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）02-0741-14中图分类号：TM721文献标志码：A学科代码：470·40基于深度学习的电力系统暂态电压与暂态功角稳定一体化超前评估史法顺，吴俊勇，季佳伸，步雨洛，李栌苏，赵鹏杰，李宝琴（北京交通大学电气工程学院，北京市海淀区100044）IntegratedAdvanceAssessmentofPowerSystemTransientVoltageandTransientAngleStabilityBasedonDeepLearningSHIFashun,WUJunyong,JIJiashen,BUYuluo,LILusu,ZHAOPengjie,LIBaoqin(SchoolofElectricalEngineering,BeijingJiaotongUniversity,HaidianDistrict,Beijing100044,China)1ABSTRACT:Transientvoltagestability(TVS)andtransientanglestability(TAS)aretheimportantbasesforthesafeoperationofthepowersystem.Withtheconstructionofthenewpowersystems,theproblemsofTVSandTAShavebeencloselycoupledandoccurredfrequently,whichneedsahighprecisionintegratedadvancedevaluationurgentlytolayasolidfoundationfortheemergencycontrol.Firstly,accordingtotheinvestigation,thecomprehensivefeaturesoftheTASandTVSareintegrated,andthefeaturecontributiondegreeismeasuredaccordingtotheextremegradientboosting(XGboost),fromwhichthefeaturesetwithdifferentialfeaturesisgeneratedastheinputoftheevaluationmodel.Secondly,amulti-scaleconvolutiongatedrecurrentunitmodelintegratingtheextrusionexcitation(SE-CGRU)isproposed.Themodelrealizestheadaptiveadjustmentofthefeatureweightchannelthroughthesqueezeandexcitation(SE)module,andfusesthedetailfeaturesandthemacrofeaturesbyusingthemulti-scaleconvolutiontorealizethehigh-precisionintegratedevaluationofthetransientpowerangleandtransientvoltagesothatthepredictionresultscanbegivenwithoutknowingthefaultclearingtimeandthesafetymarginofthesystemunderthecurrentstatecanbeoutputduringtheonlineevaluation.Byintroducingthelossfunctionwithatimeconstraintandthedynamicweighttraining,theresponsetimeisgreatlyreducedandtheadvanceevaluationisrealizedonthebasisofmaintainingtheexistingaccuracy.Themulti-criterionfusionstrategyfurtherreducesthemissedjudgmentsandmisjudgments,andimprovesthereliabilityofmodel...
