基于神经网络的电机转矩软测量研究

Measurement of electromagnetic torque by soft-sensing technology based on neural networks

径向基函数神经网络(RBFNN)具有最优逼近和全局逼近的特性,在函数拟合方面优于传统的BP网络,将在化工领域广泛使用的软测量技术应用于电机系统的转矩测量,该方法的可行性进行了论证,并运用RBF神经网络建立转矩的软测量模型。同时建立了基于BP神经网络的软测量模型,用改进的Levenberg-Marquardt算法对BP神经网络进行学习和训练,并对两种网络进行了对比。该方法只需要电流信息,辨识方法简单。研究表明,RBF神经网络辨识效果优于BP神经网络。

Since radical basis function neural networks （RBFNN） have the performances of best approximation and universal approximation, RBFNN has the better character in function simulation compared with BP neural networks. This paper extended soft-sensing technology that was used broadly in chemical industry to measure the torque of asynchronous machine, A soft-sensing model based on RBFNN was set up and the method was proved correctly. In the same time the BP neural networks was compared with the RBFNN. Simulation study shows RBFNN has higher dynamic characters.