基于卷积神经网络和Transformer的电能质量扰动分类

打开文本图片集

中图分类号：TN915.04-34；TP183；TM711 文献标识码：A 文章编号：1004-373X（2025）16-0113-10

PQDsclassificationbasedonconvolutional neural networkandTransformer

WANGGaofeng，ZHANGHao，QIANYun，GAOMan

Abstract：With the large-scaleapplication of new energy，the probabilityof power quality disturbance （PQDs）events is increasedsignificantlyandthesedisturbancescancausesignificantloses topowerdistributionsystems.Therefore，amethodof PQDclassification basedonconvolutional neuralnetwork（CNN）and Transformerisproposed，namely CTranCAM.Inthis method，theconvolutionoperationofCNNisusedtoautomaticallyextractlocalfeaturesofPQDssignal timeseries，andthen multi-headatentionmechanisminTransformerisusedtomodeltheglobalandlong-termrelationshipsofteextractedfeatures tocompensatefortheshortcomingsofCNNinprocesingglobalinformation.Therecognitionresultsareoutputbymeansoffully conectedlayer.Thesimulationof25typesofsynthesizedPQDsdataisconductedbymeansof CTranCAMmethod.Theresults showthat the classification accuracy of this method is99.60% under noise free conditions，and canreach 99.20%， 99.36% ，and 99.40% at signal-to-noiseratiosof 3OdB，40dB，and5OdB，respectively.Ithasgoodnoiseresistanceand generalization performace.Incomparison withperformanceofothermethods，theproposed methodhasgood classfication performancein different noise environments，which is an excellent PQDs classification method.

Keywords：powerquality disturbance;convolutional neural network；Transformer model; multi-headatention mechanism; featureextraction;classificationperformance

0 引言

随着资源枯竭和环境恶化问题日益严峻，可再生能源如太阳能和风能的广泛应用成为主要解决方案。（剩余11535字）