Remarks on quaternion neural network based controller with application to an inverted pendulum

In this paper, a quaternion neural network trained by using back-propagation algorithm is applied to controlling an inverted pendulum as a part of utilizing the quaternion neural network for dynamic system control. Experimental results show that the quaternion neural network can control the inverted pendulum with a large initial angle which a state feedback controller can not handle. Moreover, the quaternion framework makes the neural computing more efficient. In the control problem of the inverted pendulum, the quaternion neural network finishes its learning with a fewer number of trails compared with the conventional real number neural network which has a more complex network topology and more parameters in real number being employed. These results indicate the feasibility of using the quaternion neural network for controlling dynamic systems.