Remarks on a Recurrent Quaternion Neural Network with Application to Servo Control Systems
Kazuhiko Takahashi · 2018
This paper investigates the control system application of a fully connected recurrent neural network in which all network parameters and signals are expressed in quaternion numbers, and the training of the network is conducted using a real-time recurrent learning algorithm. The recurrent quaternion neural network (RQNN), which synthesises the control input to track the outputs of the non-linear system to the desired outputs, assumes the role of an adaptive-type servo controller in the control system. A feedback error learning method is used to train the RQNN using an online method in the control system. Numerical simulations for controlling discrete-time non-linear plants are performed to evaluate the characteristics of the RQNN-based adaptive-type controller. The simulation results demonstrate the feasibility and effectiveness of the proposed controller.