基于稳健非负矩阵分解的用电数据清洗和插补  

Electricity Consumption Data Cleansing and Imputation Based on Robust Nonnegative Matrix Factorization

作　　者：刘清蝉 钟尧 林聪 李腾斌 杨超 付志红[1] 李昕泓 LIU Qingchan;ZHONG Yao;LIN Cong;LI Tengbin;YANG Chao;FU Zhihong;LI Xinhong(School of Electrical Engineering,Chongqing University,Shapingba District,Chongqing 400044,China;Measurement Center of Yunnan Power Grid Co.,Ltd.,Kunming 650000,Yunnan Province,China)

机构地区：[1]重庆大学电气工程学院,重庆市沙坪坝区400044 [2]云南电网有限责任公司计量中心,云南省昆明市650000

出　　处：《电网技术》2024年第5期2103-2112,I0084,共11页Power System Technology

摘　　要：针对运行工况下用电数据在采集和传输过程中通常存在噪声、异常值和丢失的数据质量问题,利用单一用户用电数据时空分布的低本征维和异常值的稀疏特性,提出一种基于低秩矩阵完备的数据缺失填补、降噪和异常值剔除统一处理方法框架。首先,鉴于实际中多用户用电场景和用电特征差异巨大,仅根据单一用户用电行为的内在相似性构建具有低秩特征的数据矩阵;进而,考虑列异常和稀疏异常等加性背景噪声影响,构建低秩提升正则约束的非负矩阵完备最优化模型;最后,采用交替迭代最小二乘法方式进行最优化问题求解,实现缺失数据填补和多重背景噪声消除。通过仿真分析和实验结果验证了算法的有效性和准确性。Given the quality problems of noise,outlier,and loss in the collection and transmission of electricity consumption data under operating conditions,low eigenvalue and sparse outlier of the spatiotemporal distribution of electricity consumption data of a single user are used,and then a unified processing framework for data missing filling,noise reduction,and outlier elimination based on the low-rank matrix completion is proposed.Firstly,because of the huge differences in actual multi-user electricity consumption scenarios and characteristics,a data matrix with low-rank characteristics is constructed only according to the inherent similarity of a single user's electricity consumption behavior.Furthermore,considering the effects of additive background noise such as column and sparse anomalies,a non-negative matrix complete optimization model with low rank lifting regular constraints is constructed.Finally,the iterative least square method is used to solve the optimization problem to fill in the missing data and eliminate the multiple background noise.The effectiveness and accuracy of the proposed algorithm are verified by simulation and experimental results.

关 键 词：异常 稀疏 低秩 非负矩阵 清洗 插补 

分 类 号：TM721[电气工程—电力系统及自动化]