基于MFA-ICA的间歇过程在线监控

Batch process monitoring based on MFA-ICA

针对间歇过程独特的数据特点,提出1种将因子分析(FA)作为独立成分分析(ICA)白化预处理手段的多向因子分析白化独立成分分析(multiway factoranalysis-independent component analysis,MFA-ICA)间歇过程监控方法。因子分析充分考虑了模型误差的普遍意义,拥有优秀的噪声建模能力。将其代替主成分分析用于白化,可以更好的提取数据集的本质信息。首先将间歇过程三维数据依次按批次和变量展开得到二维数据矩阵,接着把上述方法用于展开后的数据,利用ICA的I^2统计图实现在线故障检测。该方法用于标准仿真平台Pensim,结果表明上述方法对于提高间歇过程故障检测的快速性,降低漏报率有明显效果。

A novel batch process monitoring approach based on Multiway FA-ICA was proposed.In MFA-ICA,PCA was substituted by FA as pre-whitening method of ICA.PCA is only a special case of FA,as FA take full consideration of general sense of model error.So FA have a better ability of noise building and has a better application prospect.In this article,firstly,batch process three-way data was unfolded into two-way data batch-wisely then variable-wisely.Later MFA-ICA was used to extract independent sources.At last,use I^2 statistic of ICA to achieve online fault detection.The application in the Pensim benchmark process shows this method can increase the accuracy and sensitivity of fault diagnosis.