The Studying Of Wavelet Networks And Multiple Models Adaptive Control

The developments of wavelet networks and multiple models are reviewed, analyzed and summarized. Based on wavelet networks and multiple model adaptive control theories, the identification and control methods for the complicated nonlinear dynamic systems are proposed.A new learning algorithm of the continue wavelet networks parameters is proposed, that is, hybrid parameters learning algorithm. The parameters of wavelet networks are divided into two parts, and different methods are used to training them. Compare to the traditional parameters learning method, the new method proposed in this paper has the advantages of rapid convergence speed and high approximation capability. The properties of the wavelet networks are analyzed. According to the approximation ability of wavelet networks, the nonlinear static system and the nonlinear dynamic system can be identified. The parameters of the wavelet networks are trained by using hybrid algorithm. The wavelet networks are used in identifying the stator resistance of induction motor direct torque control. The simulation results show that this method is effective and feasible. The wavelet networks are used as controller to control the nonlinear system. The direct control and indirect control methods are studied. Compare to the traditional neural networks, the wavelet control systems have the advantages of high control precision and rapid convergence rate. The controllability and stability of wavelet networks are analyzed and proved.To reduce the modeling error in the piecewise linearization of nonlinear system, a multiple models wavelet networks identification method is proposed. Compare to the single wavelet networks, the multiple models wavelet networks can identify the variety of system immediately and accurately.Multiple models wavelet networks control method is presented. Several fixed models with one adaptive model are used as sub-models in this method. And the wavelet networks controllers are designed according to the corresponding identifiers. The stability of multiple models wavelet networks is analyzed. The advantages of this method are high identification precision, good control performance and strong tracking ability. The method is particularly suitable to the uncertainty, time-varied, complicated nonlinear system.