神经网络的柔性改进及其在开关磁阻电机控制中的应用

提出三自由度柔性双极性神经网络的结构、原理以及算法,由于网络多自由度学习的特点,使得其学习能力较传统神经网络大为增强。为了全面研究网络柔性对学习能力及复杂性的影响,将三自由度网络与单自由度网络及两自由度网络进行了比较。在学习逼近开关磁阻电机非线性磁化曲线过程中,基于三自由度的神经网络表现出优良的性能,和传统神经网络及两自由度网络比,其更加柔性的特点可以使网络具有更少的神经元、更快的学习速率。基于所提三自由度神经网络,建立了开关磁阻电机转矩逆模型和磁链模型,构建了电机控制系统,有效补偿了电机的非线性特性。仿真结果表明,基于柔性神经网络的开关磁阻电机控制系统有效降低了转矩脉动,系统运行平滑。

Neural Networks Improved Flexibility and Its Application to Control Switched Reluctance Motor

A flexible bi-pole neural network structure with three degrees of freedom is proposed, and its principle and algorithms are given. Multi-degrees of freedom characteristic of the proposed network enhance its learning ability when comparing with the traditional neural network. The proposed network with three degrees of freedom is compared with the existing flexible network of two degrees of freedom and the conventional network for the purpose of studying the influences of flexibility of the network on the learning ability and complexity of the net- work. As a result, the proposed network shows us the excellent performances in their application to approach the magnetization curves of switched reluctance motor. Its high-flexibility makes the network have fewer neurons, faster learning rate. The inverse model of torque and the flux linkage model of switched reluctance motor arerespectively established by using the proposed flexible network. The neural network-based control system ef- fectively compensates the nonlinearity of the motor, and the simulation results verify that the torque control system of switched reluctance motor operates smoothly and the torque ripple is very small.