第51卷第2期电力系统保护与控制Vol.51No.22023年1月16日PowerSystemProtectionandControlJan.16,2023DOI:10.19783/j.cnki.pspc.220743基于深度学习的110kV电网监控信号语义解析及态势感知模型王洪彬1,2,周念成1,黄睿灵2,范炳昕1,王强钢1(1.输配电装备及系统安全与新技术国家重点实验室(重庆大学),重庆400044;2.国网重庆市电力公司电力科学研究院,重庆401123)摘要:新型电力系统的大力建设对电网监控信号的高效准确识别技术提出了更高的要求。首先分析了Soft-MaskedBERT语言模型的基本原理,建立了基于Soft-MaskedBERT的信号文本纠错模型。根据国家电网典型事件表梳理了包含常规与故障情况下的“信号语义—电网事件”规则字典。综合上述模型建立了基于RNN的电网态势感知模型,提出了基于深度学习的电网监控信号语义解析及态势感知求解流程。最后,以某地110kV变电站实际监控信号为测试数据,利用所提RNN模型并结合Pycorrector工具包及Pytorch软件对该地区电网监控信号进行语义解析及态势感知仿真分析,验证了模型的有效性及正确性。关键词:深度学习;电网监控信号语义解析;态势感知;RNN模型110kVsignalsemanticanalysisandsituationawarenessmodelbasedondeeplearningtheoryforapowersystemmonitoringsystemWANGHongbin1,2,ZHOUNiancheng1,HUANGRuiling2,FANBingxin1,WANGQianggang1(1.StateKeyLaboratoryofPowerTransmissionEquipment&SystemSecurityandNewTechnology(ChongqingUniversity),Chongqing400044,China;2.StateGridChongqingElectricPowerCompanyResearchInstitute,Chongqing401123,China)Abstract:Thevigorousconstructionofnewpowersystemsentailshigherrequirementsfortheefficientandaccurateidentificationtechnologyforpowergridmonitoringsignals.ThispaperfirstanalyzesthebasicprinciplesoftheSoft-MaskedBERTlanguagemodel,andestablishesasignaltexterrorcorrectionmodelbasedonSoft-MaskedBERT.AccordingtothetypicalinformationtableoftheStateGrid,theruledictionaryof"signalsemantics-gridevents"innormalandfaultconditionsisanalysed.Basedontheabovemodels,apowergridsituationawarenessmodelbasedonRNNisestablished,andasemanticanalysisofpowergridmonitoringsignalsandasituationawarenesssolutionprocessbasedondeeplearningareproposed.Finally,takingtheactualmonitoringsignalofa110kVsubstationa...