In the process monitoring and fault diagnosis of batch processes, traditional principal component analysis (PCA) and least-squares (PLS), are assuming that the process variables are multivariate Gaussian distribution. But in the practical industrial process, the data observed of process variables do not necessarily be the multivariate Gaussian distribution. Independent component analysis (ICA), as a higher-order statistical method, is more suitable for dynamic systems. Observational data are decomposed into a linear combination of the independent components under statistical significance. The higher order statistics will be extracted and the mixed signals are decomposed into independent non-Gaussian components. Traditional method of ICA has to predefine the number of independent components. This paper proposed an improved MICA method of realizing the automatically choosing the independent components through setting the threshold value of the negentropy. The method can solve the problem of predefining the number of independent components in traditional methods and meanwhile it reduces the complexity of the monitoring model. The proposed method is used to do the process monitoring and fault diagnosis of penicillin fermentation and the results verify the feasibility and effectiveness of the method.
A User-Adaptive Algorithm for Activity Recognition Based on K-Means Clustering, Local Outlier Factor, and Multivariate Gaussian Distribution
