基于概率PCA过程监控中遗失数据的重构(英文)

Estimation of missing data in process monitoring with probabilistic PCA

在实际工业过程中,PCA常常被用于数据重构。但是相比于概率PCA(PPCA),PCA无论在建模上还是在统计监控指标上都存在一些缺陷。基于此,本文提出一种基于PPCA的遗失数据重构方法。通过使样本数据点与其在PPCA模型上的投影点之问的距离最小,该方法能够有效地进行数据重构。此外,还分析了使样本数据白化值最小的数据重构方法。在田纳西-伊斯曼过程中的应用验证了其有效性。

In the real industrial process, PCA is often used for data reconstruction. However, comparing with Probebilistic PCA （ PPCA）, PCA suffers from some demerits in modeling and the monitoring indices. Due to these, the paper proposes a method to estimate missing data based on PPCA, in which a minimal distance between sample data and its projection on the PPCA model are employed to reconstruct the missing data. Furthermore, estimation of missing data by minimizing the whitened value of the sample is analyzed. The application of the method to Tennessee-Eastman （TE） process shows the validity of the proposed method.