With the development of nonlinear control theory, especially feedback linearization, nonlinear state observer design and nonlinear system control based on state observer are becoming more and more important. By means of combining intelligent control idea with nonlinear system theory, state observer design methods and observer-based nonlinear system control algorithms have been researched deeply in this paper.For a class of nonlinear system with parameter uncertainty, sufficient conditions on existence of nonlinear state observer are proposed. Furthermore, analytic solutions on observer gain are given. By combining genetic algorithm with the idea of observer design based on Lyapunov theory, a directive approach of adaptive observer design is proposed. Firstly, observer design problem is transformed to satisfiable (SAT) problem. Then further optimization is done for obtaining observer gain so as to achieve better performance of the observer. Simulations on CSTR system show the validity and practicability of the proposed approach.Dynamical recurrent neural network(DRNN) is showing good promise in modeling, identification and controlling of nonlinear systems. DRNN is not only applied to identification of nonlinear system but also used as nonlinear state observer. This paper has proved that DRNN-based observer has the ability of approximation of nonlinear states on some appropriate initial conditions. Both off-line dynamical backpropagation (DBP) algorithm and real-time fast learning algorithm are proposed, the latter is used to improve the real-time performance of system identification and state observer via DRNN. By analyzing model of neural network, programs of neural network designed with Oriented Object Programing(OOP) are proposed.As field-oriented induction motor control is considered in this paper, DRNN-based rotor flux adaptive observer is proposed, and adaptation is performed on the basis of relations between stator current and rotor flux. Simulations show that the proposed observer can track rotor flux vector in a large speed range and be robust to the variable rotor time constant.State space model of nonlinear system is reconstructed via DRNN, and states in the network are used for feedback. When input-output mapping of the neural network is close to the nonlinear system enough, decoupling control of the system is achieved. Through identification of induction motor via DRNN, equivalent model with double inputs and double outputs are achieved. Furthermore the motor is decomposed into two linear subsystems, rotor speed and stator flux, by means of state feedback. Simulations show that adaptive control based on DRNN has good performance of speed tracking even in the presence of disturbances such as load torque. |