Fault detection based on EWDKPCA-LSSVM

Aiming at the nonlinear problem of data in industrial process and the problem that dynamic conditions are difficult to detect faults in complex operation process. In order to solve these problems, this paper combines the exponentially weighted dynamic kernel principal component analysis (EWDKPCA) method with the least squares support vector machine (LSSVM) method. Firstly, the EWDKPCA method is used to detect faults in industrial processes. EWDKPCA can update the kernel principal component model in real time, and the control limit varies according to the number of principal components. Then, in order to further improve the detection accuracy, LSSVM is used for secondary fault detection. The Tennessee-Eastman (TE) chemical process data are used for simulation experiments. The experimental results show that the proposed method has high fault detection accuracy.