Position Sensorless Control System of SRM Based on Improved BP Neural Network

Abstract

Due to the shortcomings of long training time and slow convergence of BP neural network, this paper presents a new improved method that weight is no longer a constant but turned into a function of adjustable parameters. After the training of the improved BP neural network is completed, the network can map the nonlinear relationship between motor current, flux and rotor position. Based on the analysis of the unique structural properties of switched reluctance motor, this paper also proposes a method of greatly reducing the sample data to save computing time. Simulation results show that this method simplifies the complexity of the control system and improve detection accuracy, thus realize position sensorless detection of the switched reluctance motor.