The Design of BP Neural Network Modeling for Switched Reluctance Motor

Abstract

The model parameters of 8/6 poles switched reluctance motor (SRM) were determined through using the measured magnetization curve and the establish BP neural network model, selecting Sigmoid function as the hidden layer activation function and using gradient descent method to train the network. The simulated results show that the motor flux linkage model established has a good convergence rate, higher accuracy and generalization ability. It is significant to improve the reliable running and high precision speed control of SRM motor. Introduction The structural characteristics of SRM make it have the advantages of high reliability, low cost and high efficiency, and the speed regulating system composed of SRM has the advantages of both AC and DC speed regulating system, which has been widely used in the field of electrical transmission in recent years[1-2].So SRM has a promising future in this field. However, due to its double salient pole structure, the SRM drive system itself is a serious nonlinear system. The torque ripple, motor vibration, noise and other problems are particularly obvious. These defects, especially torque ripple, greatly limit its wide application in servo control and other fields. The traditional motor control method is not suitable for SRM drive system, so the suppression of motor torque ripple has become a research focus of SRM at present. Although it is difficult to study SRM torque pulsation due to the complexity of magnetic circuit structure, however, but some progress has been made by scholars over these years. To solve the torque ripple of SRM motor, the most important thing is to establish correct and reliable SRM motor model. There is a complex functional relationship between the stator flux and winding current and rotor position of SRM, and the relationship between them is nonlinear, and establishing accurate and practical magnetic chain model is a hot research today. Traditional table method[3], has high accuracy, but needs a lot of calculations, and is unable to meet the real-time control and motor rapid modeling requirements. Although the function analytical method[4-5] can improve the system performance, but it’s adaptability is poor in the change of load and parameters. With the continuous development of artificial intelligence technology, intelligent control theory in SRM modeling is used more widely. Literature [6-7], respectively, use sliding mode variable structure fuzzy neural network and RBF neural network to establish a magnetic chain model of the motor. Error Back Propagation Neural Network (BPNN) is connection type of feedback neural network, based on the kind of multilayer feedback networks, and compared with other neural network, it is the most widely used and versatile neural network model, which has better characteristics [8-10] in the aspect of classification, pattern recognition, function approximation, global convergence and generalization ability, and shows good development potential in the field of SRM modeling. BP Neural Network Modeling Based on the sample shown as magnetization curve by measuring[13], we can come up with a model which aims at SRM s BP neural network magnetic linkage model. 2nd International Conference on Electrical and Electronic Engineering (EEE 2019) Copyright © 2019, the Authors. Published by Atlantis Press. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/). Advances in Engineering Research, volume 185