Research Of The PMSM Control Based On The Fuzzy Neural Network Control Algorithm

In contrast to traditional synchronous motor of electrical excitation, permanent-magnet synchronous motor(PMSM) has more advantages, especially the occasion that requires high precision of control and high reliability, such as avigation, numerical control machine tool, processing center, robot and so on. In addition, it plays an important part in AC motor in modem times.Fuzzy neural networks combine the advantages of fuzzy logic and neural networks, so it not only can expresses the qualitative knowledge, but also have the ability of self-studying and processing quantitative data. Accordingly, it's hopeful to design a control strategy of high quality to acquire high precision control by applying fuzzy neural networks to control of PMSM servo system. Thus, this paper mainly research the application of fuzzy neural networks to PMSM control.First of all, this paper begins with the establishment of PMSM simulation stage, and fully applies environment of Matlab and Simulink to that. The motor simulation is finished in different PWM by using S function, PMSM module of power system toolbox, and mode of vector control, for the purpose of laying a strong emphasis on system A&D instead of program.Secondly, this paper proposes a neural network training scheme based on the linear least-square method through the characteristic analysis of multi-layered feed forward neural networks, and then applies that to identify a PMSM model. In this scheme, the inputs of hidden layer neurons are acquired by using the gradient descent method, and the weights and threshold of each neuron are trained using the linear least square method. The simulation results show that this scheme has characteristics of high precision and high rate of convergence.Thirdly, This paper proposes a nine-point five-state controller based on the nine-point controller, and then applies its control rules to structure design of conventional fuzzy neural networks, As a result, the structure design is greatly simplified. Parameters of fuzzy neural networks are rectified more quickly by maintaining the consistency of fifth layer weights in the fuzzy neural networks toproportional parameters in the nine-point five-state controller. At the same time, the weights of fifth layer are endowed real meaning. The simulation results prove that the controller has better effect and is easily rectified.At last, this paper designs a new learning algorithm of adaptive control based on improvement of conventional fuzzy neural networks, and then applies that to control of position in a PMSM system. As a result, expected control effect is acquired.