Quality prediction based on sub-stage LS-SVM for batch processes

For multistage, nonlinear characteristic of batch process, a substage least square support vector machines (LS SVM) method is proposed for quality prediction. Firstly, using an clustering arithmetic, PCA P loading matrices of time slice matrices is clustered according to relevance and batch process is divided into several operation stages, the most relevant stage to the quality variable is defined, and then applying correlation analysis in unfold stage data in order to get irrelevant input variables, and sub stage LS SVM models are developed in every stage for quality prediction. The proposed method easily handles the following problems: static single model; process and its model do not match; linear method may not be efficient in compressing and extracting nonlinear process data. For comparison purposes a sub MPLS quality model was establish. The results have demonstrated the effectiveness of the proposed method.