Independent neural network modeling of class analogy for classification pattern recognition and optimization

Abstract

An independent neural network modeling of class analogy (INMCA) has been proposed as a classification pattern recognition method, which combines the idea of the classical soft independent modeling of class analogy (SIMCA) with the back-propagation neural network (BPN). The INMCA can not only exclude noise samples and select useful features in the multivariate calibration of complicated chemical processes, but also provide the class centers in the non-linear space for optimization of a chemical process. The data processing of a silicon steel process, as an application example, shows this INMCA to be useful.