Compound dimensionality reduction based multi-dynamic kernel principal component analysis monitoring method for batch process with large-scale data sets

To reduce and eliminate the three problems for large-scale data sets that include high computational complexity, large storage space, and long time-consuming in complex batch process with inherent dynamics and nonlinearity, a novel approach based on multi-dynamic kernel principal component analysis (MDKPCA) by exploiting compound dimensionality reduction for fault detection is proposed. The method firstly uses discrete cosine transform (DCT) having strong energy aggregation and distance preserving property to realize dimensionality reduction without changing the essential characteristics of data. Then after the reduced dimension data is processed by inverse transformation, the dynamic kernel principal component analysis (DKPCA) model is established by combining the autoregressive moving average time series (ARMAX) model and kernel principal component analysis (KPCA) to handle the nonlinearity and dynamics in industrial process. Finally, one penicillin fermentation process case for fault monitoring is provided to test the effectiveness of the proposed method, where the comparison with multiway kernel principal component analysis (MKPCA) results is covered.