Adaptive Learning Control For A Class Of Nonlinear Systems And Its Applications

Adaptive learning control, which can online estimate the unknown information toimprove the control performance of the system, is a very effective method for dealing withsystem uncertainties. This thesis proposes the optimal control methods for uncertain systemsbased on adaptive reinforcement learning. This thesis mainly studies the following twocontents.(1) Firstly, an adaptive neural network (NN) control algorithm is designed for a class ofrobot manipulator systems based on the reinforcement learning theory. The neural networksare used to approximate the system uncertainties. The designed controller makes theperformance function reach a minimal value by using gradient descent method. However, thismethod can not guarantee the stability of the closed-loop system, when the dead zone inputexists. Therefore, we also study the optimal control problem for a class of robot manipulatorsystems with dead zone. It is assumed that the parameters of dead zone are unknown butbounded. Similarly, an optimal controller is proposed by using reinforcement learning theory.Finally, based on the Lyapunov analysis theory, all the signals of the closed-loop system areproved to be bounded. Simulation results show the effectiveness of the approach.(2) An adaptive control algorithm is designed for a class of chaotic systems with outputconstraints. The Barrier Lyapunov function approach is successful used to prevent the outputfrom violating constraint conditions. Then, an adaptive NN optimal control problem is studiedfor a class of discrete time chaotic systems by using reinforcement learning theory. Theperformance index function can reach the minimal value. Compared with the existing controlmethods of discrete chaotic systems, the optimal control problem is solved and the cost of thedesign controller is reduced. The closed-loop system is proved to be stable based on theLyapunov theory. Simulation results demonstrate the effectiveness of the proposed method.