Neural PCA controller based on multi-models

Abstract

In this paper, a new approach to design nonlinear adaptive PI multi-controllers, for SISO systems, based on neural local linear principal components analysis (PCA) models is proposed. The PCA neural networks only implements the integral term of the PI multi-controller, a proportional term is added to obtain a PI structure. A modified normalized Harris performance index is used for evaluating the controller performance. Some experimental results obtained with a nonlinear three tank benchmark model are presented, showing the adaptive PI-PCA multi-controller performance compared to neural linear PI controllers. © 2015 Springer International Publishing.