文档名：基于非侵入式负荷分解的家庭负荷两阶段超短期负荷预测模型
摘要：精细化负荷预测为制定家庭新型需求响应策略或能效管理模式提供了可靠的指导信息与理论基础,而负荷监测系统的广泛研究与发展为家庭设备层的负荷预测提供了有力的数据支撑.基于家庭负荷智能电能表集中数据,该文提出一种集分解-预测一体化的家庭负荷两阶段超短期负荷预测方法.该方法第一阶段提出了基于卷积神经网络(CNN)和双向门控单元(BiGRU)神经网络的非侵入式负荷分解(NILM)模型,解决了目前深度分解模型中特征提取不充分、分解精度低等问题.第二阶段构建了基于时间模式注意力机制(TPA)的时间卷积神经网络(TCN)负荷预测模型,深度挖掘NILM分解数据、集中负荷数据及日期特征等输入变量的深层交互信息,实现家庭设备层的负荷预测.算例部分通过UK-DALE数据集对所提方法进行验证,结果表明,该方法能够获得较高的分解精度和预测效果,为家庭负荷预测提供了良好的条件.
Abstract：Duetotheimpactoflarge-scalerenewableenergyonthesafeandstableoperationofpowersystems,thedemandforflexiblesourcesisincreasing.Thehomeenergymanagementsystemisapromisingapproachtoenhancetheflexibleregulationcapabilityofpowersystemsandimprovegridenergyefficiency.However,therandomnessofresidents'electricitybehavior,theuncertaintyofmarketinformation,andthediversityofdecision-makingsubjectsmakeitextremelychallengingforresidentstoparticipateinfastdemandresponses.Toaddresstheseissues,thispaperproposesatwo-stagehouseholdloadforecastingmethodbasedontheintegrationofloaddisaggregationandforecasting.Bylearningthecorrelationinformationfromhistoricalelectricityconsumptiondataofeachapplianceobtainedbynon-intrusiveloadmonitoring(NILM)technology,itaccuratelyrealizestheloadforecastingofhouseholdappliancesandflexibleclusterloadprediction.First,aNILMmodelbasedonconvolutionalneuralnetwork(CNN)andbi-directionalgatedunit(BiGRU)neuralnetworkisestablishedtosolvetheproblemofobtainingtheoperationdataofappliances.Subsequently,consideringtherandomnessanduncertaintyofuserbehavior,atimeconvolutionalnetwork(TCN)loadforecastingmodelbasedonthetimepatternattention(TPA)mechanismisconstructedtominethedeepinteractioninformationofinputvariables.Finally,theproposedmethodisverifiedbytheUK-DALEdataset.Theresultsshowthattheproposedmethodcanobtainhighdisaggregationaccuracyandpredictioneffect.ThispaperimplementstwosimulationsusingKeraswithaTensorFlowbackend.Thefirstoneisdesignedtomonitortheappliance-levelenergyconsumptionwiththeproposedCNN-BiGRU-enabledNILM.TheresultsshowthattheproposedNILM-Basedmodelcanaccuratelycapturethestartandendtimeoftheappliance,andhasagoodtrend-trackingeffect.WiththeNILMresults,thesecondoneisconductedtoverifytheeffectivenessoftheproposedloadforecastingmodel.Comparedwithotherdeeplearningmodelssuchaslongshort-termmemory,theeMAEandeRMSEofwashingmachines,microwaveovens,dishwashersandrefrigeratorsbasedontheproposedforecastingmethodarereducedby18.65%and4.99%,13.28%and0.72%,32.04%and5.55%,4.53%and5.70%,respectively.ThecomparisonofloadforecastingresultsbetweenthegroundtruthdataandtheNILM-baseddatashowsthattheeMAEiscontrolledwithin15%andtheeRMSEiscontrolledwithin6%.Finally,toverifytherobustnessoftheproposedmodelinclusterloadforecasting,abottom-upstrategyisappliedtoobtaintheflexibleloadpredictionofgroupusersorcommunitiesbasedonapplianceprediction.Thefollowingconclusionscanbedrawnfromthesimulationanalysis:(1)TheloadforecastingframeworkbasedonNILMreducesthedependenceonintrusivemonitoringsystems,andhasstrongadaptabilityandscalability,whichcanprovideanewmethodforthedevelopmentofhouseholdenergymanagementsystem-relatedbusinesses.(2)Comparedwithshallowneuralnetworks,CNN-BiGRUhaspowerfulmappingabilityandcanachievehighaccuracyinloaddisaggregationandeventdetection.(3)TheproposedforecastingmodelextractstemporalinformativefeaturesusingTCNandsubsequentlyimplementsdynamicweightingofinputsusingTPAtohighlighttheimpactofkeyfeatures.Thecasestudyshowsthattheproposedmethodcanimproveforecastingaccuracycomparedwithtraditionaldeeplearningmodels.
作者：李延珍 王海鑫 杨子豪 陈哲 杨俊友 Author：LiYanzhen WangHaixin YangZihao ChenZhe YangJunyou
作者单位：沈阳工业大学电气工程学院沈阳110870丹麦奥尔堡大学能源技术系奥尔堡DK-9220
刊名：电工技术学报 ISTICEIPKU
Journal：TransactionsofChinaElectrotechnicalSociety
年，卷(期)：2024, 39(11)
分类号：TM714
关键词：非侵入式负荷分解 负荷预测 卷积神经网络 双向门控单元神经网络 时间卷积网络 注意力机制
Keywords：Non-intrusiveloaddisaggregation loadforecasting convolutionalneuralnetwork bidirectionalgatedunitneuralnetwork timeconvolutionalnetwork attentionmechanism
机标分类号：TP391TM714TM933.4
在线出版日期：2024年6月18日
基金项目：高等学校学科创新引智计划资助项目基于非侵入式负荷分解的家庭负荷两阶段超短期负荷预测模型[
期刊论文] 电工技术学报--2024, 39(11)李延珍 王海鑫 杨子豪 陈哲 杨俊友精细化负荷预测为制定家庭新型需求响应策略或能效管理模式提供了可靠的指导信息与理论基础,而负荷监测系统的广泛研究与发展为家庭设备层的负荷预测提供了有力的数据支撑.基于家庭负荷智能电能表集中数据,该文提出一种...参考文献和引证文献
参考文献
引证文献
本文读者也读过
相似文献
相关博文
基于非侵入式负荷分解的家庭负荷两阶段超短期负荷预测模型 Two-Stage Ultra-Short-Term Load Forecasting Model of Household Appliances Based on Non-Intrusive Load Disaggregation
基于非侵入式负荷分解的家庭负荷两阶段超短期负荷预测模型.pdf
- 文件大小:
- 1.27 MB
- 下载次数:
- 60
-
高速下载
|
|
