基于DSP的开关磁阻电机磁链特性检测与神经网络建模  

Flux Linkage Characteristic Measurement Based on DSP and Artificial Neural Network Modeling for Switched Reluctance Motor

作　　者：薛梅[1] 夏长亮[1] 王慧敏[2] 谢细明[1] 

机构地区：[1]天津大学电气与自动化工程学院,天津300072 [2]天津工业大学电气工程与自动化学院,天津300160

出　　处：《电工技术学报》2011年第2期68-73,共6页Transactions of China Electrotechnical Society

基　　金：国家杰出青年科学基金(50825701);国家自然科学基金重点项目(51037004);国家自然科学基金(50777044);天津市应用基础及前沿技术研究计划重点项目(08JCZDJC17600)资助项目

摘　　要：基于实测磁化曲线提出一种开关磁阻电机小波神经网络磁链模型。设计了基于DSP的开关磁阻电机磁链特性检测系统,在上位机中采用离散方法计算磁链值,获得不同角位置下的电机磁化曲线族;利用实测磁链特性建立神经网络模型,选择Mexican hat小波函数作为隐层激励函数,采用下降梯度法对网络进行训练以确定模型参数。仿真结果表明,所提出的开关磁阻电机小波神经网络磁链模型具有较小的误差和较强的泛化能力,对实现开关磁阻电机在线优化控制具有实际意义。A wavelet neural network（WNN） model of flux linkage characteristic based on measured magnetization curves is presented for switched reluctance motors（SRM）.A flux linkage measuring installation is setup by using a digital signal processor（DSP）.The magnetization curves at different rotor positions are obtained by calculating flux linkage using a discrete method on an upper level computer.Then,a feed-forward neural network（FFNN） model of flux linkage characteristic is proposed.The commonly used Mexican hat wavelet is chosen as the activation function for hidden-layer neurons,and its parameters are determined by training the network with the descend gradient method.Simulation indicates that the proposed model has reduced error and improved generalization property.It shows some prospect on optimizing online control of SRM.

关 键 词：开关磁阻电机 磁链特性 小波神经网络 DSP TMS320F2812 

分 类 号：TM351[电气工程—电机]