Feedforward Neural Networks In Adaptive Inverse Control

In design of control system and regulator, Adaptive Inverse Control uses inverse of plant transfer function to control dynamic performances of system. It is an open-loop control method, so instability problem caused by feedback can be avoided. In addition, dynamic control and disturbance rejection can be individually dealt and don't affect mutually. Linear filter method is generally used to adaptive inverse control of linear plants and neural network control is used to nonlinear systems. Neural network can be widely applied to study the processes and systems described difficultly by model or formula as it has large-scale parallel structure, good ability of dealing with nonlinear and uncertain problems. Consequently the inverse control base on neural network has been developed.The thesis is supported by the Natural Science Foundation of Henan Province entitled "Robust design for adaptive inverse control systems using delta operator" and Foundation for University Key Teacher by Henan Province entitled "Delta operator approach to adaptive inverse control design and performance analysis". Basic theory of adaptive inverse control is firstly introduced. Then the problems of theory and application of adaptive inverse control based on neural network are considered. The major results are as following.(1) Adaptive filter based on feed-forward neural network is used to model nonlinear plant. The simulations show this method is much better than that of using linear filter.(2) The convergence property and quadratic stability of ε-filter and LMS algorithm are analyzed. Filtered-ε LMS algorithm using feed-forward neural network is also given. Simulation shows that the algorithm has better control effects.(3) Adaptive inverse control based on feed-forward neural network is applied to cancel earphone noise. Simulation results show that the proposed method has better performance of canceling earphone noise than that of using the linear filter.