电工材料2023No.1陈铁等：基于深度强化学习的变电站倒闸操作序列生成方法基于深度强化学习的变电站倒闸操作序列生成方法陈铁1，2，曹颖1，2，蔡东阁1，2，何思敏1，2（1.三峡大学电气与新能源学院，湖北宜昌443002；2.三峡大学梯级水电站运行与控制湖北省重点实验室，湖北宜昌443002）摘要：针对当前智能电网操作票系统通用性差、智能化程度不足等问题，提出一种基于深度强化学习的变电站倒闸操作序列生成方法。首先，采用知识图谱技术建立变电站知识图谱模型，利用知识图谱路径搜索确定操作空间，结合变电站运行规则对任务设备运行状态进行推理更新；然后，构建求解倒闸操作序列的强化学习模型；最后，应用DDQN深度强化学习算法求解倒闸操作序列。测试结果表明，该方法适用于不同的操作任务，能够根据操作任务自动生成符合倒闸操作逻辑的操作序列，无需建立复杂的规则库，通用性较强。关键词：深度强化学习；倒闸操作序列；操作票；知识图谱中图分类号：TM734DOI：10.16786/j.cnki.1671-8887.eem.2023.01.019GenerationMethodofSwitchingOperationSequenceinSubstationBasedonDeepReinforcementLearningCHENTie1,2,CAOYing1,2,CAIDongge1,2,HESimin1,2(1.CollegeofElectricalEngineeringandNewEnergy,ChinaThreeGorgesUniversity,HubeiYichang443002,China;2.HubeiProvincialKeyLaboratoryforOperationandControlofCascadedHydropowerStation,ChinaThreeGorgesUniversity,HubeiYichang443002,China)Abstract:Inviewoftheproblemssuchaspooruniversalityandinsufficientintelligenceofcurrentsmartgridoperationticketsystem,asubstationswitchingoperationsequencegenerationmethodbasedondeepreinforcementlearningispresented.Firstly,theknowledgemapmodelofsubstationisestablishedbyusingknowledgemaptechnology,theoperationspaceisdeterminedbyknowledgemappathsearch,andtheoperationstatusoftaskequipmentisinferredandupdatedwithsubstationoperationrules.Then,anintensivelearningmodelisbuilttosolvethesequenceoftheswitchingoperation.Finally,theDDQNdeepreinforcementlearningalgorithmisappliedtosolvethesequenceoftheswitchingoperation.Thetestresultsshowthatthismethodissuitablefordifferentoperationtasks,canautomaticallygenerateoperationsequenceaccordingtotheoperationtasks,doesnotneedtobuildacomplexrulebase,a...
