数据驱动PCA、ICA和KICA故障检测仿真比较

Simulation comparisons among PCA,ICA and KICA fault detection based on data-driven

针对同一非线性过程数据,分别应用多元统计过程监控技术中的主元分析(Principal Component Analysis,PCA)、独立元分析(Independent Component Analysis,ICA)和核独立元分析(Kernel Independent Component Analysis,KICA)三种方法进行了故障检测建模,并进行了仿真比较研究。通过田纳西-伊斯曼过程(Tennessee Eastman Process,TE)仿真结果表明:在处理实际工业生产中非线性、非高斯的数据方面,KICA方法具有更强的故障监测能力。

Under the identical nonlinear-data conditions, three multivariate statistical process methods-PCA, ICA and KICA are used to monitor the fault detection to compare their difference through simulation. For the characteristics of sample data, three different methods have their own characteristics. By using Tennessee Eastman process, the simulation results indicate that KICA method is superior to the other two methods because it can deal well with the actual industrial processes whose sample data are nonlinear and non-Gaussian.