人工智能驱动的电动汽车充电管理与规划：基于量化聚合异构电动汽车灵活性的可再生能源跟踪策略

ProfZYDong Departmentof ElectricalEngineering&JCSTEMLabofFutureEnergySystemsCityUniversity of HongKong香港城市大学，电机工程学系，香港赛马会未来能源系统创新实验室25Oct2025 AcknowledgementProfShuyingLaiProfYuechuanTao 香港赛馬會慈善信基金The Hong KongJockey Club Charities Trust 01 03SmartEVManagementSystem 04 Background What happened? Newpowersystem Motivation Possiblesolutions >Battery energy storage systems (BESSs)can alleviate thisdemand-sideimbalanceprobleminlarge-scalerenewableenergypenetrationsystems>Electricvehicles(EVs)canberegardedassmalldistributedenergystoragesystemprovidingflexibility.>EVsexhibit inherent flexibility.with bidirectional charging anddischargingcapabilitiesthroughvehicle-to-grid(V2G)technologies. PolicySupport 《新能源汽车产业发展规划（2021一2035年）》（2020年，国务院办公厅） 新能源汽车产业的钢领性文件。其中明确提出要促进新能源汽车与能源融合发展，加强新能源汽车与电网（V2G）能量互动，鼓励地方开展V2G应用示范。 《关于加强电网调峰储能和智能化调度能力建设的指导意见》（2024年，国家发改委、国家能源局） 明确提出要大力推动车网互动，包括制定车网互动发展规划和实施细则，全面推广有序充电，积极开展V2G应用示范，并推动电动汽车通过参与需求侧响应、辅助服务市场等获得收益 OurWork w.r.t.EVchallenges&opportunities Rapidmarketuptakeintroduceschallengesonpowergrid,esp,powerqualityrisks inurbanenetworks ·Safety&SecurityPhysical security (grid and battery)and cyber security.EVbatterysafetyforreliableandaccurateSOC/SOHestimation:EVchargingoptimisation.EV charging infrastructure planning:Supportingemissionreductionasflexibleload:Market incentives promoting EV uptake and renewables 01 02 03SmartEVManagementSystem 04 MachineLearningClassification DeepLearning LargeLanguageModels(LLMs) LargeLanguageModels(LLMs)aredeeplearningmodelstrainedonextensivetextdatatogenerateandunderstandhuman-likelanguage. Key feature: DefinitionandCoreConcept:Largelanguagemodelsaredeeplearningmodelstrainedonvastdatacapableofgeneratinghuman-liketext.UseCase:UsedintaskslikeQ&A,translation,contentwriting.andcodegeneration. AlApplicationsinEVIntegrationinSmartGirds AlfacilitatesEVintegrationinsmartgridstosupportaccurate,safety,optimization,andrapidresponse 01 02 03SmartEVManagementSystem 04 Smart EVChargingManagement 1)Customerneeds EVowners/drivers .tolocateanavailablechargingstationwithminimalwaitingtominimisecostsofchargingwithoutcompromisingtheirdrivingneedstoharnesstheenergymarketopportunitiestobeaprosumer.toensure EVbatteryhealth/safetythroughpredictivemonitoringtobecertainaboutprivacyanddatasecurity Carparkoperatorswantstomaximiseprofitby7/24charging,butlocal switchboardmaybeconstrainedPowercompanyhasnotseenmassiveEV uptakeeverbefore,theyneedensuresystemsecuritywithafriendlymarketmechanismratherthanadministrativerulestoachievepreferred EV chargingbehaviour 2)PositionofTechnology/Application Disrupter:Securegridfriendlyprosumerfocuseduniversal EVchargingservicesEnhancer:SecureAlbasedsmartEVchargingmanagement,withaprosumer-basedmarketmechanismfunctionsto ensuregrid security whilemeeting driving needs. EVChargingManagementEco-system（GTF202020192） System-ItwillenablesystemoperatorstoeffectivelymanagethepowergridtohostlargenumberofEVswithoutblackoutwhileenjoying capitaldeferral benefits: Community-ItprovidestoolsforcarparkoperatorswithEVchargingfacilitiestoavoidpotential lossofelectricitywhileallowingforEVchargingchoices;and Individuals-ItprovidesoptimalchargingstrategiesforEV ownerstosaveonchargingwhile beassuredofsufficientbattery chargeto drive. Sources - Green Tech Fund Project Profile SmartUniversalEVManagementSystem InadditiontonormalEVchargingAPPs,oursystemhasawiderangeofEVandenergysystems,smartcityplanning AuniversalplatformabletohostdifferenttypesofEVchargers; Withadvancedcontrol,incaseofblackout,lostofservices,ourplatformwillensureoverallresilience: Asauniversaltool,oursystemsupportsurbanplanningfoieconomyefficiencywhileensuringdata/cybersecurityforplanners Ourplatformcanmonitorthesystemoperations,identifyandgeneratewarningofpotentialorperspectiveshortageofsupplylostofsupply.potentialsecurityproblem. ChargingRobotControl A cloud-edge-terminal coordination-based framework is formulated, with the cloud center, the chargingrobotcontrolcenter(CRCC),i.e.MCRoperator,thetransportationsystem(TS),andtheterminal APPfordeliveringthechargingrequestsofEVscaninteractwitheachother TheCRCC obtainsall thestatedataof themobilechargingrobots(MCRs)andthepowerflowdataof theDCs,whilethetrafficflowdataareprovidedbytheTS EVChargingSchedulingbasedonDeepreinforcementlearningandLLM Human intervention improveslearningthrough: Human-machinereinforcementlearningframework:Model-free,data-drivenmethodforreal-timedecision-making >EmergencycontrolRewardramp-upGuidedexploration EVChargingSchedulingbasedonDeepreinforcementlearningandLLM LargeLanguageModel(LLM)-assistedAl-agentframework FormulatethebiddingtaskasaDRLproblem:ConditionalValue-at-Risk(CVaR)basedDRLalgorithmtohandlealeatoricuncertainty Zhang, B., Li, C., Chen, G., & Dong.Z. (2024). Large language model assistedoptimalbidding of bess infcas market:An ai-agentbased approach.arXivpreprintarXiv:2406.00974. 01 02 03SmartEVManagementSystem 04 05 Charging Infrastructure Planning Issue