Neural networks based an inverse dynamic model adaptive control

Abstract

Effective control of a complex system can often be obtained using a neural network controller. However, there are some difficulties in practical use of real-time system applications of neural network controllers because of their need for a learning strategy. A neural network inverse dynamic model adaptive control scheme is proposed in this paper, based on a feedback error learning method. In particular, online adjustment of the error-learning coefficient is provided in order to improve the performance when there are uncertainties over the values of some of the system parameters, and the saturation compensation approach is proposed to overcome the drawback of the vanishing training during the activation function saturation phases. Simulation experimental results are given to verify the robustness and real time behaviour of the proposed control scheme outperform the other traditional scheme.