Torque Control of Switched Reluctance Motors Based on Flexible Neural Network

Abstract

Application of conventional neural network (NN) in modeling and control of switched reluctance motor (SRM) has been limited due to its structure of low degree of freedom, which results in a huge network with large numbers of neurons. In this paper, a flexible neural network (FNN), which uses flexible sigmoid function, is proposed to improve the learning ability of network, and the learning algorithm is derived. It greatly simplifies the network with fewer neurons and reduces iterative learning epochs. FNN based desired-current-waveform control for SRM, where FNN provides the inverse torque model, is presented. Simulation results verify the proposed method, and show that FNN gives better performances than conventional NN and the torque output of the control system has a very small ripple.