The Research Of Nonlinear Observer Based On Quantum-behaved Particle Swarm Optimization

The design of observers for nonlinear systems has become an important issue of nonlinear feedback control theory and recently received considerable attention. Due to the shortcomings of the traditional design methods, in this thesis, some intelligent algorithms are employed for the design problem. After making a deep research on the design strategies of nonlinear observers based on dynamic recurrent neural network and moving horizon estimation method, we design and implement the intelligent observers based on two kinds of particle swarm optimizers.It is shown by the study on the principle of traditional observer design methods that the design of the normalized observer must be satisfied with extremely restrictive conditions, which confines the application area of the normalized observer. Extended linearization method is to linearize the given system and then treat it as a linear system. Because the selection of Lyapunov function is lack of construction Lyapunov design method, it can only be used as a tool for proving the stability of the observer. Thus we design the nonlinear observers with dynamic recurrent neural network and moving horizon estimation method.The dynamic recurrent neural network is a promising modeling approach, which become a attractive branches of neural network modeling, system identification and controlling. The nonlinear observer based on dynamic recurrent neural network has ability of approximating to the trajectory of a nonlinear system under a certain initial condition. A proper network weigh training algorithm is key to design such an observer. In this thesis, we employ two version of particle swarm optimization algorithm to train the neural network and the experiment results show that the algorithms can converge faster that BP network training algorithm and the resulted system has better performance.The observer based on moving horizon estimation is a design approach based on optimization algorithms, by which a cost function is minimized on a interval [t-T,t] to estimate the state of the nonlinear system. This method is simple and easy to implement and can be applied in many fields. The key problem is to select an efficient optimization algorithm. In this thesis, we employ two Particle Swarm Optimization algorithms to solve the design problem and the simulation results show that both two algorithms has fast convergence rate and high precision.