摘要
复杂化工过程采集到的数据往往夹杂着过程噪声,如何去除冗余数据、充分提取数据的有效信息是研究重点。提出了一种融合序列相关与低秩表征(LRR)的信息提取算法——序列低秩嵌入(SLRE)。通过LRR对训练样本进行低秩分解,剔除噪声点,去除数据中的冗余信息,增强了算法的鲁棒性。为了保持数据的全局-局部特征,通过计算样本间相关系数构造加权矩阵,并利用嵌入算法实现数据降维。建立T2和SPE统计量,使用核密度估计(KDE)方法估计控制限。通过数值仿真实例和田纳西-伊斯曼(TE)过程验证了本文方法的有效性。
With the rapid development of science and technology,the increasing requirements for modern industry have been receiving more and more attentions.The traditional multivariate statistical process monitoring(MSPM)methods,e.g.,principal component analysis(PCA)and partial least square(PLS),have been widely studied.However,the data collected in complex chemical processes are often mixed with process noise.It is a hot topic in industry process monitoring how to reduce redundant data and extract the effective information of data.By fusing sequence correlation with low rank representation(LRR),this paper proposes an information extraction algorithm,sequence low rank embedding(SLRE).Firstly,the training samples are decomposed by LRR,which can divide the training data matrix into a low rank part and a sparse residual part.The sparse residual part contains most of the noise of the data.By removing this part,the noise points can be largely eliminated from the sample data and the redundant information in the data can be removed such that the robustness of the algorithm can be enhanced.Meanwhile,by calculating the inter-sample correlation coefficient,the weighted matrix is constructed such that the global-local characteristics of the data can be maintained. Moreover,local weighting can effectively guarantee the integrity of the data information in the modeling and fully extract the information contained in the training dataset.Furthermore,the embedding algorithm is used to achieve the reduction of data dimensionality.And then,the T^2 and SPE statistics are established and the control limit is estimated using the kernel density estimation(KDE)method.Finally,the proposed SLRE algorithm is tested via a numerical example and the Tennessee Eastman process.It is shown from the experimental results that SLRE is feasible and effective for dealing with the problem of data noise and information redundancy,by which the characteristics of low dimension and low rank can be ensured and a more accurate process monitoring model can be obtained.
作者
张明善
杨健
侍洪波
ZHANG Ming-shan;YANG Jian;SHI Hong-bo(Key Laboratory of Advanced Control and Optimization for Chemical Processes,Ministry of Education,East China University of Science and Technology,Shanghai 200237,China)
出处
《华东理工大学学报(自然科学版)》
CAS
CSCD
北大核心
2018年第3期390-396,462,共8页
Journal of East China University of Science and Technology
基金
国家自然科学基金(61374140
61673173)
中央高校基本科研业务费重点科研基地创新基金(222201717006)
关键词
过程监控
低秩表征
过程控制
序列相关
故障检测
process monitoring
low rank representation
process control
sequence correlation
faultdetection