多特征提取与深度学习关口计量装置异常事件识别方法  

Identification method of abnormal events of gateway metering device based onmulti-feature extraction and deep learning

作　　者：徐严军 吴蒙 白佳灵 丁熠辉 谢智[1] 卢宏 肖先勇[2] XU Yanjun;WU Meng;BAI Jialing;DING Yihui;XIE Zhi;LU Hong;XIAO Xianyong(Measurement Center of Sichuan Electric Power Company,Chengdu 610045,China;College of Electrical Engineering,Sichuan University,Chengdu 610065,China)

机构地区：[1]国网四川省电力公司计量中心,四川成都610045 [2]四川大学电气工程学院,四川成都610065

出　　处：《中国测试》2021年第5期104-111,共8页China Measurement & Test

摘　　要：针对现有研究对关口计量装置定期检修会造成人力物力的浪费,以及通过指标赋权进行打分受人工干预较明显等不足,该文基于用户用电采集系统中的异常事件数据,提出一种多特征提取与深度学习相结合的关口计量装置异常事件识别的方法。通过分析不同计量异常事件反映出的数据异常形式,从不同维度提取14个特征,并将归一化后的特征作为深度学习模型的输入。在此基础上,通过无监督的预训练和有监督微调构建深度学习模型,自动学习得到输入特征与类别标签之间的非线性映射关系,构建出能识别关口计量装置异常事件的分类识别模型。通过某地区电网多个关口计量装置终端的数据对所提方法进行验证,结果证明所提方法能够准确识别出不同的异常事件,且具有良好的鲁棒性。In view of the existing research on the waste of manpower and material resources caused by the regular maintenance of the gateway metering device,as well as the deficiencies of artificially selected evaluation indicators.Based on the abnormal event data in the user's electricity collection system,this paper proposes a method using multi-feature extraction and deep learning for identifying abnormal events of the gateway metering device.14 features are selected by analyzing the data forms of different abnormal events,and the selected features are normalized and used as the input data of the deep learning model.In addition,this paper constructs a deep learning model through unsupervised pre-training and supervised fine-tuning,which can automatically learn the nonlinear mapping relationship between the input data and category labels,and a classification and recognition model is constructed to identify abnormal events of the gateway metering device.The proposed method is verified through the data of multiple gateway metering device terminals in a certain area power grid. The results prove that the proposed method can accurately identify different abnormal eventsand has good robustness.

关 键 词：关口计量装置 异常事件 特征提取 堆叠自动编码器 状态评价 

分 类 号：TM71[电气工程—电力系统及自动化]