A proposed model reference adaptive neural network controller based on Taylor development of nonlinear discrete system

This paper presents a developed model reference adaptive neural network control. It illustrates how the derivative function affects directly training speed. Indeed, the classical derivative of the sigmoid activation function decreases the learning speed because a large amount of error is not proportionally propagated back to the weights of the output layer. To overcome this difficulty, this paper proposes a Taylor development of the activation function which is capable of speeding up the learning process significantly and it can provide simultaneously stability of the learning process. The results of simulation show that using Taylor development has better effects both on response time and on tracking performance.