Nonlinear PI controller design: a neural network approach

Abstract

This article uses a hierarchical, multilayered neural network to provide parameters for a nonlinear PI controller in response to local operating conditions. The Generalized Delta Rule is adopted for use in training the connective weights of the network for subsequent on-line variation of the network-based PI controller parameters during control. Several numerical examples and one example of a highly nonlinear neutralization process are supplied to demonstrate the superior servo as well as regulatory control performance of the proposed neural net-based nonlinear PI control system.