摘要
针对化工过程数据的多尺度性和非线性特性,提出了改进多尺度核主元分析法。先利用小波变换分析测量数据的多尺度特性,然后采用核主元分析算法进行在线故障检测,对检测到的故障采用核函数梯度算法实现在线故障诊断,根据每个监控变量对统计量T2和SPE的贡献程度,绘制贡献图,用于故障的分离。在监控过程中为解决核矩阵计算困难,引入特征向量选择方法。TE过程的仿真结果表明它能有效实现故障检测、故障诊断,与主元分析方法相比,显示出更高的过程监控能力。
An improved multi-scale kernel principal component analysis method is proposed for analyzing the multi-scale and nonlinear property of chemical data.Wavelet transform is used to analyze the multi-scale property of the measurement data, while kernel principal component analysis algorithm is used to realize online fault detection. Using the gradient of kernel function, KPCA contribution plots are protracted, which represent the contribution of each monitoring variable to the statistics T2 and SPE. During the monitoring process, feature vector selection method is given to reduce the computation complexity of the kernel matrix. To demonstrate the performance, the proposed method is applied to TE process. Simulation results show that the improved MSKPCA effectively detects and diagnoses faults. Compared with PCA, the proposed method shows superior process monitoring performance.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2010年第1期51-55,共5页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金重点资助项目(60234010)
航空科学基金项目(05E52031)
江苏省高校自然科学基础研究面上项目(08KJD510016)
江苏省高校自然科学基础研究(09KJ13510005)资助项目
关键词
小波变换
核主元分析
故障检测
故障诊断
特征向量选择
贡献图
wavelet transform
kernel principal component analysis
fault detection
fault diagnosis
feature vector selection
contribution plot
