Remarks on System Identification Using a Quaternion Recurrent Neural Network Trained by Backpropagation through Time

Abstract

This study investigates the learning capability of a quaternion recurrent neural network that is trained based on a backpropagation through time algorithm extended to quaternion numbers. Computational experiments to identify nonlinear systems, e.g. a three–dimensional chaotic system and discrete–time plant, were performed, and the simulation results confirmed the feasibility of using the quaternion recurrent neural network for a control system application.