Research on Rotor Position Model for Switched Reluctance Motor Using Neural Network

Abstract

Due to the double-salient structure of the switched reluctance motor (SRM) and highly magnetic saturation, its rotor position shows highly nonlinear function with phase current and flux linkage. Therefore, the establishment of accurate and small-scale SRM rotor position model is crucial to realize SRM position sensorless control. This paper is based on SRM flux characteristics collected by actual measurement, then adopts back-propagation neural network (BPNN) with improved algorithm and radial basis function neural network (RBFNN), respectively, to build SRM rotor position model. It is concluded that the model built by BPNN based on the Levenberg–Marquardt algorithm has smaller network structure and higher accuracy than RBFNN. In order to optimize BPNN model further, two kinds of pretreatment methods are used to study SRM rotor position model, respectively, and the modeling results of these two methods are compared. Adopting pretreatment method could reduce scale and improve accuracy to achieve the model optimization. After initial rotor position was estimated accurately without rotor position sensor, experiments are performed based on a three-phase 12/8 SRM. The experimental results verify that the BPNN model of rotor position for SRM based on a pretreatment method could be applied to SRM position sensorless control.