摘要
针对电力物联感知技术推广与智能配电网建设大背景下,配电设备监测信号数据量大及类型繁多的特点,为有效缓解数据的传输及存储压力,提出基于电力物联感知的配电设备监测信号的压缩感知及异常识别算法。首先离线训练阶段中提出基于动态阈值的原子自适应奇异值分解算法,在保证重构精度的同时自适应减少稀疏字典中的原子数量。其次在线工作阶段中根据配电设备实时监测信号的稀疏系数改变矩阵健康阈值,对实时监测信号进行异常识别;正常信号用于字典的在线更新以提升重构精度,对异常信号则确定信号异常区域,对异常区域进行单独的压缩重构以实现对异常区域的精准还原。最后,将该算法与传统K-奇异值分解算法、正交匹配追踪算法进行了对比验证,结果表明该算法稀疏字典原子数量更少、运算速度更快、重构结果的峰值信噪比较大、均方根误差较小,具有更高的重构精度,且在压缩重构信号的同时能够对其进行有效的异常识别。
Under the background of the promotion of power internet of things(IOT)sensing technology and the construction of intelligent distribution network,the monitoring signal data of distribution equipment are large and various.In order to effectively alleviate the pressure of data transmission and storage,a compressed sensing and anomaly recognition algorithm of distribution equipment monitoring signal based on power IOT sensing is proposed.Firstly,in the off-line training phase,an adaptive singular value decomposition algorithm based on dynamic threshold is proposed to reduce the number of atoms in the sparse dictionary while ensuring the reconstruction accuracy.Secondly,in the online phase,according to the sparse coefficient of the real-time monitoring signal of the distribution equipment,the health threshold of the matrix is changed to identify the abnormal of the real-time monitoring signal.The normal signal is used to update the dictionary online to improve the reconstruction accuracy.For the abnormal signal,the abnormal region is determined,and the abnormal region is compressed and reconstructed separately to realize the accurate restoration of the abnormal region.Finally,the proposed algorithm is compared with the traditional K-singular value decomposition algorithm and orthogonal matching pursuit algorithm.The results show that the proposed algorithm has fewer atoms in sparse dictionary,faster operation speed,higher peak signal-to-noise ratio,smaller root mean square error and higher reconstruction accuracy.Besides,the proposed algorithm can recognize the anomaly effectively while it compresses and reconstructs the signal.
作者
王艳
李煜
赵洪山
王龄婕
赵一宇
WANG Yan;LI Yu;ZHAO Hongshan;WANG Lingjie;ZHAO Yiyu(Electrical and Electronic Engineering College,North China Electric Power University,Baoding 071003,China)
出处
《高电压技术》
EI
CAS
CSCD
北大核心
2022年第1期11-19,共9页
High Voltage Engineering
基金
国家自然科学基金(51807063)
中央高校基本科研业务费专项资金(2021MS065)。
关键词
性能监测
压缩感知
稀疏字典
稀疏系数矩阵
异常识别
performance monitoring
compressed sensing
sparse dictionary
sparse coefficient matrix
anomaly recognition
