Fuzzy Wavelet Network And Its Applications For Pmsm Control

Many complicated systems, especially nonlinear dynamic time-variable system, can not be defined by conventional methods which haven't been built effective mathematic model and control method. Recently, both fuzzy technology and neural network has become a very active branch in intelligence control theory, inspired by the capabilities of processing abstract information, strongly self-learning and parameter self-adjusting. In the meantime, wavelet analysis technology with local performance of time-frequency and multiresolution function has been rapidly developed in the past several years. So, based on the combination of wavelet analysis, fuzzy logical technique and neural network, this paper researches a new model, called fuzzy wavelet network (FWN).The internal model of the permanent magnet synchronous motor (PMSM) is a high order complicate system with parameter variable, nonlinear, strong coupling and multivariable. The motor model can be simplified to uncoupling control and keep fast response through the vector control technique. However, a high performance drive system not only requires fast response and high accuracy, but also has fast ability under unknown disturbance and parameter variable. So, this paper proposes high performance control method for PMSM based on the FWN technique.The research work of this paper includes the following respects:First, combined with wavelet analysis and fuzzy neural network, two fuzzy wavelet network models are introduced. One is to use wavelet function as the membership function, and the other adopts the wavelet neural network to the consequent parts of the fuzzy rules. The basic structures of the two FWN are given and requirement to the wavelet function are studies, respectively. The performances of FWN are researched. Motivated by wavelet network, three typical extending network models of FWN are presented. Furthermore, the properties of the extending FWN is proved when using B-spline function. From the view of the generalizing network, the initial method of FWN is introduced. Two learning algorithms of FWN are provided which is the back propagation (BP) algorithmand a mixed algorithm with extend kalman filter and the least square method. The approximation capabilities of the two FWN to nonlinear system are theoretically proved, so FWN can approximate any nonlinear function with high accuracy. Further, it builds the theory basis to the application of FWN in identification and control of nonlinear system. The simulations to the static system demonstrate that the proposed two FWN are not only reserved the multiresolution capability, but also have the advantages of simple structure, high approximation accuracy and good generalization performance to nonlinear system.Second, two dynamic recurrent fuzzy wavelet networks (DRFWN) are researched combined with dynamic recurrent neural network and the proposed static FWN. The approximation capability is proved. The BP learning algorithm for the identification of dynamic system is studied. The stability analysis of study algorithm is given. The simulation results verify the performance with multiresolution, simple structure, high approximation accuracy and good response for dynamic system.Third, the application of FWN in control system is studied. Two FWN controllers of nonlinear system are presented, the design method of two FWN controllers are researched and adaptive rate of online adjust parameter is given based on Lyapunov approach. The stability performance of the closed-loop system and the convergence of tracking error are proved. The simulation to the chaotic system and the cart-pole control system are provided to illustrate the good performance of the FWN controller.Finally, the application of FWN for PMSM is researched. The adaptive sliding controller with FWN for PMSM is proposed based on the mathematic model of the PMSM and the vector control theory. To compensate the effective of the disturbance and online track the output of the reference model, RCFWN is used as compensator utilized model reference adaptive technique. Moreover, using the mathematic model of magnetic oriented control and an extending mathematic model, the position control of PMSM is studied through the combination with sliding variable structure and optimum method. So, the robustness and the control accuracy are increased under load disturbance and the parameter variable of system, in the meantime, the system stability is proved. Further, the sliding servo system of the PMSM with FWN identification which is used to evaluate theunknown boundary is presented. This method can effectively overcome the influence of the parameter change and load disturbance to increase the robustness. Finally, the simulation and the experiments using TMS32OF240 motor control platform are completed for the proposed servo control systems of the PMSM.