Remarks on Feedforward-Feedback Controller Using Simple Recurrent Quaternion Neural Network

Abstract

In this study, a simple recurrent neural network is designed for controlling nonlinear systems. All signals and parameters of the network are quaternion numbers, and the network is trained with a real-time recurrent learning algorithm. The control system is composed of a feedforward-feedback controller based on a recurrent quaternion neural network and a feedback controller to reconcile the plant output with the desired output. A feedback error learning method is used for the online training of the feedforward-feedback controller. The numerical simulations of controlling discrete-time nonlinear plants are conducted to evaluate the characteristics of the recurrent quaternion neural network-based controller. Simulation results show the feasibility and the effectiveness of the proposed controller.