On the development of a neural network based orthogonal nonlinear principal component algorithm for process data analysis

Principal component analysis attempts to concentrate information from a data set by developing a small number of indices which describe the original information content. This paper presents an approach to the extraction of nonlinear principal components using neural networks. The basic autoassociative neural network approach used has been augmented to provide results that are consistent with features provided by linear principal component analysis. Most importantly, the orthogonal nonlinear principal components can be ranked to provide a subset of factors with minimal loss of information from the original data set. The method was applied to a data set from an industrial catalytic reforming unit. The results indicate that the proposed method can extract a reduced set of nonlinear principal components