A Novel Structured Dynamic CCA Modeling for Process Monitoring

摘要

Canonical correlation analysis (CCA) has been widely applied for data modeling and process monitoring. However, traditional CCA is not appropriate for dynamic processes, since only cross-correlations are taken into consideration. The autocorrelations are ignored. In this paper, a novel structured dynamic CCA (SDCCA) model is proposed to explore the autocorrelation information between the structured input and output matrices for dynamic processes. The new structure can improve the modeling and the interpretation of dynamic processes and enhance the performance of monitoring. First, a novel optimization criterion is established to derive the loadings for inputs and outputs. Besides, the residuals of input and output which contain the dynamic information are estimated. Then, two vector autoregressive (VAR) models are used to build the inner model for the latent variables of input and output, respectively. Subsequently, the latent variables and residuals of inputs and outputs are further analyzed by CCA method, respectively. Two monitoring statistics are developed based on the results of CCA. Finally, the closed-loop continuous stirred-tank reactor (CSTR) is employed to illustrate the effectiveness of the proposed method.