第47卷第2期电网技术Vol.47No.22023年2月PowerSystemTechnologyFeb.2023文章编号：1000-3673（2023）02-0482-11中图分类号：TM721文献标志码：A学科代码：470·40基于深度残差收缩网络的电力系统暂态频率安全集成评估王彦博，吴俊勇，季佳伸，李栌苏，李宝琴（北京交通大学电气工程学院，北京市海淀区100044）IntegratedAssessmentofPowerSystemTransientFrequencySecurityBasedonDeepResidualShrinkageNetworkWANGYanbo,WUJunyong,JIJiashen,LILusu,LIBaoqin(SchoolofElectricalEngineering,BeijingJiaotongUniversity,HaidianDistrict,Beijing100044,China)1ABSTRACT:Underthebackgroundofacceleratingtransfor-mationofChina'senergystructureandstrivingtoachievethegoalof"dualcarbon",thetraditionalpowersystemwillalsousherinstructuraltransformation.Duetotherandomness,uncertaintyandlowinertiaofrenewableenergy,aseriesofinfluencesbroughtbylarge-scalenewenergygridmakethefrequencysafetyproblemofpowersystemincreasinglyprominent.However,thetraditionaltimedomainsimulationmethodhassomedisadvantagessuchaslargeamountofcomputationandlongcalculationtime,soitisdifficulttorealizetherapidevaluationoftheactualpowersystemundertheflexibleandchangeableoperationmodeandalargeamountofmeasureddata.Inordertoquicklyassessthefrequencysecurityofpowersystems,apowersystemtransientfrequencysecurityintegratedassessmentmethodbasedondeepresidualshrinkagenetwork(DRSN)isproposed.Deepresidualshrinkagenetworkintroducesattentionmechanismbasedondeepresidualnetwork,whichcanenhanceusefulinformationandsuppressredundantinformation.Onthisbasis,thesamplesweredividedaccordingtothemaximumfrequencychangerate,andDRSNnetworkwasusedfortrainingrespectively.SimulationresultsontheIEEE39-bussystemandIEEE118-bussystemaddedwithwindpowermachinesshowthattheproposedmethodhashigheraccuracyandexcellentgeneralization、robustnessandapplicability.KEYWORDS:deeplearning;powersystem;frequencysecurity;maximumrateofchangeoffrequency;deepresidualshrinkagenetwork;attentionmechanism摘要：在我国能源结构加速转型、力争实现“双碳”目标的基金项目：国家重点研发计划项目(2018YFB0904500)；国家电...
