Study On Adaptive Inverse Control And Application

The adaptive inverse control has made great progress since it was proposed 20 years ago. It uses the inverse of the controlled plant as the cascaded controller to control the system dynamics. It is noticed that this is an open-loop control scheme, so the instability caused by feedback control is avoided. It also can control the system dynamics and the plant disturbance separately. Therefore, the adaptive inverse control is superior to the conventional feedback control. The adaptive inverse control of linear systems has been studied extensively, but there still exists many issues to be tackled. The research of nonlinear adaptive inverse control systems is less by now and needs deeply investigated.This dissertation further studies adaptive inverse control strategy and discusses the followings: (i) variable learning rate LMS algorithm based linear adaptive inverse control; (ii) adaptive inverse control of linear systems with actuator saturation; (iii) nonlinear adaptive inverse control based on variable learning rate neural network BP algorithm and its application to induction machine control.The main works of the dissertation are as follows:Construct cost function, readjust the weights update law, and propose a novel variable learning rate LMS algorithm. The chosen range of its constant is much lager than the usual fixed learning rate LMS algorithm. Because the learning rate can be adaptively adjusted according to the different input signals, the overall convergence rate is speeded up. And also the control system is robust to the noises in input signals.For the actuator saturation problem, we reconfigure the basic framework of adaptive inverse control. In the process of building optimal controller, the saturator is viewed as one part of the plant model. Then we adjust the adaptive law based on the new frame and obtain a controller both satisfying the saturation and the control target.After studying the strategy of linear adaptive inverse control, we extend it to nonlinear system and use it to control the induction machine. For the nonlinear, multi-variable induction machine, the artificial neural network is introduced into the plant modeling and inverse modeling. In order to improve the training speed, we propose a variable learning rate BP algorithm. The better performances are achieved.The simulations are carried out with MATLAB software. The results show the effectiveness of the adaptive inverse control strategies.