Direct Adaptive Neural Control with Integral-Plus-State Action

Abstract

The paper applied a Recurrent Neural Network (RNN) model in two Integral-Plus-State (IPS) schemes of real-time adaptive neural control. The proposed control modify and extend a previously published direct adaptive neural control scheme with one or two I-control terms, so to obtain a neural, IPS adaptive, offset compensational and trajectory tracking control. The control scheme contains only two RNN models (systems identificator and IPS feedback controller) and not need a third feedforward RNN control model. The good performance of the adaptive neural IPS control is confirmed by comparative simulation results, obtained using a nonlinear multi-input multi-output plant, corrupted by noise. The results exhibits good convergence and noise resistance which not depend on the magnitude of the offset.