Soft Sensing Modeling of Chemical Process Based on PCA-Interval Type-2 TSK Fuzzy Logic System Method

Abstract

ABSTRACT The critical issue in soft sensing modeling of the chemical processes is how to model multidimensional data with noise and strong nonlinearity. To handle the uncertainty of modeling and minimize the impact of uncertainty, a class of method using interval type-2 Takagi–Sugeno–Kang (TSK) fuzzy logic systems (FLS) combining the principal component analysis (PCA) algorithm are proposed. First, the linear principal components from the input variables of model can be effectively extracted by the PCA algorithm. The number of fuzzy rules is regarded as the clustering center, and the fuzzy c-means (FCM) clustering algorithm is then applied. The result of clustering centers is used as the centers of the fuzzy rule antecedent. Second, soft sensing models are established based on three types of interval type-2 TSK FLS, namely, A1-C1, A2-C0 and A2-C1, and the backpropagation(BP) algorithm is used to update the parameters of the antecedent and consequent membership function. To verify the effectiveness of the proposed method, different types of interval type-2 TSK FLS were applied to the soft sensing modeling prediction of three key product yields for industrial fluidized catalytic cracking unit. Experimental results confirm that, compared with other types of FLS methods and support vector machine, the employed Type-2 TSK FLS method with different types can achieve better prediction accuracy, among them, the A2-C1 TSK FLS method has higher modeling accuracy and faster convergence.