第46卷第1期2023年2月电子器件ChineseJournalofElectronDevicesVol.46No.1Feb．2023项目来源:国网山西省电力公司科技项目(520531200004)收稿日期:2021－08－04修改日期:2021－12－15PredictionMethodofPowerStealingBasedonＲegressionConvolutionalNeuralNetworkandLoadChaosModel*JINHaigang*，XIEZhengang，ＲENFeng(StateGridShanxiElectricPowerCompany，TaiyuanShanxi030002，China)Abstract:Accuratecalculationoflinelossandextractionofusers’electricityconsumptioncharacteristicsarethekeytoaccuratemar-ketingstrategy．However，therandomnessofrenewableenergypowergeneration，disorderlycharginganddischargingofelectricvehicles，environmentalchangesandotherfactorsmaketheuser’selectricityconsumptionbehavioreasilymutate，whichleadstothechaoticandrandomcharacteristicsofuser’selectricityconsumptiondatainthestationarea，andcannoteffectivelydetectuser’selectricitystealingbehavior．Inordertosolvethisproblem，atime-seriescorrelationchaoticmodelofpowerconsumptionforstationareausersisestab-lished，andthecharacteristicsofpowerstealingloadsamplesandnormalsamplesareextracted．Theregressionconvolutionalneuralnet-workisusedtotrainandlearnthepowerstealingsamplesandnormalsamples，andtheenhancedfeatureclassificationlearnerisob-tained，soastorealizethepowerconsumptionpredictionforpowerstealingusers．Throughthetestandanalysisofthepowerconsump-tiondataofanactualpowercompany，itshowsthatthecalculationresultsoftheproposedmethodhavehighaccuracy．Keywords:regressionconvolutionneuralnetwork;load;chaos;stealingelectricity;forecastEEACC:7220;8150doi:10．3969/j．issn．1005－9490．2023．01．038基于回归卷积神经网络和负荷混沌模型的窃电预测方法*靳海岗*，谢振刚，任峰(国网山西省电力公司，山西太原030002)摘要:精确地计算台区线损、提取用户用电特征实现窃电预测，是精准营销策略制定的关键。然而，台区可再生能源发电的随机性、电动汽车充放电无序性、环境变化等因素使用户用电行为极易突变，导致台区用户用电数据呈现混沌随机特性，无法有效检测用户窃电行为。对此，建立了台区用户用电的时序相关混沌模型，并提取窃电负荷样本与正常样本的特征，使用回归卷...
