基于FFT-DC-GRU-NLA的中长期居民用电量预测模型

打开文本图片集

中图分类号：TN911.23-34；TP18 文献标识码：A 文章编号：1004-373X（2025）16-0088-09

A medium and long-term residential electricity consumption forecasting modelbasedonFFT-DC-GRU-NLA

ZHANGCheng，SHEN Chao （Collegeof InformationandElectrical Engineering，Hebei UniversityofEngineering，Handan O56O38，China）

Abstract：Inallusion to the dificulties incomplex powerdata modeling，poor informationrepresentation capabilities，and lowpredictionacuracyinexisting mediumandlong-termresidential electricityconsumptionforecastingmodels，amediumand long-termelectricityconsumptionforecastingmodelbasedonFFT-DC-GRU-NLAisproposed.ThefastFourier transform（FFT）is usedto decompose the electrityconsumptiondata，and themulti-periodcomponentsareextracted by means of frequency domaindecompositiontoobtainasetof two-dimensionalsub-sequences.Itisusedastheinputfortheself-designedinforation representationmodule torealizemulti-scaleinformationrepresentationanddepfeatureextractionofthetwo-dimensional subsequencesbyintegratingconvolutionalneuralnetwork，gatedrecurentunit，andnon-localatentionmechanism.Thedep featuresarereconstructed bymeansoffullyconnectedlayersand iterativelypredicted bymeansof residual structure.Ona publicdatasetofresidentialelectricityconsumption，comparedwithseveraladvancedmodelsinthefieldofpowerprediction，the proposed modelhasthe highestpredictionaccuracyatpredictionlengthsof96，192，336and720.Moreover，themodelcanalso realizeexcelentpredictionaccuracyontwootherpublicdatasetsrelatedtopowerprediction.Theexperimentalresultsshowthat thismodelcanefectivelyimprovetheaccuracyofmediumandlong-termresidentialelectricityconsumptionforecastingandhas good generalization capabilities.

Keywords：mediumandlong-termelectricityconsumptionforecasting;fastFourier transform;convolutioalneuralnetwork; gated recurrent unit; non-local attention mechanism; multi-scale information; deep feature extraction

0 引言

在“双碳”政策的战略框架下，为达成“碳达峰”与“碳中和”的宏伟目标，新型电力系统的建设被确立为核心任务之一1。（剩余10762字）