Internal Model Control using Recurrent Neural Networks for Nonlinear Dynamic Systems

Abstract

This paper presents a control method using Recurrent Neural Networks (RNNs) in an Internal Model Control (IMC) framwork. and demonstrates their effectiveness of modelling and control for nonlinear dynamic systems. Unlike existing neural IMC design methods, the proposed control scheme consists of two stages. The first stage is that using two same structure of RNNs through iterative learning techniques to obtain a desired control signal to the unknown nonlinear systems. Then a RNN is used to generate the desired control signal within IMC structure. The algorithm for the RNNs is a real time iterative learning algorithm based on two-dimensional (2D) system theory. The simulation results demonstrate the proposed control method can drive unknown systems to follow the desired trajectories very well.