withthe denitionin( 5–22 )and( 5–23 ),wehave:Q k c 1 ( t ) =r1 x 2 ( k t I ) e ( tk t I )(5–24)Q k c 2 ( t ) =r2 Cx 1 ( k t I )(5–25) Weassume1 r2 t IsothatthesecondcorrectiontermQ k c 2 ( t )remainsconstantduring TI k.Thissimpliestheexpression.Inthecaseof1 r2< t I,Q k c 2 ( t )becomesastep functionwhichdropsto0att = k t I + 1 r2.Fromthedynamics( 5–21 ),x 2 ( t ) = _ T ( t ) = 1 C (Q k c 1 ( t )Q k c 2 ( t ) +Q l ( t ))(5–26) Wenowdenethediscretestates:x k 1 = x 1 ( k t I ) x k 2 = x 2 ( k t I )(5–27) 93

PAGE 94

Attimet = ( k + 1) t I,x k +1 1 = x k 1 +Z( k +1) t I k t I _ x 1 ( t ) dt = x k 1 +Z( k +1) t I k t I x 2 ( t ) dt = x k 1 +Z( k +1) t I k t I 1 C (Q k c 1 ( t )Q k c 2 ( t ) +Q l ( t )) dt = x k 1 + 1 CZ( k +1) t I k t I ( r1 x k 2 e ( tk t I ) r2 Cx k 1 +Q l ( t )) dt = (1 r2 t I ) x k 1 +r1 C( e k t I1) x k 2 +W k +1(5–28) where W k +1 := 1 CZ( k +1) t I k t IQ l ( t ) dt(5–29) Also,wehave:x k +1 2 = _ T (( k + 1) t I ) = 1 C (Q k c 1 (( k + 1) t I )Q k c 2 (( k + 1) t I ) +Q l (( k + 1) t I )) = r2 x k 1 r1 C e k t I x k 2 + 1 CQ k +1 l(5–30) Instatespaceform,wecanwrite( 5–28 )and( 5–30 )as: 264x k +1 1 x k +1 2375=2641 r2 t Ir1 C( e k t I1) r2r1 C e k t I375 264x k 1 x k 2375+264 W k +1 1 CQ k +1 l375 (5–31) Inoursimulationtheparametervaluesof r1, r2,and aregiveninTable 5-2 .Inthat case,theeigenvaluesofthestatematrixin( 5–31 )are0.52and0.04,whichshowsthe discretesystemisBIBOstable. 5.4SimulationStudy TheBASAisstudiedinsimulationsonthemodeldescribedinS ection 4.2 and 4.3 . 94

PAGE 95

5.4.1ReferenceSignals Inthissimulation,twotypesofreferencesignalsareuseda s P BA.Firstoneis theACEsignalfromPJMon05/04/2009,whichisband-passlt eredtoobtain P rbya5 th-orderButterworthlterwithapassbandoff2[1=(1hour), 1=(10minutes)].The gainoftheband-passlteris5.5106.ThesecondtypeofreferencesignalisRegA signalfromPJMon05/04/2014.ThissignalispreparedbyPJM forslowerancillary serviceproviders.TheRegAsignal-whosevaluesliebetwee n-1and1-isscaledbya factorof5KWtoobtainthereferencesignal P r.Notethatinbothcases,thereference signal P rhasamaximummagnitudeof5KW. 5.4.2BASA RecallthattheBASAhastwomaincomponents,abaselinesche dulerandaPTC, andthebaselineschedulerneedsthedata-drivenloadmodel ( 5–12 ).Historicaldata ofQ ccandQ rharecollectedfromsensorsinstalledinPughHall.Weatherd ataare collectedfrom www.wunderground.com .Twoweeksofdata(01/13/2014to01/26/2014) areusedinestimatingtheparametersofthemodel.TheBASAu sesthismodeland 24-hourregionalweatherforecastsfrom www.wunderground.com topredictthethermal loadonthebuildingusing( 5–12 ). Thepredictionfromtheloadmodeliscomparedwiththemeasu redloadforthe week(04/14/2014to04/20/2014).Figure 5-4 showsthemodelcalibrationandvalidation results.Notethatthevalidationandcalibrationdatasets arecollectedafewmonths apart.Seasonaldifferenceintheloadcontributestotheer rorseenintheprediction ofthevalidationdata.TheBASAneedstoberobusttoerrorsi nprediction,whichthe simulationresultspresentedinthenextsectionwillshow. Theotherparametersneededbythebaselineschedulerandup dater,andtheir valuesusedinthesimulation,areshowninTable 5-2 . ThecompensatorC PTCinthePTCloopisdesignedsothatthesensitivityfunction oftheclosedloop(m d atoQ cc)isS ( s ) = s s +0.1,whichmakesthetrackingerrorless 95

PAGE 96

0 24 48 72 96 120 144 168 -20 0 20 KW Measured Predicted 0 24 48 72 96 120 144 168 -20 0 20 40 HoursKW Figure5-4.Loadmodelcalibrationandvalidation.Table5-2.BASAcontrollerparameters ParameterValueParameterValue C810 8J/K r1 3.210 6J/K t I1hour r21/(1hour) t S24hoursT sp72o F than 3 dBwhen !<0.1 rad=s.ThemodelsofthecomponentsoftheHVACsystem describedinSection 4.1 areconnectedtogethertogetthefullmodel.Itislinearize d toobtainthetransferfunctionH m , cc ( s )forcontrollerdesign.ThecontrollerC PTCis calculatedfromS ( s )andH m , cc ( s ).Althoughdesignedwithalinearizedmodel,thePTC isappliedtothehighdelitynon-linearmodelinthesimula tions. 5.4.3Results Thecooling,reheating,andancillaryservicereferencesi gnalsareshownin Figure 5-5 .TheleftplotshowsthecasewithlteredACEasreferencesi gnal;the rightplotshowsthecasewithRegAasreferencesignal.P r candQ r rharecalculated andupdatedon-line.TrackingperformanceisshowninFigur e 5-6 .Thebottomplot isa2-hourclose-up;theleftplotsshowresultswithltere dACEasreferencesignal; therightplotsshowresultswithRegAasreferencesignal.T heeffectofthecontroller 96

PAGE 97

onroomclimateisshowninFigure 5-7 .Theredhorizontallinesintopgureindicate comfortabletemperaturerange;theblackhorizontallinei ntopgureindicatesthe temperatureset-point;theleftplotsshowresultswithlt eredACEasreferencesignal; theleftplotsshowresultswithRegAasreferencesignal. 0 6 12 18 24 -10 0 10 20 30 40 HoursKW P c r Q rh r d P r 0 6 12 18 24 -10 0 10 20 30 40 HoursKW P c r Q rh r d P r Figure5-5.Scheduledbaselineandancillaryservicerefer encesignal. 0 6 12 18 24 -5 0 5 HoursKW 6 6.5 7 7.5 8 -5 0 5 HoursKW Power deviation with BASA d P r 0 6 12 18 24 -5 0 5 HoursKW 6 6.5 7 7.5 8 -5 0 5 HoursKWPower deviation with BASA d P r Figure5-6.Powerdeviationtrackingperformance. TheperformancemetricsdiscussedinSection 2.2 areevaluatedandshownin Table 5-3 .TheeffectofthecontrolleronroomclimateisshowninTabl e 5-4 .Wesee thattheourcontrollerhasaperformancescoreof0.935forlteredACEsignaland0.824forRegAsignal,whichexceedsPJM'srequirement0.75.Theroomtemperature remainedincomfortablerangeduringthewholesimulation, sothetemperatureviolation 97

PAGE 98

0 6 12 18 24 70 72 74 76 Hourso F Zone temperature 0 6 12 18 24 1 2 3 4 Hourskg/s SA flow with BASA Baseline SA flow 0 6 12 18 24 70 72 74 76 Hourso F Zone temperature 0 6 12 18 24 1 2 3 4 Hourskg/s SA flow with BASA Baseline SA flow Figure5-7.Roomtemperatureandsupplyairowrateduringt hesimulation. Table5-3.Performancemetricsinexperiments. P r ^ t d r RPJMperformancescore S c S d S p S t FilteredACE 300.0290.9980.9270.8810.93500.8630.2690.968-2.816-0.530200.0270.9980.9270.8900.938400.0300.9970.9130.8740.928 RegA 300.1060.9860.7400.7450.82401.0010.3150.873-1.899-0.2367200.1200.9730.7400.7250.813400.0990.9910.7530.7600.835 D Tis0.Theaveragevariationinsupplyairowrateisaround15%ofthescheduled baseline. Table5-4.Effectsofancillaryservicecontrolleronroomc limate. P r D Tm avg(%) FilteredACE015.7RegA015.2 Duringthecontrollerdesign,thetransportdelaybetweenc oolingcoilandchiller isneeded.However,thisinformationissometimeshardtoob tainaccurately.Thus,we performedaparametricstudyontheeffectofdelaymismatch .Werunthesimulation withmultipledelayvaluesinthedesignphasethatarediffe rentfromthetruedelay.The resultsareshowninTable 5-3 .Weseethatifweassumenodelayincontrollerdesign, 98

PAGE 99

i.e.,withnoSmithpredictor,theperformanceisverypoor. However,asshowninthe table,ourcontrollercanhandle10 sofdelaymismatchinbothdirectionverywellwhen theactualdelayis30 s. 99

PAGE 100

CHAPTER6 CONCLUSIONANDFUTUREWORK Inthischapter,wesummarizetheresultsofthisworkanddis cussfuturework directions.Wedemonstratehighqualityancillaryservice incertainfrequencyrangecan beextractedfromcommercialbuildingHVACsystemswithout noticeableeffectonthe indoorclimate. InChapter 2 and 3 ,weproposetousesupplyairfanaspowersourcefortracking high-frequencyreferencesignal,morespecically,inthe rangeof[1=(10min),1=(30sec)].ThesimulationstudyinChapter 2 showcasedthepotentialofthesupply airfaninprovidingancillaryservicebyafeed-forwardcon trolstrategy.InChapter 3 , weproposeamorerobustfeedbackcontrolarchitecture.Fie ldexperimentsare conductedinPughHall.Modelsfromcontrolinputtothepowe routputareidentied fromsine-sweepexperiments.Controllersarethendesigne dbasedonthesemodels. BothlteredACEandRegDfromPJMareusedasreferencesigna linexperiments. Resultsshowthatbothcanbetrackedwellandtheeffectofou rancillaryservice controlleronindoorclimateisinsignicant. Theexperimentalinvestigationiscarriedoutinasinglezo neAHU.Inmulti-zone cases,implementingtheASCAFmaynotbestraightforward.T hesupplyairowtoeach zoneiscontrolledbyindividualVAVbox.Thecommandtomode ratetheAHUsupply airowratewillbefoughtbackbytheindoorclimatecontrol lerattheVAVlevel.One ideaistoallocatethetotaldesiredAHUsupplyairowvaria tionintoeachindividualVAV box.Testingthealgorithminarealbuildingcouldhelpexte ndthealgorithmtomulti-zone cases. InChapter 4 and 5 ,chillerisconsideredaspowersourcefortrackinglow-fre quency referencesignal–intherangeof[1=(1hour) , 1=(10min)].Comparedtosupplyair fan,chillerhaslargercapacitybutslowerdynamics.Thisp rovideslargepotentialin thelow-frequencyrange.InChapter 4 ,wedevelopedacontrollerthatcommandsthe 100

PAGE 101

supplyairowrate,whichchangesthecoolingcoilpower.Th echiller,whichproduces thechilledwaterthatthecoolingcoilusestoconditionthe air,picksupthesechangesin power.Thechillerpowerisassumedtobethecoolingcoilpow erwithatransportdelay. Inpractice,chillerpowerisalsoaffectedbythedynamicso ftherefrigerationcyclesand equipment.Howaddingthechillerdynamicstothesystemwil laffecttheperformanceof ourcontrollersisaninterestingdirectionforfuturework . Theproposedcontrollerisdesignedforrelativelyshortde laybetweenthecooling coilandchiller,suchasthecasewhenrooftopchilleruniti sused.Whenthechilleris locatedfarawayfromthebuilding,thedelayincreases.The Kalmanpredictorusedin thisworkmayleadtopoorreferencepredictioninthepresen ceoflongdelay.Dealing withlongdelayisanotherresearchtopic. Theassumptionthatbaselinepowerconsumptionisknownisu sedinthe discussionofChapter 4 .InChapter 5 ,weproposeanalgorithminwhichthebaseline isscheduledaheadoftimebasedonloadandweatherpredicti on,sothatwedonot havetoestimatethebaselineon-line.Thescheduledbaseli neisupdatedperiodically accordingtomeasurementstoensurecomfortableindoorcli mate.Simulationresults showsatisfactorytrackingperformancecanbeachievedwhi letheindoorclimate remainsinacceptablerange.ThealgorithmsproposedinCha pter 4 and 5 aretestedin simulation.Experimentalinvestigationwillbeareasonab lenextstep. Inthiswork,theancillaryservicecontrollersforthehigh -andlow-frequency referencesignalsarestudiedandtestedseparately.Study inghowthesystemwill performwhentheyareimplementedtogetherisanotheravenu eoffuturework. Thesupplyairfanandchilleraretheonlypowersourcescons ideredinthiswork. TherearemanyotherequipmentintheHVACsystemsthatcanbe potentiallyused toprovideancillaryservice.Goodexamplesincludewaterp umpsandboilers.More ancillaryservicecanbeextractfromcommercialbuildingH VACsystemsbyutilizingthe otherequipment. 101

PAGE 102

Therearealargenumberofcommercialbuildingsinapowergr id.Theproposed algorithmisdesignedforasinglebuilding,itwillbeinter estingtoseehowanaggregated populationofbuildingsbehaveswhenworkingtogethertopr ovideancillaryservice. 102

PAGE 103

REFERENCES [1] “Commercialbuildingsenergyconsumptionsurvey(CBECS): Overviewof commercialbuildings,2003,”tech.rep.,Energyinformati onadministration, DepartmentofEnergy,U.S.Govt.,December2008. [2] D.S.Callaway,“Tappingtheenergystoragepotentialinele ctricloadstodeliver loadfollowingandregulation,withapplicationtowindene rgy,” EnergyConversion andManagement ,vol.50,no.5,pp.1389–1400,2009. [3] S.Koch,J.L.Mathieu,andD.S.Callaway,“Modelingandcont rolofaggregated heterogeneousthermostaticallycontrolledloadsforanci llaryservices,”in Proc. PSCC ,pp.1–7,2011. [4] B.Biegel,P.Andersen,T.S.Pedersen,K.M.Nielsen,J.Stou strup,andL.H. Hansen,“Electricitymarketoptimizationofheatpumpport folio,”in ControlApplications(CCA),2013IEEEInternationalConferenceon ,pp.294–301,IEEE, 2013. [5] N.Lu,“Anevaluationofthehvacloadpotentialforprovidin gloadbalancing service,” SmartGrid,IEEETransactionson ,vol.3,no.3,pp.1263–1270,2012. [6] S.Kundu,N.Sinitsyn,S.Backhaus,andI.Hiskens,“Modelin gandcontrolof thermostaticallycontrolledloads,” arXivpreprintarXiv:1101.2157 ,2011. [7] R.SioshansiandP.Denholm,“Thevalueofplug-inhybridele ctricvehiclesasgrid resources,” EnergyJournal ,vol.31,no.3,pp.1–23,2010. [8] J.Tomi ´ candW.Kempton,“Usingeetsofelectric-drivevehiclesfo rgridsupport,” JournalofPowerSources ,vol.168,no.2,pp.459–468,2007. [9] S.Meyn,P.Barooah,A.Busic,andJ.Ehren,“Ancillaryservi cetothegridfrom deferrableloads:thecaseforintelligentpoolpumpsinor ida,”in 52ndConference onDecisionandControl ,IEEE,2013. [10] K.DivyaandJ.Østergaard,“Batteryenergystoragetechnol ogyforpower systemsanoverview,” ElectricPowerSystemsResearch ,vol.79,no.4, pp.511–520,2009. [11] NERCResourcesSubcommittee,“Balancingandfrequencycon trol,”tech.rep., NorthAmericanElectricReliabilityCorporation,2011. [12] K.R.KeeneyandJ.E.Braun,“Applicationofbuildingprecoo lingtoreducepeak coolingrequirements,” ASHRAEtransactions ,vol.103,no.1,pp.463–469,1997. [13] P.Xu,P.Haves,M.A.Piette,andJ.Braun,“Peakdemandreduc tionfrom pre-coolingwithzonetemperatureresetinanofcebuildin g,”2004. 103

PAGE 104

[14] S.Kiliccote,M.A.Piette,andD.Hansen,“Advancedcontrol sandcommunications fordemandresponseandenergyefciencyincommercialbuil dings,”tech.rep., LawrenceBerkeleyNationalLaboratory,2006. [15] J.GrandersonandP.Price,“Evaluationofthepredictiveac curacyofve whole-buildingbaselinemodels,”Tech.Rep.LBNL-5886E,L awrenceBerkeley NationalLaboratory,Berkeley,CA,August2012. [16] D.Claridge,“Aperspectiveonmethodsforanalysisofmeasu redenergydata fromcommercialbuildings,” Journalofsolarenergyengineering ,vol.120,no.3, pp.150–155,1998. [17] P.Zhao,G.P.Henze,S.Plamp,andV.J.Cushing,“Evaluation ofcommercial buildinghvacsystemsasfrequencyregulationproviders,” EnergyandBuildings , vol.67,pp.225–235,2013. [18] M.Maasoumy,J.Ortiz,D.Culler,andA.Sangiovanni-Vincen telli,“Flexibilityof commercialbuildinghvacfanasancillaryserviceforsmart grid,” arXivpreprint arXiv:1311.6094 ,2013. [19] H.Hao,A.Kowli,Y.Lin,P.Barooah,andS.Meyn,“Ancillarys erviceforthegridvia controlofcommercialbuildingHVACsystems,”in AmericanControlConference (ACC),2013 ,IEEE,2013. [20] H.Hao,Y.Lin,A.Kowli,P.Barooah,andS.Meyn,“Ancillarys ervicetothegrid throughcontroloffansincommercialbuildingHVACsystems ,” SmartGrid,IEEE Transactionson ,vol.5,pp.2066–2074,July2014. [21] Y.Lin,P.Barooah,andS.P.Meyn,“Low-frequencypower-gri dancillaryservices fromcommercialbuildingHVACsystems,”in SmartGridComm,2013IEEEInternationalConferenceon ,pp.169–174,IEEE,2013. [22] Y.Kim,L.Norford,andJ.Kirtley,“Modelingandanalysisof avariablespeedheat pumpforfrequencyregulationthroughdirectloadcontrol, ” PowerSystems,IEEE Transactionson ,2014. [23] T.Middelkoop,“High-resolutiondatacollectionforautom atedfaultdiagnostics,” in AutomatedDiagnostics (B.L.CapehartandM.Brambley,eds.),Atlanta,GA: FairmontPress,2014.inpress. [24] AmericanSocietyofHeating,RefrigeratingandAirConditi oningEngineers,“The ASHRAEhandbookHVACsystemsandequipment(SIEdition),”2 008. [25] PJM,“PJMmanual12:Balancingoperations,rev.27,”Decemb er2012. [26] S.Goyal,H.Ingley,andP.Barooah,“Occupancy-basedzonec limatecontrolfor energyefcientbuildings:Complexityvs.performance,” AppliedEnergy ,vol.106, pp.209–221,June2013. 104

PAGE 105

[27] AmericanSocietyofHeating,RefrigeratingandAirConditi oningEngineers,“The ASHRAEhandbookfundamentals(SIEdition),”2005. [28] K.Deng,P.Barooah,andP.G.Mehta,“Mean-eldcontrolfore nergyefcient buildings,”in AmericanControlConference(ACC),2012 ,pp.3044–3049,IEEE, 2012. [29] Y.Lin,T.Middelkoop,andP.Barooah,“Issuesinidenticat ionofcontrol-oriented thermalmodelsofzonesinmulti-zonebuildings,”in DecisionandControl(CDC), 2012IEEE51stAnnualConferenceon ,pp.6932–6937,IEEE,2012. [30] “Nationalsolarradiationdatabase(NSRDB),”2005. [31] PJM,“PJMregulationdata.” http://www.pjm.com/markets-and-operations/ancillary -services.aspx ,2014. [32] Y.Yao,Z.Lian,andZ.Hou,“Thermalanalysisofcoolingcoil sbasedonadynamic model,” Appliedthermalengineering ,vol.24,no.7,pp.1037–1050,2004. [33] Siemens,“APOGEEpowersprocesscontrollanguage(PPCL)us ermanual,” October2000. [34] C.Hubner,T.Hansemann,andH.Merz, BuildingAutomation ,pp.185–273.Signals andCommunicationTechnology,SpringerBerlinHeidelberg ,2009. [35] L.Ljung, SystemIdentication:TheoryfortheUser .PrenticeHall,2ed.,1999. [36] AmericanSocietyofHeating,RefrigeratingandAirConditi oningEngineers, “ASHRAEStandard62.1,”2004. [37] PJM,“PJMmanual28:Operatingagreementaccounting,rev.6 2,”August2012. [38] PJM,“PJMmanual11:Energyandancillaryservicesmarketop erations,rev.62,” December2013. [39] FederalEnergyRegulatoryCommission,“OrderNo.755Frequ encyRegulation CompensationintheWholesalePowerMarkets:CommentsofIS O/RTOCouncil,” May2011. [40] PJM,“PJMregulationzonepreliminarybillingdata.” http://www.pjm.com/markets-and-operations/market-se ttlements.aspx , Jan2014. [41] B.A.OgunnaikeandW.H.Ray, Processdynamics,modeling,andcontrol .Oxford UniversityPressNewYork,1994. [42] I.B.Rhodes,“Atutorialintroductiontoestimationandlt ering,” IEEETransaction onAutomaticControl ,vol.AC-16,no.6,1971. 105

PAGE 106

[43] X.ZhouandJ.E.Braun,“Asimplieddynamicmodelforchille d-watercoolingand dehumidifyingcoilsPart1:Development(RP-1194),” HVAC&RResearch ,vol.13, no.5,pp.785–804,2007. [44] A.ElmahdyandG.Mitalas,“Asimplemodelforcoolinganddeh umidifyingcoils foruseincalculatingenergyrequirementsforbuildings,” ASHRAEtransactions , vol.83,no.2,pp.103–117,1977. [45] DaikinIndustries,“DaikinMcQuayToolsSuite.” http://www.daikinmcquay.com/McQuay/DesignSolutions/ McQuayToolsEngineers , 2013. [46] S.Bendapudi,J.Braun,andE.Groll,“Adynamicmodelofavap orcompression liquidchiller,”in InternationalRefrigerationandAirConditioningConfere nce ,2002. [47] S.GoyalandP.Barooah,“Amethodformodel-reductionofnon -linearthermal dynamicsofmulti-zonebuildings,” EnergyandBuildings ,vol.47,pp.332–340, 2012. [48] D.Y.Goswami,F.Kreith,andJ.F.Kreider, Principlesofsolarengineering .CRC, 2000. [49] J.L.Mathieu, Modeling,Analysis,andControlofDemandResponseResourc es . PhDthesis,UniversityofCalifornia,Berkeley,2012. [50] D.R.KincaidandE.W.Cheney, Numericalanalysis:mathematicsofscientic computing ,vol.2.AmericanMathematicalSoc.,2002. 106

PAGE 107

BIOGRAPHICALSKETCH YashenLinwasbornin1987inBeijing,China.Hereceivedhis BachelorofScience degreeinautomationin2009fromUniversityofScienceandT echnologyBeijing.He thenjoinedUniversityofFloridatopursuehisdoctoraldeg reeundertheadvisementof Dr.PrabirBarooah.HereceivedhisMasterofSciencedegree in2012andPh.Ddegree in2014.Hisresearchinterestliesintheeldofmodelingan dintelligentcontrolof heating,ventilation,andairconditioning(HVAC)systemi nbuildingsanditsapplication inpowergridancillaryservice. 107