重构神经网络模型及开关磁阻电机恒转矩控制

Reconstructed Neural Network model and Constant Torque Control for Switched Reluctance Motor

针对开关磁阻电机(SRM)难以准确建模及计算恒转矩下的控制电流而导致的转矩脉动过大的问题,构建一种新的转矩-电流神经网络模型用于得到恒转矩下的控制电流。在新神经网络中,针对SRM转矩-电流特有的强非线性特性,设计能够描述SRM电流基本变化规律的新型激励函数,使神经网络结构更接近SRM的本质特性,有利于加快建模速度,提高建模精度。所重构神经网络模型通过在线学习计算恒转矩下对应的控制电流对SRM进行控制,实现转矩脉动的有效抑制。仿真结果表明,与通用神经网络相比,提出的重构神经网络模型能更好地描述SRM的强非线性特性,得到恒转矩下对应的控制电流,有效地抑制转矩脉动。

A novel neural network model is proposed to calculate the command current under constant torque because it′s difficult to obtain the accurate torque-current model and command current under constant torque,which results in large torque ripple.The hidden active function which is designed according to the relationship between SRM′s torque and current is able to depict the basic change law of SRM′s current,which makes the construction of the neural network more like the inherent characters of SRM.The reconstructed neural network model with self-learning ability is used to obtain the torque-current model and command current,by which the torque ripple is suppressed effectively.Simulation results show the proposed reconstructed neural network model can depict the strong nonlinearity of SRM,calculate the command current and inhibit the torque ripple better than common neural network under constant torque.