基于SVD-KDR算法的工业监测数据插补技术  

Missing Data Imputation Using SVD-KDR Algorithm in Industrial Monitoring Data

作　　者：陈帅 赵明[1] 郭栋[2] 林京[3] CHEN Shuai;ZHAO Ming;GUO Dong;LIN Jing(State Key Laboratory for Manufacturing Systems Engineering,Xi’an Jiaotong University,Xi’an 710049;Key Laboratory of Advance Manufacturing Technology for Automobile Parts(Chongqing University of Technology),Ministry of Education,Chongqing 400054;School of Reliability and Systems Engineering,Beihang University,Beijing 100083)

机构地区：[1]西安交通大学机械制造系统工程国家重点实验室,西安710049 [2]汽车零部件先进制造技术教育部重点实验室(重庆理工大学),重庆400054 [3]北京航空航天大学可靠性与系统工程学院,北京100083

出　　处：《机械工程学报》2021年第2期30-38,共9页Journal of Mechanical Engineering

基　　金：国家自然科学基金(51875434);汽车零部件先进制造技术教育部重点实验室开放课题基金(2019KLMT02)资助项目。

摘　　要：监测数据的完整性和可用性是工业大数据时代信息提取与知识发掘的前提和基础。然而由于采集中断、传输干扰、存储不当等诸多原因,监测数据的缺失问题在工业场景中频繁发生,严重影响数据价值密度。提出一种基于SVD-KDR的高精度、高鲁棒性缺失数据插补算法。该方法将一维工业数据转换为高维矩阵,弥补了传统方法直接从低维空间插补工业监测数据的维度局限。通过发掘插补过程中非缺失数据的低秩特性,借助奇异值分解理论(Singular value decomposition,SVD)建立了鲁棒性更强的SVD-KDR算法模型,有效减弱了缺失数据对参数估计精度的不利影响。试验结果表明,相比于传统插补算法,所提出算法在高缺失率下仍具有较高插补精度和稳健性。此外,该方法不仅能够有效恢复缺失数据的波形,而且能充分还原原始数据所蕴含的波动信息。提出的SVD-KDR算法可有效解决数据缺失问题,为工业大数据分析提供了数据恢复与信息处理工具。Integrity and availability of data is the basis for information extraction and knowledge discovery in the era of industrial big data. However, data missing frequently occurs in industrial scenarios due to various reasons, such as collection interruption, transmission interference and improper storage, which seriously affects the data value density. In view of this, an interpolation algorithm with high precision and high robustness based on the SVD-KDR algorithm is proposed. The one-dimensional industrial data is converted into a high-dimensional matrix, which makes up for the limitation of the traditional method of interpolating industrial monitoring data directly from the low-dimensional space. By exploring the low-rank characteristics of non-missing data in the interpolation process, a more robust SVD-KDR algorithm model is established with the help of singular value decomposition theory, which effectively reduces the adverse effect of missing data on parameter estimation accuracy. Experimental results show that the proposed algorithm has higher accuracy and robustness at high missing rates compared with the traditional interpolation algorithm. In addition, the proposed method can not only effectively recover the waveform of missing data, but also restore the fluctuating information behind the original data. The SVD-KDR algorithm can effectively solve the problem of missing data, and provides data recovery and information processing tools for industrial big data analysis.

关 键 词：缺失数据插补 相空间重构 奇异值分解 低秩逼近 

分 类 号：TH17[机械工程—机械制造及自动化]