·检测与试验·电器与能效管理技术(2023No.3)王建元(1971—),男,教授,博士,研究方向为智能配电网运行与控制、大数据分析等。刘柯辰(1998—),女,硕士研究生,研究方向为电力大数据分析。基于经验模态分解与多视角聚类的异常用电模式检测王建元,刘柯辰[现代电力系统仿真控制与绿色电能新技术教育部重点实验室(东北电力大学),吉林吉林132012]摘要:针对现有异常用电检测方法检出效率低下的问题,提出一种基于经验模态与多视角聚类的异常检测方法。遵循“经验模态分解-维度制约-多视角聚类-横向检测-纵向检测”的流程,通过多视角聚类结合初步判据,显著提高了检出率。在异常检测算法中,提出基于网格的熵离群因子(Grid-EOF)算法,并基于纵向检测给出新的判据,提高了不明显窃电行为用户的检出率。最后,用国家电网智能电表实测数据检测验证,结果表明多视角聚类和改进算法以及纵向检测的引入,能有效提高异常检测模型的检出率和准确率。关键词:异常用电检测;经验模态分解;多视角聚类;香农熵中图分类号:TM930文献标志码:A文章编号:2095-8188(2023)03-0073-08DOI:10.16628/j.cnki.2095-8188.2023.03.012AbnormalPowerConsumptionModeDetectionBasedonEmpiricalModeDecompositionandMulti-ViewClusteringWANGJianyuan,LIUKechen[KeyLaboratoryofModernPowerSystemSimulationandControl&RenewableEnergyTechnology,MinistryofEducation(NortheastElectricPowerUniversity),Jilin132012,China]Abstract:Inordertosolvethelowdetectionefficiencyoftheexistingabnormalpowerconsumptiondetectionmethods,theanomalydetectionmethodbasedonempiricalmodeandmultiviewclusteringisproposed.Followingtheprocessof"empiricalmodedecomposition-dimensionalconstraints-multi-viewclustering-horizontaldetection-verticaldetection"andcombiningthemulti-viewclusteringwiththepreliminarycriteria,thedetectionrateissignificantlyimproved.Intheanomalydetectionalgorithm,thegrid-basedentropyoutlierfactor(Grid-EOF)algorithmisproposed.Anewcriterionisgivenbasedonthelongitudinaldetection,whichcanimprovethedetectionrateofuserswithunknownelectricitytheft.Finally,itisverifiedbythemeasureddataofsmartmetersoftheStateGridofChina.Theresultsshowthattheintroductionofmulti-viewclustering,improvedalg...