基于神经网络的感应电动机特性辨识新方法

提出并论证了一种基于神经网络的感应电动机特性辨识新方法,只需测得电机两相电流数值便可以辨识出电动机转矩和转速,用改进的Levenberg-Marquardt算法对神经网络进行学习和训练,构建了适合电动机转矩转速观测的BP神经网络。由于RBF神经网络无论是在逼近能力、函数拟合和学习速度方面都优于BP网络,也利用RBF网络进行了辨识。该方法较已经提出的方法相比,需要的检测量少,辨识方法简单。仿真研究表明,RBF神经网络辨识效果优于BP神经网络。

A New Method to Identify the Characteristics of Induction Motor Based on Neural Networks

A new characteristic identification method of induction motor is set up and verified. Through the measurement of two motorphase current values, the motor speed and torque could be identified. The neural network is trained with improved Levenberg-Marquardtalgorithm and BP neural network is built for observation of motor speed. As RBF neural network features better performance than BP networkin the approaching capability, function fitting and learning speed, it is also used for the identification. Compared with existing methods, thismethod needs fewer measurements, with easy identification. The simulation study show that RBF neural network has better identificationresults than BP neural network.