Decomposition-based modeling algorithm by CCA-PLS for large scale processes

As the crucial part of predictive control, distributed modeling method is seldom studied due to the absence of efficient methods to system decomposition. In this paper, a process decomposition algorithm based on canonical correlation analysis (CCA) is proposed. The output variables of all subsystems are firstly determined by the process. And then the maximum correlation coefficient between the outputs of a subsystem and all the process variables are calculated. The variables corresponding to larger elements of axial vector extracted by the maximum correlation coefficient are selected as the input variables. After the decomposition, the sub-models are constructed by PLS algorithm and the final subsystem models are obtained. The proposed method is experimented in the modeling of typical Tennessee Eastman (TE) process and the result shows the good performance.