Application of Kernel-Based Methods to Soft Sensor Modeling of Selectivity to Acrylonitrile

Principles of kernel-based methods and two kernel-based statistical modeling techniques are introduced. Soft sensor modeling of selectivity to acrylonitrile using kernel PLS and kernel PCR is proposed to cope with the nonlinearity and multi high dimension of input and collinearity problem of process. It is found that the performance of kernel-based methods is superior to linear statistical model and that of kernel PLS is also superior to kernel PCR.