基于FPGA的永磁同步电机神经网络解耦控制设计与实现

发布时间：2018-03-20 07:39

本文选题：永磁同步电机　切入点：神经网络　出处：《电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：近年来,交流传动成为了工业电气传动的主要研究与应用的方向。而永磁同步电机(Permanent Magnet Synchronous Motor,PMSM)凭借着体积小、质量轻、功率密度大、低速输出转矩大、效率高以及维护简单等优点,一直作为电气传动方向理论与应用研究的热点。对于这样一个复杂的非线性控制对象,参数动态变化和内部状态耦合使得传统的控制方法并不能得到极佳的控制和解耦效果,因而一系列针对PMSM的智能控制研究应运而生。神经网络逆系统控制就是其中一种可行的方案。神经网络具有以任意精度逼近非线性对象的能力,结合逆系统的方法对永磁同步电机实现解耦控制,具有良好的效果。但神经网络并行处理的特性,导致它并不适合在一般的处理器上实现。基于FPGA的神经网络实现可以充分发挥这一特性,解决这一问题。本文在PMSM的神经网络解耦控制的理论基础上,针对神经网络逆系统进行了建模与仿真,验证了神经网络逆系统模型的可行性。得出了性能良好的神经网络逆系统模型的结构与参数。并进一步对神经网络逆系统模块的硬件实现方法进行了研究。将FPGA作为实现平台,构建了多种神经网络激励函数模块,比较了不同实现方法的特点。并在前面研究的基础上完成神经网络FPGA模块的建立、仿真与验证。在PMSM神经网络逆系统的模型建立过程中,根据对PMSM数学模型的分析,进行了永磁同步电机可逆性的相关推导,并建立了逆系统模型。在此基础上,依据神经网络逆系统原理,建立神经网络模型,并通过Matlab训练得到理想的神经网络模块的结构与参数。将训练好的模块带回PMSM神经网络逆系统模型中验证,反映出良好的性能效果。利用Matlab和System Generator对神经网络实现的算法进行仿真实验,在此过程中利用基于查找表的线性逼近方法和CORDIC算法分别设计了Sigmoid激励函数和Gauss激励函数,并利用建立好的激励函数完成神经元的设计与测试。按照神经网络训练实验中得到的神经网络的结构与参数建立神经网络FPGA模型,并利用解耦系统仿真获得的样本进行仿真与验证。
[Abstract]:In recent years, AC drive has become the main research and application direction of industrial electric drive, while permanent Magnet Synchronous Motor (PMSM) is a kind of permanent magnet synchronous motor with small volume, light weight, high power density and large output torque at low speed. The advantages of high efficiency and simple maintenance have always been the focus of research on the direction theory and application of electrical transmission. For such a complex nonlinear control object, Because of the dynamic change of parameters and the coupling of internal state, the traditional control method can not get the excellent control and decoupling effect. As a result, a series of intelligent control studies for PMSM have emerged. Neural network inverse system control is one of the feasible schemes. Neural network has the ability to approach nonlinear objects with arbitrary precision. The decoupling control of permanent magnet synchronous motor (PMSM) based on inverse system method has good effect. Therefore, it is not suitable to be implemented on a general processor. The neural network implementation based on FPGA can give full play to this characteristic and solve this problem. This paper is based on the theory of decoupling control of PMSM neural network. The neural network inverse system is modeled and simulated. The feasibility of the neural network inverse system model is verified. The structure and parameters of the neural network inverse system model with good performance are obtained. Furthermore, the hardware implementation method of the neural network inverse system module is studied. The FPGA is used as the implementation platform. Several kinds of neural network excitation function modules are constructed, and the characteristics of different implementation methods are compared. The establishment, simulation and verification of the neural network FPGA module are completed on the basis of the previous research. In the process of modeling the PMSM neural network inverse system, the model of the neural network inverse system is built. According to the analysis of PMSM mathematical model, the reversibility of PMSM is deduced, and the inverse system model is established. On the basis of this, the neural network model is established according to the principle of neural network inverse system. The structure and parameters of the ideal neural network module are obtained by Matlab training. The trained module is brought back to the inverse system model of PMSM neural network to verify. Matlab and System Generator are used to simulate the algorithm of neural network. In the process, the Sigmoid excitation function and Gauss excitation function are designed by using the linear approximation method based on look-up table and CORDIC algorithm, respectively. According to the structure and parameters of neural network training experiment, the neural network FPGA model is established. And the samples obtained by decoupling system simulation are simulated and verified.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM341

【参考文献】