国外电子测量技术北大中文核心期刊DOI:10.19652/j.cnki.femt.2204456基于非侵入式负荷分解的有色金属冶炼工序识别*方祖春汪繁荣(湖北工业大学电气与电子工程学院武汉430068)摘要:为进一步简化数据处理过程和提高生产工序识别准确率,提出一种基于非侵入式负荷分解的工序识别方法。首先将每种工序定义为一种用电设备,然后根据非侵入式负荷分解相关理论,分别选取双向长短期记忆网络和时间卷积网络构建负荷分解模型,选择各用电设备对应功率、总功率数据构造数据集对模型进行训练和测试,最后对测试集负荷分解结果进行相关处理得到对应的工序数据。结果表明由基于时间卷积网络的负荷分解方法构成的工序识别模型具有较高的识别准确率,针对测试集的工序识别准确率达98.83%。关键词:非侵入式负荷分解;双向长短期记忆网络;时间卷积网络;工序识别中图分类号:TM714文献标识码:A国家标准学科分类代码:510.4010Processidentificationofnon-ferrousmetalsmeltingbasedonnon-invasiveloaddecompositionFangZuchunWangFanrong(SchoolofElectricalandElectronicEngineering,HubeiUniversityofTechnology,Wuhan430068,China)Abstract:Inordertofurthersimplifythedataprocessingprocessandimprovetheaccuracyofproductionprocessidentification,aprocessidentificationmethodbasedonnon-invasiveloaddecompositionwasproposed.Firstly,eachprocesswasdefinedasakindofelectricalequipment.Then,accordingtotherelevanttheoriesofnon-invasiveloaddecomposition,bidirectionallongshort-termmemorynetworkandtemporalconvolutionnetworkwereselectedtoconstructtheloaddecompositionmodel,andthecorrespondingpowerandtotalpowerdataofeachelectricalequipmentwereselectedtoconstructthedatasetfortrainingandtestingthemodel.Finally,thecorrespondingprocessdatawasobtainedbyrelevantprocessingoftheloaddecompositionresultsofthetestset.Theresultsshowthattheprocessidentificationmodelconstructedbytheloaddecompositionmethodbasedonthetemporalconvolutionnetworkhashighrecognitionaccuracy,andtheprocessidentificationaccuracyforthetestsetis98.83%.Keywords:non-invasiveloaddecomposition;bidirectionallongshort-termmemorynetwork;temporalconvolutionnet-work;processidentification收稿日期:2022-11-01*基金项目:国家自然科学基金(61903129)项目资...
