采用少量样本的开关磁阻电机磁链SVM建模与分析  

SVM-based modeling of SRM with few flux-linkage samples

作　　者：谢勇军 赵世伟[1] 程小华[1] XIE Yong-jun;ZHAO Shi-wei;CHENG Xiao-hua(Electic Power College, South China University of Technology, Guangzhou 510640, China)

机构地区：[1]华南理工大学电力学院,广东广州510640

出　　处：《电工电能新技术》2020年第8期10-17,共8页Advanced Technology of Electrical Engineering and Energy

摘　　要：开关磁阻电机(SRM)中磁链ψ是转子位置θ和绕组电流i的二元非线性函数,常基于ψ-θ-i关系数据建立磁链的数值模型,但ψ-θ-i关系数据测量成本高。本文基于支持向量机(SVM)算法从少量ψ-θ-i关系数据中建立开关磁阻电机的磁链模型。通过改变训练集均匀采样的步长,发现若电角度采样步长超过14°,模型预测误差约增加50%;电流较小时,模型预测误差随着采样步长的增大先迅速增加后在某一值附近震荡。训练集确定后,采用原始的ψ-θ-i数据训练SVM所得磁链模型,在磁链值较小的区域,预测值的相对误差最大值超过200%。针对误差产生的原因,本文提出一种非线性映射,用于处理训练样本。采用预处理后的样本训练SVM所得模型,在磁链值较小区域预测值的相对误差小于3.5%。为进一步提高最终模型的精度,本文将由原始数据训练所得模型和由映射后数据训练所得模型集成得到最终模型。本文的方案削弱了训练样本的减少对磁链模型精度的影响,使磁链模型预测的相对误差明显减小。To SRM,it is time-consuming to get the data ofψ-θ-i,becauese the flux-linkage characteristics a binary nonlinear function of position of rotator and currenct of winding.This paper proposes a SVM-based model of SRM with few flux-linkage samples.In the region of smallψ,the relative error of prediction is large if the model is directly trained by the dataset ofψ-θ-i.The max of relative error is up to 200%.After analysing the reason of large relative error,training dataset is non-linearly mapped.By using the mapped data to training the model,the relative error is under 3.5%.Then,integrating the two kind of models to get the final model.Finally,the influence caused by reducing data and the relative error go down.

关 键 词：开关磁阻电机 磁链模型 支持向量机 少量磁链数据 非线性映射 

分 类 号：TM352[电气工程—电机]