Improved Adaptive Neuro-control Algorithms And Simulation Studies

This thesis aims to improve adaptive control schemes proposed by Professor Wang Dianhui for controlling unknown nonlinear systems using neural network. Based on the new cost function proposed in [26] for on-line modeling, a further study on improvement of the Jacobian teacher signals is given in this paper. Impact of this improved Jacobian teacher signals on both the control performance of adaptive inverse dynamics scheme and adaptive neural networks model based predictive control scheme is investigated.A mean-value filter (MVF) and a constrained linear filter (CLF) are employed to the Jacobian teacher signals proposed in [26], which further improved the quality of Jacobian teacher signals for online training of neural networks. Then we applied our proposed method to two adaptive neural control schemes, and found that the enhanced Jacobian teacher signals can help in achieving better control performance to some extent .Some theoretical aspects such as convergence of learning process and close-loop stability were revisited with the new cost function used in online modeling . Some simulations were carried out to illustrate the value of the proposed techniques in this work.