基于RBF神经网络-滑模观测器的PMSM无传感器矢量控制

Sensorless Vector Control for PMSM Based on Sliding Mode Observer Integrated with RBF Neural Network

为了解决传统滑模观测器方法应用在永磁同步电机无传感器矢量控制时所产生的抖振问题,使用RBF神经网络动态调节观测器的切换增益,即使其输入为传统滑模估计方案中的电流估计误差,输出为滑模增益;同时为了简化系统结构、提高方案可行性,将RBF神经网络设计为单输入单输出的结构,并将网络的学习和工作过程融合,使其在自身网络参数的不断优化中实时输出滑模增益,以增强系统鲁棒性。最后通过Matlab/Simulink软件对该系统进行建模仿真,并将该方法与传统滑模观测器方法进行对比。实验结果表明,该方案能够为矢量控制提供更加准确的转子位置及速度信息,提高了整个电机控制系统的稳定性。

In order to attenuate the chattering phenomenon in traditional sliding mode observer(SMO)and sensorless vector control of PMSM,a method which adaptively adjusts gain of the observer by means of RBF neural network was proposed,so the input is set to be the current error and the output is the switch gain of SMO.For purpose of simplifying structure of system and enhance feasibility of the proposed method,a single-input and single-output network by combining the neural network with the SMO was designed.What's more,procedure of training and working of the network has been integrated to meet demands of real-time work,which meanwhile enhances the robustness of the system.Lastly,simulation of the control system has been constructed on Matlab/Simulink to make a comparison between traditional SMO and proposed method.Both the simulation and experiments show that the proposed method could estimate position and speed of the rotor rather accurately and improve the stability of system at the same time.