Dichotomy-based optimal selection of KPCA kernel parameters

Kernel principal component analysis (KPCA), by aid of kernel functions, transforms the nonlinear problems in the input space into linear problems in the feature space, which has currently been widely adopted in fault detection by nonlinear process industry. The adoption of KPCA algorithm for fault detection finds the selection of kernel parameters in the kernel function an important factor affecting the accuracy and reliability of the detection results. Yet in practical applications, the values of kernel parameters are mostly selected, relying priamarily on one’s experience or cross-validation method, which fequently requires repeated adjustments of the value of the kernel parameters.. This not only hinders the automation and intelligence of fault detection, but also makes it difficult to ensure that the selected kernel parameters are the optimal values, thus to affect the final performance of fault detection. Therefore, this paper, based on the idea of dichotomy, presents an optimized method of kernel parameters selection in KPCA and applies it to the fault detection of TE process. Experimental results find that the algorithm can effectively solve the kernel parameter optimization problem of KPCA, and ensure the accuracy and reliability of fault detection results.