Non-linear Pole-placement Control with a Neural Network

Abstract

A neural network application for non-linear control is described. Mostly, neural networks are considered to produce a non-parametric model of a non-linear process. However, it is possibly to extract, very easily, a so-called gain matrix from a trained neural network model, and a partitioning of this gain matrix allows on-line estimation of the actual relevant parameters. This feature is used for a non-linear gain scheduling control of a noisy, non-linear, SISO process based on a pole-placement control design. The advantage is that the controller parameters can be changed quickly in response to process changes.